A novel feature voting model for text classification

Along with the information explosion in the Internet era, the traditional classification methods, such as KNN (k-nearest neighbor), Naive Bayes (NB), encounter bottlenecks due to the endless stream of new words. In this paper, through comparing with the Rocchio and Bayesian algorithms, it has been found that centroid-based algorithms are insufficient for text classification. Therefore, a novel feature voting model is proposed, which gives rise to a bag-of-words based feature voting algorithm for text classification. This algorithm assigns categories for each document according to the ranking of weighted sum of feature values. Experimental results have shown the efficiency of the proposed method over the other state-of-the-art methods.