def train(spam_words, unlabeled_words):
spams = list(map(features, spam_words))
unlabeled = list(map(features, unlabeled_words))
model = PositiveNaiveBayesClassifier.train(spams, unlabeled, 0.5)
data = PickleData('bayesmodel.pickle')
data.write(model)
return model
def predictor():
data = PickleData('bayesmodel.pickle')
if data.exists:
model = data.read()
else:
model = retrain()
def classify(word):
pb = model.prob_classify(features(word))
return pb.prob(1) > 0.95
return classify
def sentiment_classifier():
data = PickleData('sentiments.pickle')
if data.exists:
model = data.read()
else:
model = train_sentiments_classifier()
from snownlp import SnowNLP
def classify(word):
nlp = SnowNLP(word)
neu = 1 if (estimate_neu(nlp.tags)>0.6) else 0
senti = nlp.sentiments - 0.5
prob = model.prob_classify(features(word))
prob_senti = prob.prob(1) - prob.prob(-1)
if neu: return 0
else:
fs = (prob_senti + senti)/3
if fs >= 0.25: return 1
elif fs <= -0.25: return -1
else: return 0
return classify
def train_sentiments_classifier():
pairs = PairData('sentiments.txt', 'utf8')
model = NaiveBayesClassifier.train(pairs)
data = PickleData('sentiments.pickle')
data.write(model)
return model
