Research On Emotion Cause Analyzing Method Based On Microblog Posts

With the development of social network, especially for Sina micro-blogs, more and more people present and share their emotions and opinions by using the platform. Emotion analysis and causes extraction are important parts of emotional cognitive analysis, and they are also the key research tasks among the cross domains of natural language processing, text mining and psychology. They have important scientific and application values both on exploring the hidden sentiment and the corresponding cause events behind the natural languages.This thesis combines the relevant knowledge among the fields of computer science, cognitive linguistics and natural language processing together and presents an approach on emotion cause components detection for Chinese micro-blogs. It describes the conditions that trigger bloggers’ emotions in the progress of cognitive evaluation and also extracts the sub-events in the posts and then marks them. Then, it identifies the emotion causes in the posts and finds out the corresponding cause events. Among them, emotion analysis and causes extraction are the basis, and the calculation of cause component proportion can mine the corresponding relation between the emotion and the cause components. The main contributions are shown as follows:1) This thesis constructs the emotion model, starts from the external and the internal events extraction. On the basis of the technology of named entity recognition, dependency parsing analysis and semantic role labeling etc., this thesis does the extraction on emotion cause events from the aspects of the results of events, actions of agents and aspects of objects.2) This thesis begins with the emotional intensity of cause events by constructing the emotional lexicon and identifying different language features(e.g., emoticon feature, degree adverbs feature, negative words feature, punctuations feature and conjunctions feature), and does the calculation on proportions of different cause components based on the Bayesian probability model.3) This thesis presents a novel method on emotion classification based on language features. It combines the POS feature, syntactic structure feature and emotional words feature in context, to do the emotion classification using the machine learning methods.The experimental results validate the value of the method of emotion cause extraction in scientific research and social application, for example, the application area of crisis management. In addition, mining the relations between the emotions and the corresponding causes is important on product design. In the end, this thesis also presents the existing problems and the further research works.