-
Notifications
You must be signed in to change notification settings - Fork 0
/
util.py
884 lines (723 loc) · 31.9 KB
/
util.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
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
import os
import random, tqdm, csv
import json
import glob
from nltk import word_tokenize, sent_tokenize
from copy import deepcopy
from thefuzz import process
from xlsxwriter.workbook import Workbook
NAME_MATCHING_THRESHOLD = 80 # min: 0, max: 100
SELF_INTRO_PATTERNS = {'iam', "i'm", 'mynameis', "myname's"}
def ensure_dir(dir_fpath):
os.makedirs(dir_fpath, exist_ok=True)
def tag_paths_to_spans(paths, token_nums, vocab):
"""Convert predicted tag paths to a list of spans (entity mentions or event
triggers).
:param paths: predicted tag paths.
:return (list): a list (batch) of lists (sequence) of spans.
"""
batch_mentions = []
itos = {i: s for s, i in vocab.items()}
for i, path in enumerate(paths):
mentions = []
cur_mention = None
path = path.tolist()[:token_nums[i].item()]
for j, tag in enumerate(path):
tag = itos[tag]
if tag == 'O':
prefix = tag = 'O'
else:
prefix, tag = tag.split('-', 1)
if prefix == 'B':
if cur_mention:
mentions.append(cur_mention)
cur_mention = [j, j + 1, tag]
elif prefix == 'I':
if cur_mention is None:
# treat it as B-*
cur_mention = [j, j + 1, tag]
elif cur_mention[-1] == tag:
cur_mention[1] = j + 1
else:
# treat it as B-*
mentions.append(cur_mention)
cur_mention = [j, j + 1, tag]
else:
if cur_mention:
mentions.append(cur_mention)
cur_mention = None
if cur_mention:
mentions.append(cur_mention)
mentions.sort(key=lambda x: (x[0], x[1]))
batch_mentions.append(mentions)
return batch_mentions
def tag_path_to_spans(path):
mentions = []
cur_mention = None
for j, tag in enumerate(path):
if tag == 'O':
prefix = tag = 'O'
else:
prefix, tag = tag.split('-', 1)
if prefix == 'B':
if cur_mention:
mentions.append(cur_mention)
cur_mention = [j, j + 1, tag]
elif prefix == 'I':
if cur_mention is None:
# treat it as B-*
cur_mention = [j, j + 1, tag]
elif cur_mention[-1] == tag:
cur_mention[1] = j + 1
else:
# treat it as B-*
mentions.append(cur_mention)
cur_mention = [j, j + 1, tag]
else:
if cur_mention:
mentions.append(cur_mention)
cur_mention = None
if cur_mention:
mentions.append(cur_mention)
mentions.sort(key=lambda x: (x[0], x[1]))
return mentions
def convert_to_bio2(ori_tags):
bio2_tags = []
for i, tag in enumerate(ori_tags):
if tag == 'O':
bio2_tags.append(tag)
elif tag[0] == 'I':
if i == 0 or ori_tags[i - 1] == 'O' or ori_tags[i - 1][1:] != tag[1:]:
bio2_tags.append('B' + tag[1:])
else:
bio2_tags.append(tag)
else:
bio2_tags.append(tag)
return bio2_tags
def get_example_from_lines(sent_lines):
tokens = []
ner_tags = []
for line in sent_lines:
array = line.split()
assert len(array) >= 4
tokens.append(array[0])
ner_tags.append(array[3])
ner_tags = convert_to_bio2(ner_tags)
entities = tag_path_to_spans(ner_tags)
inst = {
'tokens': tokens,
'entity_mentions': [{'start': x[0], 'end': x[1], 'entity_type': x[2]} for x in entities]
}
return inst
def get_examples_from_bio_fpath(bio_fpath):
sent_lines = []
bio2_examples = []
nlines = 0
with open(bio_fpath) as infile:
for line in infile:
nlines += 1
line = line.strip()
if '-DOCSTART-' in line or '-docstart-' in line:
continue
if len(line) > 0:
array = line.split()
if len(array) < 4:
continue
else:
sent_lines.append(line)
elif len(sent_lines) > 0:
example = get_example_from_lines(sent_lines)
example['sent_id'] = len(bio2_examples)
bio2_examples.append(example)
sent_lines = []
if len(sent_lines) > 0:
example = get_example_from_lines(sent_lines)
example['sent_id'] = len(bio2_examples)
bio2_examples.append(example)
return bio2_examples
def generate_vocabs(datasets, coref=False,
relation_directional=False,
symmetric_relations=None):
"""Generate vocabularies from a list of data sets
:param datasets (list): A list of data sets
:return (dict): A dictionary of vocabs
"""
entity_type_set = set()
event_type_set = set()
relation_type_set = set()
role_type_set = set()
for dataset in datasets:
entity_type_set.update(dataset.entity_type_set)
event_type_set.update(dataset.event_type_set)
relation_type_set.update(dataset.relation_type_set)
role_type_set.update(dataset.role_type_set)
# add inverse relation types for non-symmetric relations
if relation_directional:
if symmetric_relations is None:
symmetric_relations = []
relation_type_set_ = set()
for relation_type in relation_type_set:
relation_type_set_.add(relation_type)
if relation_directional and relation_type not in symmetric_relations:
relation_type_set_.add(relation_type + '_inv')
# entity and trigger labels
prefix = ['B', 'I']
entity_label_stoi = {'O': 0}
trigger_label_stoi = {'O': 0}
for t in entity_type_set:
for p in prefix:
entity_label_stoi['{}-{}'.format(p, t)] = len(entity_label_stoi)
for t in event_type_set:
for p in prefix:
trigger_label_stoi['{}-{}'.format(p, t)] = len(trigger_label_stoi)
entity_type_stoi = {k: i for i, k in enumerate(entity_type_set, 1)}
entity_type_stoi['O'] = 0
event_type_stoi = {k: i for i, k in enumerate(event_type_set, 1)}
event_type_stoi['O'] = 0
relation_type_stoi = {k: i for i, k in enumerate(relation_type_set, 1)}
relation_type_stoi['O'] = 0
if coref:
relation_type_stoi['COREF'] = len(relation_type_stoi)
role_type_stoi = {k: i for i, k in enumerate(role_type_set, 1)}
role_type_stoi['O'] = 0
mention_type_stoi = {'NAM': 0, 'NOM': 1, 'PRO': 2, 'UNK': 3}
return {
'entity_type': entity_type_stoi,
'event_type': event_type_stoi,
'relation_type': relation_type_stoi,
'role_type': role_type_stoi,
'mention_type': mention_type_stoi,
'entity_label': entity_label_stoi,
'trigger_label': trigger_label_stoi,
}
def load_valid_patterns(path, vocabs):
event_type_vocab = vocabs['event_type']
entity_type_vocab = vocabs['entity_type']
relation_type_vocab = vocabs['relation_type']
role_type_vocab = vocabs['role_type']
# valid event-role
valid_event_role = set()
event_role = json.load(
open(os.path.join(path, 'event_role.json'), 'r', encoding='utf-8'))
for event, roles in event_role.items():
if event not in event_type_vocab:
continue
event_type_idx = event_type_vocab[event]
for role in roles:
if role not in role_type_vocab:
continue
role_type_idx = role_type_vocab[role]
valid_event_role.add(event_type_idx * 100 + role_type_idx)
# valid relation-entity
valid_relation_entity = set()
relation_entity = json.load(
open(os.path.join(path, 'relation_entity.json'), 'r', encoding='utf-8'))
for relation, entities in relation_entity.items():
relation_type_idx = relation_type_vocab[relation]
for entity in entities:
entity_type_idx = entity_type_vocab[entity]
valid_relation_entity.add(
relation_type_idx * 100 + entity_type_idx)
# valid role-entity
valid_role_entity = set()
role_entity = json.load(
open(os.path.join(path, 'role_entity.json'), 'r', encoding='utf-8'))
for role, entities in role_entity.items():
if role not in role_type_vocab:
continue
role_type_idx = role_type_vocab[role]
for entity in entities:
entity_type_idx = entity_type_vocab[entity]
valid_role_entity.add(role_type_idx * 100 + entity_type_idx)
return {
'event_role': valid_event_role,
'relation_entity': valid_relation_entity,
'role_entity': valid_role_entity
}
def read_ltf(path):
root = et.parse(path, et.XMLParser(
dtd_validation=False, encoding='utf-8')).getroot()
doc_id = root.find('DOC').get('id')
doc_tokens = []
for seg in root.find('DOC').find('TEXT').findall('SEG'):
seg_id = seg.get('id')
seg_tokens = []
seg_start = int(seg.get('start_char'))
seg_text = seg.find('ORIGINAL_TEXT').text
for token in seg.findall('TOKEN'):
token_text = token.text
start_char = int(token.get('start_char'))
end_char = int(token.get('end_char'))
assert seg_text[start_char - seg_start:
end_char - seg_start + 1
] == token_text, 'token offset error'
seg_tokens.append((token_text, start_char, end_char))
doc_tokens.append((seg_id, seg_tokens))
return doc_tokens, doc_id
def read_txt(path, language='english'):
doc_id = os.path.basename(path)
data = open(path, 'r', encoding='utf-8').read()
data = [s.strip() for s in data.split('\n') if s.strip()]
sents = [l for ls in [sent_tokenize(line, language=language) for line in data]
for l in ls]
doc_tokens = []
offset = 0
for sent_idx, sent in enumerate(sents):
sent_id = '{}-{}'.format(doc_id, sent_idx)
tokens = word_tokenize(sent)
tokens = [(token, offset + i, offset + i + 1)
for i, token in enumerate(tokens)]
offset += len(tokens)
doc_tokens.append((sent_id, tokens))
return doc_tokens, doc_id
def read_json(path):
with open(path, 'r', encoding='utf-8') as r:
data = [json.loads(line) for line in r]
doc_id = data[0]['doc_id']
offset = 0
doc_tokens = []
for inst in data:
tokens = inst['tokens']
tokens = [(token, offset + i, offset + i + 1)
for i, token in enumerate(tokens)]
offset += len(tokens)
doc_tokens.append((inst['sent_id'], tokens))
return doc_tokens, doc_id
def read_json_single(path):
with open(path, 'r', encoding='utf-8') as r:
data = [json.loads(line) for line in r]
doc_id = os.path.basename(path)
doc_tokens = []
for inst in data:
tokens = inst['tokens']
tokens = [(token, i, i + 1) for i, token in enumerate(tokens)]
doc_tokens.append((inst['sent_id'], tokens))
return doc_tokens, doc_id
def save_result(output_file, gold_graphs, pred_graphs, sent_ids, tokens=None):
with open(output_file, 'w', encoding='utf-8') as w:
for i, (gold_graph, pred_graph, sent_id) in enumerate(
zip(gold_graphs, pred_graphs, sent_ids)):
output = {'sent_id': sent_id,
'gold': gold_graph.to_dict(),
'pred': pred_graph.to_dict()}
if tokens:
output['tokens'] = tokens[i]
w.write(json.dumps(output) + '\n')
def mention_to_tab(start, end, entity_type, mention_type, mention_id, tokens, token_ids, score=1):
tokens = tokens[start:end]
token_ids = token_ids[start:end]
span = '{}:{}-{}'.format(token_ids[0].split(':')[0],
token_ids[0].split(':')[1].split('-')[0],
token_ids[1].split(':')[1].split('-')[1])
mention_text = tokens[0]
previous_end = int(token_ids[0].split(':')[1].split('-')[1])
for token, token_id in zip(tokens[1:], token_ids[1:]):
start, end = token_id.split(':')[1].split('-')
start, end = int(start), int(end)
mention_text += ' ' * (start - previous_end) + token
previous_end = end
return '\t'.join([
'json2tab',
mention_id,
mention_text,
span,
'NIL',
entity_type,
mention_type,
str(score)
])
def json_to_mention_results(input_dir, output_dir, file_name,
bio_separator=' '):
mention_type_list = ['nam', 'nom', 'pro', 'nam+nom+pro']
file_type_list = ['bio', 'tab']
writers = {}
for mention_type in mention_type_list:
for file_type in file_type_list:
output_file = os.path.join(output_dir, '{}.{}.{}'.format(file_name,
mention_type,
file_type))
writers['{}_{}'.format(mention_type, file_type)
] = open(output_file, 'w')
json_files = glob.glob(os.path.join(input_dir, '*.json'))
for f in json_files:
with open(f, 'r', encoding='utf-8') as r:
for line in r:
result = json.loads(line)
doc_id = result['doc_id']
tokens = result['tokens']
token_ids = result['token_ids']
bio_tokens = [[t, tid, 'O']
for t, tid in zip(tokens, token_ids)]
# separate bio output
for mention_type in ['NAM', 'NOM', 'PRO']:
tokens_tmp = deepcopy(bio_tokens)
for start, end, enttype, mentype in result['graph']['entities']:
if mention_type == mentype:
tokens_tmp[start] = 'B-{}'.format(enttype)
for token_idx in range(start + 1, end):
tokens_tmp[token_idx] = 'I-{}'.format(
enttype)
writer = writers['{}_bio'.format(mention_type.lower())]
for token in tokens_tmp:
writer.write(bio_separator.join(token) + '\n')
writer.write('\n')
# combined bio output
tokens_tmp = deepcopy(bio_tokens)
for start, end, enttype, _ in result['graph']['entities']:
tokens_tmp[start] = 'B-{}'.format(enttype)
for token_idx in range(start + 1, end):
tokens_tmp[token_idx] = 'I-{}'.format(enttype)
writer = writers['nam+nom+pro_bio']
for token in tokens_tmp:
writer.write(bio_separator.join(token) + '\n')
writer.write('\n')
# separate tab output
for mention_type in ['NAM', 'NOM', 'PRO']:
writer = writers['{}_tab'.format(mention_type.lower())]
mention_count = 0
for start, end, enttype, mentype in result['graph']['entities']:
if mention_type == mentype:
mention_id = '{}-{}'.format(doc_id, mention_count)
tab_line = mention_to_tab(
start, end, enttype, mentype, mention_id, tokens, token_ids)
writer.write(tab_line + '\n')
# combined tab output
writer = writers['nam+nom+pro_tab']
mention_count = 0
for start, end, enttype, mentype in result['graph']['entities']:
mention_id = '{}-{}'.format(doc_id, mention_count)
tab_line = mention_to_tab(
start, end, enttype, mentype, mention_id, tokens, token_ids)
writer.write(tab_line + '\n')
for w in writers:
w.close()
def normalize_score(scores):
min_score, max_score = min(scores), max(scores)
if min_score == max_score:
return [0] * len(scores)
return [(s - min_score) / (max_score - min_score) for s in scores]
def best_score_by_task(log_file, task, max_epoch=1000):
with open(log_file, 'r', encoding='utf-8') as r:
config = r.readline()
best_scores = []
best_dev_score = 0
for line in r:
record = json.loads(line)
dev = record['dev']
test = record['test']
epoch = record['epoch']
if epoch > max_epoch:
break
if dev[task]['f'] > best_dev_score:
best_dev_score = dev[task]['f']
best_scores = [dev, test, epoch]
print('Epoch: {}'.format(best_scores[-1]))
tasks = ['entity', 'mention', 'relation', 'trigger_id', 'trigger',
'role_id', 'role']
for t in tasks:
print('{}: dev: {:.2f}, test: {:.2f}'.format(t,
best_scores[0][t][
'f'] * 100.0,
best_scores[1][t][
'f'] * 100.0))
def prepare_sample_blink(data_dir):
from trankit import Pipeline
nlp = Pipeline('english')
fpaths = [os.path.join(data_dir, fname) for fname in os.listdir(data_dir) if fname.endswith('.tsv')]
random.seed(2022)
random.shuffle(fpaths)
selected_fpaths = fpaths
total_meetings, total_sent, total_speakers, total_names = 0, 0, 0, 0
found_speakers = 0
all_sentences = []
all_meetings = []
for fpath in selected_fpaths:
meeting = read_blink_file(nlp, fpath)
total_speakers += len(set([s['speakerFaceId'] for s in meeting['sentences']]))
speaker_ids = set()
for s in meeting['sentences']:
for ent in s['person_names']:
if ent['is_speaker']:
speaker_ids.add(ent['speakerId'])
found_speakers += len(speaker_ids)
all_meetings.append(meeting)
all_sentences.extend(meeting['sentences'])
total_meetings += 1
ensure_dir('datasets/blink')
total_sent = len(all_sentences)
for s in all_sentences:
total_names += len(s['person_names'])
final_meetings = []
for m in all_meetings:
final_meetings.append(m)
print('statistics:')
print('total meetings: {}, total sentences: {}, total speakers: {}, found speakers: {}, recall bound: {}, total names: {}'.format(total_meetings, total_sent, total_speakers, found_speakers, 100. * found_speakers/total_speakers, total_names))
with open('datasets/blink/all_meetings.json', 'w') as f:
json.dump(final_meetings, f)
dev = final_meetings
print('-' * 20)
print('generating data for model development')
print('dev size: {} meetings, {} sentences, {} names'.format(len(dev), sum([len(m['sentences']) for m in dev]), sum([sum([len(s['person_names']) for s in m['sentences']]) for m in dev])))
with open('datasets/blink/dev-meetings.blink.json', 'w') as f:
json.dump(dev, f)
def sample_mediasum(data_dir):
fpaths = [os.path.join(data_dir, fname) for fname in os.listdir(data_dir) if fname.endswith('.json')]
random.seed(2022)
random.shuffle(fpaths)
selected_fpaths = fpaths[:200]
total_meetings, total_sent, total_speakers, total_names = 0, 0, 0, 0
all_sentences = []
all_meetings = []
ida2idb = {}
for fpath in selected_fpaths:
meeting = read_mediasum_file(fpath)
for local_id, s in enumerate(meeting['sentences']):
ida = len(all_sentences)
idb = [len(all_meetings), local_id]
ida2idb[ida] = idb
all_sentences.append(s)
all_meetings.append(meeting)
total_meetings += 1
ensure_dir('datasets/mediasum')
total_sent = len(all_sentences)
with open('datasets/mediasum/all_sentences.json', 'w') as f:
for s in all_sentences:
f.write(json.dumps(s) + '\n')
# create input for NER model
model_identity = 'CoNLL03-English-synthetic'
input_file = 'datasets/mediasum/all_sentences.txt'
output_file = 'datasets/mediasum/all_sentences.person-entities.json'
with open(input_file, 'w') as f:
for s in all_sentences:
f.write(s['displayText'] + '\n')
if not os.path.exists(output_file):
os.system('python train.py --predict --dataset {} --pred_file {} --output {}'.format(model_identity, input_file, output_file))
with open(output_file) as f:
tagged_sentences = [json.loads(line.strip()) for line in f if line.strip()]
else:
with open(output_file) as f:
tagged_sentences = [json.loads(line.strip()) for line in f if line.strip()]
if len(tagged_sentences) < len(all_sentences):
os.system('python train.py --predict --dataset {} --pred_file {} --output {}'.format(model_identity, input_file, output_file))
with open(output_file) as f:
tagged_sentences = [json.loads(line.strip()) for line in f if line.strip()]
for ida in range(len(tagged_sentences)):
l = tagged_sentences[ida]
assert ida in ida2idb
idb = ida2idb[ida]
s = all_meetings[idb[0]]['sentences'][idb[1]]
assert s['displayText'].strip() == l['text'].strip(), '{}\n{}\n'.format(s['displayText'], l['text'])
s['tokens'] = l['tokens']
s['person_names'] = [ent for ent in l['entity_mentions'] if ent['entity_type'] == 'PER']
if len(s['person_names']) > 0:
s['has_names'] = 'True'
total_names += len(s['person_names'])
meetings_without_names = 0
final_meetings = []
for m in all_meetings:
names = 0
for s in m['sentences']:
if 'has_names' in s:
names += 1
if names == 0:
meetings_without_names += 1
else:
final_meetings.append(m)
print('statistics:')
print('total meetings: {}, meetings without names: {}, total sentences: {}, total speakers: {}, total names: {}'.format(total_meetings, meetings_without_names, total_sent, total_speakers, total_names))
with open('datasets/mediasum/all_meetings.json', 'w') as f:
json.dump(final_meetings, f)
train_size = int(0.8 * len(final_meetings))
dev_size = int(0.1 * len(final_meetings))
train = final_meetings[:train_size]
test = final_meetings[train_size: train_size + dev_size]
dev = final_meetings[train_size + dev_size:]
print('-' * 20)
print('generating data for model development')
print('train size: {} meetings, {} sentences, {} names'.format(len(train), sum([len(m['sentences']) for m in train]), sum([sum([len(s['person_names']) for s in m['sentences']]) for m in train])))
print('dev size: {} meetings, {} sentences, {} names'.format(len(dev), sum([len(m['sentences']) for m in dev]), sum([sum([len(s['person_names']) for s in m['sentences']]) for m in dev])))
print('test size: {} meetings, {} sentences, {} names'.format(len(dev), sum([len(m['sentences']) for m in dev]), sum([sum([len(s['person_names']) for s in m['sentences']]) for m in test])))
process_speaker_data(input_data=train, output_dir='datasets/mediasum/train/')
process_speaker_data(input_data=dev, output_dir='datasets/mediasum/dev/')
process_speaker_data(input_data=test, output_dir='datasets/mediasum/test/')
with open('datasets/mediasum/train-meetings.mediasum.json', 'w') as f:
json.dump(train, f)
with open('datasets/mediasum/dev-meetings.mediasum.json', 'w') as f:
json.dump(dev, f)
with open('datasets/mediasum/test-meetings.mediasum.json', 'w') as f:
json.dump(test, f)
def process_speaker_data(input_data, output_dir):
ensure_dir(output_dir)
print('-' * 20)
total_speakers = 0
total_matched_speakers = 0
total_names = 0
total_matched_names = 0
for meeting in input_data:
multiple_word_speaker_faces = set()
one_word_speaker_faces = set()
for sent in meeting['sentences']:
face = sent['speakerFaceId'].strip().split(',')[0].lower().split('(')[0].strip()
sent['face'] = face
if len(face.split()) >= 2:
multiple_word_speaker_faces.add(face)
else:
one_word_speaker_faces.add(face)
multiple_word_speaker_faces = list(multiple_word_speaker_faces)
multiple_word_speaker_faces.sort()
one_word_speaker_faces = list(one_word_speaker_faces)
one_word_speaker_faces.sort()
normalize_face = {}
for face in multiple_word_speaker_faces:
normalize_face[face] = face
for face in one_word_speaker_faces:
matching = process.extractOne(face, multiple_word_speaker_faces)
if matching is not None:
m_face, m_score = matching[0], matching[1]
else:
m_score = 0
if m_score >= NAME_MATCHING_THRESHOLD or matching is not None and (face in m_face or m_face in face):
normalize_face[face] = m_face
else:
normalize_face[face] = face
face2id = {}
for face in normalize_face:
n_face = normalize_face[face]
if n_face not in face2id:
face2id[n_face] = len(face2id)
meeting['normalize_face'] = normalize_face
meeting['face2id'] = face2id
total_speakers += len(face2id)
available_faces = list(face2id.keys())
matched_speaker_ids = set()
for sent in meeting['sentences']:
sent_text = sent['displayText']
sent_tokens = sent['tokens']
total_names += len(sent['person_names'])
for entity in sent['person_names']:
start = sent_tokens[entity['start_token']]['span'][0]
end = sent_tokens[entity['end_token'] - 1]['span'][1]
entity['text'] = sent_text[start: end]
name_lowered = entity['text'].lower()
matching = process.extractOne(name_lowered, available_faces)
m_face, m_score = matching[0], matching[1]
if m_score >= NAME_MATCHING_THRESHOLD or matching is not None and (name_lowered in m_face or m_face in name_lowered):
entity['is_speaker'] = True
entity['speakerId'] = str(face2id[m_face])
matched_speaker_ids.add(m_face)
total_matched_names += 1
else:
entity['is_speaker'] = False
entity['speakerId'] = 'N/A'
total_matched_speakers += len(matched_speaker_ids)
print('num meetings: {}, total_sentences: {}, total speakers: {}, found speakers: {}/{} ~ {}%, total_names: {}, names matching speakers: {}/{} ~ {}%'.format(
len(input_data),
sum([len(x['sentences']) for x in input_data]),
total_speakers,
total_matched_speakers,
total_speakers,
total_matched_speakers * 100. / total_speakers,
total_names,
total_matched_names,
total_names,
total_matched_names * 100./total_names))
print('writing to xlsx files...')
for meeting in input_data:
out_file = os.path.join(output_dir, meeting['doc_id'].rstrip('.json') + '.tsv')
out = []
out.append('speakerFaceName\tspeakerFaceId\tsentenceId\tdisplayText\tpersonNames\tspeakerIds')
face2id = meeting['face2id']
normalize_face = meeting['normalize_face']
for sid, sent in enumerate(meeting['sentences']):
cols = []
cols.append(sent['speakerFaceId'])
cols.append(str(face2id[normalize_face[sent['face']]]))
cols.append(str(sid))
cols.append(sent['displayText'])
cols.append(','.join([entity['text'] for entity in sent['person_names']]))
cols.append(','.join([str(entity['speakerId']) for entity in sent['person_names']]))
out.append('\t'.join(cols))
with open(out_file, 'w') as f:
f.write('\n'.join(out))
tsv_to_xlsx(out_file, out_file.rstrip('.tsv') + '.xlsx')
def tsv_to_xlsx(tsv_file, xlsx_file):
# Creating an XlsxWriter workbook object and adding
# a worksheet.
workbook = Workbook(xlsx_file)
worksheet = workbook.add_worksheet()
# Reading the tsv file.
read_tsv = csv.reader(open(tsv_file, 'r', encoding='utf-8'), delimiter='\t')
# We'll use a loop with enumerate to pass the data
# together with its row position number, which we'll
# use as the cell number in the write_row() function.
for row, data in enumerate(read_tsv):
worksheet.write_row(row, 0, data)
# Closing the xlsx file.
workbook.close()
os.system('rm -rf {}'.format(tsv_file))
def read_mediasum_file(fpath):
with open(fpath) as f:
data = json.load(f)
return {'doc_id': os.path.basename(fpath), 'sentences': [{'sentence_id': '{}-{}'.format(os.path.basename(fpath), i), 'displayText': sent['displayText'].replace('\n', ' '), 'speakerFaceId': sent['speakerFaceId']} for i, sent in enumerate(data['sentences'])]}
def read_blink_file(nlp, fpath):
with open(fpath) as f:
lines = [line.strip() for line in f if line.strip()][1:]
doc_id = os.path.basename(fpath)
data = {'doc_id': doc_id, 'sentences': []}
progress = tqdm.tqdm(total=len(lines), ncols=75,
desc='processing')
for line in lines:
progress.update(1)
feats = line.split('\t')
assert len(feats) >= 3
sent = {
'sentence_id': '{}-{}'.format(doc_id, feats[1]),
'displayText': feats[2].strip(),
'speakerFaceId': feats[0].strip()
}
if sent['speakerFaceId'].startswith('off'):
continue
sent['tokens'] = nlp.tokenize(sent['displayText'], is_sent=True)['tokens']
cid2tid = {}
for tid in range(len(sent['tokens'])):
token = sent['tokens'][tid]
for cid in range(token['span'][0], token['span'][1]):
cid2tid[cid] = tid
sent['person_names'] = []
sent['has_names'] = False
data['sentences'].append(sent)
if len(feats) > 3:
if len(feats) == 4:
feats.append(','.join(['N/A'] * len(feats[3].split(','))))
assert len(feats) == 5 and len(feats[3].split(',')) == len(feats[4].split(',')), 'file_name: {}, sentence id: {}, {}'.format(doc_id, feats[1], feats)
sent['has_names'] = True
offset = 0
for name, speaker_id in zip(feats[3].split(','), feats[4].split(',')):
new_ent = {'text': name}
start_token = 1000000
end_token = -1
start_char = offset + sent['displayText'][offset:].index(name)
assert start_char >= 0
end_char = start_char + len(name)
offset = end_char
for cid in range(start_char, end_char):
if cid not in cid2tid:
continue
start_token = min(start_token, cid2tid[cid])
end_token = max(end_token, cid2tid[cid])
end_token += 1
assert end_token > start_token >= 0, '{}'.format(sent)
new_ent['start_token'] = start_token
new_ent['end_token'] = end_token
if speaker_id == 'N/A':
new_ent['is_speaker'] = False
new_ent['speakerId'] = 'N/A'
else:
new_ent['is_speaker'] = True
new_ent['speakerId'] = speaker_id
new_ent['entity_id'] = 'entity-{}'.format(len(sent['person_names']) + 1)
sent['person_names'].append(new_ent)
progress.close()
return data
if __name__ == '__main__':
sample_mediasum('datasets/mediaSum/transcripts/')