Abstract: Sentiment classification aims to find the users views on hot issues, but now the format of the short texts on the Internet is not normative, the effect of traditional sentiment classification method is not ideal. Facing the information of the short texts on the Internet of big data, this paper puts forward a deep convolution neural network (CNNs) model of the short text on the Internet. First it uses the Skip-gram in the Word2vec training model as the feature vector, then further extracts feature vector into CNNs, finally trains the classification model of the depth convolution neural network. The experimental results show that, compared with classification methods of traditional machine learning, this method not only can effectively handle emotion classification of the short texts on the Internet, but also improves the accuracy of emotion classification significantly, the average increased by about 5%。


Key words: short texts on the Internet, sentiment classification, convolutional neural networks (CNNs), natural language processing, deep learning