KDD 2022 | how to use comparative learning in cross domain recommendation?

Paper title ：

Contrastive Cross-domain Recommendation in Matching

Thesis link ：

https://arxiv.org/pdf/2112.00999.pdf

Code ：

https://github.com/lqfarmer/CCDR

Cross domain recommendation （CDR） It is to provide better recommendation results in the target domain with the help of the source domain . However , matching （matching, Candidate generation ） Module CDR Knowledge transfer and representation learning are affected by data sparsity and popularity deviation . This paper presents a comparison of cross domain recommendations (CCDR) frame , be used for CDR Match in . say concretely , We have built a huge diversified preference network to capture a variety of information reflecting users' different interests , And designed a domain contrast learning （intra-CL） And comparative learning between the three domains （inter-CL） Mission , To better represent learning and knowledge transfer . Intra domain contrast learning achieves more effective and balanced training in the target domain through graph enhancement , And inter domain contrast learning from users 、 Different types of cross domain interactions are constructed from three aspects of classification and neighbor .