ABSTRACTIVE SUMMARIZATION OF INDIAN LEGAL DOCUMENTS USING T5 & QLoRA

This research project aims to develop an abstractive summarization system for Indian legal documents. The system leverages the power of the T5 transformer model, fine-tuned using Quantized Low-Rank Adaptation (QLoRA). The training data comprises two datasets, the Indian Legal Corpus (ILC) and IN-Abs, both containing court cases and their corresponding abstractive summaries. The system is designed to accept legal text input directly or extract it from uploaded DOCX or PDF documents. An initial extractive summary is generated using the bert-extractive-summarizer, which is subsequently fed into the fine-tuned T5 model to produce an abstractive summary. The principal result of this research is the successful implementation of a system capable of generating abstractive summaries of Indian legal documents. The system achieved a ROUGE-1 score of 46.37%, demonstrating its effectiveness. In conclusion, this research contributes to the field of legal document summarization by providing a system that can generate concise and coherent summaries, thereby aiding in the efficient comprehension of complex legal texts. This work also opens avenues for further improvements and applications in the legal tech domain.