Sequential Recommendation Method Based on Contrastive Learning and Meta-optimized Learning
Sequential recommendation is to model user interest based on historical interaction records between users and projects,and rec-ommend the next project.As a kind of side information,contrast learning(CL)can effectively improve the quality of recommendation models,but the existing sequential recommendation methods based on contrast learning are unstable and difficult to generalize by using random data enhancement.To address the above problem,a sequence recommendation method based on contrastive learning and meta-optimized learning is proposed.Firstly,in the data augmentation step,the data augmentation view with more uniform data distribution is generated for the sequence according to the time interval between the items in the sequence.Secondly,a learnable model augmentation module is constructed to capture the potential semantic information in the data augmentation view and enhance the generalization ability of the model.Finally,in order to solve the problem of different optimization objectives between the data augmentation module and the model augmentation module,the meta-optimized learning is used to optimize and update the parameters between the two modules to complete the recommendation.Experimental results on three publicly available datasets,including Beauty,Sports and Yelp,showed that CLMLRec has significantly improved in terms of recall and NDCG compared with other baseline models,indicating that the model has good recommendation performance.
国家科技期刊平台
NSTL
万方数据
