A Learning-Based POI Recommendation With Spatiotemporal Context Awareness.

Area of Science:

• Data Science
• Computer Science
• Human Mobility

Background:

• Advances in mobility techniques have spurred the development of point-of-interest (POI) services.
• Predicting user POIs is challenging due to complex sequential transitions and sparse, heterogeneous trajectory data.
• Existing POI recommendation systems often overlook the subtle time and distance relationships between consecutive POIs.

Purpose of the Study:

• To propose a novel POI recommendation system that captures complex sequential transitions.
• To incorporate time and distance irregularities into POI prediction models.
• To develop a method for dynamically weighting contextual factors in the learning process.

Main Methods:

• Developed a novel POI recommendation system incorporating time and distance irregularities.
• Introduced a dynamic weighting mechanism for decay values during model learning.
• Utilized real-world mobility datasets for performance evaluation.

Main Results:

• The proposed system effectively captures complicated sequential transitions in user movement.
• Learned awareness weights provide interpretable insights into contextual emphasis.
• Demonstrated significant performance improvements over state-of-the-art models across all metrics.

Conclusions:

• The novel POI recommendation system effectively addresses the challenges of sequential transitions and contextual irregularities.
• The dynamic weighting approach enhances model interpretability and prediction accuracy.
• The proposed methods offer a practical and effective solution for POI recommendation in real-world mobility scenarios.