GeM: Gaussian embeddings with Multi-hop graph transfer for next POI recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• Next Point-of-Interest (POI) recommendation is vital for location-based services, relying on historical user trajectory analysis.
• Current methods often use graph structures and attention mechanisms but overlook multi-hop dependencies in user movement patterns.

Purpose of the Study:

• To propose GeM, a novel framework for accurate next POI recommendation.
• To address limitations of existing methods by incorporating multi-hop graph relations and user preferences.

Main Methods:

• GeM utilizes Gaussian distribution and Mahalanobis distance for subjective user preference modeling.
• An objective module mines global trajectory graph information and multi-hop dependencies.
• Matrix factorization is employed to learn user-POI interactions.

Main Results:

• The proposed GeM framework effectively captures complex user movement patterns.
• Experiments on real-world datasets demonstrate superior performance compared to state-of-the-art methods.
• GeM achieves more accurate user behavior pattern representation.

Conclusions:

• GeM provides a unified approach to next POI recommendation by integrating subjective and objective factors.
• The framework's ability to model multi-hop dependencies enhances prediction accuracy.
• GeM represents a significant advancement in location-based recommendation systems.