Research On Personalized Recommendation Algorithm Based On Implicit Social Relationshi

With the continuous integration and development of Internet technology and ecommerce platform,the scale of users on the network is larger and the number of goods is more diverse,and it is more and more difficult for users to obtain the information they are really interested in.Studies have found that understanding users’ preferences through social networks can effectively alleviate the above problems.Many studies have directly incorporated explicit social relationships into the recommendation system for recommendation,but because the explicit relationships contain a lot of noise,the recommendation effect is difficult to meet expectations,and the recommendation accuracy is greatly reduced when facing cold start users.In fact,users who are far away from each other in social networks and have no intersection in real life are likely to have the same interests.Some studies have found that exploring implicit social relationships from heterogeneous information networks composed of rich semantic information can greatly improve the accuracy of recommendation systems.In recent years,it has shown good performance in identifying implicit friends by random walk in heterogeneous information networks and then personalized recommendation through implicit friends.However,the meta-path needs to be preset manually and requires strong domain knowledge,which makes it difficult to accurately and effectively capture semantic information.Aiming at how to find more similar implicit friends for personalized recommendation from heterogeneous information networks,this paper proposes the following two algorithms.(1)A personalized recommendation algorithm based on self-guided walking to find implicit friends.The main idea of the algorithm is to find the implicit social relationships with the same preferences through the self-guided walking,and then make personalized recommendations according to the implicit social relationships.Specifically,a guide matrix was first established to record the transition probability from user nodes to interest nodes,and implicit friends with the same preference were found by walking on heterogeneous information networks through the guide matrix.Then,the implicit friends were divided into different types according to the number of times they appeared in different sequences,and different types of implicit friends were given different weights.Finally,all the implicit friends are integrated into an enhanced social Bayesian personalized ranking model for personalized recommendation.The experimental results show that this method can effectively identify the semantic information in the heterogeneous information network,find the implicit friend relationship is more reliable,and improve the accuracy of the recommendation model.(2)Personalized recommendation algorithm for discovering implicit friends based on graph attention network.Firstly,the heterogeneous information network was composed of user-user network,user-item network and item-item network,and then the graph neural network with attention mechanism was used to intelligently aggregate the information from the social network and user-item network to mine the potential implicit social relationship,so as to obtain more comprehensive user characteristics.Then,the aggregated information of the user was input into the feedforward network to find the implicit friends in the same trust network.Finally,the implicit friends were integrated into the matrix factorization for personalized recommendation.Experimental results show that the user feature information aggregated from multiple channels by graph attention network is more comprehensive and stable,and it also effectively improves the recommendation effect.