Research On Algorithms For Identifying Users Across Multiple Online Social Networks

In recent years, diversiform online social networks (OSNs), Facebook, Twitter and Linkedln for example, keep growing at a phenomenal rate and have become an essential part of people’s online life. However, each social network is generally isolated and noncommunicating. Fragmentation of users into numerous social networks makes the mining of information about individuals difficult and becomes the major hindrances to gain a complete Internet users graph.User Identity Resolution problem (UIR) is to identify the real users who had multiple virtual profiles across multiple OSNs. Recently, there are still some problems about UIR:One hand, most of Most of the existing Identity Resolution methods are based only on attribute matching (simply using string similarity between attributes and combining such similarities in some ways). On the other hand, the relationships of non-friends are not fully utilized.In this paper, we focus on solving the local UIR problem. We investigate the problem of UIR problem over user-seed-centric ego-network and represent the mathematical model and algorithmic analysis.Based on taking advantages of profile attributes and linkage information including friends and non-friends relationship, we propose a Ranking-based Cross-Matching (RCM) algorithm for identifying users across multiple OSNs. Identity resolution consists of three processes:identity selecting, identity matching and result pruning. In the RCM algorithm, we first chose the current highest scorer for detecting candidate matching users using Profile Attributes Similarity (PAS) and User Surrounding Score (USS). Then we define the User Matching Score (UMS), which combines PAS with network structures, to determine matched users for the candidate ones. Thereafter, we utilize a novel cross-matching process inspired by Stable Marriage Problem (SMP) to further improve the matching accuracy. Finally, through a simple pruning process, we take first matching users as our final result. In this paper, experiments on Twitter and Facebook demonstrate that our method significantly improves the matching performance and outperforms the state-of-the-art algorithms.