A convolutional neural network method for Chinese document sentiment analyzing

摘要

Text sentiment analysis is composed of text feature representation and classification method, especially the former will direct effects performance of sentiment analysis. Bag-of-word, n-gram model, and word embedding are all the method translating text to computable data, so researchers in the field of neural language processing always take advantage of these methods to obtain text representation. Then, utilize classification method to analysis text sentiment polarity, such as SVM, decision tree, hierarchical classification, logistic regression and so on. In this paper, authors propose a method of extracting text features with deep learning, and use logistic regression to analysis the sentiment polarity of Chinese document. The result of experiment proves word2vec and convolution neural network effectively improve performance of prediction model.