Research On Recommendation Methods Integrating Implicit Relation

In recent years,with the advent of the digital economy era,“information overload”has become increasingly prominent.To solve this problem,personalized recommender systems have been developed rapidly.In current recommender systems,researchers usually start from the direct interaction between users and items,and rely on nonlinear reasoning techniques such as neural networks or graph neural networks to mine deep embedded representations of users and items.However,these studies ignore the importance of direct modeling of indirect interactions,resulting in failure to better capture the interaction information contained in implicit relationships,thus reducing the performance and interpretability of recommender systems.In order to make full use of the implicit relationship and its interaction information,this research proposes the following main research contents:(1)Aiming at the problems of insufficient application of interactive information of implicit relationships and poor recommendation performance in the factorization machine,this research uses the reachability matrix to directly model the implicit relationship in the user-item bipartite graph.Meanwhile,it constructs a recommendation method that integrates implicit relations and factorization machines through the organic integration of the reachability matrix and the factorization machine.Experiments on Amazon-Book,Last-FM and Yelp2018 datasets show that the method outperforms a large number of knowledge graph-based recommendation methods,which fully proves that the implicit relationship can not only alleviate the data sparsity problem in the recommendation system to a certain extent,but also improve the recommendation model’s ability to capture interactive information,and also help improve the accuracy and interpretability of the recommendation.(2)Aiming at the problem that the factorization machine cannot fully excavate the interactive information contained in the implicit relationship,this research organically integrates technologies or ideas such as generative adversarial networks,graph neural networks and reinforcement learning,and constructs a recommendation method(i.e.,GANRec)that integrates implicit relationships and generative adversarial networks.Multiple sets of experiments on six public datasets show that the recommendation performance of GANRec not only surpasses existing knowledge graph-based recommendation methods,but also outperforms state-of-the-art negative samplingbased recommendation methods.Furthermore,when no external knowledge is available,GANRec can also explore knowledge-aware,high-quality negative items and outperform many bipartite graph-based recommendation methods.At the same time,the advantages of the GANRec model in negative sampling and the excellent recommendation performance and interpretability are also verified through experiments.