摘要：As information integration based on multiple modal has to problems like complexity calculation process and low classification accuracy towards network video classification algorithm, came up with a network video online semi-supervised classification algorithm based on multiple view co-training. According to extract the features in text view and visual view, to the feature vector in each view, uses graph as basic classifier and modeling, uses linear neighborhood belief propagation to make category labels propagation in each view, and gets category prediction outcomes in this view; in different views, uses co-training method to online extract unlabeled samples to expand the training set and to incrementally update basic classifier. To the integration of different model prediction outcomes, proposed an integration method aimed at category related. Finally made detailed experimental compare with support vector machine classification algorithm, the result showed, compared with support vector machine, the performance of learner increased greatly, more suitable for large-scaled network video online semi-supervised learning.
关键词：incremental online learning;text view;visual view;multiple model integration
