A Point-of-Interest Recommendation Method Fusing User Preferences And Contextual Information

Location-Based Social Networks(LBSNs)are developing rapidly,forming a huge amount of Point-of-Interest(POI)check-in data.Such data contains rich information on user behavior preferences,which is conducive to accurate mining of user behavior patterns in order to provide better services to users.However,the lack of deep-level feature mining ability of LBSNs’ pointof-interest data leads to the dilemma that users cannot find the target point-of-interest quickly and accurately,forming a big data application.Personalized point-of-interest recommendation research has emerged in response to this,and gradually becomes one of the hot spots of current research.Point-of-interest recommendation aims to build models and predict user preferences based on the interaction data between users and points of interest,and finally recommend personalized points of interest that may be of interest to users.However,existing point-of-interest recommendation algorithms have certain shortcomings:(1)existing point-of-interest recommendation algorithms often ignore the implicit feedback features of check-in data,which limits the possibility of further improving recommendation accuracy;(2)existing point-ofinterest recommendation algorithms assume that contextual information is independent of each other and ignore the potential association between them,resulting in underutilization of deep information and difficulty in accurately capturing It is difficult to accurately capture the complex correlation between users and interest points.In this paper,we conduct an in-depth study to address the above problems,fully explore the effective information in the check-in data,explore user preferences in a multi-level and all-round way,and create a high-quality user recommendation service.The research work in this paper is as follows:(1)For the static check-in preference of users,considering the sparse check-in data and the implicit feedback characteristics,POI recommendation based on clustering and logistic matrix factorization is proposed,which uses logical matrix factorization technique to solve the problem of implicit feedback frequency and joint clustering algorithm to narrow the recommendation candidate set,avoiding the interference of invalid interest points and effectively improving the recommendation performance.(2)In response to the traditional interest point recommendation algorithm fails to fully explore the contextual information and take into account the intrinsic connection between contextual information,the intrinsic connection of contextual information for POI recommendation is proposed,which starts from four aspects: geographic location,social relationship,access sequence and time,fully considers the intrinsic connection between contexts and modeling fusion,so as to accurately extract the correlation between users and interest points.It effectively improves the recommendation performance.(3)By integrating the work in(1)and(2)above,we construct a recommendation framework for interest points that takes into account user preferences and contextual information,and realize the effective fusion of user static check-in preferences and contextual preferences.The framework takes into account the complex nonlinear relationships between models,improves the logical matrix factorization model by introducing contextual features to achieve effective fusion of high-level features,and proposes a POI recommendation fusing user preferences and contextual information based on matrix factorization,which not only combines the advantages of both algorithms,but also avoids the defects of a single algorithm and further improves the recommendation performance.(4)An example validation and prototype system development of the point-of-interest recommendation model is carried out based on open access data.In this paper,the proposed model is extensively experimented on two open access real check-in datasets and a variety of popular point-of-interest recommendation algorithms are compared.The experimental results show that the proposed method in this paper has higher performance.Meanwhile,a point-ofinterest recommendation system is designed for this purpose in order to verify the practicality of the algorithm.