Abstract:

Web spam detection is of great significance, and there only exists a small number of labeled pages. Thus, the semi-supervised co-training was used to detect the Web spam pages. The page features were divided into two views, the content view and the link view. First, the independent components of each view were extracted by  the independent component analysis, and then the co-training was used to detect the label of each Web page. Experimental results showed that this method could effectively improve the recognition accuracy of Web spam. The results also verified that two respective independent component analyses of each view were more effective than the other methods.

Key words: independent component analysis, co-training, multi-view classification, Web spam detection