Instance-Sensitive Algorithms for Pure Exploration in Multinomial Logit Bandit

Motivated by real-world applications such as fast fashion retailing and online advertising, the Multinomial Logit Bandit (MNL-bandit) is a popular model in learning operations research, has attracted much attention past decade. In this paper, we give efficient algorithms for pure exploration MNL-bandit. Our achieve instance-sensitive pull complexities. We also complement upper bounds an almost matching lower bound.