Class-Attentive Diffusion Network for Semi-Supervised Classification

Recently, graph neural networks for semi-supervised classification have been widely studied. However, existing methods only use the information of limited neighbors and do not deal with inter-class connections in graphs. In this paper, we propose Adaptive aggregation Class-Attentive Diffusion (AdaCAD), a new scheme that adaptively aggregates nodes probably same class among K-hop neighbors. To end, first novel stochastic process, called (CAD), strengthens attention to intra-class attenuates nodes. contrast diffusion transition matrix determined solely by structure, CAD considers both node features structure design our class-attentive utilizes classifier. Then, further an adaptive update leverages different reflection ratios result each depending on local class-context. As main advantage, AdaCAD alleviates problem undesired mixing caused discrepancies between labels topology. Built AdaCAD, construct simple model Network (CAD-Net). Extensive experiments seven benchmark datasets consistently demonstrate efficacy proposed method CAD-Net significantly outperforms state-of-the-art methods. Code is available at https://github.com/ljin0429/CAD-Net.