This paper treats Japanese emotion ontology building for realistic interactive CG characters. To realize realistic interactive CG characters, we have to consider their emotions because during text-based interactive communications between a human and a CG character, the facial expression of the CG character should change according to its emotion change. The emotion change should be taken based on emotion ontology. Therefore, the authors try to build emotion ontology by defining emotion words and their intensities. In this paper, the authors propose a new method to define Japanese emotion words from the knowledge on the web, i.e., BBS and Twitter and to calculate their intensities based on how many times certain emotional words appear in the documents. The built Japanese emotion ontology is represented by OWL (Web Ontology Language) and EmotionML (Emotion Markup Language). In this paper, the authors also clarify the usefulness of the emotion ontology by applying it to a facial animation system.