CODEWITHSUNDEEP
pythonnlpnltk
sachinruk
sachinruk

Reputation: 9869

Split words on boundary

I have some tweets which I wish to split into words. Most of it works fine except when people combine words like: trumpisamoron or makeamericagreatagain. But then there are also things like password which shouldn't be split up into pass and word.

I know that the nltk package has a punkt tokenizer module which splits sentences up in a smart way. Is there something similar for words? Even if it isn't in the nltk package?

Note: The example of password -> pass + word is much less of a problem than the splitting word problem.

Upvotes: 1

Views: 160

Answers (1)

RAVI
RAVI

Reputation: 3153

Ref : My Answer on another Question - Need to split #tags to text.

Changes in this answer I made are - (1) Different corpus to get WORDS and (2) Added def memo(f) to speed up process. You may need to add/use corpus depending upon Domain you are working on.

Check - Word Segmentation Task from Norvig's work.

from __future__ import division
from collections import Counter
import re, nltk
from datetime import datetime

WORDS = nltk.corpus.reuters.words() + nltk.corpus.words.words()
COUNTS = Counter(WORDS)

def memo(f):
    "Memoize function f, whose args must all be hashable."
    cache = {}
    def fmemo(*args):
        if args not in cache:
            cache[args] = f(*args)
        return cache[args]
    fmemo.cache = cache
    return fmemo

def pdist(counter):
    "Make a probability distribution, given evidence from a Counter."
    N = sum(counter.values())
    return lambda x: counter[x]/N

P = pdist(COUNTS)

def Pwords(words):
    "Probability of words, assuming each word is independent of others."
    return product(P(w) for w in words)

def product(nums):
    "Multiply the numbers together.  (Like `sum`, but with multiplication.)"
    result = 1
    for x in nums:
        result *= x
    return result

def splits(text, start=0, L=20):
    "Return a list of all (first, rest) pairs; start <= len(first) <= L."
    return [(text[:i], text[i:]) 
            for i in range(start, min(len(text), L)+1)]

@memo
def segment(text):
    "Return a list of words that is the most probable segmentation of text."
    if not text: 
        return []
    else:
        candidates = ([first] + segment(rest) 
                      for (first, rest) in splits(text, 1))
        return max(candidates, key=Pwords)

print segment('password')     # ['password']
print segment('makeamericagreatagain')     # ['make', 'america', 'great', 'again']
print segment('trumpisamoron')     # ['trump', 'is', 'a', 'moron']
print segment('narcisticidiots')     # ['narcistic', 'idiot', 's']

Sometimes, in case, word gets spilt into smaller token, there may be higher chances that word is not present in our WORDS Dictionary.

Here in last segment, it broke narcisticidiots into 3 tokens because token idiots was not there in our WORDS.

# Check for sample word 'idiots'
if 'idiots' in WORDS:
    print("YES")
else:
    print("NO")

You can add new user defined words to WORDS.

.
.
user_words = []
user_words.append('idiots')

WORDS+=user_words
COUNTS = Counter(WORDS)
.
.
.
print segment('narcisticidiots')     # ['narcistic', 'idiots']

For better solution than this you can use bigram/trigram.

More examples at : Word Segmentation Task

Upvotes: 1

Related Questions