The Prediction Of Public Sentiment Trend Based On Sentiment Commonsense Knowledge

With the development of internet, microblog becomes a popular public opinion platform for its real-time and concise. People can propagate messages, express opinions and share status on it. The sentiment analysis and prediction on it have become a hot research field. Microblog connects virtual world and real world. Analyzing the sentiment of microblog can help us master sentiment trend of current hot issues which happened in the real world. It has great significance for monitoring public opinion and even predicting recent sales of products.Microblog has the properties of context absence and feature sparseness. Sentiment analysis methods using semantic features and sentiment lexicon can’t recognize the implicit sentiment. So we construct a sentiment commonsense knowledgebase by combining with machine learning and artificial method in the perspective of cognitive and statistics. Then treat it as context to assist sentiment computation and predict the sentiment trend of events.Firstly, the sentiment commonsense knowledgebase consists of structured and unstructured commonsense knowledge. Affective schema, as a structured, is constructed by using pragmatic inference technology and pattern matching method. Binary sentiment commonsense knowledge, which is unstructured, is made by syntactic structure extracted by dependency parsing. The experiments on formal corpus suggest that they can improve the accuracy of sentiment analysis. Network idiom is an unstructured sentiment commonsense which implied network background. We employ mutual information, contextual entropy, and distributed representations of words (word2vec) to mining them and do sentiment analysis.Secondly, we present a double-layer classification based-on sentiment commonsense, which combine the rule matching and statistical classification methods. The experiments on microblog show that it has a better accuracy than previous methods. Sentiment trend prediction has been performed by using time series sentiment analysis which presents the sentiment ratio of different time pieces. The experiments show that sentiment trends are consistent with real world situations in comparison with Baidu Index curve.In conclusion, sentiment commonsense knowledge, as a universal semantic resource in the field of affective computing, can assist sentiment analysis and trend prediction effectively.