POI recommendation with queuing time and user interest awareness

Area of Science:

• Artificial Intelligence
• Data Science
• Human-Computer Interaction

Background:

• Point-of-interest (POI) recommendation systems often overlook crucial factors like user personal interests and queuing times.
• Existing attention-based models struggle with complex relationships between spatial data, user behavior, and POI attributes.

Purpose of the Study:

• To develop a novel model that incorporates user interests and queuing time predictions for more accurate next POI recommendations.
• To address the limitations of single-head attention mechanisms in capturing intricate user mobility patterns.

Main Methods:

• Proposed a multi-task, multi-head attention transformer model (TLR-M_UI) for simultaneous POI recommendation and queuing time prediction.
• Utilized POI description-based user interest modeling to address the cold-start problem for new POIs.
• Leveraged extensive experiments on six real-world datasets to validate the model's performance.

Main Results:

• The TLR-M_UI model significantly outperforms state-of-the-art baseline approaches in precision, recall, and F1-score.
• Demonstrated the model's capability to predict and minimize queuing times, enhancing user experience.
• Successfully addressed the cold-start problem by incorporating POI descriptions for interest modeling.

Conclusions:

• Integrating user interests and queuing time predictions offers a more effective approach to POI recommendation.
• The proposed multi-head attention transformer model provides a robust solution for complex mobility behavior analysis.
• The study offers a reproducible solution with publicly available code for future research.