Network Structure Embedding Method Based on Role Domain Feature

Network structure is formed by intricate connections between nodes, exploring and learning the network topological structural features has a profound impact in the field of network representation learning. Role refers to a collection of nodes with similar structural features in the network. So network representation learning that preserves node structure in a low-dimensional vector representation space is also known as role discovery, which focuses on partitioning the network into different sets of roles based on structural features. Although existing methods for network structure embedding have made some progress in role discovery tasks, most of them focus on the local structural features to generate node representations, resulting in the inability to learn multiaspect structural features of roles. Therefore, we propose a network structure embedding model URold, which uses role domain feature to enhance node structure representation capabilities and learn the proximity between roles. We conduct role discovery experiments on six real-world networks, and compare with eight state-of-the-art network structure embedding algorithms. The results show that our method URold achieves the best performance and demonstrates excellent role discovery ability.