Research On Key Technologies Of Online Novel Recommendation System

With the rapid development of the Internet,we have entered the era of information explosion.The massive amount of information brings about the dilemma of information overload,which makes the cost of obtaining high-quality and valuable information higher and higher.At the same time,the information age has given birth to a special group of "Online water army ",which is driven by interests to spread a large number of public opinions or malicious smears,which further increases the difficulty for people to obtain effective information.As a typical representative of popular literature in the information age,online novels are also plagued by the above problems,and their development is limited.Robust personalized recommendation algorithm research can solve the above problems very well.Due to the powerful data characterization ability of the graph neural network,this paper uses it as the basic model to improve the recommendation algorithm.However,the graph neural network is not robust to adversarial attacks,and its iterative propagation of information is easy to amplify the disturbance.Therefore,this paper studies the robust recommendation algorithm of graph neural network from the category of adversarial machine learning,and finally applies the algorithm to online novel recommendation.The main work of this paper is as follows:(1)Aiming at the ubiquitous occasional noise in the actual process,this paper proposes a robust graph collaborative filtering algorithm based on hierarchical attention,which includes node-level attention and graph-level attention.Through the node-level and graph-level attention,the noise reduction of the graph structure in the deep propagation process can be fully realized,ensuring that the nodes encode high-order collaborative information while minimizing the noise information carried.In this way,the robust representation of nodes is finally achieved,which promotes the improvement of recommendation results.(2)Aiming at the adversarial attack caused by false feedback from malicious users,this paper proposes a robust graph neural recommendation algorithm under malicious attack,which includes a score prediction module and a malicious user detection module.The rating prediction module follows the robust recommendation algorithm in the previous chapter,and the two modules promote each other,so as to finally achieve the dual tasks of robust score prediction and malicious user detection.(3)An online novel recommendation system is designed and implemented,two improved robust recommendation algorithms are trained offline and deployed on novel websites,and finally the robust personalized recommendation of online novels is realized.