Spectral Methods in Data Mining

by
Michelle Goodstein

The growing popularity of the Internet has created large repositories of data, with potential data mining applications in the fields of collaborative filtering and web search. However, getting meaningful results from such a vast set of data can be difficult. We are investigating the use of spectral techniques as a means of giving better answers to queries. Our research currently focuses on generalizing and simplifying previously published algorithms; proving new theorems regarding performance guarantees; testing how the addition of normalization techniques affects the accuracy of our algorithms; and empirically evaluating how the various algorithms work in practice. Joint work with Anna Karlin, Frank McSherry, Matthew Milcic, Peter Turschmid and Danny Wei.

Advised by Anna Karlin

CSE 403
Wednesday
February 16, 2005
3:30 - 4:20 pm