This paper introduces Deep Mutual Studying (DML) to enhance multi-task recommender techniques. It integrates Cross Process Characteristic Mining (CTFM) to share task-specific inputs and International Data Distillation (GKD) to refine prediction consistency. DML enhances information switch and overcomes process conflicts, validated by offline experiments and on-line A/B exams. Learn Extra
New Framework for Deep Mutual Studying (DML) to Enhance Multi-task Recommender Methods
