APR Criminal Case Database (Chinese)

Dataset Description

This dataset is constructed to support Analogical Precedent Retrieval (APR). It comprises 172,445 real-world criminal cases spanning from 2001 to 2020, covering 25 provinces in China. The extensive corpus provides sufficient geographical and temporal diversity for robust analogical retrieval research.

Acknowledgments

We would like to express our sincere gratitude to the open-source project liuhuanyong/LawCrimeMining (Law Crime Mining Based on Corpus build and content analysis by NLP methods). Our data collection pipeline was adapted based on their foundational work. The raw legal documents were primarily collected from public legal databases, including Lawlib. We crawled and processed these publicly available judicial records strictly for academic, non-commercial research purposes.

Ethical Considerations & Privacy

The dataset consists of judicial decisions that are inherently public records. To protect individual privacy and strictly comply with academic ethical guidelines, all case documents have been thoroughly anonymized to remove sensitive personal information (e.g., real names of individuals, specific identification numbers). The dataset is released solely for academic research purposes—specifically to evaluate NLP algorithms and mitigate LLM hallucinations—and strictly prohibits any commercial application or malicious use.