Abstract: Based on convolutional neural network, this paper build topic model on sentiment classification of public opinion users in social networks driven by big data. Weibo and Twitter users’ content data are extracted respectively through the crawler on hot topic of “Immigration”. Word2Vector is used to train Chinese word and GloVe to train English word vector. NLPIR and BosonNLP tools are used to participle and corpus of users’ sentiment is built based on the “Immigration” topic. Finally, sentiment classification is trained by CNN neural network. The results were compared with TimeLstm and SVM to verify the superiority of CNN classification. Results show that the proposed model can achieve effective text classification in a multilanguage environment including Chinese and English. The accuracy of model can be optimized by properly setting the activation functions and related parameters. It shows the model proposed is superior to traditional machine learning. In terms of text classification on the topic of “Immigration”, the classification effect of CNN is better than that of TimeLstm model. The study of this paper provides a preliminary research framework for the visualization analysis of the sentiment knowledge map of public opinion users in crosslanguage social networks.

Key words:

Convolutional neural network, Social networks, Sentiment classification; Topic model, Public opinim monitoring, User research