Publications

Are you an egosurfer, addicted to Googling your own name? If so, you should check out Personas, which uses data-mining techniques to answer the question: "How does the internet see you?"

New Scientist, 27 August 2009 by Peter Aldhous

Devised by Aaron Zinman, a PhD student at the Massachusetts Institute of Technology's Media Lab, Personas runs a Yahoo search on your name, extracting statements about you (such as "Peter Aldhous is San Francisco bureau chief with New Scientist magazine") from the returns.

Words from these statements are then run through an algorithm called Latent Dirichlet Allocation, which links them to 28 common topics defined by Zinman, including "online", "travel", "education" and "professional".

...

Zinman describes Personas as "a critique on data mining", intended to highlight the undue faith that some analysts place in results from computer analysis of large data sets. "It's always presented so authoritatively," he says, "when in reality what's happening behind the scenes is controlled voodoo."

Is it that a fair criticism? "It's a fun project. However, data-mining researchers and professional analysts are well aware of the danger of finding false associations," says Gregory Piatetsky-Shapiro, president of KDnuggets, a data-mining consultancy in Boston.

"Criticising data mining for some people's inability to separate false associations from true ones is like criticising the web because there are many bad pages."

Read more.

[Gregory Piatetsky, Editor: Here is how Personas sees me: a strange mix of fame, sports, and management, with sprinkling of politics and legal - only superficially relevant to what I do]