A multi-head graph attention and discriminative optimization network for rating prediction
Aims:In order to fully explore the complex interaction between users and items,a multi-head attention and discriminative optimization network was proposed for the rating prediction task.Methods:Firstly,a rating subgraph for each user-item pair was constructed.Then,the multi-head attention graph convolutional network on the subgraph was trained to predict the corresponding rating.During the message passing process,attention and aggregation modules were employed to calculate the importance of different neighbors to the central node,effectively aggregating the information from neighbors to obtain better representations of users and items.Finally,the discriminative model was introduced to optimize the predicted ratings,improving the accuracy of predicting user ratings for items by comparing them with the real ratings.Results:Comparative experiments were conducted on multiple datasets to validate the superior predictive performance of the proposed model.Conclusions:The proposed multi-head graph attention and discriminative optimization network can effectively improve the performance of rating prediction.
国家科技期刊平台
NSTL
万方数据
