Research On Cross-domain Recommendation Algorithm Based On Graph Convolutional Neural Network

With the continuous development of Internet technology,people rely more and more on the Internet to obtain information and communicate.In such a context,recommendation systems have become an important way for people to get personalized information.Traditional recommendation systems mainly recommend based on users’ historical behaviors,but such recommendation methods have some problems,such as sparse data and over-reliance on historical behaviors.Cross-domain recommendation system is born,which can correlate information from different domains and provide more comprehensive recommendations for users.It explores the potential connections between data by mining and analyzing data from different domains,and transforms them into recommendation results.The cross-domain recommendation system can not only provide more comprehensive information for users,but also provide more accurate advertising for enterprises.The implementation methods of cross-domain recommendation systems mainly include feature-based methods and representation learning-based methods.The feature-based approach mainly maps data from different domains into the same feature space and realizes cross-domain recommendation by calculating the similarity.The representation-based approach uses deep learning techniques to learn the data and realize cross-domain recommendation by the learned representation.In this paper,two cross-domain recommendation algorithms are proposed based on graph convolutional neural networks.The main work is as follows:The first main work of this paper is to propose a cross-domain recommendation method called AISGCF,which is based on graph convolutional neural networks and improved to address the shortcomings of current cross-domain graph convolutional algorithms.Unlike the existing cross-domain graph convolutional network algorithms with fixed network structure and no treatment when adding cross-domain interaction edges,AISGCF takes into account the effect of noisy edges when introducing cross-domain edges,and thus introduces an AIS module to filter the interactions to eliminate edges with inconsistent user preferences in the source and target domains.In addition,the method applies different weights to the filtered edges based on the user’s interest level to distinguish the user’s attention level for different items,which in turn improves the recommendation accuracy.This paper proposes a graph convolutional model SGCF based on attention and structural feature fusion for the relationship between users and items in recommender systems and the structural properties of graph convolutional neural networks.The model first samples the structural features in each domain using the SE module,and then exchanges and fuses the structural features of users in both domains through the MLP layer.In this paper,we further optimize the common one-way attention mechanism and propose a two-way attention mechanism.Through experiments conducted on three Amazon datasets,the results show that AISGCF and SGCF can improve the accuracy of score prediction and classification,and effectively address the sparsity problem.