Research On Interest Point Recommender System Based On Graph Convolution Network

With the development of mobile intelligent equipment and the wide spread of the Internet,more and more people are accustomed to sharing their offline consumption experiences,such as visiting scenic spots,dining in restaurants,etc.,on the Internet.Combining online browsing and offline visits has resulted in a large number of check-in data containing multiple contexts such as users,interest points,time,and location.How to use these check-in data to recommend some potential interest points to users in suitable time and location environments has become an urgent problem to be solved.However,due to the protection of their privacy or other reasons,not every visit to interest points will be checked in and shared online,which makes the recommendation of interest points face a more severe data sparse problem than traditional recommendations.Therefore,this paper improves the traditional graph convolutional network-based recommendation model,and uses various methods to extract features and integrate them into the graph convolutional network-based recommendation model according to different contexts.The experimental results show that the proposed model has significantly improved the recommendation effect and enhanced the adaptability compared with some traditional recommendation methods and other graph convolutional network-based methods when used for point of interest recommendation and other similar recommendation scenarios.Finally,based on the model proposed in this paper,an interest point recommender System is designed and implemented,which can correctly and smoothly recommend and has good application value.The main work of this paper includes:(1)For the irregular user check-in data of points of interest,a model for feature extraction using graph convolutional networks is proposed.This model adopts a multi-dimensional feature fusion method to improve the ability of graph convolutional model to mine graph information;graph sampling is used to enable the model to train large-scale bipartite graphs;attention mechanism is integrated to make the model focus on neighbor nodes in the graph convolution process;at the same time,by introducing residual neural networks,Bayesian personalized ranking training,regularization,Batch Norm layers and other methods,the original model’s difficulty in convergence is improved,the training speed and convergence effect of the model are improved,and the model convergence is more stable.(2)For the possible multi-context information in the dataset,different methods are used to integrate them into the graph convolutional network model according to the different types of context information.For example,for the text data in the context data,the Doc2 vec model is used to extract features;for the time,age and other segmentable data in the context data,the GMF model is used to extract features,which fully utilizes the multi-context information to improve the recommendation effect of the graph convolutional network model.(3)Based on the improvement of graph convolution recommendation model in this article,an interest point recommender System has been designed and implemented.The system displays a list of recommended interest points for each user based on the model,while anonymously collecting user interaction records with interest points and various types of contextual data generated during the interaction for model updates and iterations.In addition,a solution algorithm has been proposed for the cold start problem of the recommender System.The system recommends some representative interest points for users when they first log in,and initializes their feature vectors based on their feedback records and basic information,in order to provide more accurate recommendations for cold start users.Finally,the system has undergone comprehensive testing to ensure that it can provide stable and smooth high-quality interest point recommendations for users...