Recommendations as Treatments: Debiasing Learning and Evaluation

Abstract

Most data for evaluating and training recommender systems is subject to selection biases, either through self-selection by the users or through the actions of the recommendation system itself. In this paper, we provide a principled approach to handling selection biases, adapting models and estimation techniques from causal inference. The approach leads to unbiased performance estimators despite biased data, and to a matrix factorization method that provides substantially improved prediction performance on real-world data. We theoretically and empirically characterize the robustness of the approach, finding that it is highly practical and scalable.

10 pages in ICML 2016

References in corpus (4)

• Doubly Robust Policy Evaluation and Learning
• Collaborative Filtering and the Missing at Random Assumption
• Learning from Logged Implicit Exploration Data
• Reducing Offline Evaluation Bias in Recommendation Systems

Cited by in corpus (40)

• Fairness of Exposure in Rankings
• Causal Intervention for Leveraging Popularity Bias in Recommendation
• Controlling Fairness and Bias in Dynamic Learning-to-Rank
• Enhanced Doubly Robust Learning for Debiasing Post-click Conversion Rate Estimation
• A Contextual Bandit Bake-off
• Counterfactual Evaluation of Slate Recommendations with Sequential Reward Interactions
• SurvTRACE: Transformers for Survival Analysis with Competing Events
• Deconfounded Causal Collaborative Filtering
• It Is Different When Items Are Older: Debiasing Recommendations When Selection Bias and User Preferences Are Dynamic
• Debiasing Learning for Membership Inference Attacks Against Recommender Systems
• Debiasing Recommendation with Personal Popularity
• A Survey on Causal Inference
• not-MIWAE: Deep Generative Modelling with Missing not at Random Data
• Reinforcement Learning to Optimize Long-term User Engagement in Recommender Systems
• Disentangling User Interest and Conformity for Recommendation with Causal Embedding
• Unbiased Recommender Learning from Missing-Not-At-Random Implicit Feedback
• Popularity Bias Is Not Always Evil: Disentangling Benign and Harmful Bias for Recommendation
• Collaborative Filtering via High-Dimensional Regression
• On component interactions in two-stage recommender systems
• Relevance meets Diversity: A User-Centric Framework for Knowledge Exploration through Recommendations
• Learning from User Interactions with Rankings: A Unification of the Field
• Multi-Source Causal Inference Using Control Variates
• Poincare: Recommending Publication Venues via Treatment Effect Estimation
• Asymmetric Tri-training for Debiasing Missing-Not-At-Random Explicit Feedback
• Affordable Uplift: Supervised Randomization in Controlled Experiments
• An Adaptive Boosting Technique to Mitigate Popularity Bias in Recommender System
• Show Me the Whole World: Towards Entire Item Space Exploration for Interactive Personalized Recommendations
• A General Framework for Counterfactual Learning-to-Rank
• Debiasing Samples from Online Learning Using Bootstrap
• CausCF: Causal Collaborative Filtering for RecommendationEffect Estimation
• Gradient Importance Learning for Incomplete Observations
• Concept Drift Detection: Dealing with MissingValues via Fuzzy Distance Estimations
• Causal Inference in medicine and in health policy, a summary
• Causal Learning for Socially Responsible AI
• Online Evaluation Methods for the Causal Effect of Recommendations
• Learning from missing data with the Latent Block Model
• Causal Embeddings for Recommendation: An Extended Abstract
• Causality-Aware Neighborhood Methods for Recommender Systems
• Multi-Channel Sequential Behavior Networks for User Modeling in Online Advertising
• Large-scale Causal Approaches to Debiasing Post-click Conversion Rate Estimation with Multi-task Learning