BB25HLegalSum: Leveraging BM25 and BERT-Based Clustering for the Summarization of Legal Documents

Proceedings of Recent Advances in Natural Language Processing, pages 255–263 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_029 255 BB25HLegalSum: Leveraging BM25 and BERT-based clustering for the summarization of legal documents Leonardo Bonalume Informatics Institute Federal Univ. of Rio Grande do Sul Porto Alegre - Brazil lbandrade@inf.ufrgs.br Karin Becker Informatics Institute Federal Univ. of Rio Grande do Sul Porto Alegre - Brazil karin.becker@inf.ufrgs.br Abstract Legal document summarization aims to provide a clear understanding of the main points and arguments in a legal document, contributing to the efficiency of the judicial system. In this paper, we propose BB25HLegalSum, a method that combines BERT clusters with the BM25 algorithm to summarize legal documents and present them to users with highlighted impor- tant information. The process involves select- ing unique, relevant sentences from the original document, clustering them to find sentences about a similar subject, combining them to generate a summary according to three strate- gies, and highlighting them to the user in the original document. We outperformed baseline techniques using the BillSum dataset, a widely used benchmark in legal document summariza- tion. Legal workers positively assessed the highlighted presentation. 1 Introduction Pending judicial processes are a prevalent and sig- nificant issue affecting legal systems worldwide. The number of pending cases can vary significantly depending on population size, legal system, and backlog of cases. While in some countries, there may be only a few thousand pending processes, it can amount to millions in others. This scenario motivates the research of computational techniques that can help accelerate judicial analysis, select similar cases for judging in batches, or identify patterns that could lead to better decision-making. The automatic summarization of legal documents to synthesize their essence is critical in this context. The goal of automatic text summarization is to create summaries that are similar to human-created summaries (Allahyari et al., 2017). This is a chal- lenging task since natural language is complex and nuanced. Text summarization algorithms must con- sider the intended audience, the purpose of the sum- mary, as well as the type and format of the original text. Text summarization is valuable for various applications, such as news aggregation, document management, and legal document summarization. Most works use extractive summarization to gen- erate summaries, defined in (Anand and Wagh, 2019), as “the generation of a summary containing a sentence subset of the original text after iden- tifying the important sentences”. Several tech- niques were explored for extractive legal text sum- marization, including word relevance (Polsley et al., 2016), graph-based ranking models (Dalal et al., 2023; Jain et al., 2023), statistical models (Jain et al., 2022; Merchant and Pande, 2018), and deep learning (Anand and Wagh, 2019). More recently, BERT (Devlin et al., 2018) has been leveraged in the legal area (Furniturewala et al., 2021), inspired by state-of-the-art results achieved in general ex- tractive text summarization (Liu, 2019). Another approach used in the legal documents area is BM25 (Robertson et al., 2009), a ranking function commonly used in information retrieval to determine the relevance of a document …

Content truncated to 500 words