Removing Hidden Confounding in Recommendation: A Unified Multi-Task Learning Approach

In recommender systems, the collected data used for training is always subject to selection bias, which poses a great challenge for unbiased learning.