Improving the extraction of human names with nltk

I am trying to extract human names from text.

Does anyone have a method that they would recommend?

This is what I tried (code is below): I am using nltk to find everything marked as a person and then generating a list of all the NNP parts of that person. I am skipping persons where there is only one NNP which avoids grabbing a lone surname.

I am getting decent results but was wondering if there are better ways to go about solving this problem.

Code:

import nltk
from nameparser.parser import HumanName

def get_human_names(text):
    tokens = nltk.tokenize.word_tokenize(text)
    pos = nltk.pos_tag(tokens)
    sentt = nltk.ne_chunk(pos, binary = False)
    person_list = []
    person = []
    name = ""
    for subtree in sentt.subtrees(filter=lambda t: t.node == 'PERSON'):
        for leaf in subtree.leaves():
            person.append(leaf[0])
        if len(person) > 1: #avoid grabbing lone surnames
            for part in person:
                name += part + ' '
            if name[:-1] not in person_list:
                person_list.append(name[:-1])
            name = ''
        person = []

    return (person_list)

text = """
Some economists have responded positively to Bitcoin, including 
Francois R. Velde, senior economist of the Federal Reserve in Chicago 
who described it as "an elegant solution to the problem of creating a 
digital currency." In November 2013 Richard Branson announced that 
Virgin Galactic would accept Bitcoin as payment, saying that he had invested 
in Bitcoin and found it "fascinating how a whole new global currency 
has been created", encouraging others to also invest in Bitcoin.
Other economists commenting on Bitcoin have been critical. 
Economist Paul Krugman has suggested that the structure of the currency 
incentivizes hoarding and that its value derives from the expectation that 
others will accept it as payment. Economist Larry Summers has expressed 
a "wait and see" attitude when it comes to Bitcoin. Nick Colas, a market 
strategist for ConvergEx Group, has remarked on the effect of increasing 
use of Bitcoin and its restricted supply, noting, "When incremental 
adoption meets relatively fixed supply, it should be no surprise that 
prices go up. And that’s exactly what is happening to BTC prices."
"""

names = get_human_names(text)
print "LAST, FIRST"
for name in names: 
    last_first = HumanName(name).last + ', ' + HumanName(name).first
        print last_first

Output:

LAST, FIRST
Velde, Francois
Branson, Richard
Galactic, Virgin
Krugman, Paul
Summers, Larry
Colas, Nick

Apart from Virgin Galactic, this is all valid output. Of course, knowing that Virgin Galactic isn't a human name in the context of this article is the hard (maybe impossible) part.


ANSWERS:


Must agree with suggestion that "make my code better" isn't well suited for this site, but I can give you some way where you can try to dig in.

Take a look at Stanford Named Entity Recognizer (NER). Its binding has been included into NLTK v 2.0, but you must download some core files. Here is script which can do all of that for you.

I wrote this script:

import nltk
from nltk.tag.stanford import NERTagger
st = NERTagger('stanford-ner/all.3class.distsim.crf.ser.gz', 'stanford-ner/stanford-ner.jar')
text = """YOUR TEXT GOES HERE"""

for sent in nltk.sent_tokenize(text):
    tokens = nltk.tokenize.word_tokenize(sent)
    tags = st.tag(tokens)
    for tag in tags:
        if tag[1]=='PERSON': print tag

and got not so bad output:

('Francois', 'PERSON') ('R.', 'PERSON') ('Velde', 'PERSON') ('Richard', 'PERSON') ('Branson', 'PERSON') ('Virgin', 'PERSON') ('Galactic', 'PERSON') ('Bitcoin', 'PERSON') ('Bitcoin', 'PERSON') ('Paul', 'PERSON') ('Krugman', 'PERSON') ('Larry', 'PERSON') ('Summers', 'PERSON') ('Bitcoin', 'PERSON') ('Nick', 'PERSON') ('Colas', 'PERSON')

Hope this is helpful.


You can try to do resolution of the found names, and check if you can find them in a database such as freebase.com. Get the data locally and query it (it's in RDF), or use google's api: . Most big companies, geographical locations, etc. (that would be caught by your snippet) could be then discarded based on the freebase data.


For anyone else looking, I found this article to be useful:

>>> import nltk
>>> def extract_entities(text):
...     for sent in nltk.sent_tokenize(text):
...         for chunk in nltk.ne_chunk(nltk.pos_tag(nltk.word_tokenize(sent))):
...             if hasattr(chunk, 'node'):
...                 print chunk.node, ' '.join(c[0] for c in chunk.leaves())
...

This worked pretty well for me. I just had to change one line in order for it to run.

    for subtree in sentt.subtrees(filter=lambda t: t.node == 'PERSON'):

needs to be

    for subtree in sentt.subtrees(filter=lambda t: t.label() == 'PERSON'):

There were imperfections in the output (for example it identified "Money Laundering" as a person), but with my data a name database may not be dependable.


The answer of @trojane didn't quite work for me, but helped a lot for this one.

Prerequesites

Create a folder stanford-ner and download the following two files to it:

Script

#!/usr/bin/env python
# -*- coding: utf-8 -*-

import nltk
from nltk.tag.stanford import StanfordNERTagger

text = u"""
Some economists have responded positively to Bitcoin, including
Francois R. Velde, senior economist of the Federal Reserve in Chicago
who described it as "an elegant solution to the problem of creating a
digital currency." In November 2013 Richard Branson announced that
Virgin Galactic would accept Bitcoin as payment, saying that he had invested
in Bitcoin and found it "fascinating how a whole new global currency
has been created", encouraging others to also invest in Bitcoin.
Other economists commenting on Bitcoin have been critical.
Economist Paul Krugman has suggested that the structure of the currency
incentivizes hoarding and that its value derives from the expectation that
others will accept it as payment. Economist Larry Summers has expressed
a "wait and see" attitude when it comes to Bitcoin. Nick Colas, a market
strategist for ConvergEx Group, has remarked on the effect of increasing
use of Bitcoin and its restricted supply, noting, "When incremental
adoption meets relatively fixed supply, it should be no surprise that
prices go up. And that’s exactly what is happening to BTC prices.
"""

st = StanfordNERTagger('stanford-ner/english.all.3class.distsim.crf.ser.gz',
                       'stanford-ner/stanford-ner.jar')

for sent in nltk.sent_tokenize(text):
    tokens = nltk.tokenize.word_tokenize(sent)
    tags = st.tag(tokens)
    for tag in tags:
        if tag[1] in ["PERSON", "LOCATION", "ORGANIZATION"]:
            print(tag)

Results

(u'Bitcoin', u'LOCATION')       # wrong
(u'Francois', u'PERSON')
(u'R.', u'PERSON')
(u'Velde', u'PERSON')
(u'Federal', u'ORGANIZATION')
(u'Reserve', u'ORGANIZATION')
(u'Chicago', u'LOCATION')
(u'Richard', u'PERSON')
(u'Branson', u'PERSON')
(u'Virgin', u'PERSON')         # Wrong
(u'Galactic', u'PERSON')       # Wrong
(u'Bitcoin', u'PERSON')        # Wrong
(u'Bitcoin', u'LOCATION')      # Wrong
(u'Bitcoin', u'LOCATION')      # Wrong
(u'Paul', u'PERSON')
(u'Krugman', u'PERSON')
(u'Larry', u'PERSON')
(u'Summers', u'PERSON')
(u'Bitcoin', u'PERSON')        # Wrong
(u'Nick', u'PERSON')
(u'Colas', u'PERSON')
(u'ConvergEx', u'ORGANIZATION')
(u'Group', u'ORGANIZATION')     
(u'Bitcoin', u'LOCATION')       # Wrong
(u'BTC', u'ORGANIZATION')       # Wrong


 MORE:


 ? what is wrong with this code of nltk python
 ? Python NLTK interpret a fixed pattern of sentence and tokenize it
 ? NLTK Sentence boundary error
 ? NLTK / Python: Tokenising text to fixed token length
 ? NLTK / Python: Tokenising text to fixed token length
 ? NLTK / Python: Tokenising text to fixed token length
 ? NLTK tokenize questions
 ? How to print tokenized arabic text in python/nltk?
 ? Python NLTK Collocations for tagged text
 ? NLTK tokenize with collocations