Research And Application On Sentiment Fine-Grained Classification Of Text For Chat System

Nowadays, WeChat, QQ, electronic commerce, online games and distance learning develop rapidly along with high-speed propulsion of information technology and urgent increasing of interpersonal communication. Contact among friends, colleagues, and relatives, transactions between buyers and sellers, communication among players, as well as exchanges between students and trainers, however, inseparable from the instant messenger more and more serious. Communication through instant messaging has become one of the most important parts of interpersonal communication and obtained more and more attention, that's why it evolved catch up with time and perfected better and better. This thesis combined natural language processing and virtual human together and introduced them into chat system to enhance usability and interesting of chat tool. The expressions and movements of virtual human can be controlled through text corresponded to that from chat system. Control virtual human through text covered natural language processing and virtual human. The major of this thesis involved controlof expression by virtual human through text, which means that virtual human can show corresponding expression by identifying emotion of text through sentiment classification, while control of action by virtual human has been completed.The focus of this thesis is research and design personalized virtual human system that can show different facial expressions based on text from user. We divided this thesis into three aspects in detail as following:Firstly, we analyze the text in Chinese of chat system. We present interface of common chat system illustrated by the case of QQ, and then, we research features and usage of emoticons in chat system and give statistics of similarities and differences of emoticons between group chat and private chat, append with distribution of length among text. We describe processing of emoticons during classification to terminate this part.Secondly, we research the handling of text in Chinese of chat system. The handing of text in Chinese includes processing of emoticons, building of emotion dictionaries, and processing of text. The specific of text in Chinese of chat system contains segmentation, removing stop words, text representation and feature extraction. We realized segmentation by ICTCLAS and removed stop words based on list of stop words construction by ourselves. The feature of extraction can be achieved by TF-IDF based on vector space model of text.Thirdly, we study sentiment classification of text in Chinese of chat system We discuss sentiment classification based on emoticons, based on emotional dictionary, based on Naive Bayesian and multi-combination sentiment classification in detail and test classification of the four methods. The results show that the efficiency and accuracy of multi-combination sentiment classification out perform other three algorithms.Finally, we apply multi-combination sentiment classification into virtual human chat system to control expression of virtual human. We find that, the high efficiency and accuracy of multi-combination sentiment classification on text in Chinese can meet the need of both real-time arid veracity for chat system, and enhance interesting of chat system at the same time. The multi-combination sentiment classification of text in Chinese worth research deeply and apply widely.