Combining Convolutional Neural Networks and Word Topic Features for Chinese Short Text Sentiment Analysis

Abstract

The Chinese short text sentiment classification method based on word embedding and convolutional neural network has already achieved effective results. The recent classification method mainly adopts single channel CNN. In order to enhance classification performance, this study proposes a classification method based on Dual-channel CNN. The word2vec semantic feature and topic model (LDA)feature of words were used as input of Dual-channel CNN to extract sequence sentimental features which, were subsequently applied on short text sentiment classification. The experimental results show that the classification performance of Dual-channel CNN is better than those of single channel CNN and Naive Bayes on hotel review data.