Adapting Hidden Naive Bayes for Text Classification

Due to its simplicity, efficiency, and effectiveness, multinomial naive Bayes (MNB) has been widely used for text classification. As in (NB), assumption of the conditional independence features is often violated and, therefore, reduces classification performance. Of numerous approaches alleviating features, structure extension attracted less attention from researchers. To best our knowledge, only structure-extended MNB (SEMNB) proposed so far. SEMNB averages all weighted super-parent one-dependence estimators; it an ensemble learning model. In this paper, we propose a single model called hidden (HMNB) by adapting well-known NB (HNB). HMNB creates parent each feature, which synthesizes other qualified features’ influences. For learn, simple but effective algorithm without incurring high-computational-complexity structure-learning process. Our improved idea can also be improve complement (CNB) one-versus-all-but-one (OVA), resulting models are simply denoted as HCNB HOVA, respectively. The extensive experiments on eleven benchmark datasets validate effectiveness HMNB, HCNB, HOVA.