Incorporating Domain Knowledge in Multi-objective Optimization Framework for Automating Indian Legal Case Summarization

Shreya Goswami, Naveen Saini, Saurabh Shukla

Abstract: Legal cases can be lengthy and complex, often containing much information, evidence, and legal arguments. On the contrary, attorneys, judges, and legal researchers must review multiple cases to build arguments, make decisions, or conduct legal research. Summarizing cases can help legal professionals and researchers quickly grasp the essential details without reading through every word of the original documents. While various legal case summarization algorithms exist in the literature, they lack a systematic integration of the various summary quality factors necessary to create comprehensive and concise legal summaries. To address this gap, the present paper proposes an innovative extractive summarization approach by leveraging the use of legal-domain knowledge in the multi-objective optimization-based evolutionary framework. Our method simultaneously optimizes different objectives, including legal domain knowledge, tf-idf scores, and diversity, to get good-quality summaries. As per our knowledge, this work is the first of its kind in utilizing the efficacy of a multi-objective evolutionary algorithm for generating legal document summaries. For evaluation, we thoroughly conduct experiments on legal case documents from the Indian Supreme Court, accompanied by gold-standard summaries created by legal experts. The results obtained reveal that our algorithm demonstrates significant improvements of 16.00%, 12.91%, 13.65%, and 0.18%, over the state-of-the-art technique, in terms of ROUGE-2 F1, ROUGE-L F1, ROUGE-2 Recall, and ROUGE-L Recall, respectively. Further, the statistical significance of the results is also validated.

External IDs:dblp:conf/icpr/GoswamiSS24