Reproduce “STGIN: Spatial-Temporal Graph Interaction Network for Large-scale POI 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
Shaohua Liu, Yu Qi, Gen Li, Mingjian Chen, Teng Zhang, Jia Cheng, Jun Lei
Researchers from Meituan developed STGIN (Spatial-Temporal Graph Interaction Network), a machine learning model that enhances Point-of-Interest recommendation systems by capturing dynamic user preferences through a multi-view graph approach and real-time behavior integration. The model achieved gains in HitRate and Recall in experiments and demonstrated improvements of 1.1% in Click-Through Rate and 6.3% in Revenue Per Mille during its deployment in a large-scale production environment.
Key insights
- User preferences for Point-of-Interest (POI) recommendations are highly dynamic and context-dependent, varying across different spatial and temporal settings.
- A multi-view graph approach, separating global, spatial, temporal, and spatial-temporal user behavior, allows for more accurate context-aware modeling.
- Integrating real-time user interactions with attention mechanisms enables rapid adaptation to evolving preferences and trends.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.