Hello,
I'm developing nlp application for extracting named entities. The problem with
built-in nltk name entity module is that the classes are fixed.
For example, if I want to create table of entities like this:
TOOL - any mechinacal implement usable by human (hammer, axe, etc.)
QUANTITY - any human measurment (meter, pound, inch, liter, gram, etc.)
BMATERIAL - any building material (wood, brick, stone, sand, etc.)
Let's say I have some tagged data in IOB form, which would be splitted in
training and testing portions. Wordnet would be nice basis for features.
Is there a way to create custom classes and their features, and then to train
the algorithm with tagged data?
--
Nikola Jelić
EuroICC d.o.o.
Tršćanska 21
11080 Zemun, Srbija
tel: +381 11 371 36 65
fax: +381 11 371 36 66
e-mail:
nikola...@euroicc.com
web:
www.euroicc.com