A Modified Complement Naive Bayes for ChineseWeb Page Classification

摘要

The naive Bayes classifier is effective in teit classification, however, it suffers from the problem of under studied bias effect when dealing with skewed data. A complement naive Bayes classifier is raised to mitigate the effect. Nevertheless, it shows disappointing results on skewed web page data. Focusing on the poor performance of complement naive Bayes algorithm on skewed data set, this paper presents a modified complement naive Bayes algorithm by using a superior estimation for the prior class probability. Comprehensive experiments show that the modified complement naive Bayes algorithm exhibits excellent robustness to skewed data and achieves higher precision than any other naive Bayes algorithm. Furthermore, regarding the difference between web page classification and text classification, in this paper, a title weighted vector space model is presented and the effect of title weighted factor on classifier's precision is analyzed. Experimental results show that the precision is improved by 5% on average by using title weighted vector space model.