Sentiment Infomation based Model for Chinese text Sentiment Analysis

Abstract

As an important task of natural language processing, chinese text sentiment analysis aims to analyze the comprehensive sentiment polarity of chinese text. With the emergence of various deep neural network models, sentiment analysis tasks have once again made significant progress. However, these neural network models could not accurately capture sentiment information on sentiment analysis tasks, which leads to their instability. In order to enable the model to explicitly learn the sentiment knowledge in chinese text, this paper proposes a sentiment information based network model(SINM). We use Transfomer encoder and LSTM as model components. With the help of Chinese emotional dictionary, we can automatically find sentiment knowledge in chinese text. In SINM, we designed a hybrid task learning method to learn valuable emotional expressions and predict sentiment tendencies. First of all, SINM needs to learn the sentiment knowledge in the text. Under the auxiliary influence of emotional information, SINM will pay more attention to sentiment information rather than useless information. Experiments on the dataset of ChnSentiCorp and ChnFoodReviews have found that SINM can achieve better performance and generalization ability than most existing methods.