COSTA: Contrastive Spatial and Temporal Debiasing framework for next POI recommendation.

Area of Science:

• Data Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Next point-of-interest (POI) recommendation systems analyze user mobility trajectories.
• Existing methods often introduce spatial and temporal biases, misaligning recommendations with user preferences.
• These biases negatively impact user experience in location-based services.

Purpose of the Study:

• To reveal and analyze the detrimental effects of spatial and temporal biases in next POI recommendation.
• To propose a novel framework, COSTA, for mitigating these identified biases.
• To introduce new metrics for quantifying spatial and temporal bias severity.

Main Methods:

• Developed the Contrastive Spatial and Temporal Debiasing (COSTA) framework.
• Utilized user- and location-side spatial-temporal signal encoders.
• Employed contrastive learning to align user and POI representations.
• Introduced Discounted Spatial Cumulative Gain (DSCG) and Discounted Temporal Cumulative Gain (DTCG) metrics.

Main Results:

• COSTA effectively mitigates spatial and temporal biases in next POI recommendations.
• The proposed debiasing metrics (DSCG, DTCG) quantify bias severity.
• COSTA outperforms state-of-the-art methods on real-world datasets in debiasing.
• Recommendation accuracy is maintained while reducing bias.

Conclusions:

• Spatial and temporal biases are significant issues in next POI recommendation.
• COSTA offers an effective solution for debiasing recommendation systems.
• The new metrics provide valuable tools for evaluating bias in POI recommendations.