Research On Methods Of Personalized Affective Modeling

Emotions play an essential role in decision making, perception, learning and more. They influence the very mechanisms of rational thinking. Not only too much, but also too little emotion can impair decision making. The key object of Affective computing is building an intelligent computing system which can sensing, recognizing, and understanding human emotions. Emotion is changeable and influenced by many factors, so making computers intelligent is a topic full of challenge. To build an affective intelligent system, firstly a proper mathematical model should be established to describe human emotions. And how can the affective model be built? that is the research area of affective modelling.At present, affective modeling is at its infant, but there are still some stirring progresses occurred, many models were proposed, and these models can be divided into three categories, cognitive models, probability-based models, and other models. The three kinds of models have resolved some problems, but they still have some disadvantages repectively, because they involve many areas like computer science, cognitive science, psychology, behavior and physiology. A representative model of cognitive models is OCC model, according to this theory, emotions come from the appraisal of the current situation. The outcome of the appraisal depends upon how the current situation fits with one's goals and preferences. For no non-cognitive factors consideration, the disadvantage of OCC model is obvious. A representative model of probability-based models is HMM model, the model only simulated emotions from the perspective of probability, but no cognitive and non-cognitive factors consideration. The disadvantage bring on that perception results are all same in the same kind stimulus. Many improved HMM were proposed, but they can not able to fundamentally overcome the shortcoming yet. The representative models of others models is dimension-based model and layered model. Dimension-based model divided emotions into discrete categories, to simplify process of emotions, emotions just change in the small scope of emotional range in the model. For no non-cognitive factors consideration, the disadvantage of the model is obvious, too. The layered model (personality-moods- emotions), proposed by Kshirsagar S. is the first model that have established the relation between personality and emotion. For the vague treatment of mood, the model can not describe the complicated feelings of human, and lose its universal property in application.Therefore, to improve the existing affective models, this article constructs a novel intensity-based affective model that incorporates personality and some cognitive and non-cognitive factors. The model involved personality influence, emotion decay, emotion interactions, external affective stimulus, activation threshold and so on. The personalized model has great academic and practical significance on future work.To show the efficiency of the model, an emotional prediction system is designed and implemented on the basis of the model. After two contrasting group tests by 50 students, the experiment got good results. The results show that our model comply with the the nature law of human emotion and personality, and is superior than other current affective models.