MRSCN: A GNN-based Model for Mining Relationship Strength Changes Between Nodes in Dynamic Networks

Tianbao Wang, Yajun Yang, Hong Gao, Qinghua Hu

Abstract: The relationship strength between individuals in the network is an essential task in network analysis. However, existing measures of relationship strength are mostly artificially predefined, which can only reflect the relationship strength from a single perspective. To compensate for this, we propose a novel GNN-based model for Mining Relationship Strength Changes between Nodes in dynamic networks, named MRSCN, which learns the reasonable relationship strength from networks. To verify the effectiveness of our measure on the relationship strength change, we further propose a novel $$\epsilon $$ -drastic group model. We develop two group mining algorithms. We conduct extensive experiments on real-life dynamic networks to evaluate our models. The results demonstrate the effectiveness of the proposed MRSCN model and the drastic group mining method.