Research On Point Of Interest Recommendation Based On Deep Learnin

Nowadays,the location function of mobile devices such as mobile phones accumulates a lot of user track information.With the development of cloud computing and artificial intelligence,these massive data provide a new research direction for learning human behavior trajectory.That is to say,analyzing the user’s historical check-in trajectory behavior to predict the user’s next position-of-interest,point-of-interest(POI)recommendation came into being,and attracted the attention of academia and industry.However,compared with the traditional recommendation,it is more difficult to recommend the points of interest.The data sparsity of user check-in is high and there are many influencing factors.The context information and data sparsity of spatiotemporal and social factors have a great impact on the recommendation results.In the current research work,we do not consider the incorporation of temporal and spatial factors,social factors,comments and other auxiliary information to improve the recommendation performance of the system.Therefore,the main content of this paper is how to effectively use the effective information in POI recommendation data to improve recommendation performance.This dissertation focuses on the application field of interest point recommendation.For the long-standing problem of data sparsity and multi factor influence in the field of interest point recommendation,this paper uses deep learning model combined with auxiliary information to study POI recommendation.Specifically,the main contributions of this paper are as follows:1.In this dissertation,we propose a new model of POI recommendation(PRGAN)based on the generated adversary network.The model is based on SeqGAN and integrates temporal and spatial information,category and other contextual information.The main idea of PRGAN model is to treat POI recommendation as a sequence generation problem,which is mainly composed of generator and discriminator.In this dissertation,the encoder decoder structure is used as the generator to generate the recommended sequence of POIs,and then convolutional neural network is selected as the discriminator to guide the generator training in the way of policy gradient.Experiments on real datasets show that the PRGAN model presented in this dissertation performs well.2.In order to further alleviate the data sparsity and improve the recommendation effect.This dissertation attempts to use deep learning model to combine social,geographic and comment information,and propose a multi factor POI recommendation model(SRPRec).Firstly,social relation matrix and geographic location matrix are used to get social embedding feature and geographic embedding feature by self-encoder.Then we use neural network to learn the feature expression of’ user and item comments.Finally,we use the deep neural network framework to integrate social,geographical and comment factors,mining the deep and shallow interactive features of users and POIs.Sufficient experimental results show that srprec can effectively improve the accuracy of point-of-interest recommendation.