Research of Sentiment Analysis Based on Long-Sequence-Term-Memory Model

Abstract

A method based on Long-Sequence-Term-Memory model and vector embedding to analyze sentiments of the online reviews is proposed in this paper. In order to obtain the vector representation from sentence level, it uses the extraction methods of LSTM, which is based on non-liner learning, to extend the vector embedding to sentence representation, and to achieve sentence embedding ultimately. The experimental results proved the high accuracy of this method, with which could live up to 91.35% when classifying the sentiments of online reviews. The ability to be applied to variety languages and the strong scalability of large scale corpus are two of its advantages, meanwhile, feature extraction of bigram can also improve its test accuracy. As the experimental results showed that the sentiment analysis method based on the LSTM model and the principle of embedding word is a highly effective method of sentiment analysis, and it has the strong scalability and can be applied to comments of different languages.