An Ensemble Learning and Category Indicator Based Text Categorizing Method

As it is well known that the motivation of ensemble learning is to boost a strong classifier with high generalization ability from a weak classifier.However, the achievement of generalization ability is often at great cost of complexity and intense computation.In this paper an ensemble learning and category indicator based categorizing method is proposed and Adaboost.MH based mechanism is developed to adaptively compute the category indicating function at every step.Then all individual category indicating functions are combined with weight and an approximation to the expected category indicating function is obtained.Based on the combined category indicating function, a classifier, which has low computational cost, flexibility in updating with new features and suitable for real-time applications has been obtained.Furthermore it is proved that the proposed method is equivalence to ensemble classifier and thereby it has high generalization ability.Experiments on the corpus of TanCorp-12 show that the proposed method can achieve good performance in text categorizing tasks and outperform many text categorizing methods.