Research On Temporal And Spatial Social Relationship Network Modeling And Recommendation Technology Based On Graph Model

With the rapid development of mobile Internet and positioning technology,users can easily access their location information through mobile devices,which also spawns many location-based social network(LBSN)platforms.Users can share their point of interest(POI)in daily life by checking-in on the platform,and the platform also introduces recommendation services to improve users' life experience,especially POI recommendation greatly facilitates the user's exploration of POI.An excellent recommendation service can analyze the user's historical behavior through modeling at any time,and can recommend the new POI that the user is most likely to be interested in at this moment.In this paper,in order to integrate the three main factors that affect the recommendation of POI in LBSN recommendation service: social friend relationship of users,geographical and spatial location of POI,and time information to improve the effect of the recommendation of POI,we propose an POI recommendation algorithm based on the graph model,and this POI recommendation is the recommendation of time-awareness.In order to reduce the overall complexity of the recommendation algorithm,we first use the clustering algorithm to cluster users on the user graph,in which the initial clustering center is determined by the most sparse subgraph algorithm to find the graph,and then through the iterative updating and merging clustering of the clustering center,we get the final user clustering results.Then,a large-scale heterogeneous network graph including six bipartite graphs(user-user,user-POI,user time,POI-POI,POI-user,POI-time)is established in each cluster of clustering results.The heterogeneous network graph is established by integrating the above three factors that affect the recommendation of POI,and then through the training and learning of LINE model algorithm,the vertex embedding vectors representing user,POI and time are obtained.In order to speed up the training,we further use the negative sampling and Alias sampling methods to optimize the LINE model algorithm.Then we use the three vertex embedding vectors of user,POI and time to calculate the prediction score of users' access to new POI,and select top-N POI to recommende to user according to the prediction score.At last,We simulated the algorithm with Gowalla dataset.The experimental results show that our recommendation algorithm is better than the other three recommendation algorithms in time-awareness,and achieves the expected experimental results.