Empirical Analysis of Prompt-Based LLMs for Structural Classification of Indian Legal Judgements

Legal judgment is a decision formally issued by a court as a conclusion to legal proceedings, analyzing the court's findings, reasoning, and rulings on the matters brought before it. This study explores the analysis of these legal judgments and classifies these judgments with the help of large language models and incorporating prompting techniques. The dataset considered in this study facilitates comprehensive model training and evaluation by assessing performance metrics such as Accuracy, Precision, Recall, and F1-score both Macro and Weighted. The Mistral model, of version OpenHermes-2.5-Mistral-7B-16k-GGUF, achieved an accuracy score of 0.61, performing best using the 0- shot prompting technique. This research aims to provide insights into the efficacy of these Large Language Models in automating legal research tasks and streamlining document processing workflows. The findings shed light on how Large Language Models interact with and reason through documents in the complex legal domain, offering valuable implications for practitioners and researchers.