LexIssue: Benchmarking Legal Issue Identification in Chinese Civil Litigation
LexIssue introduces a benchmark for computational modelling of legal issue identification in Chinese civil litigation, addressing a significant gap in legal AI research The authors propose a legally grounded hierarchical schema combining free-form issue descriptions with structured legal categories across 27 causes of action The benchmark contains 430 real-world Chinese civil litigation cases with 1,303 expert-annotated disputed legal issues Legal issue identification is formulated as two comple
Analysis
TL;DR
- LexIssue introduces a benchmark for computational modelling of legal issue identification in Chinese civil litigation, addressing a significant gap in legal AI research
- The authors propose a legally grounded hierarchical schema combining free-form issue descriptions with structured legal categories across 27 causes of action
- The benchmark contains 430 real-world Chinese civil litigation cases with 1,303 expert-annotated disputed legal issues
- Legal issue identification is formulated as two complementary tasks: legal issue generation and legal issue classification
- A retrieval-augmented generation approach using a constructed legal issue knowledge base (441 candidate entries) consistently improves performance across diverse models
Why It Matters
This work addresses a critical gap in legal AI by focusing on legal issue identification, a foundational component of real-world litigation that has been comparatively underexplored. The benchmark and knowledge base provide a valuable resource for researchers and practitioners working on legal NLP systems, particularly for Chinese civil law contexts. The retrieval-augmented approach demonstrates practical pathways for improving AI performance in specialized legal domains.
Technical Details
- Hierarchical Schema: A legally grounded framework representing legal issues through both free-form descriptions and structured legal categories, organized across 27 causes of action
- Benchmark Construction: LexIssue contains 430 real-world Chinese civil litigation cases with 1,303 expert-annotated disputed legal issues, covering both generation and classification tasks
- Knowledge Base: An issue-centric legal knowledge base spanning 27 causes of action with 441 candidate legal issue entries designed to support retrieval-augmented reasoning
- Task Formulation: Legal issue identification is decomposed into two complementary tasks—legal issue generation (producing free-form issue descriptions) and legal issue classification (mapping to structured legal categories)
- Experimental Approach: Retrieval-augmented generation (RAG) using the constructed knowledge base was evaluated across a diverse set of models, showing consistent performance improvements in identifying disputed legal issues and their corresponding legal attributes
Industry Insight
- Legal AI systems should incorporate domain-specific knowledge bases and retrieval mechanisms rather than relying solely on fine-tuning, as demonstrated by the consistent improvements from RAG approaches
- The dual-task formulation (generation + classification) provides a practical blueprint for building legal AI systems that need to produce both natural language outputs and structured legal classifications
- Chinese civil litigation represents an underserved area in legal AI research, presenting opportunities for developing region-specific benchmarks and knowledge resources that can be adapted to other civil law jurisdictions
Disclaimer: The above content is generated by AI and is for reference only.