Context-aware POI Recommendation Based On Matrix Factorization

With the popularity of mobile devices,Location-based Social Networks(LBSNs)have been widely used and become a new form of social media in recent years.LBSNs can record rich context information,such as social networks,geographical information,POI category information,etc.,which provides a great opportunity to build personalized POI(Point-of-Interest)recommender systems.However,it faces huge challenges to model the impact of the context information and combine different kinds of information effectively in the POI recommendation.In addition,the sparsity of the check-in data poses a severe challenge for existing POI recommendation methods.Therefore,In this paper,we propose a context-aware recommendation framework based on matrix factorization.Specifically,we try to consider several aspects to model user's check-in behaviors.Firstly,we use users' check-in data(U)to model the users' own preferences,and consider the influence of POI category information(C)on the users' own preferences;Secondly,the POIs geographic information(G)has a very important impact on users' check-in behaviors.Users are more willing to visit POIs that are close to them and meet their preferences.In addition,we exploit users' social network to alleviate the data sparsity problem.Finally,we propose a unified matrix factorization model UCGSMF to effectively combine different context information,which has good scalability and relatively low time complexity.In addition,considering the local property(L)of the check-in data,we further propose a local matrix factorization model LUCGSMF,which can obtain a better low-rank approximation than the original matrix factorization by using local structure features.Main contributions of this paper are as follows:· Users' Check-in Behaviors Modeling Through the analysis of LBSN data,it is found that some context information,such as geographic information and users' social network,have a good effect on the POI recommendation,and the users' check-in behaviors are modeled from the aspects of users' preference,POI geographic information and social network,which effectively reduce the sparseness of check-in data.· Matrix Factorization Model with Fusion Context Information A unified matrix factorization model UCGSMF is proposed.The model use different fitting strategies for POIs that users have visited and have not visited respectively,which can more effectively characterize users' preference,better model users' check-in behaviors,and obtain a lower time complexity and good scalability.· Local Matrix Factorization Model with Fusion Context Information Considering the local properties of check-in data,we further propose a local matrix factorization model LUCGSMF,which can obtain better low-rank approximation than the original matrix factorization by using local structure features,so as to obtain higher recommendation performance and better distributed processing capability.