Bi-LSTM-LDA— A Topic Modelling Technique to Identify the Relevant Law for Indian Legal Cases

Abstract

AI had already changed the way of approaching the problem in many industries. Now it is changing the way of practicing the law. Already many models were created to predict the outcome of a case given the facts of the case by analysing the prior cases. Before any model starts predicting the outcome of the case, it is most important to identify the section code since punishment or judgment of a case depends on the law or article under which a specific case comes under. As far as Supreme Court of India is concerned, the section code plays the major role. Hence this paper proposes Bi-LSTM-LDA model that identifies the relevant law given only the facts of the case. The proposed model is ensemble of Bi-LSTM and LDA (Latent Dirichlet Allocation) algorithms that assigns section code given the facts of the case. The proposed model is compared with existing neural topic modelling models and the results shows Bi-LSTM-LDA outperforms than others with an accuracy of 96%.