Research On POI And User Recommendation Under Location-based Social Network

With the vigorous development of mobile Internet technology, location acquisition is becoming more and more convenient. Position localization techniques have fundamentally enhanced social networking services, which lead to a new social network generation. We refer to these social networks as location-based social network(LBSN). In recent years, with the rapid development of location based social network, the data contained in it has encountered with explosive growth. To apply the recommendation technology in location-based social network has gradually got the attention of scholars, and quickly became a hot research area.We mainly focus on POI recommendation and user recommendation in this paper. By making full use of the temporal, spatial, social and historical information provided by LBSN, we propose two targeted technical solutions to solve the two recommendations. The main research works are as follows:(1) We analyzed the related works of recommendation technologies in location-based social network, and discussed the ideas of typical algorithms. Furthermore, we pointed out the advantages and disadvantages of previous works. These works provide a theoretical basis for our researches.(2) Researches have found that human’s behavior will affect the quality of POI recommendation, so we deeply mined the temporal, historical, social and spatial information, and found the influences on human’s behavior in this paper. Finally, we combined the user’s temporal, historical and spatial behavior together, and proposed a Multi-dimensional fusion approach for POI recommendation.(3) Since vendors have great requirements to know who are potential customers. Accordingly, we proposed an approach to model users’ preferences, and elaborated a spatial-preference based reverse k Ranks query approach to do user recommendation. As we know, users’ requirements are changed over time. So in order to catch users’ real-time requirements, we proposed a tree-based pruning algorithm to improve the spatial-preference based reverse k Ranks query approach, which greatly avoid some unnecessary computation and improve the speed of recommendation. Finally, we use the tree-based pruning algorithm as our core approach, and proposed an improved spatial-preference based reverse k Ranks approach for user recommendation.In this paper, we acquired several datasets from real LBSN services which are Foursquare, Gowalla, and Brightkite. And we use these datasets to compare our proposed approaches with the state of the art approaches. Our proposed approaches show the best experimental precision and recall results under all datasets, this indicates the effectiveness our proposed algorithms.