A Transfer Learning based framework for Link's Role Discovery

Abstract

This paper aims to provide a framework for link's role discovery by using supervised information. The framework includes graph transformation, representation learning, transfer learning and role assignment. We use Edge-dual graph to regard links as nodes, struc2vec to gain links representation, adversarial learning model to transfer the target domain to the source domain to assign the roles for the target network's links. We show our proposed framework with better accuracy compared with existed method by a series of experiments on adjusted barbell graphs. Future work includes wider applications on other topology networks and real-world networks as well, and improvement on accuracy with bigger difference between the source network and the target network by updating the components used in the proposed framework.