Topic sentiment analysis using words embeddings dependency in edge social system

The topic sentiment analysis is really fundamental in detecting potential cyber threats and cyber attacks in edge social systems. We can detect potential cyber threats and cyber attacks by identifying the sentiment orientation and topics of public opinion information in the edge social system. Topic sentiment joint model is an extended model, which aims to deal with the problem of detecting sentiments and topics simultaneously from the online comment. Most existing topic sentiment joint models ignore the dependency among words so that they lose rich semantic information and the resulting distribution might be not very satisfactory. In this paper, we propose a novel topic sentiment joint model with word embeddings dependency based on recurrent neural network. The model introduces the dependency among word embedding and delivers topic information and sentiment information of words by a recurrent neural network. It fully extends the semantic information and redefines the topic sentiment-word distribution. Moreover, we obtain more accurate topic detection and sentiment analysis. Experimental results on online review data set show that the proposed model significantly improves the sentiment classification accuracy and achieved better topic detection compared with previous methods.