Research On Fused User Interest Recommender Algorithm Based On Deep Learning And Graph Structure

The recommender algorithm is of practical significance to alleviate the problem of excessive network information load.However,most of the current recommender algorithms are only based on interactive content,and most of them ignore the mining of users’ complex interests.Since people’s interests are not static and immutable,it is very important to make personalized recommendation while capturing people’s dynamic interests.Therefore,this paper combines deep learning and graph neural network,gradually studies and proposes a user interest fusion algorithm based on deep learning and graph structure,to dig the uncertain and dynamic interests of users at a deeper level.Compared with the methods based on machine learning and ordinary deep learning,this paper achieves better recommendation effect.The main work of this paper is as follows:(1)Aiming at the problems of slow speed,low robustness and poor user interest fusion in traditional recommender algorithms,a fusion user static interest recommender algorithm based on deep collaborative filtering is proposed.Firstly,the traditional factorization algorithm was improved to fully explore the interests of static users.Then the multi-layer perceptron was improved into a dual-parallel structure,which can not only fully integrate the static user interests,but also effectively avoid the interest loss caused by feature pooling.(2)Aiming at the problem of insufficient depth of user interest mining in general deep recommendation methods,a recommender algorithm based on graph convolution embedding is proposed to integrate user shallow and deep interest.It fuses the user interests at a deeper level.Firstly,graph convolutional network was introduced to enhance the fusion ability of different users’ interests.Then,the hybrid factorization machine was further improved,so as to integrate the superficial and deep interests of the user,and to achieve more recommendations based on the user’s fusion interests.(3)Aiming at the problem of changing user interests and insufficient integration degree,a graph neural network based recommender algorithm for integrating users’ long,medium and short interests is proposed.First,a session graph was constructed according to the user-item interaction information.Then,the graph neural network and linear embedding were used to obtain the low order interest of users at each stage,and the attention mechanism was used to capture the attention signals of users at each stage,and the attention signals at each stage were fused with the corresponding interests to obtain the high order interest for further integration.Finally,the fusion of deep high-level interest with the final session in the initial session layer can obtain the final deep fusion user interest.Through a large number of experiments,it is concluded that the fusion of user interest algorithm has outstanding advantages in terms of comprehensive quality,recommendation hit rate and accuracy,and also proves that the fusion of user interest is effective and practical for the actual recommendation task.