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

A self-evolving framework for legal case retrieval has been proposed, enhancing the traditional BM25 model without requiring parameter training. This framework utilizes an LLM-based agent to iteratively generate and validate query rewriting rules, demonstrating superior performance on the LeCaRD-v2 benchmark compared to non-evolutionary baselines. The approach is significant for practitioners as it leverages the LLM's ability to refine rules based on historical feedback, potentially improving retrieval accuracy in complex legal contexts.