def buildNGramDict(taggedReviews,
n=1,
applyFn=sentenceSumSentiment,
filterFn=filtering.chainFilter(filtering.lower,
filtering.removeStopwords)):
ngramDict = defaultdict(lambda: 0)
for taggedReview in taggedReviews:
for taggedSentence in taggedReview:
sentenceSentiment = applyFn(taggedSentence)
for ngram in nltk.ngrams(filterFn(filtering.tokenize(taggedSentence.sentence)), n):
ngramDict[ngram] += ngramDict[ngram]
return normalize(ngramDict)
def buildWordSentimentDict(taggedReviews,
applyFn=sentenceSumSentiment,
filterFn=filtering.chainFilter(filtering.lower,
filtering.removeStopwords)):
"""
Builds a dictionary of word sentiments from training data by taking the
running average of applying fn (defaults to sentenceSumSentiment). Filters
out words contained in the filterDict argument.
"""
sentimentDict = defaultdict(lambda: 0)
for taggedReview in taggedReviews:
for taggedSentence in taggedReview:
tokenizedSentence = filtering.tokenize(taggedSentence.sentence)
filteredSentence = filterFn(tokenizedSentence)
for word in filteredSentence:
sentimentDict[word] += applyFn(taggedSentence)
return normalize(sentimentDict)
# load static dicts
for name, path in staticSentimentDicts:
dictSym = "{}SentimentsDict".format(name)
exec("{} = util.loadWordSentimentDict(os.path.join(originalDir, '{}'))".format(dictSym,
path))
loadedSentimentDicts.append((dictSym, eval(dictSym)))
# compile learned dicts using various apply fucntions
for name, fn in taggedSentenceEvaluationFunctions:
exec("learned_{}_sentiments = util.buildWordSentimentDict(taggedReviews, applyFn={})".format(name, fn))
dictName = "learned_{}_sentiments".format(name)
loadedSentimentDicts.append((dictName, eval(dictName)))
filterFn = filtering.chainFilter(filtering.lower,
filtering.removeStopwords)
for name, sentimentDict in loadedSentimentDicts:
exec("""def total_sentiment_{0}(inp):
total = 0
for word in filterFn(filtering.tokenize(inp)):
if word in {0}:
total += {0}[word]
return total""".format(name))
fnName = "total_sentiment_{}".format(name)
definedFns.append((fnName, eval(fnName)))
exec("""def num_positive_sentiment_words_{0}(inp):
totalPos = 0
for word in filterFn(filtering.tokenize(inp)):
if word in {0} and {0}[word] > 0:
totalPos += 1
