Research on Sentiment Classification of Chinese Transition Text

Abstract

The research on text sentiment classification has been very popular. The main methods include the methods based on rules and lexicons, machine statistical learning, and neural network methods. These methods have achieved great success in text sentiment classification, but there are still many defects. For traditional methods, its versatility is poor, once the text style or format changes significantly, the rules need to be re-established. Although the neural network method is generally applicable, it is not sufficient for the application of language knowledge. In order to solve the above problems, we propose a new model combining the traditional method and neural network method. It combines transition lexicon, stacked BILSTM and attention mechanism, and uses word2vec to initialize word vectors. For sentiment classification, semantic features are necessary, stacked BILSTM can capture contextual semantics multiple times to obtain richer contextual information representation. At the same time, we consider that the polarity of the clauses before and after the transition words may be completely opposite, which will also have a significant impact on the classification results. Therefore, we propose two strategies for transition words. One is to use the transition lexicon to supervise the attention mechanism and make the attention model sensitive to the transition words. The other is to cut the sentence according to the transition words to eliminate the noise of irrelevant items in the sentence. Our method combines the advantages of traditional methods and neural network methods, and makes up for their respective shortcomings. We use the proposed method to conduct a sentimental binary classification experiment on three data sets. The experimental results show that our method is superior than the previous methods.