May 15, 2006 12:06 PM PDT
Data-mining pioneer joins Microsoft
- Related Stories
FAQ: NSA's data mining explainedMay 12, 2006
Who's who of Google hiresFebruary 27, 2006
Microsoft taps ex-Yahoo researcher to head 'Live Labs'January 25, 2006
Microsoft settles with Google over executive hireDecember 22, 2005
Search specialist stakes its claim on namesAugust 26, 2005
Privacy's random answerNovember 24, 2004
IBM's 'Marvel' to scour Net for video, audioSeptember 29, 2004
Research: From lab to marketJune 15, 2004
Old-school theory is a new forceFebruary 18, 2003
Rakesh Agrawal, who is credited with creating data mining, or the science of extracting trends from large and often disparate databases, has left IBM to become a Microsoft technical fellow in the company's Search Labs.
Large tech companies for years have tried to woo each other's top scientists, and in the search and computer science field, Google has lately been getting most of them. Google pulled away Microsoft's Kai-Fu Lee to run its China labs, which led to a lawsuit. Google also snagged search expert Udi Manber from Amazon.com. (Vint Cerf and his ceremonial hat of many colors joined Google as well.)
Agrawal--who had been an IBM fellow, the company's highest title for researchers--is one of the better-known scientists in data extraction and databases. Data mining is a hot topic because it has emerged that the federal government has begun to use the method to examine millions of phone records. Corporations, though, have exploited it for years as a way to understand customer behavior and enhance their own Web traffic.
Although not associated in the public mind with search, IBM is one of the major forces in the field. The company was one of the first to devise a video search engine, and in March, it bought Language Analysis Systems, a search company focused on finding references to individuals, even if their name gets spelled differently in different databases.
Agrawal joined Microsoft a few weeks ago, but it was not publicly announced. Microsoft created its Search Labs in January.
The idea behind data mining came up during a lunchtime conversation in the early 1990s between Agrawal and an executive from the British department store chain Marks & Spencer. The store chain had been collecting all sorts of data but didn't know what to do with it.
Agrawal and his team began devising algorithms for asking open-ended queries, eventually authoring a 1993 paper describing data mining. The paper has been cited in more than 650 other studies, making it one of the most widely cited papers of its kind.
"We were not even sure we should send it because we thought people might think it was too simple-minded," Agrawal said.
More recently, Agrawal has been working on randomization. In this technique, data is scrambled before it gets entered into a database. Nonetheless, mathematicians, by applying probabilistic computing techniques to the scrambled data, can come up with patterns that are similar to what the actual data would have shown.
Thus, a corporation can get a handle on its 18- to 24-year-old buyers while privacy is ensured; the original data is never entered in the database.
The scientific underpinnings of randomization have been the subject of a few academic papers. The short explanation is as follows: "It's the beauty of math," Agrawal said in an interview a few years ago.