Research On POI Recommendation Methods Based On Hybrid Graph Neural Networks

With the rapid development of mobile Internet technology and location technology,Location-Based Social Network(LBSN)and their services come into being.As one of the most important services of LBSN,Point-of-Interest(POI)recommendations have attracted more and more researchers’ attention.By providing users with real-time location services,LBSN makes it easy for users to record their locations of interest,share their daily lives and travel experience.By mining the characteristics of users’ historical check-in behavior,recommending the POI sequence that users may be interested in.In order to save users’ decision time and create more service value for location-associated merchants.At present,in terms of personalized POI recommendation,the existing research work is usually based on the historical check-in data of users.Based on traditional recommendation methods,deep learning methods and graph neural network methods,the user’s preference for POI is obtained,and the POI sequence is recommended according to the user’s preference.However,the existing POI recommendations usually represent the user’s interaction behavior with POI as the user’s positive preference for POI,without considering the user’s negative preference information for POI implied by the user’s noninteraction behavior with POI.In addition,the existing personalized POI recommendations do not learn modeling data transfer between users’ POIs,which affects the effect of personalized POI recommendations.In terms of group POI recommendation,the current research mainly focuses on fixed groups,while research on random groups is relatively rare.Compared with fixed groups,the interaction data between random groups and POI is extremely sparse,which leads to the cold startup problem of random groups.In addition,the preference difference of random group members is much greater than that of fixed group.Therefore,the existing POI recommendations for fixed groups cannot be directly applied to random group POI recommendations.In order to solve the above problems,the following research work has been carried out in this paper:(1)An Approach for Personalized POI Recommendation Based on Hybrid Graph Neural Network is proposed.Firstly,the user social network graph,the label interaction bipartite graph between user and POI,and the directed transition graph between POI and POI are constructed.Secondly,based on the Graph Attention Network to learn the social representation of each user,based on the Signed Bipartite Graph Neural Networks to extract users representation and POI representation including users’ POI interaction preferences,and based on the Session-based Recommendation with Graph Neural Networks to learn POI representation with users’ transfer preference.Then,the method integrates users’ social representation and users’ representation including users’ POI interaction preferences to obtain the final representation of users,and integrates POI representation including users’ POI interaction preferences and POI representation with users’ transfer preference to get the final POI representation.Finally,based on the final representation of users and the final POI representation,POIs are selected and recommended to users by the sorted prediction scores.(2)An Approach for Random Groups POI Recommendation is proposed.Firstly,the fitting feature representation of random groups is obtained based on the random group feature fitting method.Then,based on cosine similarity function,users with similar features to random group were found,and the data of similar users were used to train and optimize the model,in order to overcome the cold start problem of random group POI recommendation.Then,based on the Signed Bipartite Graph Neural Networks to learn POI characteristics that resemble user POI interaction preferences,and based on the Sessionbased Recommendation with Graph Neural Networks to learn POI characteristics of similar users’ POI transfer preferences.Finally,the two groups of POI features are merged,combined with the fitting features of the random group.POIs are recommended to the random group by calculating the predicted score for each POI.