As with all my handwritten notes, this has the usual disclaimer: these posts are just so I can use nice indexed search to find my notes, which are sufficient for me to recall talks and papers but probably not much use to anyone else.
Unfortunately, there isn’t a paper from which I can rip an abstract, so I’m going to have to type something up if I want to be able to search for this.
Facebook doesn’t publish nearly as much as Google, MS, and Yahoo, so it was nice to hear about how things are done there. One interesting tidbit was that they use Vickrey–Clarke–Groves (VCG) and not generalized second-price (GSP) (finally, a counter-example that can be raised when people say that no one uses VCG in practice) because, according to Hegeman, Click-through rate (CTR) for a user is effectively a random variable. David Reiley was skeptical; he cited the GSP paper, saying “Edelman, Ostrovsky, and Schwarz would say that it doesn’t matter — you can just average over users to get an aggregate CTR”. After some heated discussion, Susan Athey got a laugh from the crowd, citing her upcoming talk, saying “Athey would say that it matters”. It’s not surprising that they saw a big drop in revenue when switching from GSP to VCG, but I thought it was surprising that it took multiple months for revenue to recover. On the other hand, you could argue they’re lucky that companies don’t just “copy and paste” their AdSense campaigns into facebook, in which case their revenues would never have recovered.
Other interesting tidbits included a discussion on ad quality, the use of like/X on ads, how facebook handles budget constraint bidders, and why facebook doesn’t allow “manual” user ad capping, the way AdSense does.
Related link: Noam Nisan has an old post with some related links that inform the GSP vs. VCG discussion.