GraphInception: Convolutional Neural Networks for Collective Classification in Heterogeneous Information Networks

摘要: Collective classification has attracted considerable attention in the last decade, where labels within a group of instances are correlated and should be inferred collectively, instead independently. Conventional approaches on collective mainly focus exploiting simple relational features (such as count exists aggregators neighboring nodes). However, many real-world applications involve complex dependencies among instances, which obscure/hidden networks. To capture these classification, we need to go beyond extract deep between instances. In this paper, study problem Heterogeneous Information Networks (HINs), different types autocorrelations, from relations, Different conventional given explicitly by links network, autocorrelations HINs, existing hierarchical order. This is highly challenging due multiple nodes complexity features. study, proposed convolutional method, called GraphInception , learn HINs. And presented two versions models with inference styles. The methods can automatically generate hierarchy complexities. Extensive experiments four networks demonstrate that our approach improve performance considering