Research map / Probabilistic and scientific modeling / Treatment effects
Recommendations as Treatments: Debiasing Learning and Evaluation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| propensity-weighted (IPS/SNIPS) evaluation instead of naive average over observed entries (ablation) | performance estimation of rating predictions and recommendations · ML100K semi-synthetic (alpha=0.25) | MAE and DCG@50 estimation accuracy | Helped |
| propensity-weighted matrix factorization (MF-IPS) instead of unweighted MF | matrix factorization rating prediction · Yahoo R3 | MSE and MAE on MCAR test set: MSE 1.891 to 0.989; MAE 1.154 to 0.810 | Helped |
| propensity-weighted matrix factorization (MF-IPS) instead of unweighted MF | matrix factorization rating prediction · Coat | MSE and MAE on MCAR test set: MSE 1.202 to 1.093; MAE 0.920 to 0.860 | Helped |
| propensity-weighted MF with learned propensities instead of joint-likelihood HL MNAR model | matrix factorization rating prediction · Yahoo R3 | MSE and MAE: MSE 2.175 to 0.989; MAE 1.177 to 0.810 | Helped |
| propensity-weighted MF instead of joint-likelihood HL MNAR model | matrix factorization rating prediction · Coat | MAE: MAE 0.884 to 0.860; MSE 1.214 to 1.093 | Helped |
| propensity-weighted MF (MF-IPS) instead of unweighted MF across MNAR severity levels (ablation) | matrix factorization on semi-synthetic ML100K · ML100K semi-synthetic, experimental setting | MSE | Helped |
| SNIPS instead of IPS estimator (ablation) | performance estimation · ML100K semi-synthetic | MAE estimation variance and DCG@50 estimation | Mixed |
| SNIPS instead of IPS estimator (ablation) | MSE estimation · ML100K semi-synthetic, varying alpha | RMSE of MSE estimate | Helped |
| propensity estimation via Naive Bayes instead of known propensities (IPS-NB vs IPS-KNOWN) (ablation) | IPS estimator under observational setting · ML100K semi-synthetic, alpha=0.25 | MSE estimation | Helped |
| degrading propensity estimate quality (smaller MCAR sample for Naive Bayes) (ablation) | IPS/SNIPS evaluation and MF-IPS learning · ML100K semi-synthetic, alpha=0.25 | estimation error and MSE | No clear effect |
| SNIPS vs Naive when MCAR (alpha=1) (ablation) | performance estimation · ML100K semi-synthetic | estimation error | No clear effect |
Closest papers by meaning
- Alleviating Dual Biases in Recommendation (Student Abstract) (AAAI 2025 · similarity 0.90)
- Be Aware of the Neighborhood Effect: Modeling Selection Bias under Interference (ICLR 2024 · similarity 0.89)
- Addressing Hidden Confounding with Heterogeneous Observational Datasets for Recommendation (NeurIPS 2024 · similarity 0.88)
- Debiased Collaborative Filtering with Kernel-Based Causal Balancing (ICLR 2024 · similarity 0.87)
- Relaxing the Accurate Imputation Assumption in Doubly Robust Learning for Debiased Collaborative Filtering (ICML 2024 · similarity 0.86)
- TranSUN: A Preemptive Paradigm to Eradicate Retransformation Bias Intrinsically from Regression Models in Recommender Systems (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.