A Novel Naive Bayesian Text Classifier
Ding Wang, Songnian Yu, Qianfeng Wang, Jiaqi Yu, Qiang Guo · 2008
The naive Bayesian (NB) classifier is one of the simple but most efficient and stable classification methods. The great efficiency of NB is mainly because of the conditionally independence assumption among the attributes, which is problematic in practice especially while the attributes are strongly correlated. In this paper, we propose a novel NB text classifier, package and combined naive Bayesian text classifier (PC-NB) that relaxes the independence assumption. The main aim of PC-NB is to make naive Bayesian classifier be more accurate without efficiency reduction. A set of experiments were performed and the results of the analysis and experiment indicate that the proposed classifier is more accurate and powerful while the attributes of an instance are strongly correlated.