| Social recommendation aims to combine social information with user-item interactions to alleviate the data sparsity problem in recommendation prediction tasks.In view of the nature of the graph structure of social networks,more and more methods use graph neural network as the basic framework to build social recommendation models.However,the existing methods only consider the impact of users’ historical rating interaction behavior on preference propagation,lack the use of the item’s attribute information,or use the attribute information too singly,and ignore the impact of the propagation of users’ diverse preferences in the social context on user modeling.Therefore,the main research of this dissertation takes graph neural network as the basic framework to establish a social recommendation model,focuses on exploring the effective combination of user project interaction graph and social graph to enhance the modeling scheme of users and projects through the attribute information of projects,and explores the recommendation model of social diversity based on the diversity of user behavior.The main work of this dissertation is as follows:(1)Aiming at the problem that it is difficult to capture the fine-grained features of users and items under social relations due to the low degree of coupling with social networks when modeling user and item features,a social recommendation model based on attribute enhancement is explored(AE-SR),which combines user project interaction graph with social graph and divides it into four different domains,Then,the hierarchical attention mechanism is used to extract user and item features based on attribute enhancement for different domains and score prediction.(2)In view of the fact that the existing social recommendation models ignore the diversity of users’ preferences and social friends,and in order to more accurately express users’ preferences,after correcting the data with deviation in the original interaction score,a social diversity recommendation model is explored using gating mechanism and capsule network(GC-SR),Combined with the social diversity selection module,users are complementarily modeled to obtain potential representations of users and items in the context of social diversity.By conducting a large number of experiments on two real data sets commonly used in social recommendation(Ciao and Epinions)and comparing with other baseline models,it can be concluded that the mean square error of the AE-SR model is reduced by 5.4% on average;The mean squared error of the GC-SR model was reduced by 6.5%on average.Both have achieved better performance in predicting performance,which fully proves the effectiveness and rationality of the model.Figure [29] Table [5] Reference [68]... |
