When Rules Learn: A Self-Evolving Agent for Legal Case Retrieval
Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.
Researchers proposed a self-evolving LLM-based agent that improves legal case retrieval by automatically creating, testing and pruning query-rewriting rules for BM25 without parameter training. Evaluated on the Chinese LeCaRD-v2 benchmark, the framework outperformed non-evolutionary baselines such as human-designed rules and greedy rule selection, with gains tied to the LLM’s ability to use prior experimental feedback and eliminate weak rules.
Researchers developed a self-evolving AI agent that improves legal case retrieval by automatically refining search rules without additional training. The system, tested on a Chinese legal benchmark, outperformed traditional methods by using large language models to iteratively test and eliminate ineffective rules. Findings highlight the AI's ability to leverage past results and built-in knowledge to enhance search precision.