Abstract: In the semi-supervised learning process,the veracity of classification is affected because the classifier introduces the noise data to the training course.This paper proposes a kind of self-regulation and twice fake-iterative algorithm,whichstill uses the three classifier of tri-training algorithm.A small amount of manual work will be introduced under certain conditions to make the training process going on,thus,to avoid the difficulty in the classification of somelabels.The self-regulatory function is also used to reduce the noise data and noncontributory data to be added in the classification process.Mean while,the utilization and contribution of unlabeled samples is improved by using twicefake-iterative.The experiment and the results show that this algorithm can effectively improve the classification performance,and the utilization and contribution of unlabeled samples.The veracity of classification is improved obviously.

Key words: co-training, twice fake-iterative, self-regulation, contribution, manual work