Efficacy of ELECTRA-based Language Model in Sentiment Analysis

Sentiment analysis (SA), a branch of natural language processing (NLP), aims to recognize and extract the feelings behind a piece of text or speech. SA assigns the text a favourable, neutral, or negative evaluation. Social media sentiment analysis is gaining importance as it examines feelings, attitudes, and perspectives by gathering reviews and comments on several web platforms. Sentimental evaluation is significant from a business or political standpoint, where it significantly affects the strategic choice made by the customer. In this work, we have chosen movie reviews, IMDB dataset, for conducting sentiment analysis. Transformers have accelerated the performance of many NLP applications. Due to its parallel processing and lesser computation time, it has outperformed many of the existing deep learning models. Given this motivation, we have conducted experiments with various transformer models such as BERT, XLNET, RoBERTa and ELECTRA and analysed the performance of various models. Sentiment is classified as either positive or negative. Experimental results demonstrate that ELECTRA outperforms other models by giving the highest accuracy of 93.32%.