Abstract: Traditional semi-supervised text classification methods were built based on the features of root, however, the common disadvantage of neglecting the importance of semantic features resulted in low precision of classification. In order to take account of the influence of semantic on classification, a text classification model comprehensively making use of dual features of concept and root was brought forward. Under the framework of cooperative training, this algorithm considered WordNet as ontology library and built double classifiers based on both concept and root for cooperative training. Through experiments, we analyzed the accuracy rate and recall rate of new classification model, and the results showed the promotions of both accuracy rate and recall rate in new model comparing with old model. It indicates that the new algorithm based on cooperation of dual features of concept and root is more effective than the old algorithm.


Key words:  , semi-supervised； semantic； dual feature； cooperative training