Point-of-Interest Recommendation Based On Heterogeneous Information Network

In recent years,with the rapid development and popularization of Internet technologies and mobile devices,Location-Based Social Network(LBSN)services become increasingly popular.Users can explore their preferred locations,such as library and restaurant,through the “check-in” behavior provided by the LBSN services.The personalized Point-of-Interest(POI)recommendation service is designed to assist users to find their preferred places and improve the LBSN service experience by mining user preferences through check-in data.However,existing POI recommender systems face problems such as data sparsity,affecting by rich contexts and lacking negative feedback of user behavior.About the problems mentioned above,this paper deeply analyzes the influence factors of user behavior and proposes corresponding recommendation algorithms for two different application scenarios.The main contributions of this paper are as follows:(1)In order to accurately capture the user behavioral features,this paper employs Heterogeneous Information Network(HIN)to model the complex relationships in LBSN and designs meta-paths to represent different user behavioral semantics.In addition,this paper improves the accuracy and interpretability of the POI recommendation system by analyzing the contextual influence factors of users' checkin behaviors.(2)Aiming at the POI recommendation in general scenario,this paper proposes a POI recommendation algorithm based on user behavioral semantics.Firstly,user-POI semantic correlativity matrices are constructed based on a counting-based correlativity metric.Secondly,a weighted matrix factorization model considering the influence of geographical distance is designed to obtain semantic preference through the user-POI semantic correlativity matrices generated by multiple meta-paths.Finally,a rankingbased fusion method is introduced for the problem of implicit feedback of user behavior,which unifies the user semantic preferences of different meta-paths as the final preference of users.The experimental results on real-world dataset show that the proposed algorithm achieves better recommendation performance than state-of-the-art baselines.(3)Aiming at the POI recommendation in specific contexts,this paper proposes a context-aware POI recommendation algorithm.Firstly,a bidirectional influence correlativity metric based on meta-path in HIN is proposed to extract the user behavioral semantic features,and a contextual smoothing method is designed to effectively alleviate the data sparsity.Secondly,the algorithm constructs training samples based on weighted random sampling method considering contextual popularity and fuses different features through factorization machine model to predict user preferences.The experimental results show that the algorithm can effectively perceive user behavioral preferences in specific contexts,thus recommending appropriate POIs for users.