LARK：面向多模态推荐的潜在对齐推理框架
原标题：Latent-Aligned Reasoning for Multimodal Recommendation
AI 摘要
本文提出 LARK（Latent-Aligned Reasoning frameworK），一种用于多模态推荐的两阶段潜在推理框架，旨在解决视觉-语言模型在多步推理中视觉和文本信号逐渐衰减的“跨模态稀释”问题。LARK 在第一阶段将可学习潜在标记与思维链推理交错，并与冻结的视觉编码器对齐；第二阶段通过桥接 MLP 和对比学习生成嵌入，并将中间特征与思维链隐藏状态对齐。在三个公开基准和一个工业数据集上，LARK 在多种推荐架构上取得了最先进性能。
正文节选
Latent-Aligned Reasoning for Multimodal Recommendation Abstract. Multimodal Vision-Language Models (VLMs) have demonstrated remarkable capabilities in cross-modal understanding, yet a fundamental challenge persists when applying them to recommendation: as representations propagate through multi-step reasoning, both visual and textual signals progressively attenuate — a phenomenon we term cross-modal dilution. To address this, we propose LARK (Latent-Aligned Reasoning frameworK), a two-stage late