Skip to main content
QUICK REVIEW

[论文解读] Dual Metric Learning for Effective and Efficient Cross-Domain Recommendations

Li Pan, Alexander Tuzhilin|arXiv (Cornell University)|Apr 17, 2021
Recommender Systems and Techniques参考文献 71被引用 14
一句话总结

本文提出了一种新型的跨域推荐模型——双度量学习(Dual Metric Learning, DML),通过结合双学习与度量学习,仅需极少重叠用户即可提升推荐性能。通过学习域之间的潜在正交映射,并通过双向迁移迭代优化用户与物品嵌入,DML 在仅有 8 名重叠用户的情况下仍能达到最先进性能。

ABSTRACT

Cross domain recommender systems have been increasingly valuable for helping consumers identify useful items in different applications. However, existing cross-domain models typically require large number of overlap users, which can be difficult to obtain in some applications. In addition, they did not consider the duality structure of cross-domain recommendation tasks, thus failing to take into account bidirectional latent relations between users and items and achieve optimal recommendation performance. To address these issues, in this paper we propose a novel cross-domain recommendation model based on dual learning that transfers information between two related domains in an iterative manner until the learning process stabilizes. We develop a novel latent orthogonal mapping to extract user preferences over multiple domains while preserving relations between users across different latent spaces. Furthermore, we combine the dual learning method with the metric learning approach, which allows us to significantly reduce the required common user overlap across the two domains and leads to even better cross-domain recommendation performance. We test the proposed model on two large-scale industrial datasets and six domain pairs, demonstrating that it consistently and significantly outperforms all the state-of-the-art baselines. We also show that the proposed model works well with very few overlap users to obtain satisfying recommendation performance comparable to the state-of-the-art baselines that use many overlap users.

研究动机与目标

  • 解决现有跨域推荐模型在有效运行时需要大量重叠用户的问题。
  • 通过在源域与目标域之间实现用户偏好双向迁移,利用跨域推荐任务的对偶性。
  • 通过度量学习识别跨域中具有相似偏好的用户,降低对显式重叠用户的依赖。
  • 构建一个统一框架,结合双学习与度量学习,以提升推荐性能,同时最小化数据需求。
  • 在具有不同域对的真实世界工业数据集上,验证所提模型的有效性与效率。

提出的方法

  • 提出一种双学习机制,通过在两个域之间双向迭代传递用户偏好信息,以提升两个域的性能。
  • 引入一种新颖的潜在正交映射,以保留跨域用户偏好相似性,并在双向迁移过程中实现高效的逆映射。
  • 将度量学习整合到双学习框架中,以学习一个共享的潜在空间,使用户与物品嵌入能同时反映偏好关系与相似性关系。
  • 使用联合损失函数在两个域上联合优化模型,以最小化预测误差,同时在潜在空间中强制执行正交性与度量一致性。
  • 使用批量归一化与分布式训练,以确保在迭代训练过程中的可扩展性与高效优化。
  • 通过迭代优化直至收敛,训练损失在 10 个周期内稳定,确保快速且稳定的训练。
Figure 1 : Dual Metric Learning for Cross-Domain Recommendations
Figure 1 : Dual Metric Learning for Cross-Domain Recommendations

实验结果

研究问题

  • RQ1双学习能否有效应用于跨域推荐,以实现用户偏好在双向之间的迁移?
  • RQ2度量学习能否通过识别跨域中相似的用户偏好,降低对大量重叠用户的依赖?
  • RQ3双学习与度量学习的结合是否能相比最先进基线模型带来更优的推荐性能?
  • RQ4所提模型对重叠用户数量的敏感性如何?是否能在极少数重叠用户下实现强性能?
  • RQ5在大规模数据集的真实训练条件下,所提模型是否具备可扩展性与收敛性?

主要发现

  • 所提出的 DML 模型在两个大规模工业数据集的六组域对上,始终显著优于所有最先进基线模型。
  • 即使仅有 8 名重叠用户,DML 仍能达到与 SOTA 模型相当的推荐性能,证明其在低重叠场景下的有效性。
  • 模型收敛迅速,训练损失在 10 个周期内即稳定,表明优化过程快速且稳定。
  • 模型随数据规模线性扩展,在训练多达 1,000,000 条记录时仍保持高效。
  • 消融实验表明,模型在仅 8 名重叠用户时表现良好,并在不同潜在维度大小(如 16–32 维)下均保持鲁棒性。
  • 潜在正交映射能有效保留跨域用户偏好相似性,从而实现准确且高效的跨域迁移。
Figure 2 : Dual Metric Learning
Figure 2 : Dual Metric Learning

更好的研究,从现在开始

从阅读论文到最终审阅,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。