Tri-Training Based Learning from Positive and Unlabeled Data

This paper studies the problem of learning text classifier using positive and unlabeled examples with tri-training algorithm, which has been brought forward for semi-supervised learning. The key feature is that there are no negative examples. This paper proposed a new tri-training algorithm for the LPU problem that combines the step 1 of the three LPU algorithms to extract a reliable negative examples set, consequently to build an initial classifier for the tri-training and replace the bootstrap sampling procedure that has not been thought as a good method, and then iteratively use the three SVM classifiers until they convergence. Experiments on the popular Reuter21578 collection show the effectiveness of our proposed technique.