Research On Emotion Classification Based On Markov Logic Networks And Its Temporal Variations For Chinese Microblogs

With the rapid development of Web 2.0, social network media have attracted more and more people’s attentions. As an example, Sina Weibo has become a main platform for public opinion and communication in people’s lives and works. And people can freely express their ideas and moods. The research on emotion classification and its temporal variations for microblogs is not only an important research in the field of natural language processing and text mining, but also has an important academic value and application value for exploring the hidden emotion and its temporal variations behind text. This thesis adopts an emotion classification method which is combined with probability and first-order logic, and it analyzes the temporal variations of emotion combining with the time and event extraction. The main contributions are shown as follows:1) This thesis uses Markov logic networks to determine the Markov logic representations of emotion classification. On the basis of external lexicon resources, this thesis constructs emotion classification system of text. This thesis infers the emotion categories of the microblogs.2) On the basis of the natural language processing technology such as word segmentation, part-of-speech tagging, dependency syntactic paring, named entity recognition and semantic role labeling, this thesis uses parts of speech and trigger words to construct the rules of the time unit and the event extraction. This thesis recognizes time expressions and events and standardizes time expressions.3) This thesis analyzes the publishing laws and the changing emotions laws of Weibo users based on the information of the times and the events extraction. It combines the information of the events to analyze the temporal variations of emotion of Weibo users.The experimental results validate the feasibility and effectiveness of the proposed method. 1) The best precision of the method of classification can reach to at least 90%, and its performance is influenced by the size of the amount of data; 2) The precision of event information extraction can reach to at least 80%; 3) The changing emotions of Weibo users are related to the information of events. In the end, this thesis presents the existing problems and the further research works.