Exploiting and Learning Human Temperaments for Customized Information Recommendation

Human temperaments have been recognized as a predominant factor in determining the activity patterns of human behavior. In our earlier study, temperament-based filtering method has been proposed in seek of combining concept learning and content-based filtering techniques to incorporate human temperament concept into the recommendation process of an information system. In this paper, we explain the design of a prototype multiagent system, which is developed, implemented, and experimentally tested by using a group of simulated users generated from the sample users to demonstrate the effectiveness of the temperament-based filtering method for selection and customization. The notion of human factors, particularly human temperaments, is explored and learned for the representation and segmentation of an information space. Furthermore, the learned temperament concept is employed for the interpretation and measurement of the relevance extent for the classification and recommendation of the information units. The results of our preliminary experiment indicate that the accuracy of recommendation using the temperament-based filtering method exceeds that in the content-based filtering method.