def apply_tfidf(reviews_text: list):
dictionary = HashDictionary()
review_tokens = []
for result in reviews_text:
# Tokenize the reviews
review_tokens.append(tokenize(result))
# Build the dictionary
dictionary.add_documents(review_tokens)
# Convert to vector corpus
vectors = [dictionary.doc2bow(token) for token in review_tokens]
# Build TF-IDF model
tfidf = TfidfModel(vectors)
# Get TF-IDF weights
weights = tfidf[vectors]
# Get terms from the dictionary and pair with weights
freq = dict()
for doc in weights:
for pair in doc:
list_of_words = list(dictionary[pair[0]])
for word in list_of_words:
if word in freq:
freq[word] += pair[1]
else:
freq[word] = pair[1]
return freq
Download it from:
https://www.kaggle.com/benhamner/nips-papers/downloads/papers.csv/2
"""
papers = pd.read_csv('papers.csv')
corpus = list(papers['paper_text'])
print("corpus size: ", len(corpus))
# ToDo: check performance with lemmatization: gensim.utils.lemmatize
tokenized_corpus = [[
utils.to_unicode(token)
for token in utils.tokenize(corpus_item, lower=True, errors='ignore')
] for corpus_item in corpus]
hash_dictionary = HashDictionary(tokenized_corpus)
bow_corpus = [hash_dictionary.doc2bow(text) for text in corpus]
MmCorpus.serialize('nips_bow.mm', bow_corpus, progress_cnt=10000)
hash_dictionary.filter_extremes(no_below=20,
no_above=0.1,
keep_n=DEFAULT_DICT_SIZE)
hash_dictionary.save_as_text('nips_wordids.txt.bz2')
hash_dictionary.save('nips_corpura_hash.dict')
dictionary = Dictionary(tokenized_corpus)
dictionary.save('nips_corpura.dict')
