Influence-Aware Successive Point-of-Interest Recommendation

With the rapid development of mobile social networks,mobile social applications such as Instagram,Yelp,Foursquare have accumulated a large amount of user point-of-interest(POI)check-ins.Those data can be applied to plan trips for users and help them discover point-of-interest becomes a valuable task.Successive POI recommendation system can aware the migration rules of users’ visiting POIs and provide users with the next POI recommendation service.However,due to the extremely sparse data of user check-ins,there are still many challenges in the application and development of successive POI recommendation services.1)It is very difficult to capture the behavior patterns and time patterns of users visiting POIs,which makes it impossible to provide users with timely and effective successive POI recommendation services.2)Users’ personal preferences cannot be accurately captured,resulting in the inability to provide users with accurate personalized successive POI recommendation services.In response to the above problems,this paper aims at sensing the influence of users visiting POIs and modeling the temporal-spatial pattern of user behavior,which greatly improves the POI recommendation performance and effectively alleviates the challenges brought by data sparsity.The main contributions of this paper are as follows:(1)Aiming at the problem that it is difficult to capture the behavior patterns of users visiting POIs,this paper designs an influence-aware successive POI recommendation model,which mainly aware the influence of each POI in the sequence of successive POIs and predicts the user’s next action.The method integrates position encoding,multi-head self-attention mechanism and user ID embedding,and perceives the migration law of POIs in the sequence,the influence of POIs,and the user’s personal preference to provide a more accurate personalized successive POI recommendation service.(2)Aiming at the problem that it is difficult to capture the temporal patterns of users visiting POIs,this paper designs a temporal-spatial aware successive POI recommendation model based on the above contributions.This method is the first to propose a temporal-spatial encoder and a multi-head self-attention mechanism to jointly aware the user’s short-term preferences,a long-short-term memory network and an attention mechanism to jointly aware the user’s long-term preferences.It can better alleviate the impact of data sparsity,and provide users with more accurate successive POI recommendation services.(3)This paper conducts extensive experiments on the two methods proposed in this paper on the Foursquare and Instagram real-word datasets.The results of comparative experiments and ablation experiments prove that the method proposed in this paper has better performance than previous methods in the face of sparse data,and effectively improves the performance of successive POI recommendation system.