Research And Implementation Of POI Recommendation Algorithm Based On Spatio-temporal Information And Social Network

Nowadays,as the tremendous developing of location-based social network services(LBSNs)around us,POI(Point-of-Interest)recommendation has become more and more popular.No matter to customer and chest,recommendation system can send what thet need fastly.It is also widely used in to B and to C fields.In this paper,we study the POI recommendation problems based on temporal information,geographic information and social network,the main structure of this paper is:.Firstly,we study the POI recommendation algorithm based on geographical information,and propose the ABPR model.In the model,we consider the different types of user's geographical information,dividing user's activity area by clustering user's check-in records.Meanwhile,our ABPR model optimized the ranking process with Bayessian Personalized Ranking(BPR).Experiments show that this model improves the accuracy of the recommendation.Secondly,we study the POI recommendation algorithm based on temporal information and geographic information,first propose the CTPR model.In the model,we study the long-type time feature and short-time feature,combine the two types of feature with self-attention.We also do some embedding between features with XGBoost and logistic regression.Experiments show that CTPR model is better than traditional and state-of-the-art(deep interest network of Alibaba).Thirdly,we study the POI recommendation algorithm with social information,we learn the attitude of users with CNN,then learn the hidden friend relationship with random walk.Finally,we use matrix factorization to learn the hidden variable of user and POI to improve the experimental result.Experiments show that SDMF is better than the model of classic and state-of-the-art.Finally,we summarize the advance and defect of our three model,then we make a preview of future works..