An Ensemble Classification Algorithm for Imbalanced Text Data Streams

For unbalanced text data streams, an ensemble classification algorithm for unbalanced text data streams is proposed in this paper. Firstly, an improved resampling method is used to establish balanced data subsets; secondly, the topic model is used to perform topic modeling on the balanced data subsets to establish document-topic training subsets; finally, an ensemble classifier is constructed using the WE ensemble model. The algorithm sets the difference of the F-value of the neighboring data blocks to a certain threshold as the standard of updating the classifier. When the ensemble classifier is updated, the base classifier is retrained after the error positive instance is added to error set. Experimental results show that the proposed algorithm not only has good classification performance for the positive instances, but also has good classification performance for all instances. Therefore, the algorithm proposed in this paper is an effective classification algorithm for unbalanced text data streams.