Classification in Networked Data with Heterophily

Abstract

In the real world, a large amount of data can be described by networks using relations between data. The data described by networks can be called networked data. Classification is one of the main tasks in analyzing networked data. Most of the previous methods find the class of the unlabeled node using the classes of its neighbor nodes. However, in the networks with heterophily, most of connected nodes belong to different classes. It is hard to get the correct class using the classes of neighbor nodes, so the previous methods have a low level of performance in the networks with heterophily. In this paper, a probabilistic method is proposed to address this problem. Firstly, the class propagating distribution of the node is proposed to describe the probabilities that its neighbor nodes belong to each class. After that, the class propagating distributions of neighbor nodes are used to calculate the class of the unlabeled node. At last, a classification algorithm based on class propagating distribution is presented in the form of matrix operations. In empirical study, we apply the proposed algorithm to the real-world datasets, compared with some other algorithms. The experimental results show that the proposed algorithm performs better when the networks are of heterophily.