{
  "dataset_name": "LLMJRE",
  "full_name": "LLM-as-a-Judge Review and Rebuttal Evaluation Benchmark",
  "version": "1.0.0",
  "release_date": "2026-06-26",
  "license": "CC-BY-4.0",
  "language": "English",
  "source_dataset": {
    "name": "Re²",
    "description": "Consistency-ensured Dataset for Full-stage Peer Review and Multi-turn Rebuttal Discussions"
  },
  "repository": {
    "github": "https://github.com/tswj/llmjre-eval",
    "huggingface": "https://huggingface.co/datasets/tswj/llmjre"
  },
  "benchmark": {
    "master": {
      "file": "llmjre_master.jsonl",
      "papers": 19926,
      "accept": 12149,
      "reject": 7211,
      "other_or_unknown": 566,
      "description": "Canonical benchmark containing all reviewed papers with available manuscript, review, rebuttal and decision metadata."
    },
    "review": {
      "file": "llmjre_review.jsonl",
      "papers": 19255,
      "accept": 12063,
      "reject": 7192,
      "description": "Static review benchmark for LLM-human review alignment. Contains manuscript text, human reviews and valid Accept/Reject decisions."
    },
    "rebuttal": {
      "file": "llmjre_rebuttal.jsonl",
      "papers": 14633,
      "accept": 9173,
      "reject": 5460,
      "description": "Dynamic rebuttal benchmark containing papers from LLMJRE-Review that additionally include author rebuttals."
    }
  },
  "construction_pipeline": [
    "Raw Re² Dataset",
    "Paper Summary Extraction",
    "Review Extraction",
    "Rebuttal Extraction",
    "Review-Rebuttal Merge",
    "LLMJRE-Master",
    "LLMJRE-Review",
    "LLMJRE-Rebuttal"
  ],
  "benchmark_tasks": {
    "task_1": {
      "name": "Static Review Evaluation",
      "dataset": "llmjre_review",
      "objective": "Evaluate alignment between LLM reviewers and human reviewers using manuscript content and review information."
    },
    "task_2": {
      "name": "Rebuttal-Aware Evaluation",
      "dataset": "llmjre_rebuttal",
      "objective": "Evaluate whether LLM reviewers appropriately update their assessments after reading author rebuttals."
    }
  },
  "statistics": {
    "review_source_of_truth": 19926,
    "paper_summary_records": 19796,
    "rebuttal_records": 14830,
    "review_alignment_records": 19255,
    "rebuttal_benchmark_records": 14633
  },
  "notes": [
    "Human review records are the canonical source of paper identifiers.",
    "LLMJRE-Review contains only papers with manuscript text and valid Accept/Reject decisions.",
    "LLMJRE-Rebuttal is a subset of LLMJRE-Review containing available author rebuttals.",
    "Intermediate processing files are available in the companion GitHub repository."
  ]
}