NLP | Named Entity Chunker Training

Self Named entity chunker can be trained using the ieer corpus, which stands for Information Extraction: Entity Recognition. The ieer corpus has chunk trees but no part-of-speech tags for the words, so it is a bit tedious job to perform.

Named entity chunk trees can be created from ieer corpus using the ieertree2conlltags() and ieer_chunked_sents() functions. This can be used to train the ClassifierChunker class created in the Classification-based chunking.

Code #1 : ieertree2conlltags()

import nltk.tag
from nltk.chunk.util import conlltags2tree
from nltk.corpus import ieer

def ieertree2conlltags(tree, tag = nltk.tag.pos_tag):
    words, ents = zip(*tree.pos())
    iobs = []
    prev = None
    for ent in ents:
        if ent == tree.label():
            iobs.append('O')
            prev = None
        elif prev == ent:
            iobs.append('I-% s' % ent)
        else:
            iobs.append('B-% s' % ent)
            prev = ent

    words, tags = zip(*tag(words))

    return zip(words, tags, iobs)

Code #2 : ieer_chunked_sents()

import nltk.tag
from nltk.chunk.util import conlltags2tree
from nltk.corpus import ieer

def ieer_chunked_sents(tag = nltk.tag.pos_tag):
    for doc in ieer.parsed_docs():
        tagged = ieertree2conlltags(doc.text, tag)
        yield conlltags2tree(tagged)

Using 80 out of 94 sentences for training and the remaining ones for testing.

Code #3 : How the classifier works on the first sentence of the treebank_chunk corpus.

from nltk.corpus import ieer
from chunkers import ieer_chunked_sents, ClassifierChunker
from nltk.corpus import treebank_chunk

ieer_chunks = list(ieer_chunked_sents())

print ("Length of ieer_chunks : ", len(ieer_chunks))

# initializing chunker
chunker = ClassifierChunker(ieer_chunks[:80])
print("\nparsing : \n", chunker.parse(
        treebank_chunk.tagged_sents()[0]))

# evaluating
score = chunker.evaluate(ieer_chunks[80:])

a = score.accuracy()
p = score.precision()
r = score.recall()

print ("\nAccuracy : ", a)
print ("\nPrecision : ", p)
print ("\nRecall : ", r)

Output :

Length of ieer_chunks : 94

parsing : 
Tree('S', [Tree('LOCATION', [('Pierre', 'NNP'), ('Vinken', 'NNP')]),
(', ', ', '), Tree('DURATION', [('61', 'CD'), ('years', 'NNS')]),
Tree('MEASURE', [('old', 'JJ')]), (', ', ', '), ('will', 'MD'), ('join', 'VB'), 
('the', 'DT'), ('board', 'NN'), ('as', 'IN'), ('a', 'DT'), ('nonexecutive', 'JJ'),
('director', 'NN'), Tree('DATE', [('Nov.', 'NNP'), ('29', 'CD')]), ('.', '.')])

Accuracy : 0.8829018388070625

Precision : 0.4088717454194793

Recall : 0.5053635280095352

How it works ?
The ieer trees generated by ieer_chunked_sents() are not entirely accurate. There are no explicit sentence breaks, so each document is a single tree. Also, the words are not explicitly tagged, it’s guess work using nltk.tag.pos_tag().