Legal Brief Analysis - Using BERT-embeddings for Recommending Cases
Posted: 29 Jan 2021
Date Written: November 4, 2020
In the US legal system attorneys submit briefs that contain legal argument explaining why the reviewing court should affirm or reverse the lower court's judgment. Briefs typically contain a large number of arguments which are supported by citations to legal precedents i.e. controlling cases or statutory law. It is important for attorneys to make sure that they have researched all the key cases relevant to their arguments. We propose a method for mining key arguments in a brief and recommending cases supporting or opposing the identified arguments. We apply traditional machine learning techniques to identify arguments in a brief and leverage BM25 and embedding-based search techniques to recommend cases which discuss the issues matching the arguments in the attorney brief. In the paper we will discuss (1) tuning search for product requirements, (2) combining and merging results from 2 different SEARCH algorithms and (3) quality metrics for SEARCH results in our context.
Suggested Citation: Suggested Citation