Extractive and abstractive text summarization of legal documents using machine learning approaches
Ashutosh Bajpai, Varada Gupta, Anshul Verma, Prem Bahadur Rana · 2025
In the domain of Legal AI, the intricate task of summarizing legal records, particularly within the context of Indian case materials, has sparked significant research initiatives. This study undertakes a comprehensive exploration of legal text summarization by evaluating the performance of seven machine learning-based models-Luhn, BART, Legal Pegasus, LexRank, LSA, TextRank and Legal-LED - on judgment report datasets sourced from the Indian national legal portal. Our analysis encompasses both extractive and abstractive summarization models. We address crucial questions regarding the efficacy of different summarization models and their respective families when applied to the intricate structure of legal case documents. Additionally, we provide insights into the evaluation methodologies employed, considering the challenges posed by the extended length of legal documents. Noteworthy findings indicate Legal Pegasus as a standout performer in legal text summarization, surpassing other models. The outcomes of this research not only enhance our understanding of legal summarization but also offer valuable considerations for diverse applications of long document summarization.