Sampling Process Brings Additional Bias for Debiased Recommendation
Keywords: Superpopulation, Selection Bias, Recommender System
Abstract: In recommender systems, selection bias arises from the users' selective interactions with items, which poses a widely-recognized challenge for unbiased evaluation and learning for recommendation models. Recently, doubly robust and its variants have been widely studied to achieve debiased learning of prediction models. However, if the users and items in the training set are not exactly the same as those in the test set, even if the imputed errors and learned propensities are accurate, all previous doubly robust based debiasing methods are biased. To tackle this problem, in this paper, we first derive the bias of doubly robust learning methods and provide alternative unbiasedness conditions when users and items are sampled from a superpopulation. Then we propose a novel superpopulation doubly robust target learning approach (SuperDR), which is unbiased when either the imputation model or propensity model is correctly specified. We further derive the generalization error bound of the proposed method under superpopulation, and show that it can be effectively controlled by the proposed target learning approach. We conduct extensive experiments on three real-world datasets, including a large-scale industrial dataset, to demonstrate the effectiveness of our method.
Primary Area: other topics in machine learning (i.e., none of the above)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 13992
Loading