Incremental Bayesian classification for Chinese question sentences based on fuzzy feedback

Aiming at problems such as fixed training set and lacking of completed information in traditional Bayesian classification, incremental learning mechanism is introduced. Combining with the characteristics of question sentences in Chinese question answering system, Semi-Naive Bayesian model is used to construct classifier. In order to make prior distribution of samples lean to even distribution, samples whose posterior probability approach 1/n (n: number of classes) were selected and appended into training set. The approaching extent of samples is described by fuzzy set, the fuzzy distinguish result is returned to classifier, therefore a fuzzy feedback mechanism is formed. Incremental Semi-Naive Bayesian classifier based on fuzzy feedback is proposed in this paper. The results of experiments show that this Bayesian classification can improve the accuracy of classifier effectively.