Quadrianto, Novi and Smola, Alex and Caetano, Tiberio and Le, Quoc Viet
Consider the following problem: given several sets of unlabeled observations, each set with known label proportions, predict the labels of another set of observations, also with known label proportions. This problem appears in areas like e-commerce, spam filtering, and improper content detection. We present consistent estimators which can reconstruct the correct labels with high probability in a uniform convergence sense. Experiments show that our method works well in practice.
Discussion
The motivating application of giving coupons or not to people can be phrased as an offline binary contextual bandit problem. Here, the “binary” means that a policy has two actions available: give coupon or not. “offline” is because this data has been recorded, so the explore/exploit tradeoff isn't there and the goal is simply getting a good policy.