Similarity Ranking of Case Law Using Propositions as Features

In common law legal systems judges use previous similar cases, known as precedent, to determine how the case they are currently hearing should be decided. Task 1 of the 2024 Competition on Legal Information and Extraction/Entailment (COLIEE) is designed to emulate the task of retrieving these similar cases. This paper summarizes the approach of the second placed team, University of Melbourne Natural Language Processing (UMNLP) in Task 1 of the 2024 COLIEE. We have developed a pairwise similarity ranking framework. We train a feed forward neural network to perform a binary classification task, based on a multitude of features from each query-candidate case pair. Those features include the extraction and similarity matching for a novel feature which we call a ‘proposition’ (a short summary of the basis upon which a noticed case has been cited), as well as the name of the judge deciding the case, verbatim quotations from the text, and several other novel features.