BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Inference Group
SUMMARY:Hierarchical Passage Retrieval - Phil Cowans\, Inf
 erence Group
DTSTART;TZID=Europe/London:20060419T150000
DTEND;TZID=Europe/London:20060419T160000
UID:TALK4851AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/4851
DESCRIPTION:Much recent research in information retrieval has 
 concentrated on approaches which make use of langu
 age models to score documents. Typically\, languag
 e models are constructed from the collection and r
 elevance is defined in terms of the probability of
  the query string under these models. In particula
 r\, use of hierarchical Dirichlet models of the wh
 ole collection can be shown to naturally provide m
 any desirable features\, including term weighting 
 similar to the well known tf.idf scheme. \n\nIn th
 is talk I will describe the extension of this mode
 l to include the notion of a further subdivision o
 f documents into passages. This model can be used 
 to score documents while at the same time provides
  an indication of parts of the documents which are
  particularly relevant. A simple application of th
 is model is the construction of summary informatio
 n which can be displayed alongside the search resu
 lts\, although there are many other cases in which
  this information is of use. I shall present a com
 parison of this model with alternate approaches us
 ing standard test sets.
LOCATION:Ryle Seminar Room\, Cavendish Laboratory
CONTACT:Phil Cowans
END:VEVENT
END:VCALENDAR