FEUI: Fusion Embedding for User Identification across social networks

Recognizing identical users across different social networks remains challenging in recent years. Clearly, cross-platform user identification can play promising roles for many applications, such as user behavior prediction, public opinion analysis and e-commerce applications. Representation learning (RL) based methods have received more and more attention in recent years. However, most existing RL based methods only focus on the local structures (i.e., neighbors of vertices), and ignore label information and global structure patterns. Also, the current RL based methods tend to design the user identification and the embedding learning into two separate steps, which will neglect the complex correlations of different information sources. In this paper, we propose a novel approach, named as FEUI (Fusion Embedding for User Identification), by embedding the user-pair-oriented graph (UGP) through jointly integrating network structures, node attribute information and node labels to achieve robust embedding features and predict node labels simultaneously. The FEUI framework contains two modules, dual attribute embedding and joint embedding. These two modules leverage the strong representation ability of an extended auto-encoder and an one-input and two-outputs deep neural network to represent the complex correlations of different information sources. We evaluate our model on two social network datasets with collected user pairs. The experimental results show that the FEUI model can achieve better performance compared with the state-of-the-art approaches.