def get_feature_vector(self, tweet):
"""
Feature vector contains
1) 1 or 0 if word is contained in query set
2) 1 or 0 if user is contained in query set
3) missed = # words not in query set / # words in tweet
@param tweet: A tweet in the twitter json format
@return list: feature vector
"""
text = tweet['text']
hashtags, user_mentions, urls = extract_entity(tweet)
user_mentions.append(tweet['user']['str_id'])
filtered_text = []
# remove hashtags from text
text_tokenized = word_tokenize(text)
text_cleaned = clean_entity(text_tokenized)
for term in text_cleaned:
if term not in self.stop_words:
filtered_text.append(term)
filtered_text.extend(hashtags)
term_features = [1.0 if term in filtered_text else 0.0 for term in self.query_set['keyword']]
user_features = [1.0 if user in user_mentions else 0.0 for user in self.query_set['user']]
#TODO: include users as well?
missed = float(sum(term_features)) / len(filtered_text)
user_features.append(missed)
term_features.extend(user_features)
return term_features
def pos_term_selection(self):
"""
Get a positive sample of tweets. Tweets are stored
in the twitter json format.
1) Tokenize tweet
2) Remove stopwords
3) add to dict or count + 1
"""
word_count = dict(TweetCount)
user_count = dict(TweetCount)
def _doc_freq(_term):
word_count[_term]['doc_frequency'] += 1
return
def _add_term(_term, _index, _stopwords=list()):
if _term in _stopwords:
return
if _term in _index:
_index[_term] += 1
else:
_index[_term] = TweetCount(1, 0)
pos_tweets = self.get_pos_tweets()
for tweet in pos_tweets:
text = tweet['text']
text_tokenized = word_tokenize(text)
text_cleaned = clean_entity(text_tokenized)
for term in text_cleaned:
_add_term(term, word_count, self.stop_words)
hashtags, user_mentions = extract_entity(tweet)
user_mentions.append(tweet['user']['id_str'])
for hashtag in hashtags:
_add_term(hashtag, word_count, self.stop_words)
for user in user_mentions:
_add_term(user, user_count)
hashtags.extend(user_mentions)
text_cleaned.extend(hashtags)
map(_doc_freq, set(text_cleaned))
self.query_set['keyword'] = [
x for x in word_count if word_count[x]['frequency'] > 10
]
self.query_set['user'] = [
x for x in user_count if user_count[x]['frequency'] > 10
]
def pos_term_selection(self):
"""
Get a positive sample of tweets. Tweets are stored
in the twitter json format.
1) Tokenize tweet
2) Remove stopwords
3) add to dict or count + 1
"""
word_count = dict(TweetCount)
user_count = dict(TweetCount)
def _doc_freq(_term):
word_count[_term]['doc_frequency'] += 1
return
def _add_term(_term, _index, _stopwords=list()):
if _term in _stopwords:
return
if _term in _index:
_index[_term] += 1
else:
_index[_term] = TweetCount(1, 0)
pos_tweets = self.get_pos_tweets()
for tweet in pos_tweets:
text = tweet['text']
text_tokenized = word_tokenize(text)
text_cleaned = clean_entity(text_tokenized)
for term in text_cleaned:
_add_term(term, word_count, self.stop_words)
hashtags, user_mentions = extract_entity(tweet)
user_mentions.append(tweet['user']['id_str'])
for hashtag in hashtags:
_add_term(hashtag, word_count, self.stop_words)
for user in user_mentions:
_add_term(user, user_count)
hashtags.extend(user_mentions)
text_cleaned.extend(hashtags)
map(_doc_freq, set(text_cleaned))
self.query_set['keyword'] = [x for x in word_count if word_count[x]['frequency'] > 10]
self.query_set['user'] = [x for x in user_count if user_count[x]['frequency'] > 10]
def get_feature_vector(self, tweet):
"""
Feature vector contains
1) 1 or 0 if word is contained in query set
2) 1 or 0 if user is contained in query set
3) missed = # words not in query set / # words in tweet
@param tweet: A tweet in the twitter json format
@return list: feature vector
"""
text = tweet['text']
hashtags, user_mentions, urls = extract_entity(tweet)
user_mentions.append(tweet['user']['str_id'])
filtered_text = []
# remove hashtags from text
text_tokenized = word_tokenize(text)
text_cleaned = clean_entity(text_tokenized)
for term in text_cleaned:
if term not in self.stop_words:
filtered_text.append(term)
filtered_text.extend(hashtags)
term_features = [
1.0 if term in filtered_text else 0.0
for term in self.query_set['keyword']
]
user_features = [
1.0 if user in user_mentions else 0.0
for user in self.query_set['user']
]
#TODO: include users as well?
missed = float(sum(term_features)) / len(filtered_text)
user_features.append(missed)
term_features.extend(user_features)
return term_features
