Text Sentiment Analysis of Movie Reviews Based on Word2Vec-LSTM

Abstract

A hybrid model based on Word2Vec-LSTM is utilized to analyze movie review sentiment in this paper. Word2Vec integrates text context semantics to generate text vector, and LSTM extracts semantic information to classify positive and negative emotions. In order to measure the classification capacity of the Word2Vec-LSTM, Word Index and Hash Trick method are constructed as benchmark models. We combine the word index and Hash Trick with several mainstream machine learning models to obtain the Word Index-Based Classifiers and Hash Trick-Based Classifiers. The experimental results show that Word2Vec-LSTM has the best performance. The accuracy is improved by 29.12% and 18.84% compared with Word Index-Based Classifiers and Hash Trick-Based Classifiers respectively, which shows that the Word2Vec-LSTM hybrid model is more effective for the movie review sentiment analysis.