Reproduce “LegalSearchLM: Rethinking Legal Case Retrieval as Legal Elements Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chaeeun Kim, Jinu Lee, Wonseok Hwang
LBOX and the University of Illinois Urbana-Champaign introduced LegalSearchLM, a generative retrieval model for legal case retrieval that re-conceptualizes the task as generating key legal elements, and developed LEGAR BENCH, a large-scale Korean legal benchmark. LegalSearchLM achieved a Precision@5 of 0.68 on LEGAR BENCH_Standard, outperforming strong baselines like Contriever by 20 percentage points, and demonstrated robust generalization to unseen crime types.
Key insights
- Reframing legal case retrieval as a generative task, focusing on specific legal elements, aligns more closely with how legal professionals determine relevance, mitigating information loss and improving semantic precision.
- Self-supervised fine-tuning on legal elements from query cases enhances generalization across diverse and rare crime types by teaching the model to reason over legal components.
- Integrating constrained beam decoding with an FM-index ensures that all generated legal elements are directly verifiable within the target corpus, providing strong grounding and preventing hallucinations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.