TITLE: Long span contextual features for Text Classification and Entity Role Labelling

PRESENTER: Sreeram Balakrishnan, , sreevb@us.ibm.com

AFFILIATION:IBM India Research Laboratory (IRL), Indian Institute of Technology, New Delhi, http://www.research.ibm.com/irl/

DATE: Friday 21th July 2006

TIME: 12:00:00

PLACE: CSE Seminar Room, Level 1, K17

ABSTRACT:

At the IRL we are exploring two methods of finding long
span contextual features to improve the accuracy of
text classifiers and classifiers for entity role
labelling. The first method involves a modified
version of the a-priori algorithm to mine high support
context patterns directly from text, and the second
involves using Inductive Logic Programming to find
high support predicates that we then treat as
features. I will present our results for text
classification and entity role labelling and discuss
the pro and cons if these approaches.

BIOGRAPHY OF SPEAKER:

Host:

Ashesh Mahidadia

Seminar Convenor:

Van Hai Ho