BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Drupal iCal API//EN
X-WR-CALNAME:Events items teaser
X-WR-TIMEZONE:America/Toronto
BEGIN:VTIMEZONE
TZID:America/Toronto
X-LIC-LOCATION:America/Toronto
BEGIN:DAYLIGHT
TZNAME:EDT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
DTSTART:20130310T070000
END:DAYLIGHT
BEGIN:STANDARD
TZNAME:EST
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
DTSTART:20131103T060000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:682210bbad6a5
DTSTART;TZID=America/Toronto:20140226T110000
SEQUENCE:0
TRANSP:TRANSPARENT
DTEND;TZID=America/Toronto:20140226T110000
URL:/centre-pattern-analysis-machine-intelligence/event
 s/invited-talk-modeling-term-associations-probabilistic
SUMMARY:Invited Talk: Modeling Term Associations for Probabilistic Informat
 ion\nRetrieval
CLASS:PUBLIC
DESCRIPTION:Summary \n\nMODELING TERM ASSOCIATIONS FOR PROBABILISTIC INFORM
 ATION RETRIEVAL\n\nABSTRACT:\n\nTraditionally\, in many probabilistic retr
 ieval models\, query terms are\nassumed to be independent. Although such m
 odels can achieve reasonably\ngood performance\, associations can exist am
 ong terms from human\nbeing.s point of view. There are some recent studies
  that investigate\nhow to model term associations/dependencies by proximit
 y measures.\nHowever\, the modeling of term associations theoretically und
 er the\nprobabilistic retrieval framework is still largely unexplored. In 
 this\ntalk\, I will introduce a new concept named Cross Term\, to model te
 rm\nproximity\, with the aim of boosting retrieval performance. With Cross
 \nTerms\, the association of multiple query terms can be modeled in the\ns
 ame way as a simple unigram term. In particular\, an occurrence of a\nquer
 y term is assumed to have an impact on its neighboring text. The\ndegree o
 f the query term impact gradually weakens with increasing\ndistance from t
 he place of occurrence. We use shape functions to\ncharacterize such impac
 ts. Based on this assumption\, we first propose\na bigram CRoss TErm Retri
 eval (CRTER2) model as the basis model\, and\nthen recursively propose a g
 eneralized n-gram CRoss TErm Retrieval\n(CRTERn) model for n query terms w
 here n > 2. \nSpecifically\, a bigram Cross Term occurs when the correspon
 ding query\nterms appear close to each other\, and its impact can be model
 ed by the\nintersection of the respective shape functions of the query ter
 ms. For\nn-gram Cross Term\, we develop several distance metrics with diff
 erent\nproperties and employ them in the proposed models for ranking. We a
 lso\nshow how to extend the language model using the newly proposed cross\
 nterms. Extensive experiments on a number of TREC collections\ndemonstrate
  the effectiveness of our proposed models. \n\nBIOGRAPHY:\n\nJIMMY HUANG i
 s a Professor & Director at the School of Information\nTechnology and the 
 founding director of Information Retrieval &\nKnowledge Management Researc
 h Lab at the York University. 
DTSTAMP:20250512T151611Z
END:VEVENT
END:VCALENDAR