Transductive Support Vector Machine for Personal Inboxes Spam Categorization

A method based on transductive support vector machine for personalized spam filtering is proposed. Both labeled emails from the public available source and unlabeled emails in individual inbox are used as the input of the classifier. The problem of the generalizing the training data to the test data in SVM is solved. It provides a way to combine the ability of generalization and adaptation for the spam categorization. The model and parameter selection is stated in order to improve the performance of TSVM. The experiments show that the results of filtering with TSVM are better than the SVM.