Title: Department of Computer Science Seminar
Date: Wednesday, May 31, 2000
Time: 4:00 pm to 5:00 pm
Venue: Room N101, CSIT Building [108]
Speaker: Nick Craswell (Ph.D Student at DCS, ANU)
Description: "Distributed Information Retrieval on the Web"


                        Abstract

 There are tens of thousands of search servers on the Web.  Many of
 them offer the most comprehensive, or only, coverage of a useful
 document collection or database (for example at the ACM Digital
 Library, britannica.com, amazon.com).  A search broker helps users
 access such resources, by selecting a useful subset of servers,
 querying them concurrently and presenting their results in a
 relevance ranked list.  However, there has been a gap between theory
 and practice in broker research.  In theory it is possible for a
 broker to select from a large set of servers and generate a merged
 ranking using state-of-the-art ranking methods.  However, theoretical
 methods rely on special cooperation from search servers, which is not
 forthcoming on the Web.  So in practice, Web search brokers have
 addressed a few servers and used merging methods of untested
 effectiveness.

 I will present new methods which allow a broker to address any search
 server, provided that search results and documents are accessible.
 Two extensive evaluation experiments show that a Web-compatible
 broker can be as effective as any previously proposed broker (and
 more effective than existing Web brokers).  I will also relate server
 coverage limitations to a taxonomy of "Dark Matter" on the Web --
 documents which are visible to some but not all observers.


URL: http://cs.anu.edu.au/lib/seminars/seminars00/dept20000531


Biography: Nick Craswell recently submitted his PhD thesis "Methods for
 Distributed Information Retrieval" after three and a half years of
 study at the Australian National University.
 http://pastime.anu.edu.au/nick/