Point Of Interest Recommendation Method Based On Location-based Social Network

The rapid development of big data technology and mobile intelligent devices has spawned Location-Based Social Network(LBSN).In order to understand user behavior patterns and improve the accuracy of location-based services,Point-Of-Interest(POI)recommendation has become an important research task,the purpose of interest point recommendation is to mine the sign in behavior characteristics of users and predict the interest points that users may visit,so as to save decision-making time for users and create service value for location related businesses.Aiming at the problems of weak semantics,high sparsity and cold start of user check-in data in LBSN,more and more studies combine contextual information such as geography,time,text and social connections to improve the effect of point of interest recommendation.However,the rich context information also brings great challenges to the point of interest recommendation.The current point of interest recommendation methods have some problems,such as low utilization of context information,difficult to balance the richness of context information and computational complexity.In view of the above problems,taking into account the main factors affecting the three dimensions of user,time and place of check-in behavior,this thesis studies the eigenvalue calculation and construction of tensor model integrating multiple context information.On this basis,a tensor decomposition interest point recommendation method integrating context is proposed,which makes full use of the rich context information of LBSN to improve the accuracy of interest point recommendation based on Location-Based Social Networks.The main contents of this thesis are as follows:(1)User clustering method based on social relationship.Aiming at the problem of low utilization of context information in traditional interest point recommendation methods,this thesis establishes a user activity model and user similarity model for social relationship measurement,and proposes a user clustering method based on social relationship.Through user clustering,the sign in behavior characteristics of similar users are mined to improve the utilization of LBSN context information and reduce the computational complexity.(2)Tensor decomposition interest point recommendation method based on context fusion.Aiming at the problems of weak semantics,high sparsity and cold start of LBSN check-in matrix,this thesis constructs a "user location time" three-dimensional tensor model to effectively integrate contextual information such as geographical location,check-in time,user social relations and popularity of interest points,improve the problems of low utilization of contextual information and weak semantics of check-in data,and obtain personalized recommendation list of interest points through tensor decomposition,Overcome the problem of sparse user check-in data,and improve the accuracy of LBSN point of interest recommendation.(3)Experiment and analysis of interest point recommendation.Based on the public data set of the well-known LBSN platform Brightkite,through the experimental comparison and analysis with two other typical interest point recommendation methods,the correctness and effectiveness of the user clustering method based on social relationship and the tensor decomposition interest point recommendation method based on context proposed in this thesis are verified.