POI Recommendation Based On Spatial Temporal Relation Graph Embedding And Deep Learning

In recent years,with the widespread use of location-based social network(Location Based Social Network,LBSN)mobile applications,huge amounts of social network data with geographic information have been generated.For these data,interest point recommended(Point of Interest,POI)rapid development,as a typical location of social network services,by mining the user signin data activity preferences,life patterns and other information effectively model the user’s signin behavior mode,to provide users with accurate service and personalized recommendation,improve the user’s life experience.However,existing point of interest recommendation methods have not fully mined the spatio-temporal relationships and long-term dependencies in the user sign-in sequence,and the calculation of loop structure depends on the hidden state of the entire time series,making it difficult to fully capture the local information in the user sign-in sequence.In view of the above problems,starting from the POI recommendation task,we take the user checkin data in LBSN,and conduct the following research contents by constructing the space-temporal relationship diagram and LSTM-CNN model:(1)Analyze the user’s check-in behavior.Based on two open source datasets in LBSN,microblog(Micro-blog)and Foursquare,the behavior analysis of user check-in data sets from two aspects of time and geographical influence,respectively.Through a series of data preprocessing,statistical methods and probability analysis methods,the rules of user check-in and behavior trajectory patterns,so as to better establish the personalized POI position prediction model and provide data support for the POI location recommendation method proposed in this paper.(2)POI recommendation method for embedding spatiotemporal relationships with geographic influence.Geographic distance factors were added based on the introduction of the time interval,and the combination of temporal interval and geographical distance were used as weight factors to establish the spatial and temporal diagram between POI-POI.Then,the LINE embedding model is used to embed the space-temporal graph into the low-dimensional space,obtain the user continuous sign-in feature matrix by stitching the user history sign-in POI sequence vector,and input the feature matrix into the LSTM algorithm,so as to model the user continuous sign-in behavior and generate the final recommendation results.(3)Proposed a POI recommendation method based on long and short-term memory network and convolutional neural network.This method combines the advantages of CNN and LSTM,sends the original feature matrix to the LSTM layer to obtain the overall spatiotemporal and context information in the sequence,and then inputs the output of the LSTM layer to the CNN layer.After convolution and pooling,the features are connected to obtain the local information in the user sign-in sequence.Finally,the Soft Max activation function is sent to the fully connected neural network for classification to generate the final prediction results.(4)This paper conducts experimental verification and comparative analysis of the proposed method based on two data sets,Micro-blog and Foursquare.Experimental results show that the proposed POI recommendation method performs better and has high accuracy and effectiveness.This paper has 21 figures,10 tables,and 104 references.