Comparative Analysis: Sentiment Analysis for Legal Judgment Text in India’s Supreme Court Based on GloVe Pretrained Word Embedding and Deep Learning Models

Abstract

AbstractWith the rapid growth of Internet technology in recent years, an innovative legal research tool for Indian case laws has been established to save time for legal professionals. Sentiment analysis employs head notes from comparable previous instances to determine the main structure of a current case. Word embedding techniques such as GloVe and Word2Vec have shown to be especially effective at converting words into dense vectors. We use the Indian Supreme Court dataset and compare the performance of various deep learning models such as basic neural network, convolutional neural network, and recurrent neural network with GloVe embedding. GloVe + convolution neural network + attention mechanism, a new sentiment analysis model based on sentiment word embedding and integrating convolution neural network and attention mechanism, is introduced in this study analysis. The embedding vector is used to improve the sentiment features in the sentences. Then, we employ convolutional neural networks to extract the important sentiment and context-level features in the sentences, and we weight the features using the attention method. In this study, we look at 5000 positive and 5000 negative sentiments. Based on the experimental results, we conclude that our approach delivers high-quality legal reference information from full judgment texts, allowing lawyers to forecast positive and negative verdicts in a cost-effective and time-saving manner.KeywordsSentiment analysisGloVe embeddingCNNSupreme Court judgment text