Belief learning in certainty factor model and its application to text categorization

This paper describes a method of belief learning in certainty factor model and applied to a task of text categorization. The method uses a multiplicative update algorithm to perform the belief learning of rules and predicts by a certainty (plausible) inference mechanism. The key difference between the proposed method and Sleeping-experts algorithms is that the method uses the rule's combination functions instead of the weighted combination of the predictions. When applied to a text categorization task, this method can easily integrates user defined IF-THEN rules due to being compatible with expert system's rules combination framework. The initial experiments show that the performance of this method is comparable to Sleeping-experts method. Moreover, it has better time and space efficiency than Sleeping-experts method.