BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Improving Multiclass Text Classification with Error-Correcting Out
 put Coding and Sub-class Partitions - Stuart Moore
DTSTART:20110117T123000Z
DTEND:20110117T133000Z
UID:TALK28103@talks.cam.ac.uk
CONTACT:Marek Rei
DESCRIPTION:Stuart will present the following paper:\n\nImproving Multicla
 ss Text Classification with Error-Correcting Output Coding and Sub-class P
 artitions. 2010. Baoli Li and Carl Vogel.\n\nhttp://www.springerlink.com/c
 ontent/d546042162276641/\n\nError-Correcting Output Coding (ECOC) is a gen
 eral framework for multiclass text classification with a set of binary cla
 ssifiers. It can not only help a binary classifier solve multi-class class
 ification problems\, but also boost the performance of a multi-class class
 ifier. When building each individual binary classifier in ECOC\, multiple 
 classes are randomly grouped into two disjoint groups: positive and negati
 ve. However\, when training such a binary classifier\, sub-class distribut
 ion within positive and negative classes is neglected. Utilizing this info
 rmation is expected to improve a binary classifier. We thus design a simpl
 e binary classification strategy via multi-class categorization (2vM) to m
 ake use of sub-class partition information\, which can lead to better perf
 ormance over the traditional binary classification. The proposed binary cl
 assification strategy is then applied to enhance ECOC. Experiments on docu
 ment categorization and question classification show its effectiveness. \n
 \nAnyone interested in more background material might want to look at http
 ://arxiv.org/abs/cs/9501101 which was the original paper introducing this 
 method.\n
LOCATION:GS15\, Computer Laboratory
END:VEVENT
END:VCALENDAR