The Text Affective Computing Based On Semantic Resource

Affective computing has received more and more interests in the field of artificial intelligence, and its goals are that the computer holds emotions. Just like human beings, the computer could communicate amiably and naturally. With the development of Internet, textual information becomes more and more, and becomes the richest interactive resources. However, few researches have focused on affective text analysis. So this paper constructs semantic resources firstly, including affective lexicon ontology and affective corpus, and based on them, two methods of text affective computing are presented.On the procedure of constructing semantic resources, firstly, the paper analyzes the status of the emotional classification, and classification system is determined. And then the framework of ontology is filled by the combination of manual classification and acquiring the intensity automatically. About affective corpus, this paper discusses several basic questions which include the tagging criterion, tagging set, tagging tools and quality monitoring. Now, there are about 40,000 sentences in the corpus, and based on them, statistical data about emotional distribution and rules of emotional transference are available, and characters and applications of corpus are analyzed.In recognition of utilizing semantic features, the lexical emotion information and semantic features are appended into Condition Random Fields, and the emotional chain of a text document is generated. However, another method bases on Lazarus's cognitive evaluation theory and cognitive context of the cognitive pragmatic, and this new text affective model utilizes affective schema for improving precision of affective recognition. Finally the affective structure of a text document is evaluated by two different methods which are single sentence evaluation and joint sentence evaluation.The method of using CRFs is more effective on affective consistency, while the method based on cognition has higher precision by experimental results. And two methods explore the text affective computing from different prospects and have theoretical value.