Research On User Identity Linkage Across Social Networks

In order to link virtual identities of the same natural person in different social networks,the user identity linkage technology across social networks has become a research hotspot.For the field of public security,it is important to link user identity across social networks and form holographic profile of user.Especially for the overseas or niche social networks,the public security departments don't have access to the basic data of users.This technology is very helpful to the user identity verification and intelligence expansion.Therefore,the research of this technology is of great theoretical significance and practical value.The main contributions are summarized as follows.(1)A algorithm of user identity linkage based on user profile is proposed,which combines classification method and weighting method.By means of serial processing,the hierarchical and cascading machine learning model(HCML)and the result correction algorithm based on information entropy weighting are fused.This algorithm makes full use of the information carried by user profile of different dimensions,and also takes into account the statistical distribution of the comprehensive profile similarity.So,the accuracy of user identity linkage is improved.Besides,a feature extraction method is designed based on username,location,personal description and avatar.The feature extraction method is universal to the user profile in most social networks and reflects the similarity between users from different platforms more comprehensively.(2)A algorithm of user identity linkage based on network representation learning is proposed.Facing heterogeneous information networks,the traditional network representation learning algorithm LINE is improved and a network representation algorithm named CSN_LINE is proposed which is based on cross-platform priori linkage.The algorithm can be used to extract feature of user relation,converting the network topology information into lowdimensional vectors.And the multi-layer perceptron classifier is used to form user identity linkage model based on user relation.This algorithm improves the accuracy of user identity linkage through effective feature extraction.(3)A user identity linkage across social networks prototype system is designed and implemented.The system integrates the identity linkage algorithm based on user profile and user relation into a technological linkage subsystem.Besides,a hard matching linkage subsystem based on users' sensitive basic information is designed.The system can provide user identity linkage service for different types of data.At present,as an achievement of national key research and development plan ‘cyberspace security' key special project(project number: 2016YFB0801100),the system has been tested and verified in a local public security department.User data from Douban and Sina weibo is crawled in the research.The number of users on Douban is about 2,000,000 and the number of users on Sina weibo is about 800,000.Data types include text,image and network topology.Two user identity linkage algorithms and prototype system are tested by real data.The F1 values of the two algorithms reach 0.8720 and 0.8563.Compared with the traditional method,the evaluation metrics have been greatly improved,which proves the effectiveness of the identity linkage algorithms.At the same time,the prototype system can also accurately link the user identity in the test,which proves the usability of the system.