Several traffic networks offer "smart links", where you can send online traffic, and the traffic will be sent to the most profitable segment. These networks include Monetizer, Glize etc.
I.e. if you send IOS, mobile, Orange carrier in Spain, the traffic will primarily go to offer Y. If you send Android, Tablet in France it will go to offer X etc. I assume they optimize for earning per view/click. A fraction will be for testing, but the majority of the traffic will go to the optimal offer.
I wonder what the general workings of such algorithms are, and if there's a place (research papers, or books), I can learn more about this type of problem? I have a hard time finding anything directly related online, as it's not a very common problem in research papers, and a lot of the algorithms are being disclosed -- all though some mention using Bayesian statistics to optimize.
I started looking at multi-armed bandit problems, which I feel are very close to this type of problem. If you imagine the different offers are different "bandits", which bandit should you choose to maximize return. From this they can calculate a percentage weight per offer. Maybe initially they will all start at an even weight, and once there's enough data in the timespan, the weights will be shuffled (at a timed basis, i.e. every 30 min).
I think this is a very interesting problem, as I work in online marketing and also study computer science.
Thanks in advance.