Supplemental Enhancement of Action Segments: A Retrieval Optimization for Large Language Models in the Legal Domain
Abstract: Utilizing the Retrieval-Augmented Generation (RAG) framework with large language models for question answering often results in low retrieval precision and recall rates. A solution to address this issue involves retrieving external knowledge at various granularities. However, this strategy typically suffers from decreased precision in coarse-grained retrieval and omissions in fine-grained retrieval. To overcome these challenges, we introduce a novel framework designed for the legal domain, named Supplemental Enhancement of Action Segments (SEAS). SEAS utilizes few-shot prompting to extract action segments from legal texts, which are then used to enhance the retrieval of complete legal texts. In the Japanese Law Retrieval task, SEAS significantly enhances the performance of three distinct embedding models. Furthermore, in the Chinese Legal Question Answering task, SEAS outperforms all baselines across all metrics.
Paper Type: Short
Research Area: NLP Applications
Research Area Keywords: legal NLP
Contribution Types: Model analysis & interpretability
Languages Studied: English,Chinese
Submission Number: 774
Loading