SVMSemanticClassifier.py

# coding: utf-8

# In[10]:

import collections
import nltk.classify.util, nltk.metrics
from nltk.classify import SklearnClassifier
from sklearn.naive_bayes import BernoulliNB
from sklearn.svm import SVC
from nltk.corpus import movie_reviews
from nltk.metrics import scores
from nltk import precision
import itertools
from nltk.collocations import BigramCollocationFinder
from nltk.metrics import BigramAssocMeasures


# In[11]:

def evaluate_classifier(featx,collocationFunc):
    #negFiles = movie_reviews.fileids('neg')
    #posFiles = movie_reviews.fileids('pos')
    #negWordsList=[movie_reviews.words(fileids=[f]) for f in negFiles]
    #posWordsList=[movie_reviews.words(fileids=[f]) for f in posFiles]
    #negfeats = [(featx(negWords), 'neg') for negWords in negWordsList]
    #posfeats = [(featx(posWords), 'pos') for posWords in posWordsList]

    negids = movie_reviews.fileids('neg')
    posids = movie_reviews.fileids('pos')
 
    negfeats = [(featx(movie_reviews.words(fileids=[f]),collocationFunc), 'neg') for f in negids]
    posfeats = [(featx(movie_reviews.words(fileids=[f]),collocationFunc), 'pos') for f in posids]

    #lenNegFeats=min(len(negfeats),24)
    #lenPosFeats=min(len(posfeats),24)
    lenNegFeats=len(negfeats)
    lenPosFeats=len(posfeats)
    negcutoff = int(lenNegFeats*3/4)
    poscutoff = int(lenPosFeats*3/4)
 
    trainfeats = negfeats[:negcutoff] + posfeats[:poscutoff]
    testfeats = negfeats[negcutoff:lenNegFeats] + posfeats[poscutoff:lenPosFeats]
 
    #classifier = MaxentClassifier.train(trainfeats)
    classifier = SklearnClassifier(BernoulliNB()).train(trainfeats)
    refsets = collections.defaultdict(set)
    testsets = collections.defaultdict(set)
 
    for i, (feats, label) in enumerate(testfeats):
            refsets[label].add(i)
            observed = classifier.classify(feats)
            testsets[observed].add(i)
    evaluationMetrics={}
    print(classifier)
    evaluationMetrics['accuracy']=nltk.classify.util.accuracy(classifier, testfeats)
    evaluationMetrics['posPrec']=nltk.precision(refsets['pos'], testsets['pos'])
    evaluationMetrics['posRecall']=nltk.recall(refsets['pos'], testsets['pos'])
    evaluationMetrics['posF_Score']=nltk.f_measure(refsets['pos'], testsets['pos'])
    evaluationMetrics['negPrec']=nltk.precision(refsets['neg'], testsets['neg'])
    evaluationMetrics['negRecall']=nltk.recall(refsets['neg'], testsets['neg'])
    evaluationMetrics['negF_Score']=nltk.f_measure(refsets['neg'], testsets['neg'])
    return evaluationMetrics


# In[14]:

from nltk.corpus import stopwords
stopset = set(stopwords.words('english'))
evaluations=[] 
def stopword_filtered_word_feats(words,collocator):
    return dict([(word, True) for word in words if word not in stopset])
evaluations.append(evaluate_classifier(stopword_filtered_word_feats,None)) 


# In[15]:

evaluations


# In[13]:

#Bigram Collocations- Handle Cases like “not good”, here B-O-W Approach will Fail
import itertools
from nltk.collocations import BigramCollocationFinder
from nltk.metrics import BigramAssocMeasures
 
def bigram_word_feats(words, score_fn, n=200):
    bigram_finder = BigramCollocationFinder.from_words(words)
    bigrams = bigram_finder.nbest(score_fn, n)
    return dict([(ngram, True) for ngram in itertools.chain(words, bigrams)])
 
evaluations.append(evaluate_classifier(bigram_word_feats,BigramAssocMeasures.chi_sq))#Works best for this Data
evaluations.append(evaluate_classifier(bigram_word_feats,BigramAssocMeasures.jaccard))
evaluations.append(evaluate_classifier(bigram_word_feats,BigramAssocMeasures.likelihood_ratio))


# In[ ]:

from nltk.collocations import *
from nltk.probability import FreqDist
from nltk.probability import ConditionalFreqDist
word_fd = FreqDist()
label_word_fd = ConditionalFreqDist()

testNegWords = movie_reviews.words(categories=['pos'])
testPosWords = movie_reviews.words(categories=['neg'])

for word in testNegWords:
    word_fd[word.lower()]+=1
    label_word_fd['neg'][word.lower()]+=1
for word in testPosWords:
    word_fd[word.lower()]+=1
    label_word_fd['pos'][word.lower()]+=1
print(word_fd.N(),word_fd.B(),word_fd.most_common(20))
print(label_word_fd.N(),label_word_fd.conditions(),label_word_fd.items())
print(label_word_fd['pos'].N(),label_word_fd['neg'].N())


# In[ ]:

# n_ii = label_word_fd[label][word]
# n_ix = word_fd[word]
# n_xi = label_word_fd[label].N()
# n_xx = label_word_fd.N()
#         w1    ~w1
#      ------ ------
#  w2 | n_ii | n_oi | = n_xi
#      ------ ------
# ~w2 | n_io | n_oo |
#     ------ ------
#      =n_ix         TOTAL = n_xx
# A number of measures are available to score collocations or other associations. The arguments to measure 
# functions are marginals of a contingency table, in the bigram case (n_ii, (n_ix, n_xi), n_xx):
# n_ii = label_word_fd[label][word]
# n_ix = word_fd[word]
# n_xi = label_word_fd[label].N()
# n_xx = label_word_fd.N()
# Chi-Sq Contingency Table : Relating Word w1 with "pos" classification 
#         w1    ~w1
#      ------ ------
# +ve | n_ii | n_oi | = n_xi
#      ------ ------
# -ve | n_io | n_oo |
#     ------ ------
#      =n_ix         TOTAL = n_xx
# n_ix : Total Freq of word w1, n_xi: pos_word_count 
pos_word_count = label_word_fd['pos'].N()
neg_word_count = label_word_fd['neg'].N()
total_word_count = pos_word_count + neg_word_count
 
word_scores = {}

#print(word_fd.items())
for word, freq in word_fd.items():
    pos_score = BigramAssocMeasures.chi_sq(label_word_fd['pos'][word],(freq, pos_word_count), total_word_count)
    neg_score = BigramAssocMeasures.chi_sq(label_word_fd['neg'][word],(freq, neg_word_count), total_word_count)
    word_scores[word] = pos_score + neg_score
import operator
best1 = sorted(word_scores.items(), key=operator.itemgetter(1), reverse=True)[:10000]
bestwords = set([w for w, s in best1])
 
def best_word_feats(words,biGramMeasure):
    return dict([(word, True) for word in words if word in bestwords])
 
evaluations.append(evaluate_classifier(best_word_feats,BigramAssocMeasures.chi_sq))
 
def best_bigram_word_feats(words, score_fn=BigramAssocMeasures.chi_sq, n=200):
    bigram_finder = BigramCollocationFinder.from_words(words)
    bigrams = bigram_finder.nbest(score_fn, n)
    d = dict([(bigram, True) for bigram in bigrams])
    d.update(best_word_feats(words,score_fn))
    return d
evaluations.append(evaluate_classifier(best_bigram_word_feats,BigramAssocMeasures.chi_sq))


# In[ ]:

for modelEvalMetrics in evaluations:
    print(modelEvalMetrics)