Research On Recommendation Algorithm And Privacy Protection Based On Graph Neural Network

With the development of big data and privacy protection awareness,users have put forward higher requirements for the accuracy and privacy of recommendation algorithms.However,traditional recommendation algorithms are limited by sparse user interaction data,and it is difficult to guarantee the accuracy of recommendation.In business scenarios,the data used by recommendation algorithms mostly satisfy the nature of graphs.Integrating the graph neural networks into the recommendation algorithm can bring more recommendation information to the recommendation system.However,the current recommendation algorithms based on graph neural networks still have the following deficiencies: 1)they cannot provide effective negative samples that take into account both authenticity and information to alleviate the problem of missing training data that leads to limited recommendation accuracy;2)the existing recommendation strategies is difficult to guarantee the privacy of the recommendation without losing the recommendation accuracy.In view of the above problems,this thesis studies recommendation algorithms from two aspects,negative sample sampling strategy and recommendation privacy protection,based on graph neural network.The main work of this thesis is as follows:(1)Propose a negative sampling augmented recommendation model(GNSP)based on heterogeneous graph neural network.On the basis of using high-level graphs to construct user,item interaction and item attribute relationships,a reinforcement learning strategy combined with attention mechanism is used to sample negative samples that are both informative and authentic;(2)Design A privacy recommendation algorithm(KGLN)based on knowledge graph enhancement.On the basis of utilizing the rich semantic and structural information of the knowledge graph,on the one hand,a bias-based attention mechanism is introduced to improve the accuracy of recommendation,and on the other hand,a combination of adaptive gradient clipping and noise is introduced in the gradient descent process.Differential privacy mechanism to ensure recommended privacy;(3)Taking the above algorithm as the key technology,a prototype system based on graph neural network privacy-preserving recommendation algorithm is designed and implemented.In this thesis,a negative sample sampling method of both authenticity and information in the recommendation algorithm is explored,and the privacy of the recommendation algorithm is taken into account while improving the recommendation accuracy,which provides a reference for the privacy protection work of mixed recommendation based on graph neural networks.