Research On User Identification Algorithms Across Social Media Networks

With the rapid development of online social networks,more and more people have accounts in multiple social networks.However,the lack of user identity which can link people's accounts in various social networks makes it impossible to obtain complete user information.The user identification across social networks is proposed to solve this problem.User identification refers to identify multiple virtual accounts that a single user has in different social networks.Advancements in user identification could potentially impact various practical applications such as personage search,cross platform recommendation,user portrait and so on.Current studies on user identification mainly utilize three types of information: network topology information,user attribute information and user behavior information.Although some progress has been made in the research on three types of information,but there are still three aspect problems:(1)The current network topology information-based user identification algorithms only pay attention to the relationship between nodes in the network,and do not consider the non-friend relationship among the nodes in the network,which plays an important role in improving the accuracy of algorithm.(2)The current user attribute information-based user identification algorithms only pursuit the similarity of attribute values,ignoring the phenomenon that the same user has different attribute values at different time,lacking time-varying analysis of user attributes.(3)The current user trajectory information-based user identification algorithms treat each geographic location in user trajectory as a set of coordinate points,and the sequence features between geographic location in the user trajectory are ignored.Aimed at the above problems,this paper proposes three identification algorithms based on network topology,user attribute and user trajectory information in social networks.The main research contents are as follows:(1)A weighted hypergraph based user identification algorithm is proposed.First,the algorithm utilizes the weighted hypergraph to represent the friend relationship and non-friend relationship between the nodes in the network.Secondly,it combines the seed nodes whose identity is known before to represent the topology of the unmapped node,and improves the accuracy of the similarity measurement between the nodes.Finally,the cross matching algorithm is utilized to calculate the matching node pairs iteratively.Experimental results on DBLP networks and real social networks show that the proposed algorithm improves the precision and recall compared with the current algorithm.(2)An user identification algorithm based on attribute value transferring rule is proposed.Firstly,the algorithm analyzes the temporal attribute data,and obtains the transition probability between different pairs of attribute values.Then,when calculating the similarity between users' profiles,the transition probability of attribute values is combined with the traditional text similarity to achieve the goal of considering the time-varying of attribute values.The experimental results on the data set of multiple social networks show that the proposed algorithm can analyze the transfer rule of user attribute,and overcome the shortcoming of seeking only the similarity of the text of attribute values,and effectively improve the precision and F measure.(3)An user identification algorithm based on location sequence feature is proposed.First,the user trajectory is preprocessed,and the trajectory is divided according to a certain time granularity and distance scale,which makes the location feature of the user trajectory easy to extract.Secondly,the PV-DM model in the paragraph2 vec algorithm is introduced to extract the location sequence features in user trajectory,and the extracted feature is utilized to calculate the similarity between user trajectories.Then we match the trajectory from different sources using the trajectory similarity.Experiments on real spatiotemporal datasets show that the proposed method effectively improves the precision of trajectory based user identification algorithm.