Reproduce “MISSRec: Pre-training and Transferring Multi-modal Interest-aware Sequence Representation for Recommendation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jinpeng Wang, Ziyun Zeng, Yunxiao Wang, Yuting Wang, Xingyu Lu, Tianxiang Li, Jun Yuan, Rui Zhang and 2 more

MISSRec, developed by Tsinghua University and Huawei Noah’s Ark Lab, presents a pre-training and transfer learning framework that leverages multi-modal item content (text and images) to improve sequential recommendation. It addresses data sparsity, cold-start items, and domain transferability by dynamically fusing multi-modal information and discovering interest-aware sequence representations, consistently outperforming state-of-the-art baselines with up to 21.23% relative improvement on Recall@10 in transductive settings.

Key insights

  • Multi-modal content features (text and images) provide robust, ID-agnostic representations that effectively mitigate data sparsity and cold-start problems in sequential recommendation.
  • Dynamic multi-modal fusion and an interest-aware Transformer encoder-decoder are crucial for capturing nuanced user-specific modality preferences and processing complex interaction sequences by focusing on core interests.
  • Parameter-efficient pre-training with multi-modal features enables effective knowledge transfer across diverse domains, especially in transductive settings where item IDs are available.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.