Stanford Summit: Data questions

Bambi Francisco is moderating this panel about data... what is the data telling us?

The panelists are from a number of search companies. Usama Fayyad of Yahoo!, Jim Lanzone of, Peter Norvig of Google, and Mike Yavonditte of Quigo.

So, what does the data say? Here's how a few of the experts summarized their thoughts:

Usama sees three things:
(1) The notion of a portal is extremely relevant because information overload is happening.
(2) Data is used to improve the product. For example, the feature that impacted user retention the most at Y!Mail is adding a news feature box on the screen. They wouldn't have guessed that absent the data.
(3) Search is not enough.

Jim observes that:
(1) Most users are actually very lazy. The behavior Ask sees is that users like the simple box. The jump in and do a search, then review of results page in 5 seconds or less. They grab clues, and then reiterate their search. Most search sessions involve 3-5 searches.
(2) On the aggregate, they see cultural things from th data. E.g., on Monday afternoons, people do the tasks they didn't do over the weekend. On Tuesday, they start dooing searches that relate to plans for the next weekend.

Peter has these observations:
(1) Language - You can do quite a bit with linguistics and language understanding. His team at Google was able to build a translation system between Arabic and Chinese to/from English. Nobody on the team spoke those languages. They developed this through statistical analysis of the data they have.
(2) Understanding Facts about the world - you can go online and do co-learning: you learn facts and learn how they are expressed. (?) Research about birthdays. Started with 10 birthdays, but went on to learn all sorts of facts (not just facts about birthdays).
(3) Sometimes they are wrong, and get corrected by users - the search for "glacier bay" would mostly be people looking for a faucett company, not the bay in Alaska.

Mike offers comments from the advertising perspective:
(1) The data shows that context matters: serving a mortgage ad to a health/diet site is more effective than serving a Jenny Craig ad. Time of day matters too.
(2) How good the ad is, is important.
(3) Aesthetics matter. Every pixel matters. The font and background color can dramatically change the performance of an ad.

Add new comment