CAMLP: Confidence-Aware Modulated Label Propagation
Yuto Yamaguchi, Christos Faloutsos, Hiroyuki Kitagawa · 2016
How can we tell if Alice is a talkative person or a silent person? In this paper, we focus on the node classification problem on networked data such as social networks and the web. There are two open challenges with this problem: (1) we want to handle various kinds of label correlations in real-world networks such as homophily (i.e., love of the same) and heterophily (i.e., love of the different), and (2) we want to exploit the confidence of the inference results to enhance the accuracy. There is no algorithm that solves these two challenges at the same time. We tackle with these two challenges by proposing CAMLP, a novel node classification algorithm. Our contributions are three-fold: (a) Novel algorithm; our algorithm is confidence-aware and is applicable to both homophily and heterophily networks, (b) Theory; we give theoretical analyses of our algorithm, and (c) Practice; we perform extensive experiments on 5 different network datasets including homophily and heterophily networks. Our experiments show that the proposed algorithm improves the precision of major competitors not only on heterophily networks, but also on homophily networks.