Stemming unstructured text in NLTK

Question

I tried the regex stemmer, but I get hundreds of unrelated tokens. I'm just interested in the "play" stem. Here is the code I'm working with:

import nltk
from nltk.book import *
f = open('tupac_original.txt', 'rU')
text = f.read()
text1 = text.split()
tup = nltk.Text(text1)
lowtup = [w.lower() for w in tup if w.isalpha()]
import sys, re
tupclean = [w for w in lowtup if not w in nltk.corpus.stopwords.words('english')]
from nltk import stem
tupstem = stem.RegexpStemmer('az$|as$|a$')
[tupstem.stem(i) for i in tupclean] 

The result of the above is;

['like', 'ed', 'young', 'black', 'like'...]

I'm trying to clean up .txt files (all lowercase, remove stopwords, etc), normalize multiple spellings of a word into one and do a frequency dist/count. I know how to do FreqDist, but any suggestions as to where I'm going wrong with the stemming?

Answer 1

There are several pre-coded well-known stemmers in NLTK, see http://nltk.org/api/nltk.stem.html and below shows an example.

>>> from nltk import stem
>>> porter = stem.porter.PorterStemmer()
>>> lancaster = stem.lancaster.LancasterStemmer()
>>> snowball = stem.snowball.EnglishStemmer()
>>> tokens =  ['player', 'playa', 'playas', 'pleyaz'] 
>>> [porter(i) for i in tokens]
>>> [porter.stem(i) for i in tokens]
['player', 'playa', 'playa', 'pleyaz']
>>> [lancaster.stem(i) for i in tokens]
['play', 'play', 'playa', 'pleyaz']
>>> [snowball.stem(i) for i in tokens]
[u'player', u'playa', u'playa', u'pleyaz']

But what you probably need is some sort of a regex stemmer,

>>> from nltk import stem
>>> rxstem = stem.RegexpStemmer('er$|a$|as$|az$')
>>> [rxstem.stem(i) for i in tokens]
['play', 'play', 'play', 'pley']