Homophily-oriented Heterogeneous Graph Rewiring

With the rapid development of the World Wide Web (WWW), heterogeneous graphs\n(HG) have explosive growth. Recently, heterogeneous graph neural network (HGNN)\nhas shown great potential in learning on HG. Current studies of HGNN mainly\nfocus on some HGs with strong homophily properties (nodes connected by\nmeta-path tend to have the same labels), while few discussions are made in\nthose that are less homophilous. Recently, there have been many works on\nhomogeneous graphs with heterophily. However, due to heterogeneity, it is\nnon-trivial to extend their approach to deal with HGs with heterophily. In this\nwork, based on empirical observations, we propose a meta-path-induced metric to\nmeasure the homophily degree of a HG. We also find that current HGNNs may have\ndegenerated performance when handling HGs with less homophilous properties.\nThus it is essential to increase the generalization ability of HGNNs on\nnon-homophilous HGs. To this end, we propose HDHGR, a homophily-oriented deep\nheterogeneous graph rewiring approach that modifies the HG structure to\nincrease the performance of HGNN. We theoretically verify HDHGR. In addition,\nexperiments on real-world HGs demonstrate the effectiveness of HDHGR, which\nbrings at most more than 10% relative gain.\n