An email filtering method based on active learning and TCM-EKNN
Chen Long · Journal of Chongqing University of Posts and Telecommunications · 2011
The email filtering methods using machine learning and text categorization can be effectively used to solve spam.But these methods have problems which are large data in training set and characteristic vector of samples and cause curse of dimensionality and complex computation.We propose an email filtering method based on transductive confidence machines and evidence theory based K-nearest neighbors(TCM-EKNN).Moreover,the authors adopt active learning method to select the most qualified data for training classifier and make email filtering effective.Experimental results demonstrate that EKNN has good effect of filtering compared with the traditional methods and prove its validity;using active learning under the conditions of ensuring traditional methods high precision,the method could greatly reduce the number of samples,effectively improve the capability of the filter and has better performance according to the appraisal standard.