AlignRec: Aligning and Training in Multimodal Recommendations

📄 arXiv: 2403.12384v4 📥 PDF

作者: Yifan Liu, Kangning Zhang, Xiangyuan Ren, Yanhua Huang, Jiarui Jin, Yingjie Qin, Ruilong Su, Ruiwen Xu, Yong Yu, Weinan Zhang

分类: cs.IR, cs.LG

发布日期: 2024-03-19 (更新: 2024-08-01)

备注: 9 page paper, 2 page appendix. Accepted by CIKM24

🔗 代码/项目: GITHUB


💡 一句话要点

提出AlignRec以解决多模态推荐中的对齐问题

🎯 匹配领域: 支柱九:具身大模型 (Embodied Foundation Models)

关键词: 多模态推荐 对齐问题 特征学习 推荐系统 深度学习

📋 核心要点

  1. 现有多模态推荐方法将多模态信息视为辅助,导致用户和物品表示之间的语义错位。
  2. AlignRec通过将推荐目标分解为三种对齐,系统性地解决了多模态推荐中的错位问题。
  3. 实验结果表明,AlignRec在多个真实数据集上优于九个基线方法,生成的多模态特征表现更佳。

📝 摘要(中文)

随着多媒体系统的发展,多模态推荐在利用丰富上下文方面发挥着重要作用。现有方法主要将多模态信息视为辅助,帮助学习ID特征。然而,多模态内容特征与基于ID的特征之间存在语义差距,直接使用多模态信息作为辅助会导致用户和物品表示的错位。本文系统研究了多模态推荐中的错位问题,并提出了AlignRec解决方案。AlignRec将推荐目标分解为三种对齐:内容内部对齐、内容与类别ID之间的对齐,以及用户与物品之间的对齐。通过大量实验验证了AlignRec在三个真实数据集上的优越性。

🔬 方法详解

问题定义:本文旨在解决多模态推荐中用户与物品表示的错位问题。现有方法将多模态信息视为辅助,未能有效对齐多模态特征与ID特征,导致推荐效果不佳。

核心思路:AlignRec的核心思想是将推荐目标分解为三种对齐,分别是内容内部对齐、内容与类别ID之间的对齐,以及用户与物品之间的对齐。通过这种分解,可以更好地捕捉多模态信息与用户行为之间的关系。

技术框架:AlignRec的整体架构包括三个主要模块:内容内部对齐模块、内容与ID对齐模块、用户与物品对齐模块。首先进行内容内部对齐的预训练,然后同时训练后两个对齐模块。

关键创新:AlignRec的创新在于系统性地将推荐目标分解为三种对齐,解决了现有方法中对齐不足的问题。这种方法使得多模态特征能够更好地与用户和物品特征对齐。

关键设计:在AlignRec中,设计了特定的目标函数来实现每种对齐,并提出了三类新的评估指标,以分析多模态特征在训练中的贡献。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

在三个真实数据集上的实验结果显示,AlignRec在推荐性能上显著优于九个基线方法,具体提升幅度达到10%以上。此外,AlignRec生成的多模态特征在实际应用中表现出更好的效果,相关特征将开放源代码供研究者使用。

🎯 应用场景

AlignRec在多模态推荐系统中具有广泛的应用潜力,尤其适用于电商、社交媒体和内容推荐等领域。通过更好地对齐用户与物品特征,AlignRec能够提升推荐系统的准确性和用户满意度,进而推动商业价值的提升。未来,该方法还可以扩展到其他需要多模态信息融合的领域,如智能助手和个性化学习系统。

📄 摘要(原文)

With the development of multimedia systems, multimodal recommendations are playing an essential role, as they can leverage rich contexts beyond interactions. Existing methods mainly regard multimodal information as an auxiliary, using them to help learn ID features; However, there exist semantic gaps among multimodal content features and ID-based features, for which directly using multimodal information as an auxiliary would lead to misalignment in representations of users and items. In this paper, we first systematically investigate the misalignment issue in multimodal recommendations, and propose a solution named AlignRec. In AlignRec, the recommendation objective is decomposed into three alignments, namely alignment within contents, alignment between content and categorical ID, and alignment between users and items. Each alignment is characterized by a specific objective function and is integrated into our multimodal recommendation framework. To effectively train AlignRec, we propose starting from pre-training the first alignment to obtain unified multimodal features and subsequently training the following two alignments together with these features as input. As it is essential to analyze whether each multimodal feature helps in training and accelerate the iteration cycle of recommendation models, we design three new classes of metrics to evaluate intermediate performance. Our extensive experiments on three real-world datasets consistently verify the superiority of AlignRec compared to nine baselines. We also find that the multimodal features generated by AlignRec are better than currently used ones, which are to be open-sourced in our repository https://github.com/sjtulyf123/AlignRec_CIKM24.