In order to resolve the contradiction between the speed and the precision of transductive support vector machine, a semi-supervised vector machine algorithm based on information feedback is proposed. The algorithm uses the information of the number of last round, the number of reset, the number of unlabeled samples to adjust dynamically the number of labeled samples, and make a tradeoff between region labeling and pairwise tagging. While the progressive evaluating and dynamically adjusting, it can balance the contradiction between the marking speed and accuracy and reduces the transmission and accumulation of errors. The experimental results on AI data sets and UCI data sets show that the proposed algorithm can improve calculation speed on the premise of ensuring the accuracy of label precision.