Transcript Laura

Shock Group
NER Replacement
Laura Christiansen
Overview
 Currently using MetaMap for NER
 Types: Experimental Platform, Condition, Cell Type, Molecule,
Drug/Chemical Compound/Therapeutic Modality
 Errors present
 Trainable alternative
 Construct new model with ABNER
(http://pages.cs.wisc.edu/~bsettles/abner/)
Current Work
 Modified ABNER source code
 Included additional orthographic feature selection
 Updated MALLET usage and references
(http://mallet.cs.umass.edu/)
 Worked with JLex for ABNER tokenization scanner
(http://www.cs.princeton.edu/~appel/modern/java/JLex/)
 Incorporating new functionality
 Stratified cross validation
Future Work
 Finish coding (and buy more coffee)
 Test new model with stratified cross validation
 Train with modified input from MetaMap-processed abstracts
 Format output to be usable with stage 2