Logic Rules as Explanations for Legal Case Retrieval

• DOI: 10.48550/arxiv.2403.01457
• Publication Date: 2024-03-03
• Authors: Zhongxiang Sun; Kepu Zhang; Weijie Yu; Haoyu Wang; Jun Xu

In this paper, we address the issue of using logic rules to explain results from legal case retrieval. The task is critical retrieval because users (e.g., lawyers or judges) are highly specialized and require system provide logical, faithful, interpretable explanations before making decisions. Recently, research efforts have been made learn explainable models. However, these methods usually select rationales (key sentences) cases as explanations, failing faithful logically correct explanations. propose Neural-Symbolic enhanced Legal Case Retrieval (NS-LCR), a framework that explicitly conducts reasoning on matching through learning case-level law-level rules. learned then integrated into process in neuro-symbolic manner. Benefiting nature rules, NS-LCR equipped with built-in explainability. We also show model-agnostic can be plugged for multiple To showcase NS-LCR's superiority, enhance existing benchmarks by adding manually annotated introducing novel explainability metric Large Language Models (LLMs). Our comprehensive experiments reveal effectiveness ranking, alongside its proficiency delivering reliable