| Popis: |
User identity linkage (UIL) refers to linking users’ assets and identities across social networks. With the rapid growth of social media in our day-to-day life, UIL’s importance has gone beyond being just a research topic, and it has become a necessary precondition for critical tasks like fraud detection.However, only a few models have been proposed to tackle UIL in different domains than social networks. This limitation becomes even more evident in academic federated identity management (FIM) domains. Service providers (SP) deal with restricted users’ data in these environments, and often data related to connections between entities and resources (i.e., network-based information) such as friends associations or the clients’ inclinations is not available.This research addresses the account linkage (AL) problem for organizations inside federated environments with limited or no access to users’ data. In the proposed model, we focus on analyzing users’ habits and behavior during login processes by utilizing a Variational Autoencoder’s (VAE) latent space. The learned structure in this space is used to derive related accounts owned by one user.To the best of our knowledge, the proposed model is the first approach attempting to solve the AL problem inside an academic FIM domain with its high requirements regarding data security and limited users’ information. Preliminary results show that the proposed model could achieve almost 90% accuracy in linking accounts possessed by one user. |
