LawDual-Bench: A Dual-Task Benchmark for Chinese Legal Reasoning

LawDual-Bench is a dual-task benchmark designed to systematically evaluate large language models in legal document processing and judgment prediction. Created by Yuwh07 and last updated on May 18, 2025, it addresses a gap in unified evaluation frameworks for Chinese legal NLP. The dataset was constructed using a semi-automated method combining human and LLM efforts, focusing initially on insider trading cases.

Use Cases

• Benchmarking model performance on structured information extraction from legal documents, as described.
• Evaluating reasoning capabilities for case fact analysis and judgment prediction, as mentioned.
• Training or fine-tuning models for Chinese legal NLP tasks based on the described dual-task structure.

Strengths

• Designed for a dual-task evaluation framework covering document understanding and judgment prediction.
• Constructed via a described semi-automated pipeline combining human and LLM annotation.
• Last updated on 2025-05-18, indicating recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file formats are unknown, which may limit suitability assessment.
• Data may reflect bias inherent to its specific construction method and initial focus on insider trading.

Provenance

Source

huggingface, author Yuwh07

Collection Method

Semi-automated dataset construction using human+LLM annotation.

Time Range

null

Freshness

Last updated 2025-05-18 12:59:32.

Geography

Chinese legal domain