Research On Cross-domain Recommendation Based On Deep Cross Feature Transfer

In the era of information explosion,recommendation systems have been successfully applied in many fields and brought convenience to people's lives.The related research on recommender systems has made great progress,but the explosive growth of users and projects on the Internet in recent years has also exacerbated the problem of data sparsity,which has become a bottleneck for the further development of traditional recommender systems.Cross-domain recommendation as one of the solutions to this problem has received widespread attention.Although there have been many related studies to enhance the prediction performance of the target domain by using the knowledge of the source domain,these studies lack in-depth exploration of the cross-features between review information and other auxiliary information;In cross-domain scenarios,most models use overlapping users or projects between different domains to build the mapping relationship between different domains so as to realize the full use of information between different domains.This paper improves and proposes a cross-domain recommendation model DCMT-Net(Deep-Cross Module Transfer Network)based on deep cross feature transfer on the basis of MMT-Net.The model is composed of the feature cross-mode fast~1,the user and item embedding module~2,the fusion module~3,and the final scoring prediction module~4.After reasonable preprocessing of various aspects of information,the Deep&Cross network in the~1 module can simultaneously realize the construction of the displayed cross features and the nonlinear interaction between the features,so as to mine as much available information as possible,and at the same time Realize the migration of information from the source domain to the target domain without relying on users or project overlap.This paper applies the model to multiple target domains and uses RMSE and MAE to verify the effectiveness of DCMT-Net in single-domain and cross-domain recommendations through different experimental groups.