classifier_file.close()
# setup word stemmer
self.stemmer = SnowballStemmer("english")
def classify(self, tree):
words = [self.stemmer.stem(word).lower() for word in tree.leaves()]
categories = {}
for category, classifier in self.classifiers.items():
features = { feature: (feature in words) for feature in self.word_features[category] }
prob = classifier.prob_classify(features).prob(category)
categories[category] = prob
return categories
classifier = Classifier()
predictor = SentimentPredictor()
ANALYZER_VERSION = 3
# iterate over confessions and predict categories and sentiment
confessions = db.parses.find({
"analyzed": { "$ne": ANALYZER_VERSION }
}, limit=500)
threshold = 0.2
for confession in confessions:
for tree_id, raw_tree in enumerate(confession["trees"]):
if raw_tree == "None": continue
# get sentence categories
tree = Tree.fromstring(raw_tree)
categories = [(category, prob) for (category, prob) in classifier.classify(tree).items() if prob > threshold]
def score_accuracy(data):
accurate = 0
inaccurate = 0
for datum in data:
if abs(datum["gold_sentiment"]) <= 0.5: continue
s = datum["sentiment"]
if datum["gold_sentiment"] < 0:
if s < 0: accurate += 1
else: inaccurate += 1
else:
if s > 0: accurate += 1
else: inaccurate += 1
return accurate*1.0/(accurate+inaccurate)
# predict sentence sentiments
predictor = SentimentPredictor()
for (parse_id, tree_id), sentiments in sentences.items():
confession = db.parses.find_one({ "_id": ObjectId(parse_id) })
tree = confession["trees"][tree_id]
sentiment = sum(sentiments)*1.0/len(sentiments)
predictor.add_tree({
"raw_tree": tree,
"gold_sentiment": sentiment
})
predictor.run()
print score_accuracy(predictor.trees)
# predict confession sentiments
predictor = SentimentPredictor()
for parse_id, sentiments in confessions.items():
confession = db.parses.find_one({ "_id": ObjectId(parse_id) })
