def TokenizeTweets(data_path):
learner_tweet_map = {}
tok = Tokenizer(preserve_case=False)
learners = os.listdir(data_path)
for learner in learners:
tweet_path = data_path + learner + "/tweet"
if os.path.isfile(tweet_path):
tweet_file = open(tweet_path, "r")
lines = tweet_file.readlines()
individual_word_count_map = {}
individual_word_count_set = set()
num_english_tweet = 0
for line in lines:
try:
jsonObject = json.loads(line)
if jsonObject["lang"] == "en":
tweet = jsonObject["text"]
tokenized_tweet = tok.tokenize(tweet)
num_english_tweet += 1
for word in tokenized_tweet:
if word not in individual_word_count_set:
individual_word_count_set.add(word)
individual_word_count_map[word] = 0
individual_word_count_map[word] += 1
except Exception as e:
print line
learner_tweet_map[learner] = {}
learner_tweet_map[learner]["tweet"] = individual_word_count_map
learner_tweet_map[learner]["num_english_tweet"] = num_english_tweet
#if num > 30:
# break
if len(learner_tweet_map) % 100 == 0:
print len(learner_tweet_map)
output_path = os.path.dirname(os.path.dirname(data_path)) + "all_tokenized_tweets"
output_file = open(output_path, "w")
output_file.write(json.dumps(learner_tweet_map))
output_file.close()
class FeatureWorker(TextWorker):
def __init__(self):
super(FeatureWorker, self).__init__()
self.tok = Tokenizer()
def extractNgramPerTweet(self, tweet, n=1):
"""
Extract n-grams from tweet after standardizing
"""
tweet = self.shrinkSpace(tweet)
tweet = self.remove_handles(tweet)
tweet = self.remove_urls(tweet)
tokens = self.tok.tokenize(tweet)
#ngrams = Counter([" ".join(x) for x in zip(*[tokens[i:] for i in range(n)])])
ngrams = Counter([" ".join(x) for x in zip(*[tokens[n:]])])
return ngrams
def fullNGramExtract(self, tweet_list, n=1):
"""
"""
all_ngrams = Counter()
for i in range(n):
this_ngrams = Counter()
for tweet in tweet_list:
this_ngrams.update(self.extractNgramPerTweet(tweet, n))
total_ngrams = float(sum(this_ngrams.values()))
all_ngrams.update({
gram: value / total_ngrams
for gram, value in this_ngrams.items()
})
return all_ngrams
def extractLexicon(self, ngrams, lex, intercepts=None):
"""
"""
pLex = {} # prob of lex given user
for term, cats in lex.iteritems():
try:
gn = ngrams[term]
for cat, weight in cats.iteritems():
try:
pLex[cat] += float(gn) * weight
except KeyError:
pLex[cat] = float(gn) * weight
except KeyError:
pass #not in lex
if intercepts:
for cat in intercepts:
pLex[cat] += intercepts[cat]
return pLex
def tokenize_tweet(tweet, tok=None):
"""
Tokenize the tweet and discard any with fewer than 3 tokens
"""
if not tok:
tok = Tokenizer()
tweet = tweet.strip()
tokens = tok.tokenize(tweet)
if len(tokens) > 3:
tweet = " ".join(tokens)
else:
tweet = ""
return tweet
user_gender = 0 if user_gender_re.findall(file)[0] == 'female' else 1
user_industry = user_industry_re.findall(file)[0]
if user_industry in industries:
industries[user_industry] += 1
else:
industries[user_industry] = 1
user_word_count[user_id] = {}
ages[user_id] = user_age
genders[user_id] = user_gender
user_industry_map[user_id] = user_industry
sample_size += 1
f = open(file, 'r')
posts = post_re.findall(f.read())
for p in posts:
posts_count += 1
tokenized = tok.tokenize(p)
for token in tokenized:
words_count += 1
if token in user_word_count[user_id]:
user_word_count[user_id][token] += 1
else:
user_word_count[user_id][token] = 1
#
# i = 0;
# q = Queue(maxsize = 0)
# for i in range(0, len(files_list), 4):
# process1 = Process(target=run_proc, args=(files_list[i],q))
# process1.start()
#
# if i+1 < len(files_list):
tkzer = Tokenizer()
postsRe = re.compile(r'<post>(.*?)</post>',
re.DOTALL + re.I) #.*? = non-geedy match
userData = dict() #dictionary of user_id => {age, gender, industry, topics}
filesRead = 0
(numPosts, numWords, industries) = (0, 0, dict()) #for answering question one.
for file in dirFiles:
if fnmatch.fnmatch(file, '*.xml'):
user_id, gender, age, industry, zodiac = file.split('.')[:5]
industry = industry.lower()
wordCounts = dict()
totalWords = 0
currentFile = open(blogsDir + '/' + file).read()
posts = postsRe.findall(currentFile)
for post in posts:
words = tkzer.tokenize(post)
for word in words:
try:
wordCounts[word] += 1
except KeyError:
wordCounts[word] = 1
totalWords += len(words)
numPosts += 1
numWords += totalWords
#############################################
#2. CALCULATE USERS PROBABILITY OF MENTIONING A TOPIC
fTotalWords = float(totalWords) #for floating point division
pTopicGivenUser = [0] * num_topics #initial probabilities
for word, count in wordCounts.iteritems():
pWordGivenUser = count / fTotalWords
print "[LOADING CORPUS (%d files) AND CALCULATING TOPIC USAGE]" % len(dirFiles)
tkzer = Tokenizer()
postsRe = re.compile(r'<post>(.*?)</post>', re.DOTALL + re.I) #.*? = non-geedy match
userData = dict() #dictionary of user_id => {age, gender, industry, topics}
filesRead = 0
(numPosts, numWords, industries) = (0, 0, dict()) #for answering question one.
for file in dirFiles:
if fnmatch.fnmatch(file, '*.xml'):
user_id, gender, age, industry, zodiac = file.split('.')[:5]
industry = industry.lower()
wordCounts = dict()
totalWords = 0
currentFile = open(blogsDir+'/'+file).read()
posts = postsRe.findall(currentFile)
for post in posts:
words = tkzer.tokenize(post)
for word in words:
try:
wordCounts[word] += 1
except KeyError:
wordCounts[word] = 1
totalWords += len(words)
numPosts+=1
numWords+=totalWords
#############################################
#2. CALCULATE USERS PROBABILITY OF MENTIONING A TOPIC
fTotalWords = float(totalWords)#for floating point division
pTopicGivenUser = [0] * num_topics #initial probabilities
for word, count in wordCounts.iteritems():
pWordGivenUser = count / fTotalWords
user_gender = 0 if user_gender_re.findall(file)[0] == 'female' else 1
user_industry = user_industry_re.findall(file)[0]
if user_industry in industries:
industries[user_industry] += 1
else:
industries[user_industry] = 1
user_word_count[user_id] = {}
ages[user_id] = user_age
genders[user_id] = user_gender
user_industry_map[user_id] = user_industry
sample_size += 1
f = open(file, 'r')
posts = post_re.findall(f.read())
for p in posts:
posts_count += 1
tokenized = tok.tokenize(p)
for token in tokenized:
words_count += 1
if token in user_word_count[user_id]:
user_word_count[user_id][token] += 1
else:
user_word_count[user_id][token] = 1
#
# i = 0;
# q = Queue(maxsize = 0)
# for i in range(0, len(files_list), 4):
# process1 = Process(target=run_proc, args=(files_list[i],q))
# process1.start()
#
# if i+1 < len(files_list):
def parse_blogs(path):
tokenizer = Tokenizer()
users = {}
global_words_dict = {}
industry_map = {}
total_users = 0
total_blog_posts = 0
iterations = 0
topics = pd.read_csv('wwbpFBtopics_condProb.csv')
regex = r'<post>(.*?)</post>'
for filename in os.listdir(path):
iterations += 1
print "user %d" %iterations
if iterations > 50:
break
if filename.startswith("."):
continue
parts = filename.split(".")
user_attributes_map = {}
word_count_map = {}
topic_prob_map = {}
user_total_words_count = 0
user_id = (int)(parts[0])
gender = parts[1]
if gender == "male":
gender = 0
else:
gender = 1
age = (int)(parts[2])
industry = parts[3]
star_sign = parts[4]
if user_id in users:
user_attributes_map = users[user_id]
if industry in industry_map:
industry_map[industry] = industry_map[industry] + 1
else:
industry_map[industry] = 1
with open(path + filename, 'r') as user_blog:
user_blogs = user_blog.read().replace('\n', '').replace('\r', '').replace('\t', '')
all_blog_posts = re.findall(regex, user_blogs, re.DOTALL)
total_blog_posts = total_blog_posts + len(all_blog_posts)
for blog in all_blog_posts:
words = tokenizer.tokenize(blog.strip())
user_total_words_count = user_total_words_count + len(words)
if 'wc_map' in user_attributes_map:
word_count_map = user_attributes_map['wc_map']
for word in words:
if word in word_count_map:
count = word_count_map[word]
count = count + 1
word_count_map[word] = count
else:
word_count_map[word] = 1
if word in global_words_dict:
count = global_words_dict[word]
count = count + 1
global_words_dict[word] = count
else:
global_words_dict[word] = 1
for topic in range(2000):
prob_topic_given_user = 0.0
topic_dict = topics[topics['category'] == topic]
for row in topic_dict.itertuples():
word = row[1]
prob_topic_given_word = row[3]
if word in word_count_map:
count_user_word = word_count_map[word]
prob_word_given_user = count_user_word/user_total_words_count
cur = prob_topic_given_word * prob_word_given_user
prob_topic_given_user = prob_topic_given_user + cur
topic_prob_map[topic] = prob_topic_given_user
user_attributes_map['wc_map'] = word_count_map
user_attributes_map['age'] = age
user_attributes_map['industry'] = industry
user_attributes_map['star_sign'] = star_sign
user_attributes_map['user_id'] = user_id
user_attributes_map['topic_prob_map'] = topic_prob_map
user_attributes_map['total_count'] = user_total_words_count
user_attributes_map['gender'] = gender
users[user_id] = user_attributes_map
return (users, global_words_dict, industry_map, total_blog_posts)
class Preprocessor:
def __init__(self, delimiter=" "):
# delimiter to split tweets into tokens
self.DELIM = delimiter
self.tokenizer = Tokenizer()
def tokenize(self, tweet):
return " ".join(self.tokenizer.tokenize(tweet))
def replace_user_tags(self, tweet, remove=False):
"""
Replace mentions to usernames with "@USER"
if remove=True removes the user mentions
>>> p=Preprocessor()
>>> p.replace_user_tags("@maya yes this is cool1@ did b@ @augyyz")
'@USER yes this is cool1@ did b@ @USER'
>>> p.replace_user_tags("@maya yes this is cool1@ did b@ @augyyz",remove=True)
'yes this is cool1@ did b@'
"""
if remove:
return self.DELIM.join(
[w for w in tweet.split(self.DELIM) if not w.startswith("@")])
else:
return self.DELIM.join([
"@USER" if w.startswith("@") else w
for w in tweet.split(self.DELIM)
])
def replace_urls(self, tweet, remove=False):
"""
Replace urls with @URL
if remove=True removes them
>>> p=Preprocessor()
>>> p.replace_urls("@maya yes this is cool1@ did b@ @augyyz http://www.bitly")
'@maya yes this is cool1@ did b@ @augyyz @URL'
>>> p.replace_urls("@maya yes this is cool1@ did b@ @augyyz http://www.bitly",remove=True)
'@maya yes this is cool1@ did b@ @augyyz'
"""
if remove:
return self.DELIM.join([
w for w in tweet.split(self.DELIM) if not w.startswith("http")
])
else:
return self.DELIM.join([
"@URL" if w.startswith("http") else w
for w in tweet.split(self.DELIM)
])
def replace_hashtags(self, tweet, remove=False):
"""
Replace hashtags with @HASHTAG
if remove=True removes them (any number of # at token start)
>>> p=Preprocessor()
>>> p.replace_hashtags("yes #cool we are in #miami ###yes")
'yes @HASHTAG we are in @HASHTAG @HASHTAG'
>>> p.replace_hashtags("yes #cool we# are in #miami ###yes",remove=True)
'yes we# are in'
>>> p.replace_hashtags("yes #cool we# are in #miami ###yes bar . #wishiwere in italy .")
'yes @HASHTAG we# are in @HASHTAG @HASHTAG bar . @HASHTAG in italy .'
"""
if remove:
return self.DELIM.join(
[w for w in tweet.split(self.DELIM) if not w.startswith("#")])
else:
return self.DELIM.join([
"@HASHTAG" if w.startswith("#") else w
for w in tweet.split(self.DELIM)
])