diff --git "a/-tE4T4oBgHgl3EQfDwuq/content/tmp_files/load_file.txt" "b/-tE4T4oBgHgl3EQfDwuq/content/tmp_files/load_file.txt" new file mode 100644--- /dev/null +++ "b/-tE4T4oBgHgl3EQfDwuq/content/tmp_files/load_file.txt" @@ -0,0 +1,766 @@ +filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf,len=765 +page_content='Sharpening Ponzi Schemes Detection on Ethereum with Machine Learning Letterio Galletta Fabio Pinelli IMT School for Advanced Studies Lucca, Italy Abstract—Blockchain technology has been successfully ex- ploited for deploying new economic applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' However, it has started arousing the interest of malicious users who deliver scams to deceive honest users and to gain economic advantages.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Among the various scams, Ponzi schemes are one of the most common.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Here, we present an automatic technique for detecting smart Ponzi contracts on Ethereum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We release a reusable data set with 4422 unique real-world smart contracts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Then, we introduce a new set of features that allow us to improve the classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Finally, we identify a small and effective set of features that ensures a good classification quality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Index Terms—blockchain fraud, Ponzi scheme, smart contracts I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' INTRODUCTION Blockchain is revolutionizing how individuals and compa- nies exchange digital assets without the control of a central authority.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' This technology has been successfully exploited for deploying new economic applications, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=', cryptpcurren- cies [1] and Decentralized Finance [2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' However, soon after this technology became widespread and its economic value increased, it has started arousing the interest of malicious users who are eager to take some advantages due to the pseudonymity of these platforms and the lack of regulation [3]: on the one hand, they exploit cryptocurrencies to transfer currency without being tracked by authorities;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' on the other hand, they deliver scams to deceive honest users willing to make revenues through cryptocurrencies.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Nowadays, many types of scams can be found on blockchain platforms, such as exploits, hacks, and phishing [4].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Estimates say that more than 7 million USD was gathered by scams in Bitcoin [5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Among the various scams, Ponzi schemes have approached the blockchain world, first on Bitcoin [5] and more recently on Ethereum [6].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Ponzi schemes are fraudulent investment operations where older investors obtain returns from money paid by new investors rather than from legitimate business activities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Although the actual conditions to gain money de- pend on the specific rules of the scheme, a common feature is that participants who want to redeem their investments have to make new participants join the scheme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Participants who join later are the most likely to lose their money.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Thus, the development of automatic techniques to counter these scams is required to protect average users and to allow them to participate safely in the blockchain economy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In this paper, we focus on Ethereum and smart contracts to deliver Ponzi schemes, called smart Ponzi contracts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Our goal is to implement an automatic technique for classifying smart contracts, which can be used as the backbone for developing new detection tools.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' More precisely, we provide a threefold contribution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' First, we address the problem of the unavailability of public data sets to train effective automatic classifiers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Thus, we release a reusable data set that collects 4422 unique real-world smart contracts, where 3749 (84.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='78%) are not-Ponzi, and 673 (15.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='22%) are Ponzi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Our data set contains both information about the transaction history of the contracts as well as their bytecode.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Then, we implement a binary classification model to detect smart Ponzi contracts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Our experiments show that the proposed model performs better than the models proposed in the litera- ture when considering the AUC as a metric and achieves high accuracy for practical use.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Finally, we address the issue of identifying a small and effective set of features that ensures a good quality of the classification process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' To find this set, we proceed as follows: first, we introduce new features w.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='r.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' the literature, and we show through experiments that they improve the classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Then, we consider the union of our features and those from the literature and identify which ones can be safely removed because they do not contribute to the classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We adopt eXplainable Artificial Intelligence (XAI) techniques to inves- tigate how the identified features contribute to classification.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In summary, the main contributions of this paper are: a reusable and publicly available data set that collects 4422 real-world smart contracts where 3749 are not Ponzi, and 673 are Ponzi;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' a binary classifier to detect smart Ponzi contracts that per- forms better than the classifiers proposed in the literature when considering the AUC as a metric;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' a small and effective set of features that ensures a good classification quality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Structure of the paper: We proceed as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In Sec- tion II, we introduce Ponzi schemes and smart Ponzi contracts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In Section III, we discuss the connection between our work and the literature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In Section IV, we describe our data set and the features we use.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In Section V, we build our binary classifier and perform an experimental evaluation to study its quality and the impact and importance of the features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Availability: The data set and the notebooks used for the experiments presented in this paper are available online.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='1 II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' BACKGROUND: SMART PONZI CONTRACTS Ponzi schemes are classic frauds concealed as “high-yield” investment programs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The initiator of the scheme generates 1https://github.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='com/fpinell/ponzi_ml arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='04872v1 [cs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='CR] 12 Jan 2023 returns for existing investors through revenue paid by new investors rather than from legitimate business activities or profits of financial trading.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' More in general, the U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Securities and Exchange Commission (SEC)2 defines Ponzi schemes as: A Ponzi scheme is an investment fraud that in- volves the payment of purported returns to exist- ing investors from funds contributed by new in- vestors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Ponzi scheme organizers often solicit new investors by promising to invest funds in opportu- nities claimed to generate high returns with little or no risk.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' With little or no legitimate earnings, Ponzi schemes require a constant flow of money from new investors to continue.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Ponzi schemes inevitably collapse, most often when it becomes difficult to recruit new investors or when a large number of investors ask for their funds to be returned.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Although the actual conditions to gain money depend on the specific rules of the scheme, a common feature is that a user who wants to redeem her investment has to make new users join the scheme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In this way, the schemes create a pyramid of investors, where the initiator is at the top, and the investors at level l + 1 compensate for the investment of those at level l.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Once a scheme collapses because no more investors join it, those at the top levels of the pyramid gain money, while those at the bottom lose it.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The spread of cryptocurrency and smart contracts has cre- ated new opportunities to deploy this kind of fraud.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Indeed, it is possible to find samples of smart contracts implementing Ponzi schemes, called smart Ponzi contracts, deployed on the main blockchain platforms like Ethereum and Bitcoin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' This paper focuses on the Ethereum platform and smart contracts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' According to the literature [7] smart Ponzi contracts have several attractive features to be used for scams: 1) The initiator of a smart Ponzi could stay anonymous, since deploying the contract on the blockchain and withdrawing money from it only requires an Ethereum account that does not reveal her real identity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 2) Once deployed on the blockchain, smart contracts are “unmodifiable” and “unstoppable”.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Thus, no central authority could terminate the execution of the scheme, seize the money and refund the victims.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 3) Since the code of smart contracts is public, immutable, and its execution is automatically enforced by the blockchain platform, investors may believe that no one can take advantage of their money and that they could eventually gain the declared interests.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The most significant feature of a smart Ponzi is the policy used to redistribute new investments among participants, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=', how the money flows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' This requires a smart Ponzi to maintain a data structure storing participants’ information and to implement a strategy for redistributing dividends.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Identifying redistribution behaviour is crucial to classify a contract as a smart Ponzi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Also, it is a challenging task because 2See https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='sec.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='gov/spotlight/enf-actions-ponzi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='shtml many other kinds of contracts, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=', gambling games, may have similar behaviour, which may induce many false positives.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Bartoletti et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [6] proposed the following four requirements to classify a smart contract as a Ponzi scheme: R1 the contract redistributes money to the investors according to a given logic;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' R2 the contract receives money only from the investors;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' R3 each investor makes a profit if a certain number of investors join subsequently the contract investing a certain amount of money;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' R4 the later an investor joins the contract, the higher the risk of losing her money.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' A smart contract is classified as a smart Ponzi when it satisfies all four requirements.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Note that requirement R1 rules out contracts that provide users with some assets but do not im- plement a logic to distribute them to participants, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=', tokens;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' requirement R2 ensures that a participant invests a certain amount in joining the contract;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' requirement R3 demands a constant flow of new investments for investors to make a profit;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' requirement R4 characterizes the fraudulent nature of smart Ponzi contracts because it reflects the fact that making a profit for investors is likely impossible after a certain point in time: too many victims must join the scheme for the contract to have enough money to reward all the participants.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Thus, the scheme collapses when this happens.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Note that the requirements above impose no condition on the money received or not by the initiator of the scheme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We will study the need for such a condition in our experimental evaluation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Typically, smart Ponzi contracts are categorized into four types according to their redistribution strategies: Tree-scheme, Chain-scheme, Waterfall-scheme and Handover-scheme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Since we do not consider these categories here, we refer the inter- ested reader to the relevant literature [6].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' III.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' RELATED WORK Since the inception of Bitcoin in 2009, cryptocurrencies and blockchain systems have attracted the attention of cybercrim- inals, who exploit them to carry out potentially untraceable scams.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Since the full transaction history is publicly available and provides accurate records of user behaviours, several pa- pers [4], [8], [9] have proposed machine learning techniques to detect possible frauds and scams.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Below, we discuss proposals similar to ours.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Bartoletti et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [7] study the problem of identifying Ponzi schemes in Bitcoin using data mining techniques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In particular, they released a public data set of Bitcoin addresses and an open-source tool to build such a data set;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' then, they apply different classification algorithms (Random Forest, Bayes Net- work, and Ripper) and systematically evaluate them to identify the best discriminating features for detecting Ponzi schemes in Bitcoin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Our work shares their goal of providing a public data set of smart Ponzi contracts and a good classifier and of identifying the most discriminating features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' However, we tar- get Ethereum smart contracts and consider different classifiers (Random Forest, XGBoost, Decision Trees).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Moreover, we use XAI techniques to study the contributions of the features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 2 In a subsequent paper, Bartoletti et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [6] consider smart Ponzi contracts in Ethereum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' First, they define four criteria based on behavioural aspects to identify a contract as a smart Ponzi and produce a data set with several contracts satisfying such criteria.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Then, they perform several analyses by hand on some contracts, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=', about the security of their code and the fairness of the distribution policy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In our work, we adopted their criteria, enriched their data set by considering new features, and added new contracts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Moreover, we trained and experimented with different classifiers to automatically determine if a smart contract is a Ponzi scheme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Chen et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [10] provide a reusable data set with real- world samples and evaluate different classifiers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' They use two different classes of features: the account features taken from the transaction history and code features extracted from the contract’s bytecode.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We follow the same approach for building and evaluating different classifiers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Moreover, we reused and extended their data set with new features and contracts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In contrast, we provide a more accurate analysis and an explanation of how the different features impact the classification process, and we show that our best classifier outperforms theirs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Chen et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [11] propose SADPonzi, a detection approach based on symbolic execution.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' This approach analyzes the bytecode of contracts to extract semantic information and to identify investor-related transfer behaviours and the distribu- tion strategies adopted by the scheme.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' SADPonzi performs the classification only by looking at the code, not the transactions’ history.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We include in our dataset some contracts identified by this tool.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The main difference with our work is that we resort to machine learning techniques to classify contracts, and we do not consider bytecode but only the information about the transaction history.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Wang et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [12] propose an approach based on Long-short Term Memory Network to detect smart Ponzi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In contrast, our classifier is not based on neural networks, and it is obtained after an experimental evaluation that tests different models and various parameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Moreover, we provide a precise account of how the various features impact the classification, and our dataset is publicly available for future investigation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Fan et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [13] propose PonziTect, a smart Ponzi contract detection method based on ordered boosting that classifies contracts considering only their bytecode.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' They adopt data augmentation as a method to solve the problem of imbalanced data by increasing the the proportion of smart Ponzi contracts at the boundary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Here, we do not consider the problem of imbalanced data because the algorithms we apply in our experiments are not affected too much by this issue.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Another difference with our paper is that we focus only on transaction data instead of bytecode.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Moreover, we release the data set used for training and testing our models.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Lou et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [14] use convolutional neural networks to detect smart Ponzi contracts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' They focus mainly on bytecode fea- tures, and the pipeline they propose is quite standard: first, they transform smart contracts into single-channel images and then adopt the spatial pyramid pooling method to ensure that the generated images have the same size.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In contrast, our classifier is not based on neural networks, considers only account features, and allows us to understand how the various features impact the classification process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Moreover, our data set is publicly available.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Ibba et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [15] builds a machine learning model that uses account features and the bytecode of the contracts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' They also take into account Solidity source code and apply text classi- fication techniques.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' They tested their approach with decision trees, support vector machines, and naive Bayes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In contrast, our classifier considers only account features, and we study how the various features impact the classification process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Moreover, we made our data set publicly available.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' IV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' DATA SET CONSTRUCTION This section describes the data and features we use to detect smart Ponzi on Ethereum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We build two data sets from the same sources but based on different features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Below, we proceed as follows: first, we describe our data and features;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' then, we present a qualitative analysis of some of the features to understand how they distribute across the two classes of contracts and how much they discriminate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We focus on the newly defined features, on the ones that result to be the most important for the classifier, and on those that perform poorly in helping the classifier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Feature description As described in Section III, previous works [6], [10], [11], made their data sets publicly available with the list of smart contracts, the extracted features, and the classification labels.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' On the one hand, we reuse these data sets that contain 4422 smart contracts in total, where 3749 (85.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='23%) are labelled as not-Ponzi and 673 (14.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='77%) as Ponzi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' On the other hand, we introduce new features that capture new specific characteristics and allow the improvement of our classifier (see Section V).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Also, we update the values of the previous features through the API provided by etherscan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='3 Below, we report the list of the features for each contract: 1) Address: the address of the smart contract (not used in the classification task);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 2) Balance: the difference between the amount of ETH in input and the ETH in output;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 3) Lifetime: the difference between the time of the first and the last transaction made or received;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 4) Tx_in: the number of input transactions;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 5) Tx_out: the number of output transactions;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 6) Investment_in: the number of input transactions that deposit an amount of ETH in the contract;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 7) Payment_out: the number of output transactions paying an amount of ETH;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 8) #addresses_paying_contract: the number of distinct ad- dresses that paid the contract;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 9) #addresses_paid_by_contract: the number of distinct addresses paid by the contract;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 3https://docs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='etherscan.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='io/ 3 10) Mean_v1: the average of the differences of the number of input/output transactions from/to the same address;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 11) Mean_v2: the average of the differences of the amount of ETH received and paid by the contract involving the same address;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 12) Sdev_v1: the standard deviation of the differences in the number of input and output transactions involving the same address;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 13) Sdev_v2: the standard deviation of the differences be- tween the amount of ETH in and out involving the same address;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 14) Paid_rate: the ratio between Tx_in and Tx_out;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 15) Paid_one: the ratio between the number of investors paid many times and the number of total investors;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 16) Know_rate: the proportion of receivers who have in- vested before payment;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 17) N_maxpayment: the max number of payments to all participants;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 18) Skew_v1: the skewness of the differences of the number of input and output transactions involving the same address;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 19) Skew_v2: the skewness of the differences of the amount of ETH in and out involving the same address;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 20) Investment_in/Tx_in: the ratio between Investment_in and Tx_in;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 21) Payment_out/Tx_out: the ratio between Payment_out and Tx_out;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 22) Percentage_some_tx_in: the percentage of active days with at least one input transaction during the contract lifetime;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 23) Sdev_tx_in: the standard deviation of the number of transactions per day;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 24) Percentage_some_tx_out: the percentage of active days with at least one output transaction during the contract lifetime;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 25) Sdev_tx_out: the standard deviation of the number of transactions in output per day;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 26) Initiator_get_eth_wo_investing: this feature is 1 if the contract initiator has earned ETH without any invest- ment, 0 otherwise;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 27) Initiator_get_eth_investing: this feature is 1 if the ini- tiator has earned ETH investing in the contract, 0 otherwise;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 28) Initiator_no_eth: this feature is 1 if the Initiator has obtained no ETH investing in the contract, 0 otherwise.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The first 19 features are inherited from previous works, while the last nine are new ones introduced by us.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' For example, Feature 15 estimates how often a contract interacts with accounts it already knows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' A high value of this feature means more interactions (we expect it to happen for smart Ponzi).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Features 20 and 21 aim to capture requirement R1 of Section II by measuring the percentage of transactions dis- tributing ether among Ethereum addresses.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Since we expect that not Ponzi contracts present a lower percentage than Ponzi Table I A SUMMARY OF THE FEATURES INCLUDED IN THE VARIOUS DATA SETS.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' THE NEW FEATURES ARE IN BOLD.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='D1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='D2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='D3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Address ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Balance ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Lifetime ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Tx_in ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Tx_out ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Investment_in ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Payment_out ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='#addresses_paying_contract ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='#addresses_paid_by_contract ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Mean_v1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Mean_v2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Sdev_v1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Sdev_v2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Paid_rate ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Paid_one ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Known_rate ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='N_maxpayment ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Skew_v1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Skew_v2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Investment_in/tx_in ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Payment_out/tx_out ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Percentage_some_tx_in ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Sdev_tx_in ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Percentage_some_tx_out ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Sdev_tx_out ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Initiator_gets_eth_Wo_investing ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Initiator_gets_eth_investing ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='Initiator_no_eth ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x13 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='\x17 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='ones,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' these features are uniquely based on Ether exchanges.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Features 22 and 24 monitor the number of input and output transactions per day over time.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' A small value of Feature 22 (respectively 24) indicates that the contract was active for a few days considering input transactions (output, respectively, for Feature 24).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' On the contrary, a high value means that the contract presented a more regular activity over its lifetime.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We also consider the standard deviation of the number of transactions per day to capture the variability during the contract life.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Feature 18 and Feature 20 consider input and output transactions, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' These two features measure if a contract sent or received transactions only in a few days or regularly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In particular, we expect that Ponzi contracts will have a short lifetime in which they will receive several user investments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Indeed, since making a profit for investors is likely impossible after a certain time, the scheme collapses, and the contract will no longer receive new investments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' These features try to capture this behaviour, also taking into account what is specified by requirements R2, R3, and R4 of Section II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The last three features, 26, 27, and 28, verify whether the initiator received money from the contract.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Indeed, in smart Ponzi contracts, the initiator usually receives a certain amount of money, even without an initial investment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We add these features to study whether receiving a certain amount of money identifies this kind of fraud.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Therefore, we expect most contracts labelled as Ponzi to have Feature 26 and Feature 27 equal to 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' On the contrary, Feature 28 is equal to 1 for not- Ponzi contracts since it is quite unusual to find the initiator of a smart Ponzi that receives no Ether.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 4 We build two data sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The first one, called D1, uses all the features listed above, whereas the second one, called D2, contains the features used by Chen et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We use these data sets to evaluate how the different features impact the classification quality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In particular, we use the classifier of Chen et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [10] as a baseline to measure the improvement.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Table I summarizes which features are present in the various data sets: symbol \x13 means that the feature is included, whereas \x17 that the feature does not appear.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Note that how we generate the data set D3 will be explained in Section V, and that features lifetime, tx_in, tx_out, #addresses_paying_contract, and #addresses_paid_by_contract are not present in Chen et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' but are present in other papers in the literature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Qualitative analysis of the features We report below a qualitative analysis of how the new features distribute across the two classes and which seems to be the most characteristic for each class.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We also consider Features 4 and 7 that will play a role during our experiments in Section V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Figure 1 shows the cumulative distributions for the new continuous features (Features 20-25 above): the orange line represents the Ponzi behaviour, while the blue line the not-Ponzi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' For visualization purposes, we discard the 1st and the 99th percentile in the plots;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' however, the shape of the cumulative distributions is preserved.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The two populations present a different behaviour for what concerns the input transactions (Tx_in): the plot of the cumu- lative distribution presents a very different shape.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Typically, smart Ponzi presents a small number of input transactions with few exceptions with many transactions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The same hap- pens for features Investment_in/TX_in, Payment_out/TX_out where smart Ponzi generally present larger values than the other class.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We expect that these features provide the clas- sifier with a relevant contribution to discriminate between the two classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' These differences are smoothed when we consider the features Payment_out, Percentage_some_tx_in, Percentage_some_tx_out, Sdev_tx_in, and Sdev_tx_out where the cumulative distributions are very similar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Therefore, we expect that these features provide a marginal contribution to discriminating between the two classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Figure 2 shows how the new binary features Initia- tor_gets_eth_Wo_investing, Initiator_gets_eth_investing, Ini- tiator_no_eth are distributed across the two classes in per- centage.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' From the plot, we can see that the number of Ponzi contracts having the feature Initiator_no_eth equal to 0 and 1 is similar.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Moreover, most not-Ponzi contracts have this feature’s value set to 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We expect this feature does not help the classifier discriminate between the two classes, but that can cause it to make mistakes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Also, the feature Initiator_gets_eth_investing may not help the classifier since most contracts of the two classes have the value of this feature set to 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The feature Initiator_gets_eth_Wo_investing could positively impact the classification because the number of not- Ponzi contracts with a value equal to 0 is low.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' EXPERIMENTS In this section, we build binary classifiers for detecting smart Ponzi contracts, and we perform an experimental evaluation to study how the new features of Section IV impact classification and the quality of the obtained classifiers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In particular, we answer the following three research questions: RQ1 Do the new features 15-22 improve the classifier’s quality?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' RQ2 Which are the most relevant features among them?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Can we find the best set of features?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' RQ3 Which are the characteristics of the best classifier?' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Below, we report the experiments we performed to answer each question and the relative results.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We carried out our experiments with Python and Scikit-lean library.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' RQ1: Impact of the new features We take the data sets D1 and D2 and study the performances of classifiers trained with them.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' (Recall that D2 uses the same features of Chen et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=') In particular, we consider Decision Tree [16], Random Forest [17], and Light Gradient Boosting Machine Classifier (LGBMC) [18] as classifiers and perform a grid search procedure with cross-validation to fine- tune the hyper-parameters of each classifier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We consider AUC as the metric to be optimized.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We split the data sets into an 80% training set (3537 samples) and a 20% test set (885 samples) stratified on the target variable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Thus, given a training set, a classifier, and a combination of hyper-parameters, the cross-validation splits the training data into 5 folds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Then, the model (i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=', classifier and relative hyperparameters) is trained and tested 5 times, varying the fold used as a validation set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' An average of the metric to be optimized over the 5 tests is performed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The optimal classifier is the one with the highest mean score.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In our case, the selected classifier is the one with the highest mean AUC.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Once we have selected the best values for the hyper-parameters for each data set and classifier, we compute the standard metrics Accuracy, AUC, F1, Precision, and Recall on the test set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Table II reports a summary of the performances of the considered classifiers on the two data sets.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' According to the AUC metric, the best model for both datasets is the LGBM classifier but with different values for the hyperparameters.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' For D1, the classifier has the following hyper-parameters: 80 estimators, max depth equal to 20, learning rate 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='1, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='8 the subsample of columns, regulation alpha equal to 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='2 (L1 regularization term), and regulation lambda set to 1 (L2 regularization term).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' For the dataset D2, 100 estimators, max depth equal to 15, learning rate 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='1, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='5 the subsample of columns, regulation alpha equal to 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='1, and regulation lambda set to 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Moreover, the table shows that the model trained with D1 presents a better AUC than the one trained with D2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Therefore, we can answer positively to RQ1: the new features do improve the classifier’s quality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Notice that for both datasets, we perform the grid search using the same set of values for the hyperparameters, applying, thus, the same procedure to select the best-performing classifier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Below, we analyze the results obtained by the LGBM classifier on D1and D2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Figure 3 shows the Receiver Operating 5 Tx_in Payment_out Investment_in/Tx-in Payment_out/Tx_out Percentage_some_tx_in Percentage_some_tx_out Sdev_tx_in Sdev_tx_out Figure 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The cumulative distributions for the two classes considering some continuous features: the distributions for the not Ponzi smart contracts are in orange, in blue for the Ponzi ones.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Figure 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The percentage of Ponzi and not Ponzi smart contracts for the new binary features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Characteristic (ROC) curve that describes the diagnostic ability of a classifier by comparing the true positive and the false positive rate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The first one is the proportion of observations that are correctly predicted to be positive out of all positive observations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The second rate is the proportion of observations that are incorrectly predicted to be positive out of all negative observations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Typically, a curve closer to the top-left corner indicates that the true positive rate is greater than the false one, meaning a good performance of the classifier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' This happens in Figure 3 where the curve of D1 presents higher values than the one of D2, as summarized in the fourth column of Table II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We also study the confusion matrices of the two models (see Figure 4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The confusion matrix summarizes the number of correctly classified samples, namely true positive (TP);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' the number of false positives (FP);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' the number of false negatives (FN);' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' and the number of true negatives (TN).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' From the matrices in the figure, we observe that the classifier trained with D1 presents a higher number of TP and TN, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' of smart contracts that are correctly classified, and a lower number of FP and FN, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' of smart contracts that are incorrectly classified.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Thus, D1 performs better than D2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Table II RESULTS OF THE GRID SEARCH COMPARING THREE CLASSIFIERS ON DATA SETS D1 AND D2(TOP).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' THE PERFORMANCES ON THE BEST CLASSIFIER ON D3 (BOTTOM).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Metric Accuracy AUC F1 Precision Recall Data set Classifier D1 Decision Tree 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='855 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='733 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='496 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='529 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='467 LGBM 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='904 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='883 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='608 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='805 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='489 Random Forest 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='896 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='875 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='562 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='787 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='437 D2 Decision Tree 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='861 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='674 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='481 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='559 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='422 LGBM 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='876 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='788 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='444 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='698 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='326 Random Forest 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='873 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='770 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='462 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='658 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='356 D3 LGBM 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='908 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='884 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='620 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='846 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='489 Finally, we consolidate our results through the McNemar test [19], a well-known approach to analyzing the statistical significance of the differences in classifier performances.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Our null hypothesis is that none of the two classifiers performs better than the other, whereas the alternative hypothesis is that the classifiers’ performances are unequal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We set the significance threshold to 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='05 and compute the p-value, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=', the probability of the observations where the null hypothesis is true.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The test results in a p-value equal to 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0007, thus lower than our significance threshold of 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='05.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' This means that the null hypothesis is rejected and confirms that the classifier trained on D1 outperforms the one trained on D2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' This strengthens the answer to RQ1 since the new features improve the classifier’s quality, and the improvement is statistically significant and not due to some randomness in the data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' RQ2: Detect most relevant features We take our best model on D1 and investigate which features contribute most to classification and which mislead the classifiers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' First, we compare the new features to determine their importance.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Each classifier adopts different metrics to determine the importance of features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The LGBM classifier considers the number of times a feature is used to split the 6 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='4 Class Ponzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='2 Not Ponzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0 2000 4000 6000 8000 100001.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='8 Probability 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='4 Class Ponzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='2 NotPonzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0 2000 4000 00091.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='4 Class Ponzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='2 Not Ponzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='01.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 Class Ponzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='8 NotPonzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='01.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='6 Class Ponzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='2 Not Ponzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='5 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='5 20 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='51.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='8 Probability 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='4 Class Ponzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='2 Not Ponzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='5 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='5 201.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='8 robability 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='6 Class Ponzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='2 Not Ponzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0 500 1000 15001.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='8 robability 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='6 Class Ponzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='2 NotPonzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0 100 200 300Initiator-no-eth=1 Initiator-no-eth=0 PONZI Initiator-gets-eth-investing=1 NOT PONZI Initiator-gets-eth-investing=0 Initiator-gets-eth-Wo-investing=0 Initiator-gets-eth-Wo-investing=1 0 20 40 60 80 Percentage of Smart-contractsFigure 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The ROC curve on the test set for the three best classifiers, one for each dataset.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Predicted class N P N 734 16 P 69 66 D1 Predicted class N P N 731 19 P 91 44 D2 Predicted class N P N 738 12 P 69 66 D3 Figure 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Confusion matrices of the best classifier on our data sets, where we indicate with N and P the not Ponzi and Ponzi class respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' data across all ensemble trees: the more a feature is used, the more important it is.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Figure 5 shows the new features sorted by their importance, where the x-axis represents the number of splits.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We see that Sdev_tx_in is the most important, whereas Initiator_no_eth and Initiator_gets_eth_investing are less important.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Then, we consider the problem of determining if there exists a subset of the features of D1 that improves the quality of the classifier, and we proceed as follows.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We perform a grid search procedure with cross-validation with the optimiza- tion of the AUC and, in this experiment, the set of hyper- parameters also includes the number of features.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' To tune this last hyper-parameter, we start considering all the features in D1, we perform a grid search and 5-fold cross-validation procedure with the other hyper-parameters and obtain the best- performing classifier, then we adopt the Recursive Feature Elimination algorithm to remove the less important feature.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Given the new reduced dataset, the procedure is repeated until the number of features to be evaluated is equal to the number of features of D2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' As for the previous experiments, we split the dataset into an 80% training set (3537 samples) and a 20% test set (885 samples).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' As a result, we obtain the highest mean AUC with the data set D3 which includes 25 features (reported in Table I).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The LGBM classifier trained on D3 presents better performances and improves the best classifier on D1 as reported in Table II.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The other hyper-parameters are the following: 120 estimators, max depth equal to 15, learning rate 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='1, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='5 the subsample of columns, regulation alpha equal to 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='1, and regulation lambda set to 10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In D3 the following features are removed in this order Initiator_no_eth, Figure 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The importance of the new set of features included in dataset D1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Figure 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Distribution of the probability to be a Ponzi scheme estimated using the best performing classifier on D3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We report the distribution of both classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Initiator_gets_eth_investing, and Payment_out.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' This confirms what was anticipated in Section IV.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Indeed, Initiator_no_eth and Initiator_gets_eth_investing do not contribute to discrim- inating between the two classes because creators of legit smart contracts may receive or not an amount of money from contracts they created.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' This is in line with requirements R1-R4 of Section II that do not consider the money received by the initiator as a discriminating factor to be a Smart Ponzi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Finally, Payment_out does not contribute to classification because every smart contract transfers money during its lifecycle.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Thus, the number of out transactions on its own seems not to capture requirement R1 of Section II, so it is not a good estimator to detect Ponzi schemes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' RQ3: Characteristics of the best classifier As described above, the best classifier is LGBM considering 25 features among all the available.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Figure 4 (right) shows the corresponding confusion matrix.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' As we can see, 4 false positive contracts are now classified correctly.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Figure 6 dis- 7 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 True Positive Rate (Recall) 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='8 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='4 D1 AUC=0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='883 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='2 D2 AUC=0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='788 D3 AUC=0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='884 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 FalsePositiveRate (Eall-OutSdev_tx in Percentage_some tx_in Investment_in/Tx in Percentage_some tx out Sdev_tx ino Initiator_gets eth_Wo_investing Payment_out/Tx out Initiator_no eth Initiator_gets eth_investing 0 50 100 150 200 Feature importance0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='8 Ponzi Not Ponzi 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='7 - 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='5 - 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='4 - 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='3 - 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='2 a1 - 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='2 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='4 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='6 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='8 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='0 Estimated Probability being PonziFigure 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' The feature importance of the top 10 features in terms of SHAP values.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' plays histograms of the probability of being in one of the two classes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Note that if we consider a threshold equal to 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='5, the classifier may assign a lower probability to some smart Ponzi contracts.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' To understand how the classifier works on D3, we adopt the eXplanaible Artificial Intelligence (XAI) library SHAP4 to visually investigate how the most important features impact the classification process.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' In Figure 7, we show the beeswarn plot of the feature importance based on SHAP values for the test set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Notice that the plot might differ from the one obtained directly from the classifier library, since the importance of the features is computed with a different logic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Moreover, it is designed to display an information-dense summary of how the top features in the dataset impact the model output.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Each instance is represented by a single dot whose position on the x-axis is determined by the SHAP value of the feature it represents;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' the dots are “pile up” along each feature row to show the density.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Colour is used to display the original value of a feature: blue corresponds to a lower value, while red a higher value.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' From the plot, we can understand the positive or negative effect of each feature w.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='r.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' the prediction outcome.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We observe that the number of transactions in input (Tx_in) negatively impacts when it presents a high value.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' On the contrary, a higher value of the ratio Investment_in/tx_in indicates a positive effect on the classifier to label the instance as a smart Ponzi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Moreover, note that 3 of our newly defined features belong to the top 10 features of the figure.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' VI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' CONCLUSION In this paper, we presented an automatic technique for detecting smart Ponzi contracts on Ethereum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We released a reusable data set with 4422 unique real-world smart contracts that can be used for future research.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Then, we introduced a new set of features that allowed us to improve the classification and outperform previous efforts in the literature [10].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Finally, we identified a small and effective set of features that ensures a good classification quality.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 4https://shap.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='readthedocs.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='io/en/latest/index.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='html In future work, we plan to extend our model in different directions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' We intend to improve the procedure to optimize the best set of features, we would like to take into account also the bytecode of contracts present in our dataset but not used by our classifier and possibly apply deep learning techniques to minimize the feature engineering effort.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Analogously, our approach can be applied to detect other forms of scams on Ethereum, and phishing is one of the most promising.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' REFERENCES [1] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Nakamoto, “Bitcoin: A peer-to-peer electronic cash system,” https: //bitcoin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='org/bitcoin.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='pdf, 2008.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [2] E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Napoletano, “Decentralized finance is building a new financial system,” https://www.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='nasdaq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='com/articles/ decentralized-finance-is-building-a-new-financial-system-2021-04-02, 2021, (last access 2022).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [3] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Moore, “The promise and perils of digital currencies,” International Journal of Critical Infrastructure Protection, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 6, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 3, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 147–149, 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [4] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Bartoletti, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Lande, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Loddo, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Pompianu, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Serusi, “Cryp- tocurrency scams: Analysis and perspectives,” IEEE Access, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 9, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 148 353–148 373, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [5] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Vasek and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Moore, “There’s no free lunch, even using bitcoin: Tracking the popularity and profits of virtual currency scams,” in Financial Cryptography and Data Security.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Springer, 2015, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 44–61.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [6] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Bartoletti, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Carta, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Cimoli, and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Saia, “Dissecting ponzi schemes on ethereum: Identification, analysis, and impact,” Future Gener.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Syst.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=', vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 102, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 259–277, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [7] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Bartoletti, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Pes, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Serusi, “Data mining for detecting bitcoin ponzi schemes,” 2018 Crypto Valley Conference on Blockchain Technol- ogy (CVCBT), pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 75–84, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [8] X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Liu, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='-J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Jiang, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='-H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Liu, and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Tse, “Knowledge discovery in cryptocurrency transactions: A survey,” IEEE Access, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 9, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 37 229– 37 254, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [9] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Trozze, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Kamps, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Akartuna, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Hetzel, B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Kleinberg, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Davies, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Johnson, “Cryptocurrencies and future financial crime,” Crime Science, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 11, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 1–35, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [10] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Chen, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Zheng, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='-H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Ngai, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Zheng, and Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Zhou, “Exploiting blockchain data to detect smart ponzi schemes on ethereum,” IEEE Access, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 7, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 37 575–37 586, 2019.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [11] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Chen, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Li, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Sui, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' He, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Wang, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Wu, and X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Luo, “Sadponzi: Detecting and characterizing ponzi schemes in ethereum smart contracts,” Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' ACM Meas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Anal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Comput.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Syst.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=', vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 5, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 2, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [12] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Wang, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Cheng, Z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Zheng, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Yang, and X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Zhu, “Ponzi scheme detection via oversampling-based long short-term memory for smart contracts,” Knowledge-Based Systems, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 228, p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 107312, 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [13] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Fan, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Fu, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Xu, and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Zhu, “Expose your mask: Smart ponzi schemes detection on blockchain,” in International Joint Conference on Neural Networks (IJCNN), 2020, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 1–7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [14] Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Lou, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Zhang, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Chen, “Ponzi contracts detection based on improved convolutional neural network,” in 2020 IEEE International Conference on Services Computing (SCC), 2020, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 353–360.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [15] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Ibba, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Pierro, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Di Francesco, “Evaluating machine- learning techniques for detecting smart ponzi schemes,” in 2021 IEEE/ACM 4th International Workshop on Emerging Trends in Software Engineering for Blockchain (WETSEB), 2021, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 34–40.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [16] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Quinlan, “Induction of decision trees,” Machine learning, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 1, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 81–106, 1986.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [17] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Breiman, “Random forests,” Machine learning, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 45, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 1, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 5–32, 2001.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [18] G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Ke, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Meng, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Finley, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Wang, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Chen, W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Ma, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Ye, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content='- Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' Liu, “Lightgbm: A highly efficient gradient boosting decision tree,” Advances in neural information processing systems, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 30, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' [19] Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' McNemar, “Note on the sampling error of the difference between correlated proportions or percentages,” Psychometrika, vol.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 12, no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 2, pp.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 153–157, 1947.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'} +page_content=' 8 Hgh Tx-in nvestment-infTx-in Lifetime Known-rate Paid-one Sdev-tx-in Percentage-some-tx-in Meal-vl Sdev-v2 Paid-rate 2 1 0 2 SHAP value' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/-tE4T4oBgHgl3EQfDwuq/content/2301.04872v1.pdf'}