Reproduce “MoE-TransMov: A Transformer-based Model for Next POI Prediction in Familiar & Unfamiliar Movements” 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
Ruichen Tan, Jiawei Xue, Kota Tsubouchi, Takahiro Yabe, Satish V. Ukkusuri
The MoE-TransMov model, a Transformer-based architecture enhanced with a Mixture-of-Experts framework, predicts a user's next Point of Interest by adaptively modeling movements in both familiar and unfamiliar regions. It achieved superior prediction accuracy, particularly for less predictable unfamiliar movements, on a large-scale real-world dataset.
Key insights
- Human mobility patterns are fundamentally different in familiar versus unfamiliar regions, necessitating context-aware modeling.
- A Mixture-of-Experts framework effectively addresses this heterogeneity by allowing specialized sub-models to adapt to different movement contexts.
- Combining a Transformer for sequence modeling with heterogeneous experts (LSTM and Transformer) within the MoE framework maximizes prediction performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.