Sentiment Analysis of Movie Reviews using ELMO Word Representation on Hybrid Neural Network

Abstract

Sentiment analysis is an active research area in natural language process ing where underlying emotions expressed in textual data are analysed and identified. It can either be categorized as positive and negative or mul ticlass such as happiness, sadness, anger, fear, etc that enables to under stand the social sentiment of a particular brand, product or service while monitoring online conversations. In addition, it can be used for studying psychological problems, conducting emotional marketing and for improv ing customer experience. The overload of data in today’s environment makes it impossible to analyze it manually and makes systematic senti ment analysis even more relevant. Many approaches were used earlier for sentiment analysis like ruled - based analysis and machine learning based analysis. Recent research indicates that deep learning techniques shows better performance for sentiment anal ysis. Hence the hybrid neural network model consisting of CNN and BiL STM is introduced. Initially, text are converted into vector format us ing embedding languages such as Word2Vec, GloVe and ELMo language model. Then CNN allows extracting local features of text vectors while global features are extracted by BiLSTM. The features extracted by the two models are then used together for performing the sentiment analy sis. Experiments are conducted on IMDB movie reviews dataset. The trained hybrid neural network can then automatically classify the sen tences achieving an accuracy rate of 91% for text categorization compared to other state-of-art approaches.