Movie-LSTM and Lexicon Technique-Based Movie Review Analysis

As movies and entertainment influence a big chunk of our life, there is a pressing need to understand user behaviors and opinions expressed toward movies. This study aims to address the challenge of identifying suitable classification algorithms for analyzing such data. Deep learning models for sentiment analysis offer promising capabilities in terms of detailed representation and improved performance compared to traditional feature-based techniques. This work focuses on enhancing sentiment classification performance through a custom deep learning model that incorporates an advanced word embedding technique and leverages a movie-long short-term memory (Movie-LSTM) network. Additionally, the work proposes a sequential model that combines the proposed baseline classifier with other state-of-the-art classifiers commonly employed in sentiment analysis. The contributions of this paper can be defined into two categories, the first one is to establish a robust framework that leverages word embedding and the second one is proposed network helps to learn contextual relationships between words. This allows the proposed model to comprehend previously unseen or infrequently encountered words, which is responsible for capturing the internal meaning of particular words with respect to a particular context. The derived model is compared with the existing state of the arts using accuracy. The proposed model predicts sentiments of testing data with 88.3% accuracy for the IMDB dataset.