TheFinAI/gr_general_domain_laws_and_regulations_combined

This dataset is a combination of various smaller datasets containing regulations and laws in Greek. The "source" column helps distinguish the source of each row. In the case of a "source" being something like 20160100001.pdf, this means that it's a law from 2016 taken from the National Printing House of Greece. All non-finance and related laws from 2016 onwards were taken from the National Printing House. In case the "source" is "multi_eurlex" it means the text of the law is taken from the Multi_Eurlex dataset, which are EU laws. Finally, when "source" is "raptarchis" then it means that the law was taken by the greek legal code archive Raptarchis, which includes Greek state laws from 1834 to 2015. "tokens" column contains an integer which is the token count for that specific row, using the GPT-4 tokenizer.

Cleaning process: The Raptarchis and Multi_Eurlex datasets are not ours, so they have already been preprocessed and we use their version. We do some cleaning on the Greek laws and regulations from the National Printing House. Namely, the steps were:

1. Some lines contained numerical values surrounded by asterisks. This come from the text extractor extracting QR codes. These lines were removed.
2. Removed lines which include a single page number.
3. Removed footer lines and other artifacts from the extraction, which are lines that exist in every page and come from things like logos etc.
4. Removed lines which are entirely in English and are artifacts which are extracted along the text and are mainly from things such as digital signage included in the PDF.
5. Some words, when they reach the end of the line, are hyphenated and continued on the next line due to spacing issues. These were found programmatically and concatenated in order to produce valid Greek words.

The total number of tokens, after cleaning, counted using the Llama-3.1-8B tokenizer were: 0.25B and using the GPT-4 tokenizer: 0.5B