AutoDebias: Learning to Debias for Recommendation

Abstract

Recommender systems rely on user behavior data like ratings and clicks to build personalization model. However, the collected data is observational rather than experimental, causing various biases in the data which significantly affect the learned model. Most existing work for recommendation debiasing, such as the inverse propensity scoring and imputation approaches, focuses on one or two specific biases, lacking the universal capacity that can account for mixed or even unknown biases in the data. Towards this research gap, we first analyze the origin of biases from the perspective of \textit{risk discrepancy} that represents the difference between the expectation empirical risk and the true risk. Remarkably, we derive a general learning framework that well summarizes most existing debiasing strategies by specifying some parameters of the general framework. This provides a valuable opportunity to develop a universal solution for debiasing, e.g., by learning the debiasing parameters from data. However, the training data lacks important signal of how the data is biased and what the unbiased data looks like. To move this idea forward, we propose \textit{AotoDebias} that leverages another (small) set of uniform data to optimize the debiasing parameters by solving the bi-level optimization problem with meta-learning. Through theoretical analyses, we derive the generalization bound for AutoDebias and prove its ability to acquire the appropriate debiasing strategy. Extensive experiments on two real datasets and a simulated dataset demonstrated effectiveness of AutoDebias. The code is available at \url{https://github.com/DongHande/AutoDebias}.

Accepted by SIGIR 2021

References in corpus (7)

• Theoretical Models of Learning to Learn
• Fairness of Exposure in Rankings
• Correcting for Selection Bias in Learning-to-rank Systems
• ESAM: Discriminative Domain Adaptation with Non-Displayed Items to Improve Long-Tail Performance
• Privacy-Aware Recommendation with Private-Attribute Protection using Adversarial Learning
• How to Retrain Recommender System? A Sequential Meta-Learning Method
• LambdaOpt: Learn to Regularize Recommender Models in Finer Levels

Cited by in corpus (20)

• Debiased Contrastive Learning for Sequential Recommendation
• Self-Guided Learning to Denoise for Robust Recommendation
• Graph-less Collaborative Filtering
• Invariant Collaborative Filtering to Popularity Distribution Shift
• Unbiased Knowledge Distillation for Recommendation
• Balancing Unobserved Confounding with a Few Unbiased Ratings in Debiased Recommendations
• SIGformer: Sign-aware Graph Transformer for Recommendation
• Distillation Matters: Empowering Sequential Recommenders to Match the Performance of Large Language Model
• Understanding and Modeling Passive-Negative Feedback for Short-video Sequential Recommendation
• How Do Recommendation Models Amplify Popularity Bias? An Analysis from the Spectral Perspective
• Leveraging Watch-time Feedback for Short-Video Recommendations: A Causal Labeling Framework
• Deep Generative Imputation Model for Missing Not At Random Data
• Pareto Invariant Representation Learning for Multimedia Recommendation
• LabelCraft: Empowering Short Video Recommendations with Automated Label Crafting
• The Importance of Causality in Decision Making: A Perspective on Recommender Systems
• Treatment Effect Estimation for User Interest Exploration on Recommender Systems
• Addressing Correlated Latent Exogenous Variables in Debiased Recommender Systems
• Rankformer: A Graph Transformer for Recommendation based on Ranking Objective
• Uncertainty Calibration for Counterfactual Propensity Estimation in Recommendation
• Breaking the Top-$K$ Barrier: Advancing Top-$K$ Ranking Metrics Optimization in Recommender Systems