I saw someone from DeepMind mention using RL for Recommender Systems at Google at their ICML 2016 presentation, but they were very tight lipped when I asked for

I saw someone from DeepMind mention using RL for Recommender Systems at Google at their ICML 2016 presentation, but they were very tight lipped when I asked for more details. Bandit algorithms are pretty popular in the ads world, and there is probably some use in extending that to the full RL setup.