Fine-Grained Cross-Domain Recommendation via Two-Tier Attention and Three-Channel Learning

Cross-Domain Recommendation (CDR) algorithms, aimed at alleviating the long-standing data sparsity and cold-start problems by transferring information collected from the source domains to the target domains, have attracted increasing attention recently. Other than ratings, existing works on CDR mostly consider side information like tags, reviews, contents etc., yet cannot make full use of texts (i.e. reviews and contents etc.) efficiently or fuse these side information with ratings deeply. Inspired by the advantages shown in review-based recommendations and aspect-based ones, we propose to model fine-grained user preference transfer at aspect level. To achieve this goal, we propose an end2end CDR framework via aspect transfer network with two-tier attention and three-channel learning (named TATCL). TATCL is devised to extract aspects to represent each user or item from their reviews by a review encoder and a subsequent user/item encoder with two-tier attention, and learn accurate aspect correlations across domains with three-channel learning. In addition, we enhance the user and item representation with auxiliary reviews and item contents. Experimental results on datasets demonstrate that, under certain condition, the proposed TATCL has superior predictive performance than existing models in terms of rating prediction accuracy.

[1]  Hongbo Deng,et al.  CATN: Cross-Domain Recommendation for Cold-Start Users via Aspect Transfer Network , 2020, SIGIR.

[2]  Chenliang Li,et al.  Cross-Domain Recommendation via Preference Propagation GraphNet , 2019, CIKM.

[3]  Yang Xu,et al.  Deeply Fusing Reviews and Contents for Cold Start Users in Cross-Domain Recommendation Systems , 2019, AAAI.

[4]  Xing Xie,et al.  Neural News Recommendation with Attentive Multi-View Learning , 2019, IJCAI.

[5]  Li Peng,et al.  A Capsule Network for Recommendation and Explaining What You Like and Dislike , 2019, SIGIR.

[6]  Chuhan Wu,et al.  Hierarchical User and Item Representation with Three-Tier Attention for Recommendation , 2019, NAACL.

[7]  Qiang Yang,et al.  Transfer Meets Hybrid: A Synthetic Approach for Cross-Domain Collaborative Filtering with Text , 2019, WWW.

[8]  Shafiq R. Joty,et al.  ANR: Aspect-based Neural Recommender , 2018, CIKM.

[9]  Yang Xu,et al.  Cross-Domain Recommendation for Mapping Sentiment Review Pattern , 2018, KSEM.

[10]  Philip S. Yu,et al.  Aspect-Level Deep Collaborative Filtering via Heterogeneous Information Networks , 2018, IJCAI.

[11]  Feng Zhu,et al.  A Deep Framework for Cross-Domain and Cross-System Recommendations , 2018, IJCAI.

[12]  Yu Zhang,et al.  CoNet: Collaborative Cross Networks for Cross-Domain Recommendation , 2018, UMCit@KDD.

[13]  Philip S. Yu,et al.  Cross-Domain Recommendation for Cold-Start Users via Neighborhood Based Feature Mapping , 2018, DASFAA.

[14]  Xiaolong Jin,et al.  Cross-Domain Recommendation: An Embedding and Mapping Approach , 2017, IJCAI.

[15]  Philip S. Yu,et al.  Review-Based Cross-Domain Recommendation Through Joint Tensor Factorization , 2017, DASFAA.

[16]  Lei Zheng,et al.  Joint Deep Modeling of Users and Items Using Reviews for Recommendation , 2017, WSDM.

[17]  Julian J. McAuley,et al.  Ups and Downs: Modeling the Visual Evolution of Fashion Trends with One-Class Collaborative Filtering , 2016, WWW.

[18]  Jeffrey Dean,et al.  Distributed Representations of Words and Phrases and their Compositionality , 2013, NIPS.

[19]  Guandong Xu,et al.  Personalized recommendation via cross-domain triadic factorization , 2013, WWW.

[20]  Xindong Wu,et al.  Cross-Domain Collaborative Filtering over Time , 2011, IJCAI.

[21]  Geoffrey J. Gordon,et al.  Relational learning via collective matrix factorization , 2008, KDD.