[Nrg-l] BU/CS Colloquium by Boaz Patt-Shamir: Tuesday August 12, 2008

Azer Bestavros best at cs.bu.edu
Wed Aug 6 14:53:33 EDT 2008

Boston University
Computer Science Colloquium Series 


 Speaker: Boaz Patt-Shamir
          Associate Professor
          Tel-Aviv University
    When: Tuesday August 12, 2008 @ 11:00am
   Where: MCS-135 


In recommendation systems (e.g., for books or movies), the system
tracks which product each user liked in the past, and tries to deduce
which other products the user is likely to be satisfied with.
Recommendation systems can help users who share many preferences with
many other users by means of "collaborative filtering." Most current 
approaches to on-line recommendation systems employ algebraic
techniques such as Singular Value Decomposition (SVD), which are
computationally intensive and, more important, often applicable only
under some additional strong conditions. We overview a new approach
which demonstrates that simple combinatorial algorithms can make good
recommendations that guarantee that the cost per user is only
polylogarithmic factor over optimal, without restricting the allowed


Host: Azer Bestavros

More information about the Nrg-l mailing list