A network splitting multi-role network embedding by quantum walk

Abstract

Network embedding is a technique that maps nodes of the network to low-dimensional vector spaces. Role, as an essential notion in social networks, holds significant importance in comprehending relationships between nodes and their attributes. Thus, role-based network embedding has emerged as the latest tool for analyzing networks and achieving network embedding applications. To address the problems of existing network embedding methods, such as 1) inefficient network wandering, 2) only neighborhood information is considered while ignoring role, and 3) multiple roles were not being addressed, this paper proposes the method called Network Splitting Multi-Role Network Embedding by Quantum walk (MRSQ). It firstly splits the multi-role nodes, and then obtains role structure by quantum walk utilizing its superposition ability. Next, the model fuses the neighborhood information of the nodes using a weighted characteristic function to obtain the feature of roles. Finally, the model designs variational auto-encoder to reduce the noise, which improves the quality of the role network embedding. The node classification experiments on real network datasets show that the F1 and AUC scores of MRSQ outperform the baseline method by up to 18.3%, thus reflecting the superiority and robustness of the model.