Domain-Invariant Latent Representation Discovers Roles.

Discovering the roles of nodes in a network is important for solving various social issues. Role discovery aims to infer nodes’ roles from a network structure, and it has received considerable attention recently. The conventional methods of role discovery mainly use unsupervised learning, but due to the lack of information, it is difficult to discover the roles we want or to ascertain the results. In this paper, we attempt to improve accuracy through using supervised information. Specifically, we adopt transfer learning using adversarial learning. As a result of computational experiments, we show that the proposed model discovers a node’s role more effectively than do the conventional methods. Furthermore, we found that domain-invariant features lead to higher accuracy, the proposed method discovers roles better even with different network sizes, and the proposed method works well even if the networks have nodes of various structures.