Learning Interpretable Legal Case Retrieval via Knowledge-Guided Case Reformulation

TL;DR

Legal case retrieval faces challenges from long, structured documents and the need for domain-specific knowledge. The paper introduces KELLER, which uses legal knowledge to reformulate cases into concise crime-article sub-facts via a two-step LLM prompting process, then performs interpretable sub-fact level matching with a MaxSim-Sum aggregation. It couples this relevance modeling with dual-level contrastive learning (case-level and sub-fact-level) to improve robustness on complex queries, achieving state-of-the-art results on LeCaRD and LeCaRDv2 in both zero-shot and fine-tuned settings. The work delivers interpretable retrieval by exposing which sub-facts drive ranking and highlights practical trade-offs and ethical considerations for deploying knowledge-guided legal IR systems in real-world settings.

Abstract

Legal case retrieval for sourcing similar cases is critical in upholding judicial fairness. Different from general web search, legal case retrieval involves processing lengthy, complex, and highly specialized legal documents. Existing methods in this domain often overlook the incorporation of legal expert knowledge, which is crucial for accurately understanding and modeling legal cases, leading to unsatisfactory retrieval performance. This paper introduces KELLER, a legal knowledge-guided case reformulation approach based on large language models (LLMs) for effective and interpretable legal case retrieval. By incorporating professional legal knowledge about crimes and law articles, we enable large language models to accurately reformulate the original legal case into concise sub-facts of crimes, which contain the essential information of the case. Extensive experiments on two legal case retrieval benchmarks demonstrate superior retrieval performance and robustness on complex legal case queries of KELLER over existing methods.

Figures (8)

• Figure 1: The query case and candidate document case examples. The query case typically contains only partial content since it has not been adjudicated. Extractable crimes and law articles are highlighted in red.
• Figure 2: Overview of KELLER. We first perform legal knowledge-guided prompting to reformulate the legal cases into a series of crucial and concise sub-facts. Then, we directly model the case relevance based on the sub-facts. The model is trained at both the coarse-grained case level and the fine-grained sub-fact level via contrastive learning.
• Figure 3: Evaluation on different query types. We evaluate four models on (a) LeCaRD and (b) LeCaRDv2.
• Figure 4: An example of the interpretability of KELLER. We can observe that each sub-fact of the query finds a correct match in the candidate document (in red).
• Figure 5: Comparison of the original text, naive summarization, and our proposed knowledge-guided case reformulation. The original text is manually abbreviated due to its length. Important sentences are marked in red.