Skip to Main Content
Collective classification, which is represented to classify unobserved nodes simultaneously in networked data is becoming an important research area with applications in several domains, such as the classification of documents, image processing. Most algorithms are based on the hypothesis that nearby nodes tend to have the same label. However, there are many networks that do not necessarily satisfy this hypothesis. In this paper, we present a new method based on random walk and link pattern of the network. It adopts the pseudoinverse laplacian matrix of the graph as similarity measure to identify nearby nodes and assigns an initial label for each unlabeled node, then iteratively update the label of unlabeled nodes based on the link pattern. The experimental results on two real world datasets demonstrate that the proposed method outperforms the other state-of-art approaches for this problem.