Off-Policy Evaluation for Large Action Spaces via Embeddings

TL;DR

This paper introduces a new off-policy evaluation estimator for large action spaces that uses action embeddings to reduce bias and variance, enabling more reliable policy evaluation in complex systems.

Contribution

The paper proposes a novel OPE estimator leveraging action embeddings, with theoretical analysis and empirical results showing improved performance over existing methods in large action spaces.

Findings

Significant reduction in bias and variance with the new estimator

Theoretical conditions where embeddings improve statistical efficiency

Empirical results demonstrate reliable OPE in large action spaces

Abstract

Off-policy evaluation (OPE) in contextual bandits has seen rapid adoption in real-world systems, since it enables offline evaluation of new policies using only historic log data. Unfortunately, when the number of actions is large, existing OPE estimators -- most of which are based on inverse propensity score weighting -- degrade severely and can suffer from extreme bias and variance. This foils the use of OPE in many applications from recommender systems to language models. To overcome this issue, we propose a new OPE estimator that leverages marginalized importance weights when action embeddings provide structure in the action space. We characterize the bias, variance, and mean squared error of the proposed estimator and analyze the conditions under which the action embedding provides statistical benefits over conventional estimators. In addition to the theoretical analysis, we find that the empirical performance improvement can be substantial, enabling reliable OPE even when existing estimators collapse due to a large number of actions.