When Rules Learn: A Self-Evolving Agent for Legal Case Retrieval
Researchers have developed a novel self-evolving agent framework designed to enhance legal case retrieval systems. This agent iteratively refines rewriting rules for the BM25 baseline by utilizing an LLM within an automatic evaluation environment. The framework demonstrates improved performance on the Chinese legal case retrieval benchmark LeCaRD-v2, outperforming methods that rely on human-designed rules or greedy selection. The study highlights the LLM's crucial role in leveraging experimental feedback and its inherent ability to eliminate ineffective rules, thereby refining the rule set through self-evolution. AI
IMPACT This research could lead to more accurate and efficient legal information retrieval systems by automating the refinement of search query rules.