The National Institute of Statistical Sciences (NISS) hosted a workshop on computational advertising in November of 2009. This area comprises a new set of statistical challenges stimulated by the Internet becoming the medium of choice for many advertisers.
ASA Members Part of $1M Win
ASA members Bob Bell and Chris Volinsky—along with their team, BellKor’s Pragmatic Chaos—won the $1 million Netflix prize for coming up with an algorithm that improved the online movie rental service’s recommendation system.
Other members of the winning team include Martin Chabbert, Michael Jahrer, Yehuda Koren, Martin Piotte, and Andreas Toscher. All team members attended the awards ceremony in New York City on September 21, 2009.
Netflix launched the competition in October of 2006 and made available to contestants 100 million anonymous movie ratings ranging from one to five stars. All personal information identifying individual Netflix members was removed from the prize data, which contained only movie titles, star ratings, and dates. No text reviews were included. More than 40,000 teams from around the world competed.
The team prepared a system description consisting of three papers, which can be downloaded here. Click here for more information.
The workshop focused on three topics that are critical to the business models of Yahoo, Google, Microsoft, and other corporations:
Display advertising. How can ads be matched to web page content and reader interests, needs, and preferences based on data collected by a web site?
Search engines, especially sponsored search. How do data inform economic models of the auctions by which keywords and link placement are sold, taking into account the multiple perspectives of search engine operators, advertisers, and users?
Recommender systems. Think Netflix: How can users be provided useful recommendations given the extreme sparsity of the data? Most viewers rate only a few movies, and most movies are rated by only a few customers.
Kishore Papineni of Yahoo gave the workshop’s introductory talk, setting up the big-picture perspective that informed the talks and discussion. Silviu-Petru Cucerzan of Microsoft described a spelling corrector that uses nearly correct spellings to correct grossly incorrect ones. Daniel Ford of Google described optimal refresh rates for search engine indexes of web sites. Deepak Agarwal of Yahoo spoke about multi-armed bandit models for content placement on the Yahoo front page. Carl Mela of Duke University spoke about behavioral game theory and models of auctions for keywords and advertisement placement. Daryl Pregibon of Google spoke about advertiser graphs that indicated which companies compete for the same keywords and how mining this information leads to improvement for consumers, advertisers, and Google. Charles Elkan of the University of California, San Diego, and Bob Bell of AT&T Labs Research both spoke about aspects of the Netflix competition. Elkan addressed it from the standpoint of recommender systems in general, and Bell shared his experience as a member of the team that won the Netflix prize.
Leave a Reply