type=int)
required.add_argument('--output', help="Output directory", required=True)
args = parser.parse_args()
input = args.input
vec_size = args.size
output = args.output
if not input.endswith('/'):
input = input + '/'
if not output.endswith('/'):
output = output + '/'
if not os.path.exists(output):
os.makedirs(output)
sentences = get_sentences(input)
tagged_data = [
TaggedDocument(words=word_tokenize(_d.lower()), tags=[str(i)])
for i, _d in enumerate(sentences)
]
max_epochs = 100
alpha = 0.025
model = Doc2Vec(size=vec_size,
alpha=alpha,
min_alpha=0.00025,
min_count=1,
dm=1)
sentence = args.sentence
alpha = args.alpha
eta = args.eta
bigram = args.bigram
max_df = args.max_df
min_df = args.min_df
if not input_dir.endswith('/'):
input_dir = input_dir + '/'
if not output_dir.endswith('/'):
output_dir = output_dir + '/'
if sentence:
logging.info('Reading sentences of messages...')
# Get sentences of messages
messages = get_sentences(input_dir)
else:
logging.info('Reading messages...')
# Get the whole messages
messages = get_messages(input_dir)
logging.info('Run lda with {} number of clusters...'.format(n_clusters))
# Initialise the count vectorizer
if bigram:
count_vectorizer = CountVectorizer(analyzer='word',
stop_words=STOP_WORDS,
ngram_range=(1, 2),
max_df=max_df,
min_df=min_df)
else:
def index(word):
return jsonify(get_sentences(word))
min_cl = args.min_cl
max_cl = args.max_cl
samples = args.samples
keywords = args.keywords
sentence = args.sentence
vector_path = args.vector_path
if not input.endswith('/'):
input = input + '/'
if not output.endswith('/'):
output = output + '/'
if sentence:
logging.info('Reading sentences of emails...')
# Get sentences of emails
emails = get_sentences(input)
else:
logging.info('Reading emails...')
# Get emails
emails = get_emails(input)
# Max number of clusters is always n_samples/2 if not specified
if max_cl is None:
max_cl = len(emails) // 2
# Min number of clusters is greater than 1.
if min_cl < 2:
sys.exit('Minimum number of clusters should be greater than 1.')
if max_cl > len(emails) - 1:
sys.exit('Maximum number of clusters should be less than n_samples.')
if min_cl > max_cl:
sys.exit('Minumum number of clusters should be less than maximum')
def get_feature_vector(feature_data):
headline, content = feature_data
print(headline)
flat_content = ' '.join(content)
content_sentences = [get_sentences(paragraph) for paragraph in content]
qm_count_headline, em_count_headline, digit_count = counts(headline,0)
qm_count_content, em_count_content = counts(flat_content,1)
headline = headline.split()
content = [paragraph.split() for paragraph in content]
num_contractions_headline = num_contractions(headline)
num_contractions_content = sum([num_contractions(paragraph) for paragraph in content])
num_stopwords = sum([count_stopwords(paragraph) for paragraph in content])
# print(content[0])
headline = [removePunctuation(word) for word in headline]
# content = [[word.replace('\'','') for word in paragraph] for paragraph in content]
content = [[removePunctuation(word) for word in paragraph] for paragraph in content]
content_words = [word.split()[0] for paragraph in content for word in paragraph if word.split() != []]
num_words = len(content_words)
num_tokens = token_count(' '.join(content_words))
avg_length_words, longest_word = word_lengths(content_words,num_words)
avg_length_sentences = num_words/len(content_sentences)
starts_with_question_word = int(starts_with_q_word(headline))
stopword_ratio = num_stopwords/num_words
contraction_ratio = num_contractions_content/num_words
headline = ' '.join(headline)
POS_tags_headline = getPOSTags(headline)
POS_tags_content = getPOSTags(' '.join(content_words))
adverb = int(contains_adverb(POS_tags_headline))
super_adj_count, super_adv_count = superlative_adj_adv_count(POS_tags_content)
POS_counts = posTagFeatures(POS_tags_content)
BERT_keywords = ' '.join(get_key_words(content_words))
document_sim = get_similarity(BERT_keywords,headline)
sentence_sims = []
paragraph_sims = []
for paragraph in content:
for sent in paragraph:
sentence_sims.append(get_similarity(BERT_keywords,' '.join(sent)))
paragraph_sims.append(get_similarity(BERT_keywords,' '.join([' '.join(sent) for sent in paragraph])))
sentence_sims = sum(sentence_sims)/len(sentence_sims)
paragraph_sims = sum(paragraph_sims)/len(paragraph_sims)
vector = [qm_count_headline, em_count_headline, digit_count, qm_count_content, em_count_content,
num_contractions_headline, num_contractions_content, num_stopwords, num_words, num_tokens,
avg_length_words, longest_word, avg_length_sentences, starts_with_question_word, stopword_ratio,
contraction_ratio, adverb, super_adj_count, super_adv_count, document_sim, sentence_sims,
paragraph_sims]
for item in POS_counts:
vector.append(int(item))
print(headline,' DONE')
return (feature_data[0],vector)
