[1J4-J-3] Data mining: social mining

[1J4-J-3-02] Role discovery of the links based on the network structure

〇Shumpei Kikuta¹, Fujio Toriumi¹, Mao Nishiguchi¹ (1. The University of Tokyo)

Keywords:social network, role discovery, representation learning

This paper aims at the discovery of link roles in order to understand links on the network. This work presents a flexible, general framework including graph transformation, representation learning, role assignment, and sense-making. We use Edge-dual graph to regard links as nodes and struc2vec to embed links based on roles. We show our model successfully embed the similar links into the low 2-dimensional space on visualization task. Furthermore, we assign roles to links and conclude the structure is critical for a better understanding of links. Future work includes the automatic algorithm to decide the number of clusters and apply our method to real-world datasets.