Improving Training Data for sentiment analysis with NLTK





So now it is time to train on a new data set. Our goal is to do Twitter sentiment, so we're hoping for a data set that is a bit shorter per positive and negative statement. It just so happens that I have a data set of 5300+ positive and 5300+ negative movie reviews, which are much shorter. These should give us a bit more accuracy from the larger training set, as well as be more fitting for tweets from Twitter.

I have hosted both files here, you can find them by going to the downloads for the short reviews. Save these files as positive.txt and negative.txt.

Now, we can build our new data set in a very similar way as before. What needs to change?

We need a new methodology for creating our "documents" variable, and then we also need a new way to create the "all_words" variable. No problem, really, here's how I did it:

short_pos = open("short_reviews/positive.txt","r").read()
short_neg = open("short_reviews/negative.txt","r").read()

documents = []

for r in short_pos.split('\n'):
    documents.append( (r, "pos") )

for r in short_neg.split('\n'):
    documents.append( (r, "neg") )


all_words = []

short_pos_words = word_tokenize(short_pos)
short_neg_words = word_tokenize(short_neg)

for w in short_pos_words:
    all_words.append(w.lower())

for w in short_neg_words:
    all_words.append(w.lower())

all_words = nltk.FreqDist(all_words)

Next, we also need to adjust our feature finding function, mainly tokenizing by word in the document, since we didn't have a nifty .words() feature for our new sample. I also went ahead and increased the most common words:

word_features = list(all_words.keys())[:5000]

def find_features(document):
    words = word_tokenize(document)
    features = {}
    for w in word_features:
        features[w] = (w in words)

    return features
	
featuresets = [(find_features(rev), category) for (rev, category) in documents]
random.shuffle(featuresets)

Other than this, the rest is the same. Here's the full script just in case you or I missed something:

This process will take a while.. You may want to just go run some errands. It took me about 30-40 minutes to run it in full, and I am running an i7 3930k. For the typical processor in the year I am writing this (2015), it may be hours. This is a one and done process, however.

import nltk
import random
from nltk.corpus import movie_reviews
from nltk.classify.scikitlearn import SklearnClassifier
import pickle

from sklearn.naive_bayes import MultinomialNB, BernoulliNB
from sklearn.linear_model import LogisticRegression, SGDClassifier
from sklearn.svm import SVC, LinearSVC, NuSVC

from nltk.classify import ClassifierI
from statistics import mode

from nltk.tokenize import word_tokenize


class VoteClassifier(ClassifierI):
    def __init__(self, *classifiers):
        self._classifiers = classifiers

    def classify(self, features):
        votes = []
        for c in self._classifiers:
            v = c.classify(features)
            votes.append(v)
        return mode(votes)

    def confidence(self, features):
        votes = []
        for c in self._classifiers:
            v = c.classify(features)
            votes.append(v)

        choice_votes = votes.count(mode(votes))
        conf = choice_votes / len(votes)
        return conf
        
short_pos = open("short_reviews/positive.txt","r").read()
short_neg = open("short_reviews/negative.txt","r").read()

documents = []

for r in short_pos.split('\n'):
    documents.append( (r, "pos") )

for r in short_neg.split('\n'):
    documents.append( (r, "neg") )


all_words = []

short_pos_words = word_tokenize(short_pos)
short_neg_words = word_tokenize(short_neg)

for w in short_pos_words:
    all_words.append(w.lower())

for w in short_neg_words:
    all_words.append(w.lower())

all_words = nltk.FreqDist(all_words)

word_features = list(all_words.keys())[:5000]

def find_features(document):
    words = word_tokenize(document)
    features = {}
    for w in word_features:
        features[w] = (w in words)

    return features

#print((find_features(movie_reviews.words('neg/cv000_29416.txt'))))

featuresets = [(find_features(rev), category) for (rev, category) in documents]

random.shuffle(featuresets)

# positive data example:      
training_set = featuresets[:10000]
testing_set =  featuresets[10000:]

##
### negative data example:      
##training_set = featuresets[100:]
##testing_set =  featuresets[:100]


classifier = nltk.NaiveBayesClassifier.train(training_set)
print("Original Naive Bayes Algo accuracy percent:", (nltk.classify.accuracy(classifier, testing_set))*100)
classifier.show_most_informative_features(15)

MNB_classifier = SklearnClassifier(MultinomialNB())
MNB_classifier.train(training_set)
print("MNB_classifier accuracy percent:", (nltk.classify.accuracy(MNB_classifier, testing_set))*100)

BernoulliNB_classifier = SklearnClassifier(BernoulliNB())
BernoulliNB_classifier.train(training_set)
print("BernoulliNB_classifier accuracy percent:", (nltk.classify.accuracy(BernoulliNB_classifier, testing_set))*100)

LogisticRegression_classifier = SklearnClassifier(LogisticRegression())
LogisticRegression_classifier.train(training_set)
print("LogisticRegression_classifier accuracy percent:", (nltk.classify.accuracy(LogisticRegression_classifier, testing_set))*100)

SGDClassifier_classifier = SklearnClassifier(SGDClassifier())
SGDClassifier_classifier.train(training_set)
print("SGDClassifier_classifier accuracy percent:", (nltk.classify.accuracy(SGDClassifier_classifier, testing_set))*100)

##SVC_classifier = SklearnClassifier(SVC())
##SVC_classifier.train(training_set)
##print("SVC_classifier accuracy percent:", (nltk.classify.accuracy(SVC_classifier, testing_set))*100)

LinearSVC_classifier = SklearnClassifier(LinearSVC())
LinearSVC_classifier.train(training_set)
print("LinearSVC_classifier accuracy percent:", (nltk.classify.accuracy(LinearSVC_classifier, testing_set))*100)

NuSVC_classifier = SklearnClassifier(NuSVC())
NuSVC_classifier.train(training_set)
print("NuSVC_classifier accuracy percent:", (nltk.classify.accuracy(NuSVC_classifier, testing_set))*100)


voted_classifier = VoteClassifier(
                                  NuSVC_classifier,
                                  LinearSVC_classifier,
                                  MNB_classifier,
                                  BernoulliNB_classifier,
                                  LogisticRegression_classifier)

print("voted_classifier accuracy percent:", (nltk.classify.accuracy(voted_classifier, testing_set))*100)

Output:

Original Naive Bayes Algo accuracy percent: 66.26506024096386
Most Informative Features
              refreshing = True              pos : neg    =     13.6 : 1.0
                captures = True              pos : neg    =     11.3 : 1.0
                  stupid = True              neg : pos    =     10.7 : 1.0
                  tender = True              pos : neg    =      9.6 : 1.0
              meandering = True              neg : pos    =      9.1 : 1.0
                      tv = True              neg : pos    =      8.6 : 1.0
                 low-key = True              pos : neg    =      8.3 : 1.0
              thoughtful = True              pos : neg    =      8.1 : 1.0
                   banal = True              neg : pos    =      7.7 : 1.0
              amateurish = True              neg : pos    =      7.7 : 1.0
                terrific = True              pos : neg    =      7.6 : 1.0
                  record = True              pos : neg    =      7.6 : 1.0
             captivating = True              pos : neg    =      7.6 : 1.0
                portrait = True              pos : neg    =      7.4 : 1.0
                 culture = True              pos : neg    =      7.3 : 1.0
MNB_classifier accuracy percent: 65.8132530120482
BernoulliNB_classifier accuracy percent: 66.71686746987952
LogisticRegression_classifier accuracy percent: 67.16867469879519
SGDClassifier_classifier accuracy percent: 65.8132530120482
LinearSVC_classifier accuracy percent: 66.71686746987952
NuSVC_classifier accuracy percent: 60.09036144578314
voted_classifier accuracy percent: 65.66265060240963

Yeah, I bet that took you a while, so, in the next tutorial, we're going to talk about pickling everything!

The next tutorial:






  • Tokenizing Words and Sentences with NLTK
  • Stop words with NLTK
  • Stemming words with NLTK
  • Part of Speech Tagging with NLTK
  • Chunking with NLTK
  • Chinking with NLTK
  • Named Entity Recognition with NLTK
  • Lemmatizing with NLTK
  • The corpora with NLTK
  • Wordnet with NLTK
  • Text Classification with NLTK
  • Converting words to Features with NLTK
  • Naive Bayes Classifier with NLTK
  • Saving Classifiers with NLTK
  • Scikit-Learn Sklearn with NLTK
  • Combining Algorithms with NLTK
  • Investigating bias with NLTK
  • Improving Training Data for sentiment analysis with NLTK
  • Creating a module for Sentiment Analysis with NLTK
  • Twitter Sentiment Analysis with NLTK
  • Graphing Live Twitter Sentiment Analysis with NLTK with NLTK
  • Named Entity Recognition with Stanford NER Tagger
  • Testing NLTK and Stanford NER Taggers for Accuracy
  • Testing NLTK and Stanford NER Taggers for Speed
  • Using BIO Tags to Create Readable Named Entity Lists