Semantic Legal Searcher: Neural Information Retrieval-based Semantic Search for Case Law

Anonymous

Abstract: This study aims to build a highly performant semantic search model in the field of law by applying neural information retrieval techniques. With classical keyword-based search models, it is difficult for users without domain knowledge of the law to obtain information by searching with appropriate legal terms. In order to solve this problem, we propose a Semantic Legal Searcher (SLS), a neural information retrieval-based case law search model. It enables users to search and gain access to legal information even with simple queries rather than professional legal terms. Specifically, the SLS process starts with generating good-quality embeddings from a pre-trained language model we created. Next, latent keywords are extracted by a parallel clustering-based topic modeling and then relevance between input queries and legal documents and keywords is estimated by a multi-interactions paradigm we developed. Lastly, the SLS provides users with semantic similar case laws based on the estimated scores. Experimental results demonstrate that our semantic search model provides relevant precedents for users by understanding legal text and is a powerful tool for information retrieval. The SLS can be useful for a lot of real-life applications and allows the general public to easily access legal information.

Paper Type: long

Research Area: Semantics: Sentence-level Semantics, Textual Inference and Other areas