| --- |
| license: mit |
| task_categories: |
| - reinforcement-learning |
| - text-classification |
| language: |
| - en |
| tags: |
| - rule-based reasoning |
| - logical reasoning |
| size_categories: |
| - 10K<n<100K |
| --- |
| # RuleCollection-32K for Rule-centric Logical Reasoning |
|
|
| This repository contains datasets for rule-based reasoning tasks, organized into two main categories: |
|
|
| GitHub Repository: https://github.com/bigai-nlco/RuleReasoner |
|
|
| Paper: https://openreview.net/forum?id=MQV4TJyqnb |
|
|
| arXiv: arxiv.org/abs/2506.08672 |
|
|
| ## In-Distribution (ID) Datasets |
|
|
| - **ar_lsat**: Analytical Reasoning from LSAT |
| - **clutrr**: CLUTtRR (Compositional Language Understanding and Text-based Relational Reasoning) |
| - **folio**: FOLIO (First-Order Logic in Natural Language) |
| - **logic_nli**: Logic-based Natural Language Inference |
| - **logical_deduction**: Logical Deduction tasks |
| - **logiqa**: LogiQA (Logical Reasoning QA) |
| - **prontoqa**: ProntoQA (Prompt-based Question Answering) |
| - **proofwriter**: ProofWriter (Proof Generation and Verification) |
| |
| ## Out-of-Distribution (OOD) Datasets |
| |
| - **bigbench_extra_hard**: BigBench Extra Hard tasks (logical problems subset) |
| - **bigbench_hard**: BigBench Hard tasks (logical problems subset) |
| - **proverqa**: ProverQA (Automated Theorem Proving) |
|
|
| ## File Format |
|
|
| Each dataset contains: |
| - `train.json` / `train.parquet`: Training data |
| - `test.json` / `test.parquet`: Test data |
|
|
| Data is provided in both JSON and Parquet formats for convenience. |
|
|
| ## Usage |
|
|
| ```python |
| from datasets import load_dataset |
| |
| # Load a specific dataset |
| dataset = load_dataset("RuleReasoner/rule-reasoning", data_files="id/folio/train.json") |
| |
| # Or load all files from a specific dataset |
| dataset = load_dataset("RuleReasoner/rule-reasoning", data_files="id/folio/*.json") |
| ``` |
|
|
| ## Citation |
|
|
| If you use these datasets in your research, please cite the original papers for each dataset. |
| ```latex |
| @inproceedings{ |
| liu2026rulereasoner, |
| title={RuleReasoner: Reinforced Rule-based Reasoning via Domain-aware Dynamic Sampling}, |
| author={Yang Liu and Jiaqi Li and Zilong Zheng}, |
| booktitle={The Fourteenth International Conference on Learning Representations}, |
| year={2026}, |
| url={https://openreview.net/forum?id=MQV4TJyqnb} |
| } |
| ``` |
