User Identification Cross Multiple Social Media Platform with Revised Input Output Network Embedding Framework

The rapid development of various online applications has brought the network into the era of socialization, which facilitate people's life. Plenty of people own multiple accounts on diverse social media network (SMN), like Facebook, Twitter and Instagram. In order to grasp the user's behavior characteristics comprehensively, it is necessary to identify the target user across SMNs. In this paper, we develop a model to matchup accounts from different SMNs, based on the Input-Output Network Embedding Algorithm. We firstly represent the nodes from SMNs into the embedding space with the follower-ship and followee-ship of users. Second, with the hard constraint from anchor links and soft constraint from the candidate anchor users across SMNs, we align nodes from two networks in the embedding space. We also use Non-negative Matrix Factorization (NMF) to generate empirical counterparts of the node representation required in Input Output Network Embedding (IONE) framework without using additional features of users besides their relationship in SMN. To reduce time complexity, we set learning rate automatically adjusted in stochastic gradient descent. Based on the data collected from real world SMNs, we proceed several experiments and the results show the improvement of the user identification efficiency the model provided.