The Sentiment And Sematic Analysis Of Text For Avatar

In recent years, vivid avatars are more and more popular. More and more beautiful avatars appear in movies, games, and on the Internet. How to interact with the avatar is an interesting topic and lots of researches are achieved by vision, speech or haptic devices. But these implementations are expensive. The way that avatar controlled by text costs less and is more widely available. You can add that on network instant communication tools like QQ, Fetion, Ali Wangwang, which makes the communication more interesting. The project of avatars controlled by text needs the natrual language processing technology and virtual technology.Our main objective is to design an individual avatar system. The system can change body movements and facial expressions according to the users' input text. This paper can be divided into three parts by its contents.First, we achieve the action semantic analysis of text. The system uses HowNet to calculate Chinese word similarity and uses WordNet to calculate English word similarity. The two dictionaries solve the problems of ambiguity and synonyms. Then we use the latent semantic analysis and the singular value decomposition of the matrix to get the action semantic keywords.Second, we achieve the sentiment classification. The imporved K-nearest neighbor (KNN) method is used. We use KNN to divide the text into six categories:anger, disgust, fear, joy, sadness and surprise. KNN first choose classified texts for training. It uses mutual information (MI) method and statistic method to select the text characteristics. When constructing the training text feature vectors, we improved term frequency-inverse document frequency (TFIDF) weight, combining with MI and statistic method, and increase the accuracy of sentiment classification.Finally, we can control the avatar using the results of the action semantic analysis and the sentiment classification. The avatar changes its body movements and facial expressions. We programmed a3D virtual chat system. In the chat room, the avatar fulfills its body changes.