An interpretable framework for investigating the neighborhood effect in POI recommendation

Area of Science:

• Geographic Information Science
• Computer Science
• Data Science

Background:

• Geographical features enhance Point-of-Interest (POI) recommendations.
• Current POI recommendation systems primarily consider travel cost (time/money).
• The neighborhood effect, user preference for POIs within preferred neighborhoods, is understudied.

Purpose of the Study:

• To develop an interpretable framework for studying the neighborhood effect in POI recommendations.
• To identify and represent various aspects of the neighborhood effect.
• To propose a novel deep learning recommendation framework leveraging the neighborhood effect.

Main Methods:

• Feature engineering to represent different facets of the neighborhood effect.
• Utilizing the Yelp dataset for empirical evaluation.
• Developing and applying a deep learning-based recommendation model.

Main Results:

• Analysis of how distinct neighborhood effect features influence user POI visiting patterns.
• Demonstrated superior performance of the proposed deep learning framework.
• Outperformed two state-of-the-art matrix factorization-based POI recommendation techniques.

Conclusions:

• The neighborhood effect is a crucial, underutilized factor in POI recommendation.
• The proposed deep learning framework effectively incorporates the neighborhood effect for improved recommendations.
• This research offers a more comprehensive approach to understanding user preferences in location-based services.