Text Sentiment Analysis of Film Reviews Using Bi-LSTM and GRU

• Published in: 2023 4th International Conference on Electronics and Sustainable Communication Systems (ICESC) pp. 1379 - 1386
• Main Authors: Mouthami, K., Yuvaraj, N., Thilaheswaran, K.K., Lokeshvar, K.J.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 06.07.2023
• Subjects: Bidirectional long short-term memory; Computational efficiency; Gated recurrent units; Logic gates; Machine learning; Motion pictures; movie review; Natural language processing; Sentiment analysis; Training; Writing
• DOI: 10.1109/ICESC57686.2023.10193121

Sentiment analysis is a vast subject to explore in natural language processing (NLP) techniques. The film reviews were analyzed and segregated into positive, neutral, and negative reviews. The proposed model examines two distinct datasets, one with multi-class labels and the other with binary-type class labels. The preprocessing is done by using the bag of words and skip-gram word2vec, variety of classifiers, including Two state Gated recurrent units (TS-GRU), was used for binary classification, and the Bidirectional Long Short-Term Memory (Bi-LSTM for the multi-class situation was implemented. Bi-LSTM was utilized in the model, in which symmetric Bi-LSTM replaces the LSTM approaches to get around the high computational cost of training the normal LSTM. The Bi-LSTM has equivalent accuracy to the LSTM while significantly reducing computational costs. Sentiment analysis offers a glimpse of how consumers feel about certain movies. This endeavor is undertaken via the study of natural language processing, in which sentiment analysis is a branch which examines how words are arranged and utilized to extract meaning from user writing. This helps to investigate further the methods for deciphering such data thanks to significant advancements in machine learning techniques; with Keras API and the Internet Movie Database (IMDb) dataset, this study compares single and multi-branch CNN with Bidirectional LSTMs with different kernel sizes. GRU achieved 98.24% of accuracy, and bi-LSTM achieved 98.65% of accuracy.