When Rules Learn: A Self-Evolving Agent for Legal Case Retrieval

Short summary

Researchers propose a self-evolving agent framework that autonomously creates and refines retrieval rules for legal documents without parameter training. The LLM-based agent learns from iterative experiments, eliminating ineffective rules based on feedback. Testing on a Chinese legal benchmark shows outperformance over static rule baselines.

• •Self-evolving agent creates and tests retrieval rules iteratively without manual design
• •Works with BM25 baseline, improves accuracy through rule elimination and optimization
• •Demonstrates LLM's ability to leverage prior experimental results for systematic refinement

Generated with AI, which can make mistakes.