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:NLIP Seminar Series
SUMMARY:A Weakly-supervised Approach to Argumentative Zoni
 ng of Scientific Documents - Yufan Guo\, Universit
 y of Cambridge
DTSTART;TZID=Europe/London:20110718T120000
DTEND;TZID=Europe/London:20110718T130000
UID:TALK32129AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/32129
DESCRIPTION:Argumentative Zoning (AZ) – analysis of the argume
 ntative structure of a scientific paper – has prov
 ed useful for a number of information access tasks
 . Current approaches to AZ rely on supervised mach
 ine learning (ML). Requiring large amounts of anno
 tated data\, these approaches are expensive to dev
 elop and port to different domains and tasks. A po
 tential solution to this problem is to use weaklys
 upervised ML instead. We investigate the performan
 ce of four weakly-supervised classifiers on scient
 ific abstract data annotated for multiple AZ class
 es. Our best classifier based on the combination o
 f active learning and selftraining outperforms our
  best supervised classifier\, yielding a high accu
 racy of 81% when\nusing just 10% of the labeled da
 ta. This result suggests that weakly-supervised le
 arning could be employed to improve the practical 
 applicability and portability of AZ across differe
 nt information access tasks. 
LOCATION:FW09\, Computer Laboratory
CONTACT:Thomas Lippincott
END:VEVENT
END:VCALENDAR