Named Entity Recognition in Legal Domain : A Multilingual Study

Abstract

This thesis presents an investigation into the development of Named Entity Recog nition (NER) models for Indian Legal datasets in both English and Hindi languages. Thestudy explores the application of some efficient neuralnetworks,includingRoBERTa1, RoBERTa-GCN, Text-GCN, and BiLSTM-CRF, in conjunction with different classi fiers such as softmax and CRF. The primary objective of this task is to address the need for NER models as a crucial component in building AI applications in the legal domain.The thesis outlines the methodology employed for training and evaluating the NER models on the Indian Legal dataset. It discusses the preprocessing steps undertaken to prepare the data for training and highlights the specific challenges and nuances encountered in the legal domain.Furthermore, the thesis presents the architecture and configurations of the different NER models utilized, along with the rationale behind their selection. It provides a detailed analysis of the model efficacy in terms of precision, recall, and F1 score, comparing the effectiveness of each model and classifier combination.The findings demonstrate the efficacy of the proposed NER models for Indian Legal datasets, with certain models exhibiting superior performance in capturing named entities accurately. The thesis also discusses the limitations and potential areas for future research and improvement.Overall, the study contributes to the advancement of NER techniques for Indian Legal datasets in both English and Hindi languages, laying the foundation for the development of AI applications in the legal domain. The insights gained from this research can aid in automating legal information ex traction, document analysis, and other related tasks, thereby enhancing efficiency and productivity in the legal industry.