Near-Optimal MNL Bandits Under Risk Criteria

TL;DR

This paper develops algorithms for MNL bandits optimized for various risk criteria, achieving near-optimal regret and demonstrating strong empirical performance on synthetic and real data.

Contribution

It introduces a unified approach to handle multiple risk criteria in MNL bandits, extending beyond expected revenue optimization.

Findings

Algorithms achieve near-optimal regret bounds.

Empirical results show strong performance on synthetic data.

Algorithms perform well on real-world datasets.

Abstract

We study MNL bandits, which is a variant of the traditional multi-armed bandit problem, under risk criteria. Unlike the ordinary expected revenue, risk criteria are more general goals widely used in industries and bussiness. We design algorithms for a broad class of risk criteria, including but not limited to the well-known conditional value-at-risk, Sharpe ratio and entropy risk, and prove that they suffer a near-optimal regret. As a complement, we also conduct experiments with both synthetic and real data to show the empirical performance of our proposed algorithms.