Researchers propose SPACE to fix POI recommendation bias
A new AI model called SPACE boosts visibility for long-tail businesses in location apps
Deep Dive
Researchers introduced SPACE, a model-agnostic framework designed to promote provider fairness in next POI recommendation. SPACE works by generating virtual users under explicit feasibility and supply constraints, helping long-tail POIs gain more exposure. Tested on three real-world datasets, it substantially improved provider fairness while maintaining and often improving recommendation accuracy across multiple backbone models.
Key Points
- SPACE (Supply- and Physics-Aware Conditional Embedding generation) increases long-tail POI exposure by 30% in next POI recommendations
- The framework uses three-stage processing: community inference, optimal-transport allocation, and latent diffusion to generate virtual user embeddings
- Tested across three real-world datasets, SPACE improves provider fairness while maintaining recommendation accuracy across multiple models
Why It Matters
This research could transform how location apps distribute business visibility, creating fairer opportunities for small businesses in digital maps