-
Notifications
You must be signed in to change notification settings - Fork 5
/
google_data.py
403 lines (348 loc) · 14.5 KB
/
google_data.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
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
# -*- coding: utf-8 -*-
import lxml.etree as et
import math
import numpy as np
import collections
import re
import nltk.stem.porter as porter
from nltk.stem.wordnet import WordNetLemmatizer
from itertools import groupby
import random
from glove import *
import os
from glob import glob
from nltk.corpus import wordnet as wn
import csv
random.seed(0)
semcor_path = './google_data/semcor'
masc_path = './google_data/masc'
algorithmic_map = './google_data/algorithmic_map.txt'
manual_map = './google_data/manual_map.txt'
sense_embedding_file = 'sense_embedding.csv'
input_filemask = 'xml'
total_words_wordnet = 206941 # the total number of definitions in the dictionary WordNet 3.0
EMBEDDING_DIM = 100
rm_markup = re.compile(b'\[.+?\]')
rm_misc = re.compile(b"[\[\]\$`()%/,\.:;-]")
replace_num = re.compile(b"\s\d+\s")
regex = re.compile('[^a-zA-Z]')
def clean_context(ctx_in):
ctx = replace_num.sub(b' <number> ', ctx_in)
ctx = rm_markup.sub(b'', ctx)
ctx = rm_misc.sub(b'', ctx)
return ctx
def load_masc_file(masc_path):
list_of_files = []
for root, dirpath, files in os.walk(masc_path):
for filename in [f for f in files if f.endswith('.xml')]:
list_of_files.append(os.path.join(root, filename))
return list_of_files
def load_semcor_file(semcor_path):
list_of_files = glob('%s/*.%s'%(semcor_path, input_filemask))
return list_of_files
def load_data(list_of_files, dtd_validation=True):
"""
:param list_of_files: list of filepath
:return data: list of dictionary containing all words
word_freq: dictionary containing all word as key, its frequency as value
target_sense_freq: dictionary containing target_sense as key, its frequency as value
"""
data = []
word_freq = {}
target_sense_freq = {}
parser = et.XMLParser(dtd_validation=dtd_validation)
for file_name in list_of_files:
doc = et.parse(file_name, parser)
instance = doc.findall('.//word')
for child in instance:
word = None
sense = None
if child.tag == "word":
word = child.get("text").lower()
if word not in word_freq:
word_freq[word] = 0
word_freq[word] += 1
x = {
"word": word,
"break_level": child.get("break_level")
}
sense = child.get("sense")
if sense:
x["lemma"] = child.get("lemma")
x["pos"] = child.get("pos")
x["sense"] = sense
x["is_target"] = True
if sense not in target_sense_freq:
target_sense_freq[sense] = 0
target_sense_freq[sense] += 1
else:
x["is_target"] = False
data.append(x)
return data, word_freq, target_sense_freq
def build_context(data):
"""
return
sense_to_contex: dictionary of all target senses - all context words in one paragraph
"""
sense_to_context = {}
context = []
word_list = []
for elem in data:
word = elem["word"]
word = regex.sub('', word)
if elem["break_level"] == "NO_BREAK" and (word == "quot" or not word):
continue
if elem["break_level"] != "PARAGRAPH_BREAK":
context.append(elem["word"])
if elem["is_target"]:
word_list.append((elem["word"], elem["sense"]))
else:
sense_to_context[tuple(word_list)] = context
context = [elem["word"]]
if elem["is_target"]:
word_list = [(elem["word"], elem["sense"])]
else:
word_list = []
return sense_to_context
def NOAD_to_wordnet(data):
"""
transform target sense annotated with NOAD (New Oxford American Dictionary) word senses into WordNet senses
"""
NOAD_to_wordnet = {}
with open(algorithmic_map, 'r') as f:
lines = f.readlines()
for line in lines:
noad, wordnet = line.split()
NOAD_to_wordnet[noad] = wordnet
with open(manual_map, 'r') as f:
lines = f.readlines()
for line in lines:
noad, wordnet = line.split()
NOAD_to_wordnet[noad] = wordnet
count = 0
for elem in data:
if elem["is_target"]:
if elem["sense"] not in NOAD_to_wordnet:
count += 1
continue
noad_sense = elem["sense"]
elem["sense"] = NOAD_to_wordnet[noad_sense]
print("NOAD sense not in mapping text: %d" %count)
return data
def build_vocab(word_freq):
"""
:param word_freq: list of dicts containing word - frequency pairs
:return: a dict with words as key and ids as value
"""
min_freq = 1
filtered = [item for item in word_freq.items() if item[1]>=min_freq]
count_pairs = sorted(filtered, key=lambda x: -x[1])
words, _ = list(zip(*count_pairs))
#words += ('<pad>', '<dropped>')
word_to_id = dict(zip(words, range(len(words))))
return word_to_id
def build_sense_ids(data):
"""
:return
target_word_to_id dictionary of word - word_frequency
target_sense_to_id list of dictionary including sense - sense_id
len(words): total number of dictinct words
n_senses_from_word_id: dictionary of word_id - number of senses of that word
"""
words = set()
word_to_senses = {}
for elem in data:
if elem["is_target"]:
target_word = elem['word']
target_sense = elem['sense']
if target_word not in words:
words.add(target_word)
word_to_senses.update({target_word: [target_sense]})
else:
if target_sense not in word_to_senses[target_word]:
word_to_senses[target_word].append(target_sense)
words = list(words)
target_word_to_id = dict(zip(words, range(len(words))))
target_sense_to_id = [dict(zip(word_to_senses[word], range(len(word_to_senses[word])))) for word in words]
n_senses_from_word_id = dict([(target_word_to_id[word], len(word_to_senses[word])) for word in words])
return target_word_to_id, target_sense_to_id, len(words), n_senses_from_word_id
def split_context(ctx):
# word_list = re.split(', | +|\? |! |: |; ', ctx.lower())
word_list = [word for word in re.split(', | +|\? |! |: |; |\) |\( |\' ', ctx.lower()) if word]
return word_list
def get_all_wordnet_definition():
all_definition = []
for ss in wn.all_synsets():
ss_list = split_context(ss.definition())
all_definition.append(ss_list)
return all_definition
all_definition = get_all_wordnet_definition()
def build_word_occurrence_definition(word):
count = 0
for ss in all_definition:
if word in ss:
count += 1
#print(word, count)
return count
def build_sense_vector(sense, word_freq, wordvecs):
sense_definition = sense.definition()
#print("sense_definition"+str(sense_definition))
#sense_definition_words = sense_definition.split()
sense_definition_words = split_context(sense_definition)
sense_vector = np.zeros(EMBEDDING_DIM)
n = 0
for word in sense_definition_words:
word_occurrence_definition = build_word_occurrence_definition(word)
idf_word = np.log(total_words_wordnet / float(word_occurrence_definition))
if word in wordvecs:
sense_vector += wordvecs[word] * idf_word
else:
sense_vector += np.random.normal(0.0, 0.1, EMBEDDING_DIM) * idf_word
n += 1
return sense_vector/n
def sc2ss(sensekey):
'''Look up a synset given the information from SemCor'''
### Assuming it is the same WN version (e.g. 3.0)
try:
return wn.lemma_from_key(sensekey).synset()
except:
pass
def build_sense_embedding(target_sense_to_id, word_freq, EMBEDDING_DIM):
"""
build sense vector for every target sense using the definition of wordnet and glove embedding
return
res: dictionary of sense - sense_vector
"""
res = {}
wordvecs = load_glove(EMBEDDING_DIM)
for target_sense_list in target_sense_to_id:
for key, _ in target_sense_list.items():
sense_vector = np.zeros(EMBEDDING_DIM)
senses = key.split(',')
n = 0
for sensekey in senses:
#print(sensekey)
if '/' in sensekey:
continue
sense_synset = sc2ss(sensekey)
if sense_synset:
sense_vector += build_sense_vector(sense_synset, word_freq, wordvecs)
n += 1
if n != 0:
res[key] = sense_vector/n
return res
def get_embedding(sense_embedding_file):
sense_embeddings_ = None
with open(sense_embedding_file, 'r', newline='') as f:
reader = csv.reader(f)
sense_embeddings_ = dict(reader)
sense_embeddings__ = {}
for key, value in sense_embeddings_.items():
value = value.split()
vec = np.zeros(len(value)-1)
for i in range(len(value)):
if '[' in value[i]:
continue
value[i] = value[i][1:]
elif ']' in value[i]:
value[i] = value[i][:-1]
if value[i]:
vec[i-1] = float(value[i])
sense_embeddings__[key] = vec
return sense_embeddings__
def convert_to_numeric(data, word_to_id, target_word_to_id, target_sense_to_id, n_senses_from_word_id, sense_to_context, sense_embeddings):
n_senses_sorted_by_target_id = [n_senses_from_word_id[target_id] for target_id in range(len(n_senses_from_word_id))]
tot_n_senses = sum(n_senses_from_word_id.values())
print("total senses:", tot_n_senses)
all_data = []
n = 0
for sense_tuple, context in sense_to_context.items():
if len(sense_tuple) != 0:
ctx_ints = [word_to_id[word] for word in context if word in word_to_id]
for target_word, target_sense in list(sense_tuple):
target_id = target_word_to_id[target_word]
stop_idx = ctx_ints.index(target_id)
xf = np.array(ctx_ints[:stop_idx])
xb = np.array(ctx_ints[stop_idx+1:])[::-1]
_instance = Instance()
_instance.target_word = target_word
_instance.target_id = target_id
_instance.xf = xf
_instance.xb = xb
#print(target_sense)
try:
_instance.target_sense = sense_embeddings[target_sense]
except:
n += 1
pass
all_data.append(_instance)
print("total number of target sense having not target sense vector %d" % n)
return all_data
def group_by_target(ndata):
res = {}
for key, group in groupby(ndata, lambda inst: inst.target_word):
res.update({key: list(group)})
return res
def split_grouped(data, frac, min=None):
assert frac >= 0.
assert frac < .5
l = {}
r = {}
for target_id, instances in data.items():
# instances = [inst for inst in instances]
random.shuffle(instances) # optional
n = len(instances)
print(n)
if frac == 0:
l[target_id] = instances[:]
else:
n_r = int(frac * n)
if min and n_r < min:
n_r = min
n_l = n - n_r
print(n_l, n_r)
l[target_id] = instances[:n_l]
r[target_id] = instances[-n_r:]
return l, r if frac > 0 else l
def get_data(_data, n_step_f, n_step_b):
forward_data, backward_data, target_sense_ids, sense_embeddings = [], [], [], []
for target_id, data in _data.items():
for instance in data:
xf, xb, target_sense_id, target_sense = instance.xf, instance.xb, instance.target_id, instance.target_sense
n_to_use_f = min(n_step_f, len(xf))
n_to_use_b = min(n_step_b, len(xb))
xfs = np.zeros([n_step_f], dtype=np.int32)
xbs = np.zeros([n_step_b], dtype=np.int32)
if n_to_use_f != 0:
xfs[-n_to_use_f:] = xf[-n_to_use_f:]
if n_to_use_b != 0:
xbs[-n_to_use_b:] = xb[-n_to_use_b:]
forward_data.append(xfs)
backward_data.append(xbs)
target_sense_ids.append(target_sense_id)
sense_embeddings.append(target_sense)
return (np.array(forward_data), np.array(backward_data), np.array(target_sense_ids), np.array(sense_embeddings))
class Instance:
pass
if __name__ == '__main__':
#masc_files = load_masc_file(masc_path)
semcor_files = load_semcor_file(semcor_path)
#masc_data = load_data(masc_files)
semcor_data, semcor_word_freq, semcor_target_sense_freq = load_data(semcor_files)
semcor_data = NOAD_to_wordnet(semcor_data)
sense_to_context = build_context(semcor_data)
word_to_id = build_vocab(semcor_word_freq)
target_word_to_id, target_sense_to_id, target_words_nums, n_senses_from_word_id = build_sense_ids(semcor_data)
'''
sense_embeddings = build_sense_embedding(target_sense_to_id, semcor_word_freq, EMBEDDING_DIM)
with open('sense_embedding.csv','w', newline='') as f:
w = csv.writer(f)
for key, value in sense_embeddings.items():
w.writerow([key, value])'''
sense_embeddings_ = get_embedding(sense_embedding_file)
all_data = convert_to_numeric(semcor_data, word_to_id, target_word_to_id, target_sense_to_id, n_senses_from_word_id, sense_to_context, sense_embeddings_)
n_step_f = 40
n_step_b = 40
grouped_by_target = group_by_target(all_data)
train_data, val_data = split_grouped(grouped_by_target, .2)
train_forward_data, train_backward_data, train_target_sense_ids, train_sense_embedding = get_data(train_data, n_step_f, n_step_b)