Research On Group Recommendation Algorithm Based On Attention Mechanism And Graph Neural Network

In the era of big data,recommendation systems can effectively solve the problem of information overload and bring huge commercial value to enterprises.As the service targets of the recommendation system gradually expand on a single user to a group user,group recommendation has become one of the current research hotspots in the field of recommendation systems.Compared with traditional recommendation systems,group recommendation aims to provide group users with accurate and personalized recommendations for group activities,which can effectively to integrate the user preferences of all group members and alleviate preference conflicts within the group.However,most existing group recommendation systems used predefined fusion strategies based on experience of predefined member influence weights,and then weighted and fused into group preferences,but they do not take into account the potential preferences of members in group decision-making for different items.In addition,the existing methods also ignore the influence of potential interactions between members on the modeling of group preferences.Limited by the above-mentioned problems,the existing methods are difficult to accurately and comprehensively mine group preferences,thus limiting the recommendation effect for complex groups.In response to the above problems,this paper proposes an Attention and Graph Neural Network Recommendation Model(AGNN)based on the attention mechanism and graph neural network.On the one hand,considering the importance of group member preferences in different item recommendations,the AGNN model introduces graph neural network technology,and first uses the space-based graph convolution method to mine the project-based perspective from the interaction graph between the user and the item The user's potential interest preferences are then combined with the attention mechanism to learn the influence weights of different member preferences in different group decisions from the project perspective,thereby effectively modeling the group preference expression based on the project perspective.On the other hand,considering the influence of interaction between members on group preference modeling,the AGNN model also uses the space-based graph convolution method to effectively mine the potential interaction between members from the interaction graph between the group and the members,and then Accurately generate a group preference representation based on the user's perspective.Finally,the AGNN model uses the gating module to fuse the above-mentioned group preference representation based on the two perspectives of users and items to obtain a more comprehensive and accurate group preference representation.Then,the group preference representation and the feature representation of the candidate items are input together into In the prediction layer,more accurate group recommendation results are generated.In order to evaluate the group recommendation performance of the model,this paper conducts extensive comparison experiments on three real data sets.The experimental results show that the AGNN model achieves a better recommendation effect than the benchmark model.In addition,this article also sets up an ablation experiment to compare the effects of different modules on recommendation performance,and verifies the effectiveness and necessity of the group recommendation task for the differences in the importance of interaction between members and member preferences in different item recommendations.Finally,this paper also developed a corresponding group recommendation system interface for the model,which verified the usability of the AGNN model in real-life recommendation applications.