A Semi-supervised Text Classification Method Based on Incremental EM Algorithm
Xinghua Fan, Zhiyi Guo · 2010
In the standard EM-based semi-supervised text classification, the classification performance is not well when the initial labeled samples are a few. How to improve the performance is an important issue. In view of this, a semi-supervised method based on incremental EM algorithm is proposed. This method makes full use of the useful information of intermediate classifier. On the one hand, this method verifies the feasibility of division existed in unlabeled samples, and uses the division mechanism to enhance the reliability of new incremental samples by dividing the unlabeled samples scientifically; on the other hand, a feedback learning mechanism is proposed, and it is used to decrease the probability of adding misclassified samples. Experimental results show that the classification performance is improved in our method.