BSOGCN: Brain Storm Optimization Graph Convolutional Networks Based Heterogeneous Information Networks Embedding

Abstract

Recently, Graph Convolutional Networks (GCNs) have shown great potential in the field of graph embedding. They map the nodes of the graph into the low dimensional vectors by aggregating the neighbor nodes’ features information. However, most existing GCNs only focus on the homogeneous information networks instead of the heterogeneous information networks (HINs) with multiple types of nodes which are more common in the real world. Because the different types of neighbor nodes could have different impacts on the target nodes, it is difficult to manually design a proper neighbor nodes’ features information aggregating weights. To address this problem, we propose a novel HINs embedding algorithm based on the brain storm optimization (BSO) algorithm and the graph convolutional network (GCN), called BSOGCN, which utilizes BSO to optimize the neighbor nodes’ features information aggregating weights. It can be applied to the various HINs under various scenarios without any prior knowledge. The proposed method has been evaluated on both inductive and transductive node multi-class classification tasks on three real-world HINs datasets. The experimental results demonstrate that BSOGCN is competitive against other state-of-the-art methods.