mlsanityData to 14 Oct 2026

CASP: Support-Aware Offline Policy Selection for Two-Stage Recommender Systems

arXiv 2604.23022 ↗First posted 24 Apr 2026cs.IR cs.LG stat.MLcited by 0

Not on the map. Why?

Abstract (arXiv)

Two-stage recommender systems first choose a candidate generator and then rank items within the generated set. Because the generator decides which items are available to the ranker, changing the generator changes both the policy value and the data support used to estimate that value. This creates an offline selection problem that standard single-stage objectives do not capture: a policy may look good under a retrieval score or a raw off-policy value estimate, but still be unreliable if it depends on weakly supported generator-item pairs. We propose CASP (Coupled Action-Set Pessimism), a support-aware offline selector for finite libraries of two-stage recommender policies. CASP combines doubly robust value estimation with a support-burden penalty. We show that stagewise rules that ignore downstream continuation value can be arbitrarily suboptimal, and we derive population, finite-class, and reconstructed-propensity guarantees for conservative selection. In simulations and a reconstructed MovieLens 1M application, CASP selects lower-burden policies when estimated value and support credibility are in tension.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 19 of its 39 references that are in the map. First posted on arXiv 24 Apr 2026.

Builds on

19 of 39 references.

References · 19

  1. 2020Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problemscited by 1,920
  2. 2016Recommendations as Treatments: Debiasing Learning and Evaluationcited by 248
  3. 2018Top-K Off-Policy Correction for a REINFORCE Recommender Systemcited by 182
  4. 2015Doubly Robust Policy Evaluation and Optimizationcited by 158
  5. 2016Optimal and Adaptive Off-policy Evaluation in Contextual Banditscited by 111
  6. 2016Off-policy evaluation for slate recommendationcited by 103
  7. 2018Offline A/B testing for Recommender Systemscited by 96
  8. 2019Doubly robust off-policy evaluation with shrinkagecited by 87
  9. 2020Open Bandit Dataset and Pipeline: Towards Realistic and Reproducible Off-Policy Evaluationcited by 77
  10. 2020Off-policy Bandits with Deficient Supportcited by 61
  11. 2022Off-Policy Evaluation for Large Action Spaces via Embeddingscited by 54
  12. 2020Counterfactual Evaluation of Slate Recommendations with Sequential Reward Interactionscited by 35
7 more
  1. 2023Off-Policy Evaluation for Large Action Spaces via Conjunct Effect Modelingcited by 34
  2. 2021Evaluating the Robustness of Off-Policy Evaluationcited by 32
  3. 2019Reinforcement Learning for Slate-based Recommender Systems: A Tractable Decomposition and Practical Methodologycited by 27
  4. 2023A Survey on Causal Inference for Recommendationcited by 22
  5. 2021Control Variates for Slate Off-Policy Evaluationcited by 13
  6. 2024Multi-Objective Recommendation via Multivariate Policy Learningcited by 10
  7. 2018A General Framework for Counterfactual Learning-to-Rankcited by 3

Built on by

Not cited yet.