Research On Social Recommendation Incorporating User Behavior Information

Recommendation systems,as effective tools to alleviate the information overload problem,have been commonly used in major online platforms.Among them,social recommendation systems,as one of their important research directions,can improve recommendation performance by incorporating auxiliary information.In recent years,social functions in online platforms have been evolving,making users’ behaviors on the platform more diverse.The diverse behaviors of users in the platform show their preferences for items and other users,including user-item interacted behavior,user-item description behavior,user-user individual social behavior,and user-user group social behavior.The continuous mining and effective utilization of user behavior information has greatly contributed to the flourishing of social recommendation systems.However,there are still some problems that need to be solved in the current social recommendation systems,such as simple user-item interactions can hardly explain the reasons for users’ preferences on items,the mutual influence between users’ social relationships and users’ preferences is not fully utilized,and users would like to follow group decision-making opinions but the opinions of members in the group are not consistent.To this end,this research analyzes and explores users’ behaviors towards items and other users,which are based on the information of user behavior generated on online platforms.To address the shortcomings of existing social recommendations,we focus on social recommendations incorporating tags in user-item description behavior,social recommendations incorporating trust and distrust relationships in user-user individual social behavior,and social recommendations incorporating groups in user-user group social behavior.The specific research contents and contributions are described as follows.(1)To address the problem that simple user-item interactions can hardly explain the reasons for users’ preferences on items in recommendation systems,this research incorporates tags in user-item description behavior into the recommendation systems,and proposes a recommendation method based on graph neural networks to mine users’ finegrained preferences.The method first identifies the importance of different tags to <user,item> based on the tag preference of users and the tag relevance of items,and thus mines the potential reasons that induce the user’s preference for the item and obtains a finergrained preference representation of the user.Next,based on user-item ratings,the influence of different rating levels on users’ preferences is learned,to further optimize users’ preferences and match them with item features to achieve recommendations.The results on the Movie Lens dataset show that the recommendation method for exploring users’ fine-grained preferences by incorporating user-item tag information can improve the effectiveness of social recommendation,and that user-item tag information plays an important role in exploring users’ preferences.(2)To address the problem that the mutual influence between users’ social relationships and users’ preferences is not fully utilized in recommendation systems,this research proposes a recommendation method based on graph neural networks considering dynamic interactions by analyzing the trust and distrust relationships in user-user individual social behavior and designing a joint learning framework.The method firstly considers the specificity of the distrust relationship based on the trust relationship,and uses the balance theory to explore users’ potential social relationships,which further provides rich information for predicting user preferences.Secondly,based on the mutual influence between users’ social relationships and users’ preferences,a dynamic item-aware mutual influence mechanism is designed to jointly learn item recommendation and trust prediction tasks,which enables the performance of the recommendation system to be improved in the process of the mutual influence.The results on the Epinions dataset show that the joint learning method of item recommendation and trust prediction incorporating users’ trust and distrust relationships can improve the effectiveness of social recommendations,and that users’ trust and distrust relationships have a positive impact on predicting user preferences.(3)To address the problem that users would like to follow group decision-making opinions but the opinions of members in the group are not consistent,this research analyzes group information in user-user group social behavior and proposes a group recommendation method based on probability matrix factorization and evidential reasoning rule.The method first uses group information to mine members’ potential preferences,decomposes the user-item rating matrix and user-group relationship matrix jointly based on the probability matrix factorization method,and combines item content information to predict the individual preference ratings of group members for items.Next,according to the importance and trustworthiness of members’ contribution to the group,different weights and reliability are assigned to group members,and the individual preference ratings of group members are aggregated by the evidential reasoning rule to generate group-oriented recommendation results.The results on the Cite ULike dataset show that the group-oriented recommendation method incorporating users’ social group information can improve the effectiveness of social recommendation,and the users’ social group information has an important influence on mining latent user preferences.This research expands the ideas of social recommendation methods,enriches the research content of social recommendation incorporating user behavior information,and plays an important role in promoting theoretical research of social recommendation problems.Moreover,this research has a wide scope of application in practice,providing rational suggestions and insights for enterprises from the perspective of data analysis,which are of high reference value for enterprises to build social recommendation systems that meet users’ needs and maintain a strong business chain.