Research On POI Deep Dynamic Recommendation Algorithm Based On Potential Friends Mining In LBSN

With the widespread application of mobile smart devices and location services,Location-Based Social Network(LBSN)has gradually penetrated into everyone's life,so personalized points of interest(POI)recommendation research is conducted for users It becomes more and more important.Recommending suitable POIs for users by learning their actual preferences for POIs can not only help users find their favorite POIs quickly,but also help businesses find more potential customers to improve economic benefits.In LBSN,it contains a large amount of user history check-in information and social information.How to choose appropriate relevant information and a reasonable integration strategy to improve the recommendation effect is the focus of current research.Most of the current researches improve the performance of POI recommendation by mining the geographic and social relationships between POI and users.Few researches consider the time characteristics of user check-in behavior.To this end,this paper deeply analyzes the impact of temporal and geographic factors on user check-in behavior,and combine these two influencing factors with deep learning technology to propose a temporal-geographic aware POI deep recommendation model(TG-DAE).First,use the Conditional Restricted Boltzmann Machine(CRBM)to model the temporal information in the check-in data to obtain the temporal correlation of the user's check-in;then construct a deep autoencoder based on CRBM to mine the deep features of the user check-in data;finally,in the deep model add geographic weight connections in the input and output layers to model geographic information between POIs,further enhance the recommendation effect.In addition,in view of the insufficient use of social relationship information by existing recommendation algorithms based on social factors,this paper studies the influence of different social relationships on user check-in behavior,and proposes a model learning method based on potential friend mining.Under this method,realize the above-mentioned TG-DAE model to obtain a POI deep recommendation model(TGS-DAE)that integrates the influence of multiple factors.Specifically,in order to expand the scope of social influence,this article uses the advanced hypergraph embedding technology LBSN2 vec to mine the user's potential friend relationship;after that,a social influence Regularization term containing explicit and potential friends is established and added to the previously proposed TG-DAE In the model,mining the check-in associations between friends,while improving the robustness of the model.Finally,the rationality and effectiveness of the model proposed in this paper are verified on two real datasets,and compared with related algorithms.Through the analysis of the experimental results in this article,it can be concluded that compared with several comparison algorithms,the proposed TGS-DAE model can achieve significant improvement in accuracy and recall indicators,and can effectively generate POI recommendations.The combination of temporal,geography,and social influence factors play an important role in improving the performance of the model.The TGS-DAE model can handle these influence factors well and ultimately improve recommendation performance.