as the reviewers were happy with the motivation and practicality of our work. believe that our novelty is in proposing Thompson sampling latent bandit algorithms using offline-learned graphical