Datasets:

rishia2220
/

scopebench-pr

license: cc-by-4.0
language:
  - en
tags:
  - benchmark
  - peer-review
  - llm-evaluation
  - fairness
pretty_name: ScopeBench-PR
size_categories:
  - 10K<n<100K
configs:
  - config_name: papers
    data_files: papers/papers.parquet
  - config_name: review_runs
    data_files: review_runs/review_runs.parquet
  - config_name: score_pairs
    data_files: score_pairs/score_robustness_pairs.parquet
  - config_name: scope_items
    data_files: scope_items/scope_detection_items.parquet
  - config_name: rebuttal_items
    data_files: rebuttal_items/rebuttal_items.parquet
  - config_name: human_validation
    data_files: human_validation/human_validation.parquet

ScopeBench-PR

Counterfactual benchmark for scope fairness in LLM peer review.

Code: https://github.com/SPELLAILab/ScopeBench-PR
Workshop: https://sites.google.com/view/globalsouthai-ijcai26/home

from datasets import load_dataset

load_dataset("rishia2220/scopebench-pr", "papers")
load_dataset("rishia2220/scopebench-pr", "scope_items")

Tables are stored as Parquet. See the GitHub repository for schemas, prompts, and analysis scripts.