Konum temelli sosyal ağlarda güven farkında konum önerisi

Users can share their location with other social network users through location-embedded information in LBSNs (Location-Based Social Network). LBSNs contain useful resources, such as user check-in activities, for building a personalized recommender system. Trust in social networks is another important concept that has been integrated into a recommendation system in various settings. In this thesis, we propose two novel techniques for location recommendation, TLoRW and SgWalk, to improve recommendation performance through integrated trust information. In both of the algorithms, the elements of LBSN and their relationships (user-user, user-location) are represented by using a graph model. For trust modeling, we develop a method to generate trust scores of LBSN users. With the developed method, the global trust score of a user is predicted with respect to the check-in history. The trust model is integrated into the LBSN graph model to be used within the proposed location recommendation algorithms. The first algorithm, TLoRW, generates location recommendations based on the user's current location by exploiting the friendships, experts, and trusted users traversing the region of user's spatial context through a random walk based algorithm. This region is constructed as the subgraph of the user according to the current location. In the second recommendation algorithm, SgWalk, we consider user subgraph as a heterogeneous information network and propose a novel HIN embedding technique. The location recommendation is generated by the proximity between users and locations based on their corresponding node embedding. SgWalk is differentiated from the previous node embedding techniques relying on meta-path or bi-partite graphs by utilizing the user subgraphs generated based on spatial context. By this way, it is aimed to capture the relationship between the entities with respect to the spatial context.The recommendation performance of TLoRW and SgWalk is analyzed through extensive experiments conducted on benchmark datasets by evaluating the accuracy in top-k location recommendations. The experiments reveal that trust information has a significant effect on improving the location recommendation performance. The performance evaluation results show a substantial improvement compared to baseline techniques and the state-of-the-art trust-aware recommendation and heterogeneous graph embedding techniques in the literature.