-
Notifications
You must be signed in to change notification settings - Fork 0
/
Sense2Vec.py
46 lines (42 loc) · 1.88 KB
/
Sense2Vec.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
from X2Vec import X2Vec
from pywsd import disambiguate
class Sense2Vec(X2Vec):
"""
A word embedding model that generates different embeddings for identical words based on their meaning.
"""
def tokenize_corpus(self, corpus, tokenize=True):
"""
Method that tokenizes the corpus prior to training. For each word in the corpus we compute the sense of that
word and change it with word_sense. For example: cat can become cat_n.01
:param corpus: the corpus as a string.
:param tokenize: True if should tokenize the corpus beforehand.
:return: the tokenized corpus.
"""
# convert the corpus to be sentence
corpus = [' '.join(sentence) for sentence in corpus]
if not tokenize:
return corpus
print('Starting to tag corpus')
corpus_tags = []
counter = 0.0
for sentence in corpus:
if (counter % 100000) == 0:
print(counter/len(corpus)*100, " percent complete \r",)
try:
# get the sense of each word in the sentence
tagged_sentence = disambiguate(sentence)
corpus_tags.append(tagged_sentence)
except IndexError:
print("pywsd can't handle the sentence: " + sentence)
counter += 1
# create a dictionary of each word and all the senses it was mapped to
for sentence in corpus_tags:
for tag in sentence:
if tag[1] is None:
continue
cur_set = self.token_dict.get(tag[0], set())
cur_set.add(tag[1].name())
self.token_dict[tag[0]] = cur_set
# create the tagged corpus in a format ready for training
tagged_corpus = [[word[1].name() for word in sentence if word[1] is not None] for sentence in corpus_tags]
return tagged_corpus