Role-oriented network embedding via anonymous walks

Article type: Research Article

Authors: Qiu, Yutan | Zhou, Qing^{; *}

Affiliations: College of Computer Science, Chongqing University, Chongqing, China

Correspondence: [*] Corresponding author. Qing Zhou, College of Computer Science, Chongqing University, Chongqing 400044, China. E-mail: [email protected].

Abstract: Role-oriented network embedding aims to preserve the structural similarity of nodes so that nodes with the same role stay close to each other in the embedding space. Role-oriented network embeddings have wide applications such as electronic business and scientific discovery. Anonymous walk (AW) has a powerful ability to capture structural information of nodes, but at present, there are few role-oriented network embedding methods based on AW. Our main contribution is the proposal of a new framework named REAW, which can generate the role-oriented embeddings of nodes based on anonymous walks. We first partition a number of anonymous walks starting from a node into the representative set and the non-representative set. Then, we leverage contrastive learning techniques to learn AW embeddings. We integrate the learned AW embeddings with AW’s empirical distribution to obtain the structural feature of the node, and finally we generate the node’s embedding through message passing operations. Extensive experiments on real network datasets demonstrate the effectiveness of our framework in capturing the role of nodes.

Keywords: Network embedding, network structure, role-oriented, anonymous walk

DOI: 10.3233/JIFS-222712

Journal: Journal of Intelligent & Fuzzy Systems, vol. 44, no. 2, pp. 2729-2739, 2023