Uncertainty-Calibrated Recommendations for Low-Active Users
Researchers have developed a new framework to improve recommender systems by quantifying model uncertainty. This approach differentiates strategies for low-active users (LAUs) and high-active users (HAUs) to boost engagement and satisfaction. For LAUs, the system suppresses unreliable recommendations, while for HAUs, it encourages content exploration. Tested on a livestream platform, the framework improved retention and satisfaction for LAUs and increased interest diversity for HAUs. AI
IMPACT Enhances user engagement and content diversity in recommender systems by intelligently managing model uncertainty.