A Bi-level Framework for Debiasing Implicit Feedback with Low Variance

TMLR Paper608 Authors

17 Nov 2022 (modified: 26 May 2023)Rejected by TMLREveryoneRevisionsBibTeX
Abstract: Implicit feedback is easy to collect and contains rich weak supervision signals, thus is broadly used in recommender systems. Recent works reveal a huge gap between the implicit feedback and the user-item relevance due to the fact that users tend to access items with high exposures but these items may not be necessarily relevant to users' preferences. To bridge the gap, existing methods explicitly model the item exposure degree and propose unbiased estimators to improve the relevance. Unfortunately, these unbiased estimators suffer from the high gradient variance, especially for long-tail items, leading to inaccurate gradient updates and degraded model performance. To tackle this challenge, we propose a bi-level framework for debiasing implicit feedback with low variance. We first develop a low-variance unbiased estimator from a probabilistic perspective, which effectively bounds the variance of the gradient. Unlike previous works which either estimate the exposure via heuristic-based strategies or use a large biased training set, we propose to estimate the exposure via an unbiased small-scale validation set. Specifically, we parameterize the user-item exposure by incorporating both user and item information, and propose to construct the unbiased validation set only from the biased training set instead of using random policy at the cost of degrading user experience. By leveraging the unbiased validation set, we adopt a bi-level optimization framework to automatically update exposure-related parameters along with recommendation model parameters during the learning. Experiments on two real-world datasets and two semi-synthetic datasets verify the effectiveness of our method. Our code is available at \url{https://anonymous.4open.science/r/TMLR-Biff/README.md}.
Submission Length: Regular submission (no more than 12 pages of main content)
Assigned Action Editor: ~Jeremie_Mary1
Submission Number: 608
Loading