Learning on heterogeneous graphs using high-order relations
This addresses the challenge of handling heterogeneity in graph learning for tasks like vertex classification, offering a more robust alternative to meta-path-based methods.
The paper tackles the problem of learning on heterogeneous graphs without relying on meta-paths, which can be sensitive to suboptimal choices, by decomposing the graph into homogeneous relation-type graphs and combining them with attention mechanisms. The result is a model that outperforms state-of-the-art baselines in vertex classification on three datasets.
A heterogeneous graph consists of different vertices and edges types. Learning on heterogeneous graphs typically employs meta-paths to deal with the heterogeneity by reducing the graph to a homogeneous network, guide random walks or capture semantics. These methods are however sensitive to the choice of meta-paths, with suboptimal paths leading to poor performance. In this paper, we propose an approach for learning on heterogeneous graphs without using meta-paths. Specifically, we decompose a heterogeneous graph into different homogeneous relation-type graphs, which are then combined to create higher-order relation-type representations. These representations preserve the heterogeneity of edges and retain their edge directions while capturing the interaction of different vertex types multiple hops apart. This is then complemented with attention mechanisms to distinguish the importance of the relation-type based neighbors and the relation-types themselves. Experiments demonstrate that our model generally outperforms other state-of-the-art baselines in the vertex classification task on three commonly studied heterogeneous graph datasets.