def static_classifier_test(testText):
count_vect = CountVectorizer(stop_words="english") # initialize the vectorizer
tfidf_trans = TfidfTransformer() # initialize our tfidf transformer
trainingArticles = app.getTrainingSet(40, 0) # get the latest 50 articles
trainingText = [x.text for x in trainingArticles]
trainingLabels = [x.category for x in trainingArticles]
trainingCounts = count_vect.fit_transform(trainingText)
trainingTfidf = tfidf_trans.fit_transform(trainingCounts)
testCounts = count_vect.transform(testText)
testTfidf = tfidf_trans.transform(testCounts)
clf = SVC(kernel="linear").fit(trainingTfidf, trainingLabels) # train classifier
return clf.predict(testTfidf)
def getSimilarArticles(target, numOfDays, numOfNeighbors):
articles = app.getTrainingSet(500, 70)
neigh = KNeighborsClassifier()
count_vect = CountVectorizer(stop_words='english', ngram_range=(1,2))
tfidf_trans = TfidfTransformer()
trainingTitle = [x.title for x in articles]
trainingLabels = [1 for x in articles]
targetTitleCounts = count_vect.fit_transform([target.title])
targetCounts = count_vect.transform([target.text]) + targetTitleCounts
trainingCounts = count_vect.transform(trainingTitle)
print count_vect.get_feature_names()
trainingCountsTfidf = tfidf_trans.fit_transform(trainingCounts)
targetCountsTfidf = tfidf_trans.transform(targetTitleCounts)
print targetCounts
print 'After weighted by tfidf:'
targetCounts = targetCounts.multiply(targetCountsTfidf)
print targetCounts
neigh.fit(trainingCounts, trainingLabels)
similar_articles_index = neigh.kneighbors(targetCounts, numOfNeighbors, False)
similar_articles = []
for index in similar_articles_index[0]:
similar_articles.append(articles[index].title)
return similar_articles
tfidf_trans = TfidfTransformer()
trainingTitle = [x.title for x in articles]
trainingLabels = [1 for x in articles]
targetTitleCounts = count_vect.fit_transform([target.title])
targetCounts = count_vect.transform([target.text]) + targetTitleCounts
trainingCounts = count_vect.transform(trainingTitle)
print count_vect.get_feature_names()
trainingCountsTfidf = tfidf_trans.fit_transform(trainingCounts)
targetCountsTfidf = tfidf_trans.transform(targetTitleCounts)
print targetCounts
print 'After weighted by tfidf:'
targetCounts = targetCounts.multiply(targetCountsTfidf)
print targetCounts
neigh.fit(trainingCounts, trainingLabels)
similar_articles_index = neigh.kneighbors(targetCounts, numOfNeighbors, False)
similar_articles = []
for index in similar_articles_index[0]:
similar_articles.append(articles[index].title)
return similar_articles
trainingArticles = app.getTrainingSet(1, 69)
target = trainingArticles[3]
print 'Target article title:'
print target.title
print '--------------------------------------'
similar_articles = getSimilarArticles(target, 6, 10)
print "Similar articles' title:"
for article in similar_articles:
print article
# -*- coding: utf-8 -*- import time from db import app import numpy as np from StringIO import StringIO from sklearn.feature_extraction.text import CountVectorizer from sklearn.feature_extraction.text import TfidfTransformer count_vect = CountVectorizer(stop_words='english') #initialize the vectorizer tfidf_trans = TfidfTransformer() #initialize our tfidf transformer trainingArticles = app.getTrainingSet(50, 0) # get the latest 50 articles trainingTitle = [x.title for x in trainingArticles] trainingText = [x.text for x in trainingArticles] trainingLabels = [x.category for x in trainingArticles] trainingCounts = count_vect.fit_transform(trainingText) trainingTfidf = tfidf_trans.fit_transform(trainingCounts) vocab = count_vect.vocabulary_ vocabValue=vocab.keys() vocabIndex=vocab.values() tfidfArray = trainingTfidf.toarray() article1 = tfidfArray[0] articleCount = 0 for article in tfidfArray: wordIndex = 0 myKeyword = [] for wordTfidf in article:
def getKeywords(texts):
# Given a list of texts, will extract the keywords for each and return that!
# NOTICE: Don't send more than 100 articles at a time, as we can't insure heterogeneity
count_vectUni = CountVectorizer(stop_words='english', ngram_range=(1,1)) # for unigrams
count_vectBi = CountVectorizer(stop_words='english', ngram_range=(2,2)) # for bigrams
tfidf_trans1 = TfidfTransformer()
tfidf_trans2 = TfidfTransformer()
trainingArticles = app.getTrainingSet(50, 20) # get a training set to mix up the TfIdf
trainingText = [x.content for x in trainingArticles]
totalText = trainingText + texts # merge
# training unigram and bigram TFIDF
trainingTfidf1 = tfidf_trans1.fit_transform(count_vectUni.fit_transform(totalText))
trainingTfidf2 = tfidf_trans2.fit_transform(count_vectBi.fit_transform(totalText))
# extract vocab value and index. This helps link column to a word ( 13 = 'banana')
vocabValue1 = count_vectUni.vocabulary_.keys()
vocabIndex1 = count_vectUni.vocabulary_.values()
vocabValue2 = count_vectBi.vocabulary_.keys()
vocabIndex2 = count_vectBi.vocabulary_.values()
tfidfArray1 = trainingTfidf1.toarray()
tfidfArray2 = trainingTfidf2.toarray()
keywordList = [] # prepare for output
articleIndex = len(trainingText)
while articleIndex < len(totalText):
thisTfidfUni = tfidfArray1[articleIndex]
thisTfidfBi = tfidfArray2[articleIndex]
i = 0
keywords = []
wordScores = []
for wordTfidf in thisTfidfUni:
if wordTfidf > 0.15:
thisWord = vocabValue1[vocabIndex1.index(i)].encode('utf-8')
keywords.append(thisWord)
wordScores.append(wordTfidf)
i = i + 1
wordIndex2 = 0
bigrams = []
bigramScores = []
for wordTfidf2 in thisTfidfBi:
if wordTfidf2 > 0.15:
thisBigram = vocabValue2[vocabIndex2.index(wordIndex2)].encode('utf-8')
bigrams.append(thisBigram)
bigramScores.append(wordTfidf2)
wordIndex2 = wordIndex2 + 1
splitBigrams = []
splitBigramTfidf = []
numBigram = {}
for bigram in bigrams:
index = bigrams.index(bigram)
wordSplit = bigram.split()
splitBigrams.append(wordSplit[0])
splitBigrams.append(wordSplit[1])
splitBigramTfidf.append(bigramScores[index])
splitBigramTfidf.append(bigramScores[index])
if wordSplit[0] not in numBigram: #add first word in bigram
numBigram[wordSplit[0]] = 1
else:
numBigram[wordSplit[0]] += 1
if wordSplit[1] not in numBigram: #add second word in bigram
numBigram[wordSplit[1]] = 1
else:
numBigram[wordSplit[1]] += 1
unigramsAndScores = sorted(zip(wordScores, keywords), reverse=True)
unigramSorted = [e[1] for e in unigramsAndScores]
unigramTfidfSorted = [e[0] for e in unigramsAndScores]
#0 bigrams - keep unigram
#1 bigram - compare values, keep higher
#>1 bigram - keep unigram, keep highest bigram
#loop thru unigrams in increasing tfidf value
for unigram in reversed(unigramSorted): #loop through backwords
if unigram in numBigram:
if numBigram[unigram] == 1: # unigram appears in 1 bigram
indexUni = unigramSorted.index(unigram)
unigramValue = unigramTfidfSorted[indexUni]
indexSplitBi = splitBigrams.index(unigram)
bigramValue = splitBigramTfidf[indexSplitBi]
if unigramValue > bigramValue:
# delete both words of bigram
if indexSplitBi % 2 == 0: # unigram is first word in the split
splitBigrams.pop(indexSplitBi)
splitBigramTfidf.pop(indexSplitBi)
# if other word is also in bigram, update bigram numbers
if splitBigrams[indexSplitBi] in numBigram:
numBigram[splitBigrams[indexSplitBi]] -=1
splitBigrams.pop(indexSplitBi)
splitBigramTfidf.pop(indexSplitBi)
else: # unigram is second word in the split
splitBigrams.pop(indexSplitBi)
splitBigramTfidf.pop(indexSplitBi)
# if the other word is also in bigram, update bigram numbers
if splitBigrams[indexSplitBi - 1] in numBigram:
numBigram[splitBigrams[indexSplitBi - 1]] -= 1
splitBigrams.pop(indexSplitBi - 1)
splitBigramTfidf.pop(indexSplitBi - 1)
else:
# delete unigram
unigramSorted.pop(indexUni)
unigramTfidfSorted.pop(indexUni)
elif numBigram[unigram] > 1: # unigram appears in more than 1 bigram
highestValue = 0
for index in range(len(splitBigrams)): #find value of highest bigram
if splitBigrams[index] == unigram and splitBigramTfidf[index] > highestValue:
highestValue = splitBigramTfidf[index]
index = len(splitBigrams) - 1
while (index >= 0): # loop through index and delete bigrams
if splitBigrams[index] == unigram and splitBigramTfidf[index] < highestValue:
if index % 2 == 0: # unigram is first word in the split
splitBigrams.pop(index)
splitBigramTfidf.pop(index)
# if other word is also in bigram, update bigram numbers
if splitBigrams[index] in numBigram:
numBigram[splitBigrams[index]] -= 1
splitBigrams.pop(index)
splitBigramTfidf.pop(index)
else: # unigram is second word in the split
splitBigrams.pop(index)
splitBigramTfidf.pop(index)
# if other word is also in bigram, update bigram numbers
if splitBigrams[index - 1] in numBigram:
numBigram[splitBigrams[index - 1]] -= 1
splitBigrams.pop(index - 1)
splitBigramTfidf.pop(index - 1)
index = index - 1
index = index - 1
# Checking for plural words using stemming
stemmed = []
delList = []
stemmer = PorterStemmer()
for item in unigramSorted:
try:
stemmed.append(stemmer.stem(item))
except:
#print "Not a word, breh"
pass
i = 0
for stem in stemmed:
j = i
while j + 1 < len(stemmed):
if stem == stemmed[j + 1]:
# pray for no octopus/octopi
if len(unigramSorted[i]) < len(unigramSorted[j+1]):
delList.append(j+1)
else:
delList.append(i)
j += 1
i += 1
for index in delList:
if index in unigramSorted:
del(unigramSorted[index])
if index in unigramTfidfSorted:
del(unigramTfidfSorted[index])
# Concatenate both lists
for index in range(len(splitBigrams)):
if (index % 2 == 0):
word = "".join([splitBigrams[index], " ", splitBigrams[index + 1]])
unigramSorted.append(word)
value = splitBigramTfidf[index]
unigramTfidfSorted.append(value)
keywordTfidf = zip(unigramTfidfSorted, unigramSorted)
keywordTfidf.sort(reverse=True) # sort from most relevant to least relevant
keywordSorted = [e[1] for e in keywordTfidf]
keywordList.append(keywordSorted)
articleIndex = articleIndex + 1
return keywordList
