1.National Institute of Excellence Engineers, Zhejiang University, Hangzhou 310015, China
2.Hello Inc, Shanghai, 201199, China
3.School of Computer Science and Technology, Dalian University of Technology, Dalian 116024, China
4.School of Public Administration and Policy, Dalian University of Technology, Dalian 116024, China
Show less
文章历史+
Received
Accepted
Published
2025-09-20
2025-11-17
2026-05-25
Issue Date
2026-08-25
PDF (1878K)
摘要
为解决现有基于映射的跨域推荐算法中,兴趣迁移过程忽略目标域情境信息、模型性能对重叠用户规模高度敏感两个问题,该文提出一种端到端优化的基于跨域元学习的兴趣迁移网络(Cross-domain Meta-learning Driven Interest Transfer Network,CMITN)。CMITN引入目标域感知的跨域注意力机制,依据候选物品特性重构源域兴趣表示;设计卷积元学习网络动态生成个性化映射函数,缓解数据稀疏依赖;构建双目标联合学习范式,同时优化推荐任务与表示对齐,实现全域兴趣空间的一致性映射。利用亚马逊与哈啰两类数据集构建了5个跨域推荐任务并进行广泛实验,实验结果表明,该文提出的算法在AUC(Area Under Curve)与Recall两个指标上均优于其他基线模型。在哈啰生活服务业务场景下,该算法对新用户带来了4.5%的点击率(Click Through Rate, CTR)提升与7.0%的人均商品交易总额(Gross Merchandise Volume, GMV)提升。
Abstract
To address two issues in existing mapping-based cross-domain recommendation algorithms—namely, the neglect of target domain contextual information during interest transfer and the model's high sensitivity to the scale of overlapping users—a novel end-to-end optimized Cross-domain Meta-learning Driven Interest Transfer Network (CMITN) is proposed. CMITN introduces a cross-domain attention mechanism that is sensitive to the target domain, which reconstructs the source domain's interest representaion based on candidate item characteristics. A convolutional meta-learning network is designed to dynamically generate personalized mapping functions, alleviating data sparsity dependence. A dual-objective joint optimization paradigm is constructed to optimize both the recommendation task and the representation alignment, ensuring a consistent mapping of the full interest space. Five cross-domain recommendation tasks were constructed using Amazon and Hello datasets, followed by extensive experiments. The experimental results show that the proposed algorithm outperforms other baseline models in both AUC (Area Under Curve) and Recall. In the business scenario of Hello Life Services, the algorithm led to a 4.5% increase in CTR (Click Through Rate) and a 7.0% improvement in GMV (Gross Merchandise Volume) per user for new users.
在此背景下,跨域推荐(Cross-domain Recommendation, CDR)成为解决用户冷启动问题的有效路径[7]。其核心思想是通过用户在相关源域(如电影评分)的丰富行为数据,为目标域(如书籍推荐)的冷启动用户提供迁移知识支持。在诸多相关研究中,基于域间嵌入映射的方法通过学习映射函数,将冷启动用户映射到目标域特征空间(如Embedding and Mapping Framework for Cross-domain Recommendation, EMCDR[8]),取得了不错的效果[9-10]。然而,此类方法高度依赖域间重叠用户的数量,当重叠用户不足时,映射函数的泛化能力急剧下降[11]。此外,当前方法(如Personalized Transfer of User Preferences for Cross-domain Recommendation, PTUPCDR[12])多将用户兴趣压缩为单一静态向量,忽略了行为序列的时序特征与目标域候选物品的动态相关性,导致兴趣迁移的粒度粗糙、灵活性不足。
基于此,本文提出一种端到端优化的基于跨域元学习的兴趣迁移网络(Cross-domain Meta-learning Driven Interest Transfer Network,CMITN)。具体而言,引入目标域候选物品驱动的注意力模块,自适应调整源域历史行为的权重,实现细粒度兴趣表征;同时,设计卷积元学习网络(Conv-Meta),通过卷积层捕捉用户行为序列的局部时序模式,结合元学习动态生成跨领域映射函数,减少对重叠用户数据的依赖;最后,融合任务导向损失(如点击率预测)与映射约束损失,通过协同训练提升模型全局优化能力,避免分阶段优化的误差累积。在公开数据集与工业数据集上的实验表明,该方法能够有效建模用户源域到目标域的兴趣迁移,缓解用户冷启动问题。
MAOQ, XIEW C, QIAOY T, et al. Survey on Solving Cold Start Problem in Recommendation Systems[J]. J Front Comput Sci Technol, 2024, 18(5): 1197-1210. DOI: 10.3778/j.issn.1673-9418.2308044 .
[4]
LIL H, CHUW, LANGFORDJ, et al. A Contextual-bandit Approach to Personalized News Article Recommendation[C]//Proceedings of the 19th International Conference on World Wide Web. Raleigh: ACM, 2010: 661-670. DOI: 10.1145/1772690.1772758 .
[5]
PANF Y, CAIQ P, TANGP Z, et al. Policy Gradients for Contextual Recommendations[C]//The World Wide Web Conference. New York: ACM, 2019: 1421-1431. DOI: 10.1145/3308558.3313616 .
[6]
YAOH X, LIUY D, WEIY, et al. Learning from Multiple Cities: A Meta-learning Approach for Spatial-temporal Prediction[C]//The World Wide Web Conference. New York: ACM, 2019: 2181-2191. DOI: 10.1145/3308558.3313577 .
[7]
DONGM Q, YUANF, YAOL N, et al. MAMO: Memory-augmented Meta-optimization for Cold-start Recommendation[C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. New York: ACM, 2020: 688-697. DOI: 10.1145/3394486.3403113 .
[8]
ZANGT Z, ZHUY M, LIUH B, et al. A Survey on Cross-domain Recommendation: Taxonomies, Methods, and Future Directions[J]. ACM Trans Inf Syst, 2023, 41(2): 1-39. DOI: 10.1145/3548455 .
[9]
MANT, SHENH W, JINX L, et al. Cross-domain Recommendation: An Embedding and Mapping Approach[C]//International Joint Conference on Artificial Intelligence. Melbourne: IJCAI, 2017, 17: 2464-2470. DOI: 10.24963/ijcai.2017/343 .
[10]
WANGJ, YUANF J, CHENGM Y, et al. TransRec: Learning Transferable Recommendation from Mixture-of-modality Feedback[M]//Web and Big Data. Singapore: Springer Nature Singapore, 2024: 193-208. DOI: 10.1007/978-981-97-7235-3_13 .
XUJ, WANGX, WANGY Q, et al. Cross-domain Recommendation Method Based on Aggregation of Intra-domain and Inter-domain Meta-paths[J]. Appl Res Comput, 2025, 42(8): 2374-2382. DOI: 10.19734/j.issn.1001-3695.2025.01.0022 .
[13]
KANGS, HWANGJ, LEED H, et al. Semi-supervised Learning for Cross-domain Recommendation to Cold-start Users[C]//Proceedings of the 28th ACM International Conference on Information and Knowledge Management. Beijing: ACM, 2019: 1563-1572. DOI: 10.1145/3357384.3357914 .
[14]
ZHUY C, TANGZ W, LIUY D, et al. Personalized Transfer of User Preferences for Cross-domain Recommendation[C]//Proceedings of the Fifteenth ACM International Conference on Web Search and Data Mining. New York: ACM, 2022: 1507-1515. DOI: 10.1145/3488560.3498392 .
[15]
XIER B, QIUZ J, RAOJ, et al. Internal and Contextual Attention Network for Cold-start Multi-channel Matching in Recommendation[C]//Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence. International Joint Conferences on Artificial Intelligence Organization. Yokohama: IJCAI, 2020: 2732-2738. DOI: 10.24963/ijcai.2020/379 .
WANGY G, LIUD N. Cross-domain Recommendation Algorithm Combining Multi-personalized Bridges and Self-supervised Learning[J]. J Front Comput Sci Technol, 2024, 18(7): 1792-1805. DOI: 10.3778/j.issn.1673-9418.2305022 .
[20]
SINGHA P, GORDONG J. Relational Learning via Collective Matrix Factorization[C]//Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Las Vegas: ACM, 2008: 650-658. DOI: 10.1145/1401890.1401969 .
[21]
PANW K, XIANGE, LIUN, et al. Transfer Learning in Collaborative Filtering for Sparsity Reduction[J]. Proc AAAI Conf Artif Intell, 2010, 24(1): 230-235. DOI: 10.1609/aaai.v24i1.7578 .
[22]
ZHUY C, XIER B, ZHUANGF Z, et al. Learning to Warm up Cold Item Embeddings for Cold-start Recommendation with Meta Scaling and Shifting Networks[C]//Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval. New York: ACM, 2021: 1167-1176. DOI: 10.1145/3404835.3462843 .
[23]
HAOX B, LIUY D, XIER B, et al. Adversarial Feature Translation for Multi-domain Recommendation[C]//Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. New York: ACM, 2021: 2964-2973. DOI: 10.1145/3447548.3467176 .
LIUY F, WANGS Q, ZHENGS, et al. Cold-start User Representation Learning Method Based on Cross-domain Meta-learning Framework[J]. J Shandong Univ Eng Sci, 2024, 54(6): 29-37. DOI: 10.6040/j.issn.1672-3961.0.2023.100 .
[26]
HUANGL, LIW T, ZHANGC R, et al. EXIT: an EXplicit Interest Transfer Framework for Cross-domain Recommendation[C]//Proceedings of the 33rd ACM International Conference on Information and Knowledge Management. New York: ACM, 2024: 4563-4570. DOI: 10.1145/3627673.3680055 .
WUG D, LIUX X, BIH J, et al. Review of Personalized Recommendation Research Based on Meta-learning[J]. Comput Eng Sci, 2024, 46(2): 338-352. DOI: 10.3969/j.issn.1007-130X.2024.02.016 .
[29]
HOSPEDALEST, ANTONIOUA, MICAELLIP, et al. Meta-learning in Neural Networks: a Survey[J]. IEEE Trans Pattern Anal Mach Intell, 2022, 44(9): 5149-5169. DOI: 10.1109/tpami.2021.3079209 .
[30]
RAVIS, LAROCHELLEH. Optimization as a Model for Few-shot Learning[C]//International Conference on Learning Representations. Singapore: OpenReview, 2017.
[31]
KOCHG, ZEMELR, SalakhutdinovR. Siamese Neural Networks for One-shot Image Recognition[C]//ICML Deep Learning Workshop. Lille: JMLR, 2017, 2(1): 1-30.
[32]
LEEH, IM J, JANGS, et al. MeLU: Meta-learned User Preference Estimator for Cold-start Recommendation[C]//Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. New York: ACM, 2019: 1073-1082. DOI: 10.1145/3292500.3330859 .
[33]
ZHOUG R, ZHUX Q, SONGC R, et al. Deep Interest Network for Click-through Rate Prediction[C]//Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. New York: ACM, 2018: 1059-1068. DOI: 10.1145/3219819.3219823 .
[34]
NIJ M, LIJ C, MCAULEYJ. Justifying Recommendations Using Distantly-Labeled Reviews and Fine-grained Aspects[C]//Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). Stroudsburg: ACL, 2019: 188-197. DOI: 10.18653/v1/d19-1018 .
RENDLES, FREUDENTHALERC, GANTNERZ, et al. BPR: Bayesian Personalized Ranking from Implicit Feedback[EB/OL]. (2012-05-09)[2025-06-01].
[37]
HSIEHC K, YANGL Q, CUIY, et al. Collaborative Metric Learning[C]//Proceedings of the 26th International Conference on World Wide Web. Perth: International World Wide Web Conferences Steering Committee, 2017: 193-201. DOI: 10.1145/3038912.3052639 .