Research Of Link Prediction In Location-based Social Networks

With the rapid development of social networks and the increasing popularity of mobile smart terminals,location-based social networks have gradually become an ideal platform for people to maintain social relationships and share location-based content information.However,the amount of data on the Internet has also shown an explosive growth,resulting in information overload.The link prediction research in LBSN can help users to discover the potential link relationships of users from massive data and recommend other "interesting" users or locations,which has important research significance and application value to grasp the evolutionary trends and development rules of LBSN structure and increase the loyalty of users to the LBSN.The main research contents of this thesis are as follows:The existing link prediction methods in LBSN mainly consider the single factors such as society,location,time and category.When the data sparseness is large,the performance of the algorithm is reduced.Aiming at the problem,a supernetwork link prediction method based on spatio-temporal relation is proposed in LBSN.First of all,aiming at the heterogeneity of network and the spatio-temporal relation among users in LBSN,the network was divided into "spatio-temporal,user,location,category" four layers supernetworks to combine multidimensional factors.Secondly,considering the impact of edge weights on the network,a four-layer weighted supernetwork model was built by mining user influence,implicit association,user preference and node degree information.Finally,the super edges and weighted super-edge structures were defined to predict the relation among users.The experimental results show that the proposed method can effectively predict the link relationship between users,alleviate the problem of sparseness and improve the performance of link prediction.The existing link prediction methods in LBSN treats the user link prediction and location link prediction in isolation and adopt a unified standard to model user's temporal and spatial characteristics.However,in practice,there is a certain internal relationship between user links and location links,both of which can promote each other,in addition,different individuals have different performances in time and space,so individuals need to be modeled.Aiming at the problem,a Cooperation based Personalized link Prediction(CPP)algorithm is proposed in LBSN.Firstly,according to the user's personalized characteristics,the kernel density estimation method was used to model the user's time and spatial.Secondly,the users were reorganized based on the interest groups,and the personalized user link prediction was performed through communities,friends,and check in relationships.Finally,based on the result of personalized user link prediction,a personalized link relationship between users and locations was predicted from the community by a random walk with restarts algorithm.The CPP algorithm improves the performance by the iteration of the user link prediction and the location link prediction.The experimental results show that the CPP algorithm has better prediction performance than the baseline algorithms.