PCDe: A personalized conversational debiasing framework for next POI recommendation with uncertain check-ins

Area of Science:

• Computer Science
• Human-Computer Interaction
• Data Science

Background:

• Next point-of-interest (POI) recommendations face challenges with uncertain check-ins at collective POIs (e.g., shopping malls).
• Uncertain check-ins introduce scale bias (favoring collective over individual POIs) and exacerbate popularity bias (favoring popular over unpopular POIs).
• These biases significantly impact the fairness of POI recommendation systems.

Purpose of the Study:

• To propose a novel framework for mitigating scale and popularity biases in next POI recommendation with uncertain check-ins.
• To enhance the fairness and accuracy of POI recommendation systems by addressing personalized user preferences.
• To introduce a debiasing mechanism that accounts for user behavior in collective POIs.

Main Methods:

• Developed a Personalized Conversational Debiasing (PCDe) framework leveraging conversational techniques.
• Implemented an inquiry component using personalized information entropy to mitigate scale bias.
• Introduced a rewarding component with Jensen-Shannon divergence for a debiasing reward mechanism to address popularity bias.

Main Results:

• The PCDe framework effectively mitigates both scale and popularity biases in next POI recommendation.
• Experimental results demonstrate the superiority of PCDe over state-of-the-art methods.
• PCDe enhances recommendation accuracy by addressing personalized user preferences and biases.

Conclusions:

• Personalized debiasing strategies are crucial for fair and accurate POI recommendations, especially with uncertain check-ins.
• Conversational techniques offer a promising approach to capture dynamic user preferences and mitigate recommendation biases.
• The proposed PCDe framework provides an effective solution for improving the fairness and performance of POI recommendation systems.