CATN: Cross-Domain Recommendation for Cold-Start Users via Aspect Transfer Network

SIGIR '20: The 43rd International ACM SIGIR conference on research and development in Information Retrieval Virtual Event China July, 2020, pp. 229-238, 2020.

Cited by: 0|Bibtex|Views127|DOI:https://doi.org/10.1145/3397271.3401169
EI
Other Links: arxiv.org|dl.acm.org|dblp.uni-trier.de|academic.microsoft.com
Weibo:
Our key focus is on the transfer of user preference derived from source domain to target domain, for effective and explainable recommendation

Abstract:

In a large recommender system, the products (or items) could be in many different categories or domains. Given two relevant domains (e.g., Book and Movie), users may have interactions with items in one domain but not in the other domain. To the latter, these users are considered as cold-start users. How to effectively transfer users' pref...More

Code:

Data:

0
Introduction
  • Traditional collaborative filtering methods recommend items to users mainly based on their historical feedbacks.
  • These approaches become less effective for new users, i.e., cold-start users, who have no historical feedbacks.
  • These users are considered as cold-start users.
  • As the two domains are relevant, feedbacks in the source domain could be leveraged to provide meaningful recommendations in target domain
Highlights
  • Recommender systems play vital roles in various e-commerce platforms
  • Traditional collaborative filtering methods recommend items to users mainly based on their historical feedbacks
  • Given two relevant domains (e.g., Book and Movie), users may have historical interactions in one domain, but not the other. These users are considered as cold-start users
  • We propose a cross-domain recommendation framework for cold-start users via aspect transfer network, named CATN
  • Our key focus is on the transfer of user preference derived from source domain to target domain, for effective and explainable recommendation
  • Through extensive experiments conducted on three pairs of real-world datasets, we demonstrate that CATN performs significantly better than state-of-the-art (SOTA) alternatives
  • Instead of following the existing framework to first learn user/item representations in the source and target domains, learn the mapping, we propose an end-to-end learning strategy
Methods
  • The authors conduct recommendation by leveraging their corresponding reviews in the source domain directly, after training the model purely on the target domain
Results
  • Results and Discussion

    The overall results of all methods over the three cross-domain recommendation scenarios are reported in Table 3.
  • CATN outperforms all baselines significantly on all cross-domain recommendations, and in terms of different ratios of overlapping users in all settings.
  • This result demonstrates the superiority of the review-based recommendation for cold-start users in cross-domain setting.
  • Learning user representations by factorizing a joint matrix is not adequate, which is consistent with what has been observed in earlier studies [9, 22, 29].
  • CDLFM makes some improvements to the user factors learning and the
Conclusion
  • The authors study the problem of review-based cross-domain recommendation for cold-start users.
  • The authors' key focus is on the transfer of user preference derived from source domain to target domain, for effective and explainable recommendation.
  • Instead of following the existing framework to first learn user/item representations in the source and target domains, learn the mapping, the authors propose an end-to-end learning strategy.
  • The authors show that the CATN model outperforms all existing models for cross-domain recommendation tasks.
  • Inspired by [13], the authors may investigate for more possibilities on graph-based CDR in the future
Summary
  • Introduction:

    Traditional collaborative filtering methods recommend items to users mainly based on their historical feedbacks.
  • These approaches become less effective for new users, i.e., cold-start users, who have no historical feedbacks.
  • These users are considered as cold-start users.
  • As the two domains are relevant, feedbacks in the source domain could be leveraged to provide meaningful recommendations in target domain
  • Objectives:

    The authors aim to exploit user/item reviews for cross-domain aspect correlations. Since the authors aim to map the aspect across domains, Vs and Vt are shared in their corresponding domains respectively.
  • Methods:

    The authors conduct recommendation by leveraging their corresponding reviews in the source domain directly, after training the model purely on the target domain
  • Results:

    Results and Discussion

    The overall results of all methods over the three cross-domain recommendation scenarios are reported in Table 3.
  • CATN outperforms all baselines significantly on all cross-domain recommendations, and in terms of different ratios of overlapping users in all settings.
  • This result demonstrates the superiority of the review-based recommendation for cold-start users in cross-domain setting.
  • Learning user representations by factorizing a joint matrix is not adequate, which is consistent with what has been observed in earlier studies [9, 22, 29].
  • CDLFM makes some improvements to the user factors learning and the
  • Conclusion:

    The authors study the problem of review-based cross-domain recommendation for cold-start users.
  • The authors' key focus is on the transfer of user preference derived from source domain to target domain, for effective and explainable recommendation.
  • Instead of following the existing framework to first learn user/item representations in the source and target domains, learn the mapping, the authors propose an end-to-end learning strategy.
  • The authors show that the CATN model outperforms all existing models for cross-domain recommendation tasks.
  • Inspired by [13], the authors may investigate for more possibilities on graph-based CDR in the future
Tables
  • Table1: Statistics of the three datasets in Amazon
  • Table2: Statistics of the three cross-domain recommendation scenarios. η donotes the ratio of overlapping users included in the training set
  • Table3: Performance comparison on three recommendation scenarios in terms of MSE. The best and second best results are highlighted in boldface and underlined respectively. % denotes the relative improvement of CATN over the best SOTA algorithm. All reported improvements over baseline methods are statistically significant at a 0.05 level
  • Table4: Performance comparison of the three model variants on three recommendation scenarios
  • Table5: Example study of three user-item pairs from three recommendation scenarios at η = 50%
  • Table6: Top-5 words for each aspect in Example 1. The “Aspect Labels” are manually generated based on our own interpretation
Download tables as Excel
Related work
  • Our work is related to two subareas of recommender systems: crossdomain recommendation, and aspect-based recommendation. Next, we briefly review the works in each subarea.

    2.1 Cross-Domain Recommendation

    By leveraging relevant source domain as auxiliary information, a surge of solutions are proposed to address the data sparsity and cold-start problems for the target domain. At the very beginning, CMF [25] proposes to achieve knowledge integration across domains by concatenating multiple rating matrices and sharing user factors across domains. Then Temporal-Domain CF [18] shares the static group-level rating matrix across temporal domains. Later, CDTF [16] is proposed to capture the triadic relation of user-itemdomain by tensor factorization. These collaborative filtering based works suffers severely from the data sparsity problem when considering different domains as a whole.
Funding
  • This work was supported by Alibaba Group through Alibaba Innovative Research Program and National Natural Science Foundation of China (No 61872278)
Reference
  • Konstantin Bauman, Bing Liu, and Alexander Tuzhilin. 2017. Aspect Based Recommendations: Recommending Items with the Most Valuable Aspects Based on User Reviews. In KDD. 717–725.
    Google ScholarLocate open access versionFindings
  • Xu Chen, Zheng Qin, Yongfeng Zhang, and Tao Xu. 2016. Learning to Rank Features for Recommendation over Multiple Categories. In SIGIR. 305–314.
    Google ScholarLocate open access versionFindings
  • Zhiyong Cheng, Ying Ding, Lei Zhu, and Mohan S. Kankanhalli. 2018. AspectAware Latent Factor Model: Rating Prediction with Ratings and Reviews. In WWW. 639–648.
    Google ScholarLocate open access versionFindings
  • Jin Yao Chin, Kaiqi Zhao, Shafiq R. Joty, and Gao Cong. 2018. ANR: Aspect-based Neural Recommender. In CIKM. 147–156.
    Google ScholarLocate open access versionFindings
  • Qiming Diao, Minghui Qiu, Chao-Yuan Wu, Alexander J. Smola, Jing Jiang, and Chong Wang. 2014. Jointly modeling aspects, ratings and sentiments for movie recommendation (JMARS). In KDD. 193–202.
    Google ScholarLocate open access versionFindings
  • Xin Dong, Lei Yu, Zhonghuo Wu, Yuxia Sun, Lingfeng Yuan, and Fangxi Zhang. 2017. A Hybrid Collaborative Filtering Model with Deep Structure for Recommender Systems. In AAAI. 1309–1315.
    Google ScholarFindings
  • Ali Mamdouh Elkahky, Yang Song, and Xiaodong He. 2015. A Multi-View Deep Learning Approach for Cross Domain User Modeling in Recommendation Systems. In WWW. 278–288.
    Google ScholarLocate open access versionFindings
  • Aleksandr Farseev, Ivan Samborskii, Andrey Filchenkov, and Tat-Seng Chua. 2017. Cross-Domain Recommendation via Clustering on Multi-Layer Graphs. In SIGIR. 195–204.
    Google ScholarLocate open access versionFindings
  • Wenjing Fu, Zhaohui Peng, Senzhang Wang, Yang Xu, and Jin Li. 201Deeply Fusing Reviews and Contents for Cold Start Users in Cross-Domain Recommendation Systems. In AAAI. 94–101.
    Google ScholarLocate open access versionFindings
  • Chen Gao, Xiangning Chen, Fuli Feng, Kai Zhao, Xiangnan He, Yong Li, and Depeng Jin. 2019. Cross-domain Recommendation Without Sharing User-relevant Data. In WWW. 491–502.
    Google ScholarFindings
  • Ruining He and Julian McAuley. 2016. Ups and Downs: Modeling the Visual Evolution of Fashion Trends with One-Class Collaborative Filtering. In WWW. 507–517.
    Google ScholarLocate open access versionFindings
  • Xiangnan He, Tao Chen, Min-Yen Kan, and Xiao Chen. 2015. TriRank: Reviewaware Explainable Recommendation by Modeling Aspects. In CIKM. 1661–1670.
    Google ScholarFindings
  • Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang. 2020. Simplifying and Powering Graph Convolution Network for Recommendation. In SIGIR.
    Google ScholarFindings
  • Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 2017. Neural Collaborative Filtering. In WWW. 173–182.
    Google ScholarLocate open access versionFindings
  • Guangneng Hu, Yu Zhang, and Qiang Yang. 2018. CoNet: Collaborative Cross Networks for Cross-Domain Recommendation. In CIKM. 667–676.
    Google ScholarLocate open access versionFindings
  • Liang Hu, Jian Cao, Guandong Xu, Longbing Cao, Zhiping Gu, and Can Zhu. 2013. Personalized recommendation via cross-domain triadic factorization. In WWW. 595–606.
    Google ScholarLocate open access versionFindings
  • SeongKu Kang, Junyoung Hwang, Dongha Lee, and Hwanjo Yu. 2019. SemiSupervised Learning for Cross-Domain Recommendation to Cold-Start Users. In CIKM. 1563–1572.
    Google ScholarFindings
  • Bin Li, Xingquan Zhu, Ruijiang Li, Chengqi Zhang, Xiangyang Xue, and Xindong Wu. 2011. Cross-Domain Collaborative Filtering over Time. In IJCAI. 2293–2298.
    Google ScholarFindings
  • Chenliang Li, Cong Quan, Li Peng, Yunwei Qi, Yuming Deng, and Libing Wu. 20A Capsule Network for Recommendation and Explaining What You Like and Dislike. In SIGIR. 275–284.
    Google ScholarLocate open access versionFindings
  • Donghua Liu, Jing Li, Bo Du, Jun Chang, and Rong Gao. 2019. DAML: Dual Attention Mutual Learning between Ratings and Reviews for Item Recommendation. In KDD. 344–352.
    Google ScholarLocate open access versionFindings
  • Muyang Ma, Pengjie Ren, Yujie Lin, Zhumin Chen, Jun Ma, and Maarten de Rijke. 2019. π -Net: A Parallel Information-sharing Network for Shared-account Cross-domain Sequential Recommendations. In SIGIR. 685–694.
    Google ScholarLocate open access versionFindings
  • Tong Man, Huawei Shen, Xiaolong Jin, and Xueqi Cheng. 2017. Cross-Domain Recommendation: An Embedding and Mapping Approach. In IJCAI. 2464–2470.
    Google ScholarFindings
  • Tomas Mikolov, Ilya Sutskever, Kai Chen, Gregory S. Corrado, and Jeffrey Dean. 2013. Distributed Representations of Words and Phrases and their Compositionality. In NIPS. 3111–3119.
    Google ScholarFindings
  • Sungyong Seo, Jing Huang, Hao Yang, and Yan Liu. 2017. Interpretable Convolutional Neural Networks with Dual Local and Global Attention for Review Rating Prediction. In RecSys. 297–305.
    Google ScholarFindings
  • Ajit Paul Singh and Geoffrey J. Gordon. 2008. Relational learning via collective matrix factorization. In KDD. 650–658.
    Google ScholarLocate open access versionFindings
  • Tianhang Song, Zhaohui Peng, Senzhang Wang, Wenjing Fu, Xiaoguang Hong, and Philip S. Yu. 2017. Review-Based Cross-Domain Recommendation Through Joint Tensor Factorization. In DASFAA (Lecture Notes in Computer Science), Vol. 10177. 525–540.
    Google ScholarLocate open access versionFindings
  • Yi Tay, Anh Tuan Luu, and Siu Cheung Hui. 2018. Multi-Pointer Co-Attention Networks for Recommendation. In KDD. 2309–2318.
    Google ScholarFindings
  • Nan Wang, Hongning Wang, Yiling Jia, and Yue Yin. 2018. Explainable Recommendation via Multi-Task Learning in Opinionated Text Data. In SIGIR. 165–174.
    Google ScholarFindings
  • Xinghua Wang, Zhaohui Peng, Senzhang Wang, Philip S. Yu, Wenjing Fu, and Xiaoguang Hong. 2018. Cross-Domain Recommendation for Cold-Start Users via Neighborhood Based Feature Mapping. In DASFAA. 158–165.
    Google ScholarLocate open access versionFindings
  • Yaqing Wang, Chunyan Feng, Caili Guo, Yunfei Chu, and Jenq-Neng Hwang. 2019. Solving the Sparsity Problem in Recommendations via Cross-Domain Item Embedding Based on Co-Clustering. In WSDM. 717–725.
    Google ScholarLocate open access versionFindings
  • Libing Wu, Cong Quan, Chenliang Li, and Donghong Ji. 2018. PARL: Let Strangers Speak Out What You Like. In CIKM. 677–686.
    Google ScholarLocate open access versionFindings
  • Libing Wu, Cong Quan, Chenliang Li, Qian Wang, Bolong Zheng, and Xiangyang Luo. 2019. A Context-Aware User-Item Representation Learning for Item Recommendation. ACM Trans. Inf. Syst. 37, 2 (2019), 22:1–22:29.
    Google ScholarLocate open access versionFindings
  • Yao Wu and Martin Ester. 2015. FLAME: A Probabilistic Model Combining Aspect Based Opinion Mining and Collaborative Filtering. In WSDM. 199–208.
    Google ScholarFindings
  • Haifeng Xia, Zengmao Wang, Bo Du, Lefei Zhang, Shuai Chen, and Gang Chun. 2019. Leveraging Ratings and Reviews with Gating Mechanism for Recommendation. In CIKM. 1573–1582.
    Google ScholarFindings
  • Feng Yuan, Lina Yao, and Boualem Benatallah. 2019. DARec: Deep Domain Adaptation for Cross-Domain Recommendation via Transferring Rating Patterns. In IJCAI. 4227–4233.
    Google ScholarFindings
  • Yang Zhang, Fuli Feng, Chenxu Wang, Xiangnan He, Meng Wang, Yan Li, and Yongdong Zhang. 2020. How to Retrain a Recommender System? A Sequential Meta-Learning Approach. In SIGIR.
    Google ScholarFindings
  • Yongfeng Zhang, Guokun Lai, Min Zhang, Yi Zhang, Yiqun Liu, and Shaoping Ma. 2014. Explicit factor models for explainable recommendation based on phrase-level sentiment analysis. In SIGIR. 83–92.
    Google ScholarLocate open access versionFindings
  • Cheng Zhao, Chenliang Li, and Cong Fu. 2019. Cross-Domain Recommendation via Preference Propagation GraphNet. In CIKM. 2165–2168.
    Google ScholarFindings
  • Lei Zheng, Vahid Noroozi, and Philip S. Yu. 2017. Joint Deep Modeling of Users and Items Using Reviews for Recommendation. In WSDM. 425–434.
    Google ScholarLocate open access versionFindings
  • Feng Zhu, Yan Wang, Chaochao Chen, Guanfeng Liu, Mehmet A. Orgun, and Jia Wu. 2018. A Deep Framework for Cross-Domain and Cross-System Recommendations. In IJCAI. 3711–3717.
    Google ScholarFindings
Full Text
Your rating :
0

 

Tags
Comments