Integrating Higher-Order Features for Structural Role Discovery

The role of node is able to denote its function and effect in the network, and can represent personal identity or status in real-world complex systems. It is defined on the local connective patterns and structural similarities. Compared to the community detection, the task of role discovery is independent to the node proximity which is generally related to the distance of density in the network. It is more likely to be determined by structural similarity, and the structural node representations have achieved great success in this field. Some existing methods focus on the local structural features to generate role-oriented node representations, but they consider too much on local structures and fail to learn multi-aspects representations of structural roles. More specifically, the local, global, and higher-order structures can denote different type of roles, and there are varying dependencies between them, leading to the difficulty to effectively integrate them. Thus, we propose a novel model HORD to integrate higher-order features into structural role discovery, aiming to learn multi-aspects structural node representations of roles. We leverage higher-order and local features and utilize the unified graph neural network (GNN) framework to organically combine them to generate structural node representations. We conduct extensive experiments on several real-world networks and the results demonstrate that our model is better than state-of-the art methods.