Research On POI Recommendation Method Based On Graph Neural Network

With the rapid development of the Internet,there are more and more mobile device applications.On Location Based Social Network(LBSN),people share users’ location information and user experience with the programs of mobile devices,making LBSN platform possess a large number of users’ location information.Through these location information,LBSN can recommend the Point of Interest(POI)of the next check-in that users are interested in and meet their personalized needs,thus forming the POI recommendation.It allows people to become more connected with LBSN,and it has important practical significance for the recommendation of tourism and catering industry.Therefore,in recent years,POI recommendation has become one of the most concerned research directions.Previous researchers have done a lot of work in the field of POI recommendation,but there are still some problems.On the one hand,due to the sparsity of data,although most researches analyze and model various contextual information such as time,space and semantics,they fail to mine sufficient auxiliary information and integrate it effectively.On the other hand,users’ interest in check-in often changes dynamically,but the influence of users’ long-term and short-term preferences on users’ check-in has not been effectively taken into account in the past studies,which makes it difficult to understand users’ real interest preferences.In order to solve the above problems,this paper proposes a POI recommendation method based on graph neural network.The main research work is as follows:(1)A POI recommendation method based on social relationships and location popularity is proposed in order to mine sufficient auxiliary information and integrate it effectively.Firstly,a graph is constructed based on the historical check-in records of users,and the nonlinear relationship between users is captured through the graph neural network to obtain effective neighbor information.Then,a dynamic popularity calculation method is proposed to evaluate the influence of location popularity on user check-in in different time and space.Finally,the user sequence is combined with the neighbor information and the selfattention mechanism is used to learn the dependence of the check-in sequence,so as to obtain the user’s sequence characteristics.Then,the influence of location popularity is fused into the sequence characteristics,so as to obtain the accurate interest preference for recommendation.Compared with the existing research methods on real data sets,the experimental results show that the proposed method can effectively improve the recommendation performance.(2)A POI recommendation method based on long-term and short-term preferences based on graph neural network is proposed to obtain more comprehensive interest preferences of users for POI recommendation.Firstly,based on the historical check-in records of users themselves and all users,the local and global graphs are constructed respectively,and the graph neural network is used to learn the user’s check-in track information and the relationship information between locations from the local and global graphs respectively.Then,the feature representation of the user check-in trajectory is obtained by integrating the information of the local graph and the global graph.Finally,the attention mechanism is used to assign different weights to the positions in the trajectory to get a new feature representation,from which the user’s long-term and short-term preferences are obtained for recommendation.Experiments on real data sets show that the recommended performance of this method is better than that of multiple comparison baselines.