A Common Topic Transfer Learning Model for Crossing City POI Recommendations.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• Location-aware devices generate vast amounts of user check-in data, fueling research in personalized Point of Interest (POI) recommendations.
• Existing recommendation systems struggle with cross-city scenarios, failing to suggest POIs in new urban environments based on past user behavior.

Purpose of the Study:

• To develop a novel machine learning model for effective Point of Interest (POI) recommendations in new cities.
• To address the challenge of transferring user preferences across different geographical locations in recommendation systems.

Main Methods:

• Proposed the Common-Topic Transfer Learning Model (CTLM), a graphical model designed for cross-city POI recommendations.
• Separated city-specific features from common features to enable accurate transfer of user interests.
• Incorporated spatial influence by considering regional accessibility to model user-POI co-occurrence patterns.

Main Results:

• The CTLM model demonstrated superior performance in cross-city POI recommendation tasks.
• Experimental results on Foursquare and Twitter datasets validated the model's effectiveness over state-of-the-art methods.
• The model successfully mitigated the ill-matching problem by distinguishing common user interests from city-specific attributes.

Conclusions:

• The proposed CTLM effectively enables personalized POI recommendations in previously unvisited cities.
• Transfer learning, by separating common and city-specific features, is a viable approach for cross-city recommendation challenges.
• Integrating spatial factors enhances the accuracy and relevance of POI recommendations across diverse urban environments.