Reproduce “Spatial-Temporal Knowledge Distillation for Takeaway 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
Shuyuan Zhao, Wei Chen, Boyan Shi, Liyong Zhou, Shuohao Lin, Huaiyu Wan
Researchers at Beijing Jiaotong University developed STKDRec, a model that improves takeaway recommendations by integrating fine-grained spatial-temporal knowledge through a knowledge distillation framework. This approach consistently outperforms state-of-the-art baselines across three real-world datasets and efficiently transfers complex graph-based insights to a more lightweight sequential Transformer model.
Key insights
- Accurately capturing dynamic user preferences in location-based recommendations requires integrating fine-grained geospatial information (both spatial regions and distances) into sequential models.
- Knowledge distillation provides an effective and computationally efficient mechanism for fusing heterogeneous knowledge by transferring rich graph-based, high-order relationships to a sequence-based Transformer model.
- A two-stage teacher-student framework effectively leverages the strengths of both graph structures (for comprehensive relational knowledge) and sequential models (for dynamic preference modeling) in a synergistic manner.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.