Bounding System-Induced Biases in Recommender Systems with a Randomized Dataset

Debiased recommendation with a randomized dataset has shown very promising results in mitigating system-induced biases. However, it still lacks more theoretical insights or an ideal optimization objective function compared the other well-studied routes without dataset. To bridge this gap, we study debiasing problem from new perspective and propose to directly minimize upper bound of function, which facilitates better potential solution First, formulate Second, according prior constraints that adopted loss may satisfy, derive two different bounds function: generalization error triangle inequality separability. Third, show most existing related methods can be regarded as insufficient these bounds. Fourth, novel method called approximate ( DUB ) dataset, achieves sufficient Finally, conduct extensive experiments on public real product verify effectiveness our DUB.