[NRG] PhD Defense by Giorgos Zervas on June 16 @ 3pm: Data-driven Analysis of Electronic Commerce Systems

Bestavros, Azer best at bu.edu
Thu Jun 9 17:49:35 EDT 2011


________________________________

Boston University -- Computer Science Department

P H D    D E F E N S E

Thursday June 16, 2011
3:00pm - 4:30pm
Location TBD

________________________________


DATA-DRIVEN ANALYSIS OF ELECTRONIC COMMERCE SYSTEMS

Giorgos Zervas
Abstract: We contribute to the large body of work studying electronic commerce by proposing and evaluating a method of inquiry that complements standard game-theoretic and econometric methods, which simplify systems to strategic interactions of rational, utility-maximizing agents. In our framework, economic agents continuously reveal their preferences through their interactions with e-commerce systems that ``leak'' data. Such data can be seen as an ex-post realization of strategic behavior. We demonstrate that by collecting and analyzing this information, we are able to gain new insights on system-level properties as well as the participating agents' operational strategies. Our approach thereby complements economic modeling both in informing the design of more accurate models and in assessing their validity.
The contributions arise from three case studies in electronic commerce: pay-per-bid auctions, daily discount coupons, and Internet advertising. In the context of pay-per-bid auctions we extend previous modeling work which, despite evidence to the contrary, predicts profit-free equilibria to offer an explanation for the dramatic profits firms can derive from even rational, risk-neutral players. In the context of discount coupons, we undertake a study of Groupon, the market leader. We develop regression-based models of deal outcomes that demonstrate the extent to which Groupon customers are sensitive to incentives other than price such as the deal's location, duration and starting weekday. Finally, in the context of Internet advertising we develop algorithms to estimate valuations of search terms from highly aggregated and noisy click-stream data. We experimentally demonstrate that using our algorithms, search term valuations can be efficiently learned, leading to properly informed bidding strategies.

Examination Committee:.

*         John Byers (First Reader and Major Advisor)

*         Michael Mitzenmacher (Second Reader)

*         Steve Homer (Committee Member)

*         Azer Bestavros (Committee Chair)


________________________________



-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://cs-mailman.bu.edu/pipermail/nrg-l/attachments/20110609/d46ffbf8/attachment.html 


More information about the NRG-L mailing list