Research On Point Of Interest Recommendation Based On Deep Learnin

With the rapid development of social network,users can more easily share their visited locations in real time through location-based services on mobile devices.The resulting huge amount of check-in data contain information about users’ movement patterns and behaviors.How to effectively obtain information from these data has become a hot topic of current research.Among the various types of information extraction research,the point-of-interest(POI)recommendation task,as research to analyze the potential movement pattern of users based on check-in data and then predict their future visiting behavior,has attracted a lot of attention from industry and academia for its great commercial application.However,the high sparsity of data severely limits the performance of POI recommendation models,and the existing work ignores the variability in the impact of temporal information on visiting behavior,thus fails to achieve accurate recommendations.To address the above problems,this paper proposes two POI recommendation models based on deep learning methods.The specific research works are as follows:(1)To address the problem caused by sparse check-in data,this paper proposes a POI recommendation model(Graph Convolutional Network and Neighbor Discovery,GCNND)based on graph convolutional and neighbor discovery mechanism.First,for the problem of sparse check-in records,the extremely sparse user data is filtered out by data pre-processing.Then,to address the sparse check-in location problem,this paper theoretically expands the range of users’ visited POIs by mining the neighbor POI features of visited POIs.In this paper,the user’s check-in trajectory is divided into the current trajectory and historical trajectory,and the recurrent layer is used to capture the sequence features in the current trajectory,and the graph convolutional layer is used to capture the spatial correlation features between the user’s visited POIs in the historical trajectory.Finally,the generated features of all modules are fused to complete the POI recommendation work.(2)To address the problem caused by variability in the impact of temporal information on visiting behavior,this paper proposes a POI recommendation model based on convolutional neural network and attention mechanism.This paper labels the check-in trajectory into weekday trajectory and holiday trajectory,and further divides the trajectory into current trajectory and historical trajectory on this basis.The proposed model mainly consists of a spatial-temporal relationship module and geographical relationship module,in which the spatial-temporal relationship module uses the long short-term memory to extract the spatial-temporal features in the current trajectory,and the geographical relationship module uses the convolutional neural network and attention mechanism to capture the local geographical preferences of users and generate the POI score matrix,respectively.Finally,the fused features are used to predict the next location to be visited.Through extensive comparison experiments with existing methods on real datasets,it is demonstrated that our models have better recommendation results in POI recommendation task.The methods in this paper achieve better performance in evaluation metrics such as accuracy rate and normalized discounted cumulative gain,and have higher accuracy in terms of POI recommendation.