A Label Inference Method Based on Maximal Entropy Random Walk over Graphs

With the rapid development of Internet, graphs have been widely used to model the complex relationships among various entities in real world. However, the labels on the graphs are always incomplete. The accurate label inference is required for many real applications such as personalized service and product recommendation. In this paper, we propose a novel label inference method based on maximal entropy random walk. The main idea is that a small number of vertices in graphs propagate their labels to other unlabeled vertices in a way of random walk with the maximal entropy guidance. We give the algorithm and analyze the time and space complexities. We confirm the effectiveness of our algorithm through conducting experiments on real datasets.