Bayesian Optimization based Score Fusion of Linguistic Approaches for Improving Legal Document Summarization

Abstract

Due to the lengthy and complex nature of legal documents, automatic summarization has very high applicability in this domain. Recently, several researchers have proposed extractive summarization based ideas for legal documents. Extraction of important sentences via appropriate sentence scoring is the main idea behind extractive summarization, for which several linguistic approaches are proposed in the text summarization literature. However, the field of legal document summarization needs proper exploration in terms of the interaction among these approaches. This work proposes a Bayesian Optimization (BO) based Score Fusion approach that performs an efficient weighted combination of several statistical and semantic sentence scoring techniques. Experimental evaluation on two publicly available datasets and one privately accessed dataset suggests that the efficient combination of heterogeneous linguistic features for sentence scoring can improve the summarization performance for ROUGE-1, ROUGE-2, and ROUGE-L metrics. Moreover, the proposed BO-based Score Fusion approach can effectively summarize lengthy legal documents outperforming the current state-of-the-art techniques.