Research On Graph Model Recommendation Algorithm Based On User Behavior And Trust Degree

With the rapid development of the Internet information technology in the new era,the emergence of the recommendation system has brought great relief to the problem of information overload.The current popular recommendation method has good performance in practical applications.However,in the process of user interest modeling,a large amount of subjective data is required such as user ratings,reviews,and object preferences that clearly reflect the preferences of the users' interests.But the difference in subjective factors and the size of training data will affect recommendation performance of the users' interest model,resulting in data sparseness of the recommendation system and the accuracy of the recommended items.However,the implicit feedback data is more easily obtained in the website service,and it contains more potential interest features of the user and can more accurately learn the interest model.Therefore,for the emergence of these problems,this paper mainly aims at the users' implicit feedback data to do the following research work:1.Utilize the auxiliary information of social relationships between users to incorporate the users' implicit feedback behaviors and attention relationships into the graph recommendation algorithm.The addition of the attention relationship between users can make up for the insufficiency of information resources data.For sparseness problems,the recommendation algorithm can also use the newly added user-interest relationship to find potential interest content for the target user.Users have certain exclusivism about how many users they care about,there are some concerns that do not reflect the similarities between users.For this reason,this paper proposes a method based on Probabilistic Neural Network(PNN)to effectively focus on relationships to eliminate users' attention to user relationships affecting the recommendation effect in the network,and then use the attention network to complete personalized recommendations.The algorithm performs evaluation experiments on Epinions and YouKu datasets.2.Graph-based recommendation method based on user implicit feedback behavior and trust degree.As the most reliable form of relationship in social networks,trust relationship is more inclined to accept recommendation information from trusted friends than random behaviors such as concern relationships.To this end,based on basic graph model recommendation as the basis for research,a method for integrating user trust in a social network into a graph model to calculate user similarity between interests is proposed.The potential interest is passed to the trust relationship.Target users can not only mine more target interest content,but also improve the performance of content recommendation.The method was evaluated on four data sets such as Ciao and FilmTrust.Through in-depth analysis of implicit feedback behaviors and socialization of users,the experiment validates the effectiveness of the proposed method and basically meets our expectations.