A LSTM-based Sentiment Analysis Network on Classifying Natural Language Comments

Machine learning techniques, especially deep learning, are currently widely studied in the sentiment analysis task. Automatically classifying natural language comments to positive or negative is an essential and challenging task. In this paper, the sentiment analysis network based on LSTM is utilized to predict human emotion via natural language. Firstly, several variants of LSTM networks are introduced and developed. Then, the proposed network is designed to be able to inference and classify five levels of rating via human comments as input. In addition, a movie reviews benchmark dataset is utilized to evaluate our proposed network with other baseline algorithms. Finally, the results demonstrate promising and competitive performance against other classical methods with the accuracy of 0.8743.