An enhanced EM method of semi-supervised classification based on Naive Bayesian

Semi-supervised learning (SSL) based on Naïve Bayesian and Expectation Maximization (EM) combines small limited numbers of labeled data with a large amount of unlabeled data to help train classifier and increase classification accuracy. With the aim of improving the efficiency problem of the basic EM algorithm, an enhanced EM method is proposed. Firstly, a feature selection function of strong category information is constructed to control the dimension of feature vector and preserve useful feature terms. Secondly, an intermediate classifier gradually transfers unlabeled documents of maximum posterior category probability to labeled collection during each iteration process of the EM algorithm. The iteration number of the enhanced EM is obviously less than the basic EM. Finally, experiments shows that the improved method obtains very effective performance in terms of macro average accuracy and algorithm efficiency.