README.md · Daoze/ReviewRebuttal at main

metadata

license: apache-2.0
task_categories:
  - text-generation
language:
  - en
size_categories:
  - 100K<n<1M

Introduction

This dataset is the largest real-world consistency-ensured dataset for peer review, which features the widest range of conferences and the most complete review stages, including initial submissions, reviews, ratings and confidence, aspect ratings, rebuttals, discussions, score changes, meta-reviews, and final decisions.

Comparison with Existing Datasets

The comparison between our proposed dataset and existing peer review datasets is given below. Only the datasets in [bold] can guarantee all provided papers are initial submissions, which is critical for the consistency and quality of the review data. For the “Task” column, AP is the abbreviation for Acceptance Prediction, SP for Score Prediction, RA refers to Review Analysis tasks, RG stands for Review Generation, and MG denotes Meta-review Generation. In the three columns representing numbers, “-” means please refer to the original dataset for the number.

Dataset Name	Data Source	Task	Reviews	Aspect Scores	Rebuttal	Score Changes	Meta-reviews	Final Decision	#Paper	#Review Comments	#Rebuttal
PeerRead	ICLR 17, ACL 17, NeurIPS 13-17	AP, SP	✓	✓	✓	✓	✓	✓	14,700	10,700	0
DISAPERE	ICLR 19-20	RA	✓	✓	✓				0	506	506
PEERAssist	ICLR 17-20	AP	✓	✓					4,467	13,401	0
[MReD]	ICLR 18-21	MG	✓	✓		✓	✓	✓	7,894	30,764	0
ASAP-Review	ICLR 17-20, NeurIPS 16-19	RG	✓	✓	✓	✓			8,877	28,119	0
[NLPeer]	ACL 17, ARR 22, COLING 20, CONLL 16	RA	✓	✓	✓	✓			5,672	11,515	0
PRRCA	ICLR 17-21	MG	✓	✓	✓	✓	✓	✓	7,627	25,316	-
Zhang et al.	ICLR 17-22	RA	✓	✓	✓	✓	✓	✓	10,289	36,453	68,721
[ARIES]	OpenReview	RG	✓	✓	✓	✓	✓	✓	1,720	4,088	0
AgentReview	ICLR 20-23	RG, MG	✓	✓	✓	✓	✓	✓	500	10,460	-
Reviewer2	ICLR 17-23, NeurIPS 16-22 (PeerRead & NLPeer)	RG	✓	✓	✓	✓	✓	✓	27,805	99,727	0
RR-MCQ	ICLR 17 (from PeerRead)	RG	✓	✓					14	55	0
ReviewMT	ICLR 17-24	RG	✓	✓	✓	✓	✓	✓	26,841	92,017	0
Review-5K	ICLR 24	RG	✓	✓	✓	✓	✓	✓	4,991	16,000	0
DeepReview-13K	ICLR 24-25	RG	✓	✓	✓	✓	✓	✓	13,378	13,378	0
[Our dataset]	45 venues from 2017 to 2025	RG, MG, AP, SP	✓	✓	✓	✓	✓	✓	19,926	70,668	53,818

papers

The papers folder, located in ./papers, contains the latest versions of the papers before review, collected from 45 venues between 2017 and 2025, in both .tex and .md formats. For the .tex files, the references and appendices have been filtered out.

REVIEWS

We organize the reviews corresponding to 19,926 papers into a dictionary keyed by each peer paper, and then merge each paper-review dictionary into a list. This list is split into two files: REVIEWS_train and REVIEWS_test. Specifically, for the Re²-Review dataset, we sample 1,000 papers along with their reviews to form the test set, while the remaining papers and their reviews are used as the training set.

Review Data Format

The format of the review data is below.

paper_id refers to the unique identifier of the paper on OpenReview
initial_score is the score before the rebuttal
final_score is the final score after the rebuttal
Fields ending with _unified represent the scores after unification
review_initial_ratings_unified and review_final_ratings_unified are the lists of all review scores before and after the rebuttal, respectively

[
    {
        "paper_id": "m97Cdr9IOZJ",
        "conference_year_track": "NeurIPS 2022 Conference",
        "reviews": [
            {
                "reviewer_id": "Reviewer_UuMf",
                "review_title": "null",
                "review_content": "<review_content>",
                "initial_score": {
                    "rating": "7: Accept: Technically solid paper, with high impact...",
                    "confidence": "3: You are fairly confident...",
                    "aspect_score": "soundness: 3 good\npresentation: 4 excellent\ncontribution: 3 good\n"
                },
                "final_score": {
                    "rating": "7: Accept: Technically solid paper, with high impact...",
                    "confidence": "3: You are fairly confident...",
                    "aspect_score": "soundness: 3 good\npresentation: 4 excellent\ncontribution: 3 good\n"
                },
                "initial_score_unified": {
                    "rating": "7 accept, but needs minor improvements",
                    "confidence": "The reviewer is fairly confident that the evaluation is correct"
                },
                "final_score_unified": {
                    "rating": "7 accept, but needs minor improvements",
                    "confidence": "The reviewer is fairly confident that the evaluation is correct"
                }
            },
            {
                "reviewer_id": "Reviewer_15hV",
                "review_title": "null",
                "review_content": "<review_content>",
                "initial_score": {
                    "rating": "5: Borderline accept: Technically solid paper where...",
                    "confidence": "1: Your assessment is an educated guess...",
                    "aspect_score": "soundness: 3 good\npresentation: 3 good\ncontribution: 3 good\n"
                },
                "final_score": {
                    "rating": "5: Borderline accept: Technically solid paper where...",
                    "confidence": "1: Your assessment is an educated guess...",
                    "aspect_score": "soundness: 3 good\npresentation: 3 good\ncontribution: 3 good\n"
                },
                "initial_score_unified": {
                    "rating": "5 marginally below the acceptance threshold",
                    "confidence": "The reviewer's evaluation is an educated guess"
                },
                "final_score_unified": {
                    "rating": "5 marginally below the acceptance threshold",
                    "confidence": "The reviewer's evaluation is an educated guess"
                }
            },
            {
                "reviewer_id": "Reviewer_hkZK",
                "review_title": "null",
                "review_content": "<review_content>",
                "initial_score": {
                    "rating": "7: Accept: Technically solid paper, with high impact..",
                    "confidence": "3: You are fairly confident in your assessment...",
                    "aspect_score": "soundness: 3 good\npresentation: 3 good\ncontribution: 3 good\n"
                },
                "final_score": {
                    "rating": "7: Accept: Technically solid paper, with high impact...",
                    "confidence": "3: You are fairly confident in your assessment...",
                    "aspect_score": "soundness: 3 good\npresentation: 3 good\ncontribution: 3 good\n"
                },
                "initial_score_unified": {
                    "rating": "7 accept, but needs minor improvements",
                    "confidence": "The reviewer is fairly confident that the evaluation is correct"
                },
                "final_score_unified": {
                    "rating": "7 accept, but needs minor improvements",
                    "confidence": "The reviewer is fairly confident that the evaluation is correct"
                }
            }
        ],
        "review_initial_ratings_unified": [
            7,
            5,
            7
        ],
        "review_final_ratings_unified": [
            7,
            5,
            7
        ],
        "metareview": "The paper proves universal approximation in...",
        "decision": "Accept"
    }
]

REBUTTALS

As for the rebuttal, we model the rebuttal and discussion data as a multi-turn conversation between reviewers and authors. This conversational setup enables the training and evaluation of dynamic, interactive LLM-based reviewing assistants, which can offer authors more practical and actionable guidance to improve their work before submission. To facilitate this, we split the rebuttal data into two files: REBUTTAL_train and REBUTTAL_test. Specifically, for the Re²-Rebuttal dataset, we select 500 papers along with their rebuttals as the test set, while the remaining data is used for training.

Rebuttal Data Format

The format of the review data is below.

paper_id refers to the unique identifier of the paper on OpenReview
messages is formatted as a multi-turn conversation
final_score is the final score after the rebuttal
When the role is set to system, it defines the overall context for the entire rebuttal multi-turn dialogue
The first message with the role of user serves to trigger the review process, providing the model with the paper information and confirming its identity as a reviewer
Subsequent messages with the role of user serve as the author's responses
Messages with the role of assistant serve as the reviewer's comments or replies during the discussion

[
    {
        "paper_id": "N-eul1pdagX",
        "conference_year_track": "ICLR 2023 Conference",
        "reviewer_id": "Reviewer_s4Ez",
        "messages": [
            {
                "role": "system",
                "content": "\nYou are an experienced academic paper reviewer. You will receive a conference paper in the field of computer science or AI. Please analyze the paper content carefully and output your review content.\nMore importantly, the author will further explain or refute the review comments you gave, and you need to further discuss with the author based on the author's response. \nThe purpose of this rebuttal and discussion is:\n- Clarification: During the review process, the reviewer (you) may not accurately understand the contribution or method of the paper due to unclear expression or misunderstanding of certain details. Through rebuttal, the author has the opportunity to clarify these misunderstandings and further illustrate the effectiveness of their innovations or methods.\n- Respond to criticisms and suggestions: The reviewer (you) may make criticisms or suggestions on the paper, and the author can use the discussion session to provide specific responses, explain why certain criticisms may not apply, or how to improve and optimize certain aspects of the paper.\nFinally, you and the author will work together to enhance the persuasiveness of the paper and improve the quality of the paper.\n"
            },
            {
                "role": "user",
                "content": "Now your reviewer ID is assigned as Reviewer_s4Ez, and the following are the paper you need to review:\n```<<N-eul1pdagX>>```"
            },
            {
                "role": "assistant",
                "content": "<review_content>"
            },
            {
                "role": "user",
                "content": "Title: Response to Reviewer s4Ez\nThank you for the extensive review. We address the questions and concerns piecewise:\n\n**Q1**: Fig. 1 - can you add some details on the sampling?\n\n**Response**: As suggested, we have now included some clarifying text..."
            },
            {
                "role": "assistant",
                "content": "Title: Post Review Discussion\nThank you for clarifying.\n\nI did not notice any changes to the text after equation 5, maybe accidentally not added?..."
            },
            {
                "role": "user",
                "content": "Title: Response to Follow-up and Post Reviewer-Discussion Update\nThank you for following up.\n\n> I did not notice any changes to the text after equation 5, maybe accidentally not added?\n\nApologies, indeed, we created this sentence, but it was not added in the upload above due to a version control error on our part..."
            },
            {
                "role": "assistant",
                "content": "It was a pleasure to engage in the above discussion with the author, which has added considerable value to this work."
            }
        ],
    }
]