-
Notifications
You must be signed in to change notification settings - Fork 0
/
processing.py
841 lines (675 loc) · 32.4 KB
/
processing.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
import os
from os.path import join
import warnings
from tqdm import tqdm
from collections import Counter
import numpy as np
import sklearn
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
import torch
from torchtext.data.utils import get_tokenizer
from torchtext.vocab import build_vocab_from_iterator
import utils
import settings
import logging
logger = logging.getLogger(__name__)
logging.basicConfig(level=logging.INFO, format='%(asctime)s %(message)s') # include timestamp
warnings.simplefilter(action='ignore', category=FutureWarning)
def gen_short_text_pair_sim_features(e1, e2, tfidf_vec):
e1 = utils.normalize_text(e1)
e2 = utils.normalize_text(e2)
tokens1 = e1.split()
tokens2 = e2.split()
overlap = Counter(tokens1) & Counter(tokens2)
jaccard = sum(overlap.values()) / (len(tokens1) + len(tokens2) - sum(overlap.values()))
tfidf_vectors = tfidf_vec.transform([e1, e2])
cos_sim = cosine_similarity(tfidf_vectors, tfidf_vectors)
tfidf_cos_sim = cos_sim[0, 1]
return [jaccard, tfidf_cos_sim]
def gen_short_texts_sim_stat_features(entity_type="aff"):
cur_data_dir = join(settings.DATA_DIR, entity_type)
pairs_train = utils.load_json(cur_data_dir, "{}_alignment_train_pairs.json".format(entity_type))
pairs_test = utils.load_json(cur_data_dir, "{}_alignment_test_pairs.json".format(entity_type))
text_corpus = set()
for pair in pairs_train + pairs_test:
text_corpus.add(utils.normalize_text(pair["{}1".format(entity_type)]))
text_corpus.add(utils.normalize_text(pair["{}2".format(entity_type)]))
tfidf_vec = TfidfVectorizer()
tfidf_vec.fit(list(text_corpus))
print(tfidf_vec.vocabulary_)
print(tfidf_vec.idf_)
feat_train = []
labels_train = []
for pair in tqdm(pairs_train):
cur_vec = gen_short_text_pair_sim_features(pair["{}1".format(entity_type)], pair["{}2".format(entity_type)], tfidf_vec)
feat_train.append(cur_vec)
labels_train.append(pair["label"])
feat_test = []
labels_test = []
for pair in tqdm(pairs_test):
cur_vec = gen_short_text_pair_sim_features(pair["{}1".format(entity_type)], pair["{}2".format(entity_type)], tfidf_vec)
feat_test.append(cur_vec)
labels_test.append(pair["label"])
cur_out_dir = join(settings.OUT_DIR, entity_type)
os.makedirs(cur_out_dir, exist_ok=True)
np.save(join(cur_out_dir, "{}_sim_stat_features_train.npy".format(entity_type)), np.array(feat_train))
np.save(join(cur_out_dir, "{}_sim_stat_features_test.npy".format(entity_type)), np.array(feat_test))
np.save(join(cur_out_dir, "{}_labels_train.npy".format(entity_type)), labels_train)
np.save(join(cur_out_dir, "{}_labels_test.npy".format(entity_type)), labels_test)
def gen_each_author_pair_struct_feature(pids1, pids2, pids_map):
overlap = Counter(pids1) & Counter(pids2)
jaccard = sum(overlap.values()) / (len(pids1) + len(pids2) - sum(overlap.values()))
vids1 = ["/".join(pids_map[x].split("/")[:-1]) for x in pids1]
vids2 = ["/".join(pids_map[x].split("/")[:-1]) for x in pids2]
print("vids1", vids1)
print("vids2", vids2)
overlap2 = Counter(vids1) & Counter(vids2)
jaccard2 = sum(overlap2.values()) / (len(vids1) + len(vids2) - sum(overlap2.values()))
return [jaccard, jaccard2]
def gen_author_sim_struct_features():
cur_data_dir = join(settings.DATA_DIR, "author")
pairs_train = utils.load_json(cur_data_dir, "author_alignment_train_pairs.json")
pairs_test = utils.load_json(cur_data_dir, "author_alignment_test_pairs.json")
paper_aid_to_did = utils.load_json(cur_data_dir, "aminer_to_dblp_paper_map.json")
aminer_author_info = utils.load_json(cur_data_dir, "aminer_ego_author_attr_dict.json")
dblp_author_info = utils.load_json(cur_data_dir, "dblp_ego_author_attr_dict.json")
feat_train = []
feat_test = []
for pair in tqdm(pairs_train):
aid = pair["aminer"]
name_d = pair["dblp"]
cur_vec = gen_each_author_pair_struct_feature(aminer_author_info[aid]["pubs"], dblp_author_info[name_d]["pubs"], paper_aid_to_did)
feat_train.append(cur_vec)
for pair in tqdm(pairs_test):
aid = pair["aminer"]
name_d = pair["dblp"]
cur_vec = gen_each_author_pair_struct_feature(aminer_author_info[aid]["pubs"], dblp_author_info[name_d]["pubs"], paper_aid_to_did)
feat_test.append(cur_vec)
out_dir = join(settings.OUT_DIR, "author")
os.makedirs(out_dir, exist_ok=True)
entity = "author"
np.save(join(out_dir, "{}_sim_stat_features_train.npy".format(entity)), np.array(feat_train))
np.save(join(out_dir, "{}_sim_stat_features_test.npy".format(entity)), np.array(feat_test))
np.save(join(out_dir, "{}_labels_train.npy".format(entity)), [x["label"] for x in pairs_train])
np.save(join(out_dir, "{}_labels_test.npy".format(entity)), [x["label"] for x in pairs_test])
def build_tokenizer(entity_type, pairs_train, pairs_test):
tokenizer = get_tokenizer('basic_english')
def yield_tokens(texts):
for text in texts:
yield tokenizer(text)
text_corpus = set()
for pair in tqdm(pairs_train + pairs_test):
text_corpus.add(utils.normalize_text(pair["{}1".format(entity_type)]))
text_corpus.add(utils.normalize_text(pair["{}2".format(entity_type)]))
vocab = build_vocab_from_iterator(yield_tokens(list(text_corpus)))
def text_pipeline(x):
return vocab(tokenizer(x))
return vocab, text_pipeline
def calc_keywords_seqs(x1, x2):
N = len(x1)
x1_keywords = []
x2_keywords = []
for i in tqdm(range(N)):
item1 = x1[i].tolist()
item2 = x2[i].tolist()
overlap = Counter(item1) & Counter(item2)
item1_new = []
item2_new = []
for w in item1:
if w in overlap:
item1_new.append(w)
for w in item2:
if w in overlap:
item2_new.append(w)
x1_keywords.append(torch.LongTensor(item1_new))
x2_keywords.append(torch.LongTensor(item2_new))
return x1_keywords, x2_keywords
def process_rnn_match_pair(entity_type, max_seq1_len=10, max_seq2_len=5, shuffle=True, seed=42):
from keras.preprocessing.sequence import pad_sequences
file_dir = join(settings.DATA_DIR, entity_type)
pairs_train = utils.load_json(file_dir, "{}_alignment_{}_pairs.json".format(entity_type, "train"))
pairs_test = utils.load_json(file_dir, "{}_alignment_{}_pairs.json".format(entity_type, "test"))
vocab, text_pipeline = build_tokenizer(entity_type, pairs_train, pairs_test)
x1 = []
x2 = []
labels = []
for pair in tqdm(pairs_train):
item1 = utils.normalize_text(pair["{}1".format(entity_type)])
item2 = utils.normalize_text(pair["{}2".format(entity_type)])
item1 = text_pipeline(item1)
item2 = text_pipeline(item2)
x1.append(torch.LongTensor(item1))
x2.append(torch.LongTensor(item2))
labels.append(pair["label"])
# test
x1_test = []
x2_test = []
labels_test = []
for pair in tqdm(pairs_test):
item1 = utils.normalize_text(pair["{}1".format(entity_type)])
item2 = utils.normalize_text(pair["{}2".format(entity_type)])
item1 = text_pipeline(item1)
item2 = text_pipeline(item2)
x1_test.append(torch.LongTensor(item1))
x2_test.append(torch.LongTensor(item2))
labels_test.append(pair["label"])
x1_keywords, x2_keywords = calc_keywords_seqs(x1, x2)
x1_test_keywords, x2_test_keywords = calc_keywords_seqs(x1_test, x2_test)
vocab_size = len(vocab) + 2
x1 = pad_sequences(x1, maxlen=max_seq1_len, value=len(vocab) + 1)
x2 = pad_sequences(x2, maxlen=max_seq1_len, value=len(vocab) + 1)
x1_keywords = pad_sequences(x1_keywords, maxlen=max_seq2_len, value=len(vocab) + 1)
x2_keywords = pad_sequences(x2_keywords, maxlen=max_seq2_len, value=len(vocab) + 1)
x1_test = pad_sequences(x1_test, maxlen=max_seq1_len, value=len(vocab) + 1)
x2_test = pad_sequences(x2_test, maxlen=max_seq1_len, value=len(vocab) + 1)
x1_test_keywords = pad_sequences(x1_test_keywords, maxlen=max_seq2_len, value=len(vocab) + 1)
x2_test_keywords = pad_sequences(x2_test_keywords, maxlen=max_seq2_len, value=len(vocab) + 1)
if shuffle:
x1, x2, x1_keywords, x2_keywords, labels = sklearn.utils.shuffle(
x1, x2, x1_keywords, x2_keywords, labels,
random_state=seed
)
out_dir = join(settings.OUT_DIR, entity_type, "rnn")
os.makedirs(out_dir, exist_ok=True)
N = len(labels)
if entity_type == "aff":
n_train = int(N*0.75)
n_valid = int(N*0.25)
elif entity_type == "venue":
n_train = int(N / 5 * 4)
n_valid = int(N / 5)
else:
raise NotImplementedError
train_data = {}
train_data["x1_seq1"] = x1[:n_train]
train_data["x1_seq2"] = x1_keywords[:n_train]
train_data["x2_seq1"] = x2[:n_train]
train_data["x2_seq2"] = x2_keywords[:n_train]
train_data["y"] = labels[:n_train]
train_data["vocab_size"] = vocab_size
print("train labels", len(train_data["y"]))
valid_data = {}
valid_data["x1_seq1"] = x1[n_train:(n_train + n_valid)]
valid_data["x1_seq2"] = x1_keywords[n_train:(n_train + n_valid)]
valid_data["x2_seq1"] = x2[n_train:(n_train + n_valid)]
valid_data["x2_seq2"] = x2_keywords[n_train:(n_train + n_valid)]
valid_data["y"] = labels[n_train:(n_train + n_valid)]
print("valid labels", len(valid_data["y"]))
utils.dump_large_obj(train_data, out_dir, "{}_rnn_train.pkl".format(entity_type))
utils.dump_large_obj(valid_data, out_dir, "{}_rnn_valid.pkl".format(entity_type))
test_data = {}
test_data["x1_seq1"] = x1_test
test_data["x1_seq2"] = x1_test_keywords
test_data["x2_seq1"] = x2_test
test_data["x2_seq2"] = x2_test_keywords
test_data["y"] = labels_test
print("test labels", len(test_data["y"]))
utils.dump_large_obj(test_data, out_dir, "{}_rnn_test.pkl".format(entity_type))
def process_rnn_author_match_pair(max_seq1_len=10, max_seq2_len=5, shuffle=True, seed=42):
from keras.preprocessing.sequence import pad_sequences
cur_data_dir = join(settings.DATA_DIR, "author")
pairs_train = utils.load_json(cur_data_dir, "author_alignment_train_pairs.json")
pairs_test = utils.load_json(cur_data_dir, "author_alignment_test_pairs.json")
node_list = []
with open(join(cur_data_dir, "large_cross_graph_nodes_list.txt")) as rf:
for i, line in enumerate(rf):
node_list.append(line.strip())
node_to_idx = {n: i for i, n in enumerate(node_list)}
aminer_ego_author_attr = utils.load_json(cur_data_dir, "aminer_ego_author_attr_dict.json")
dblp_ego_author_attr = utils.load_json(cur_data_dir, "dblp_ego_author_attr_dict.json")
aminer_aid_to_vids = utils.load_json(cur_data_dir, "aminer_aid_to_vids.json")
dblp_aid_to_vids = utils.load_json(cur_data_dir, "dblp_aid_to_vids.json")
x1 = []
x2 = []
x1_keywords = []
x2_keywords = []
for pair in tqdm(pairs_train):
aid = pair["aminer"]
did = pair["dblp"]
pubs_a = aminer_ego_author_attr.get(aid, [])["pubs"][: max_seq1_len]
pubs_d = dblp_ego_author_attr.get(did, [])["pubs"][: max_seq1_len]
pubs_a_idx = [node_to_idx[x + "-pp"] for x in pubs_a]
pubs_d_idx = [node_to_idx[x + "-pp"] for x in pubs_d]
x1.append(pubs_a_idx)
x2.append(pubs_d_idx)
vids_a = aminer_aid_to_vids.get(aid, [])[: max_seq2_len]
vids_d = dblp_aid_to_vids.get(aid, [])[: max_seq2_len]
vids_a_idx = [node_to_idx[x["id"] + "-vv"] for x in vids_a]
vids_d_idx = [node_to_idx[x["id"] + "-vv"] for x in vids_d]
x1_keywords.append(vids_a_idx)
x2_keywords.append(vids_d_idx)
# test
x1_test = []
x2_test = []
x1_keywords_test = []
x2_keywords_test = []
for pair in tqdm(pairs_test):
aid = pair["aminer"]
did = pair["dblp"]
pubs_a = aminer_ego_author_attr.get(aid, [])["pubs"][: max_seq1_len]
pubs_d = dblp_ego_author_attr.get(did, [])["pubs"][: max_seq1_len]
pubs_a_idx = [node_to_idx[x + "-pp"] for x in pubs_a]
pubs_d_idx = [node_to_idx[x + "-pp"] for x in pubs_d]
x1_test.append(pubs_a_idx)
x2_test.append(pubs_d_idx)
vids_a = aminer_aid_to_vids.get(aid, [])[: max_seq2_len]
vids_d = dblp_aid_to_vids.get(aid, [])[: max_seq2_len]
vids_a_idx = [node_to_idx[x["id"] + "-vv"] for x in vids_a]
vids_d_idx = [node_to_idx[x["id"] + "-vv"] for x in vids_d]
x1_keywords_test.append(vids_a_idx)
x2_keywords_test.append(vids_d_idx)
n_nodes = len(node_list)
x1 = pad_sequences(x1, maxlen=max_seq1_len, value=n_nodes + 1)
x2 = pad_sequences(x2, maxlen=max_seq1_len, value=n_nodes + 1)
x1_keywords = pad_sequences(x1_keywords, maxlen=max_seq2_len, value=n_nodes + 1)
x2_keywords = pad_sequences(x2_keywords, maxlen=max_seq2_len, value=n_nodes + 1)
x1_test = pad_sequences(x1_test, maxlen=max_seq1_len, value=n_nodes + 1)
x2_test = pad_sequences(x2_test, maxlen=max_seq1_len, value=n_nodes + 1)
x1_test_keywords = pad_sequences(x1_keywords_test, maxlen=max_seq2_len, value=n_nodes + 1)
x2_test_keywords = pad_sequences(x2_keywords_test, maxlen=max_seq2_len, value=n_nodes + 1)
labels = [x["label"] for x in pairs_train]
labels_test = [x["label"] for x in pairs_test]
if shuffle:
x1, x2, x1_keywords, x2_keywords, labels = sklearn.utils.shuffle(
x1, x2, x1_keywords, x2_keywords, labels,
random_state=seed
)
out_dir = join(settings.OUT_DIR, "author", "rnn")
os.makedirs(out_dir, exist_ok=True)
N = len(pairs_train)
n_train = int(N / 9 * 8)
n_valid = int(N / 9)
train_data = {}
train_data["x1_seq1"] = x1[:n_train]
train_data["x1_seq2"] = x1_keywords[:n_train]
train_data["x2_seq1"] = x2[:n_train]
train_data["x2_seq2"] = x2_keywords[:n_train]
train_data["y"] = labels[:n_train]
train_data["vocab_size"] = n_nodes
print("train labels", len(train_data["y"]))
valid_data = {}
valid_data["x1_seq1"] = x1[n_train:(n_train + n_valid)]
valid_data["x1_seq2"] = x1_keywords[n_train:(n_train + n_valid)]
valid_data["x2_seq1"] = x2[n_train:(n_train + n_valid)]
valid_data["x2_seq2"] = x2_keywords[n_train:(n_train + n_valid)]
valid_data["y"] = labels[n_train:(n_train + n_valid)]
print("valid labels", len(valid_data["y"]))
utils.dump_large_obj(train_data, out_dir, "author_rnn_train.pkl")
utils.dump_large_obj(valid_data, out_dir, "author_rnn_valid.pkl")
test_data = {}
test_data["x1_seq1"] = x1_test
test_data["x1_seq2"] = x1_test_keywords
test_data["x2_seq1"] = x2_test
test_data["x2_seq2"] = x2_test_keywords
test_data["y"] = labels_test
print("test labels", len(test_data["y"]))
utils.dump_large_obj(test_data, out_dir, "author_rnn_test.pkl")
def complete_sentence_pair_to_matrix(t1, t2, mat_size, pretrain_emb=None):
matrix = -np.ones((mat_size, mat_size))
for i, word1 in enumerate(t1[: mat_size]):
for j, word2 in enumerate(t2[: mat_size]):
v = -1
if word1 == word2:
v = 1
elif pretrain_emb is not None:
v = cosine_similarity(pretrain_emb[word1].reshape(1, -1),
pretrain_emb[word2].reshape(1, -1))[0][0]
matrix[i][j] = v
return matrix
def sentences_overlap_to_matrix(t1, t2, mat_size):
overlap = set(t1).intersection(t2)
new_seq_t1 = []
new_seq_t2 = []
for w in t1:
if w in overlap:
new_seq_t1.append(w)
for w in t2:
if w in overlap:
new_seq_t2.append(w)
matrix = -np.ones((mat_size, mat_size))
for i, word1 in enumerate(t1[: mat_size]):
for j, word2 in enumerate(t2[: mat_size]):
v = -1
if word1 == word2:
v = 1
matrix[i][j] = v
return matrix
def process_cnn_match_pair(entity_type, matrix_size_1=10, matrix_size_2=5, shuffle=True, seed=42):
file_dir = join(settings.DATA_DIR, entity_type)
pairs_train = utils.load_json(file_dir, "{}_alignment_{}_pairs.json".format(entity_type, "train"))
pairs_test = utils.load_json(file_dir, "{}_alignment_{}_pairs.json".format(entity_type, "test"))
vocab, text_pipeline = build_tokenizer(entity_type, pairs_train, pairs_test)
n_matrix = len(pairs_train)
x_long = np.zeros((n_matrix, matrix_size_1, matrix_size_1))
x_short = np.zeros((n_matrix, matrix_size_2, matrix_size_2))
y = np.zeros(n_matrix, dtype=np.long)
count = 0
for i, pair in enumerate(pairs_train):
if i % 100 == 0:
print('pairs to matrices', i)
cur_y = pair["label"]
item1 = utils.normalize_text(pair["{}1".format(entity_type)])
item2 = utils.normalize_text(pair["{}2".format(entity_type)])
item1 = text_pipeline(item1)
item2 = text_pipeline(item2)
matrix1 = complete_sentence_pair_to_matrix(item1, item2, matrix_size_1)
x_long[count] = utils.scale_matrix(matrix1)
matrix2 = sentences_overlap_to_matrix(item1, item2, matrix_size_2)
x_short[count] = utils.scale_matrix(matrix2)
y[count] = cur_y
count += 1
n_test = len(pairs_test)
x_test_long = np.zeros((n_test, matrix_size_1, matrix_size_1))
x_test_short = np.zeros((n_test, matrix_size_2, matrix_size_2))
y_test = np.zeros(n_test, dtype=np.long)
count = 0
for i, pair in enumerate(pairs_test):
if i % 100 == 0:
print('pairs to matrices', i)
cur_y = pair["label"]
item1 = utils.normalize_text(pair["{}1".format(entity_type)])
item2 = utils.normalize_text(pair["{}2".format(entity_type)])
item1 = text_pipeline(item1)
item2 = text_pipeline(item2)
matrix1 = complete_sentence_pair_to_matrix(item1, item2, matrix_size_1)
x_test_long[count] = utils.scale_matrix(matrix1)
matrix2 = sentences_overlap_to_matrix(item1, item2, matrix_size_2)
x_test_short[count] = utils.scale_matrix(matrix2)
y_test[count] = cur_y
count += 1
print("shuffle", shuffle)
if shuffle:
x_long, x_short, y = sklearn.utils.shuffle(
x_long, x_short, y,
random_state=seed
)
N = len(y)
if entity_type == "aff":
n_train = int(N*0.75)
n_valid = int(N*0.25)
elif entity_type == "venue":
n_train = int(N / 5 * 4)
n_valid = int(N / 5)
else:
raise NotImplementedError
out_dir = join(settings.OUT_DIR, entity_type, "cnn")
os.makedirs(out_dir, exist_ok=True)
train_data = {}
train_data["x1"] = x_long[:n_train]
train_data["x2"] = x_short[:n_train]
train_data["y"] = y[:n_train]
print("train labels", len(train_data["y"]))
valid_data = {}
valid_data["x1"] = x_long[n_train:(n_train + n_valid)]
valid_data["x2"] = x_short[n_train:(n_train + n_valid)]
valid_data["y"] = y[n_train:(n_train + n_valid)]
print("valid labels", len(valid_data["y"]), valid_data["y"])
test_data = {}
test_data["x1"] = x_test_long
test_data["x2"] = x_test_short
test_data["y"] = y_test
print("test labels", len(test_data["y"]), test_data["y"])
utils.dump_large_obj(train_data, out_dir, "{}_cnn_train.pkl".format(entity_type))
utils.dump_large_obj(test_data, out_dir, "{}_cnn_test.pkl".format(entity_type))
utils.dump_large_obj(valid_data, out_dir, "{}_cnn_valid.pkl".format(entity_type))
def process_cnn_author_match_pair(matrix_size_1=10, matrix_size_2=5, shuffle=True, seed=42):
file_dir = join(settings.DATA_DIR, "author")
pairs_train = utils.load_json(file_dir, "author_alignment_{}_pairs.json".format("train"))
pairs_test = utils.load_json(file_dir, "author_alignment_{}_pairs.json".format("test"))
labels_test = [x["label"] for x in pairs_test]
pretrain_emb = np.load(join(file_dir, "large_cross_graph_node_emb.npy"))
pretrain_emb = np.concatenate((pretrain_emb, np.zeros(shape=(2, 128))), axis=0)
node_list = []
with open(join(file_dir, "large_cross_graph_nodes_list.txt")) as rf:
for i, line in enumerate(rf):
node_list.append(line.strip())
node_to_idx = {n: i for i, n in enumerate(node_list)}
aminer_ego_author_attr = utils.load_json(file_dir, "aminer_ego_author_attr_dict.json")
dblp_ego_author_attr = utils.load_json(file_dir, "dblp_ego_author_attr_dict.json")
aminer_aid_to_vids = utils.load_json(file_dir, "aminer_aid_to_vids.json")
dblp_aid_to_vids = utils.load_json(file_dir, "dblp_aid_to_vids.json")
n_matrix = len(pairs_train)
x_long = np.zeros((n_matrix, matrix_size_1, matrix_size_1))
x_short = np.zeros((n_matrix, matrix_size_2, matrix_size_2))
y = [x["label"] for x in pairs_train]
count = 0
for pair in tqdm(pairs_train):
aid = pair["aminer"]
did = pair["dblp"]
pubs_a = aminer_ego_author_attr.get(aid, [])["pubs"][: matrix_size_1]
pubs_d = dblp_ego_author_attr.get(did, [])["pubs"][: matrix_size_1]
pubs_a_idx = [node_to_idx[x + "-pp"] for x in pubs_a]
pubs_d_idx = [node_to_idx[x + "-pp"] for x in pubs_d]
matrix1 = complete_sentence_pair_to_matrix(pubs_a_idx, pubs_d_idx, matrix_size_1, pretrain_emb)
x_long[count] = utils.scale_matrix(matrix1)
vids_a = aminer_aid_to_vids.get(aid, [])[: matrix_size_2]
vids_d = dblp_aid_to_vids.get(aid, [])[: matrix_size_2]
vids_a_idx = [node_to_idx[x["id"] + "-vv"] for x in vids_a]
vids_d_idx = [node_to_idx[x["id"] + "-vv"] for x in vids_d]
matrix2 = complete_sentence_pair_to_matrix(vids_a_idx, vids_d_idx, matrix_size_2, pretrain_emb)
x_short[count] = utils.scale_matrix(matrix2)
count += 1
x_test_long = np.zeros((len(labels_test), matrix_size_1, matrix_size_1))
x_test_short = np.zeros((len(labels_test), matrix_size_2, matrix_size_2))
y_test = labels_test
count = 0
for pair in tqdm(pairs_test):
aid = pair["aminer"]
did = pair["dblp"]
pubs_a = aminer_ego_author_attr.get(aid, [])["pubs"][: matrix_size_1]
pubs_d = dblp_ego_author_attr.get(did, [])["pubs"][: matrix_size_1]
pubs_a_idx = [node_to_idx[x + "-pp"] for x in pubs_a]
pubs_d_idx = [node_to_idx[x + "-pp"] for x in pubs_d]
matrix1 = complete_sentence_pair_to_matrix(pubs_a_idx, pubs_d_idx, matrix_size_1, pretrain_emb)
x_test_long[count] = utils.scale_matrix(matrix1)
vids_a = aminer_aid_to_vids.get(aid, [])[: matrix_size_2]
vids_d = dblp_aid_to_vids.get(aid, [])[: matrix_size_2]
vids_a_idx = [node_to_idx[x["id"] + "-vv"] for x in vids_a]
vids_d_idx = [node_to_idx[x["id"] + "-vv"] for x in vids_d]
matrix2 = complete_sentence_pair_to_matrix(vids_a_idx, vids_d_idx, matrix_size_2, pretrain_emb)
x_test_short[count] = utils.scale_matrix(matrix2)
count += 1
print("shuffle", shuffle)
if shuffle:
x_long, x_short, y = sklearn.utils.shuffle(
x_long, x_short, y,
random_state=seed
)
N = len(y)
n_train = int(N / 9 * 8)
n_valid = int(N / 9)
out_dir = join(settings.OUT_DIR, "author", "cnn")
os.makedirs(out_dir, exist_ok=True)
train_data = {}
train_data["x1"] = x_long[:n_train]
train_data["x2"] = x_short[:n_train]
train_data["y"] = y[:n_train]
print("train labels", len(train_data["y"]))
valid_data = {}
valid_data["x1"] = x_long[n_train:(n_train + n_valid)]
valid_data["x2"] = x_short[n_train:(n_train + n_valid)]
valid_data["y"] = y[n_train:(n_train + n_valid)]
print("valid labels", len(valid_data["y"]), valid_data["y"])
test_data = {}
test_data["x1"] = x_test_long
test_data["x2"] = x_test_short
test_data["y"] = y_test
print("test labels", len(test_data["y"]), test_data["y"])
utils.dump_large_obj(train_data, out_dir, "author_cnn_train.pkl")
utils.dump_large_obj(test_data, out_dir, "author_cnn_test.pkl")
utils.dump_large_obj(valid_data, out_dir, "author_cnn_valid.pkl")
def gen_author_paired_subgraphs():
author_dir = join(settings.DATA_DIR, "author")
pairs_train = utils.load_json(author_dir, "author_alignment_{}_pairs.json".format("train"))
pairs_test = utils.load_json(author_dir, "author_alignment_{}_pairs.json".format("test"))
labels_train = [x["label"] for x in pairs_train]
labels_test = [x["label"] for x in pairs_test]
adjs = []
vertex_ids = []
vertex_types = []
adjs_test = []
vertex_ids_test = []
vertex_types_test = []
n_nodes_ego = 382
aminer_ego_author_attr = utils.load_json(author_dir, "aminer_ego_author_attr_dict.json")
aminer_aid_to_sorted_pubs = utils.load_json(author_dir, "aminer_aid_to_sorted_pubs.json")
dblp_ego_author_attr = utils.load_json(author_dir, "dblp_ego_author_attr_dict.json")
dblp_aid_to_sorted_pids = utils.load_json(author_dir, "dblp_aid_to_sorted_pubs.json")
aminer_aid_to_vids = utils.load_json(author_dir, "aminer_aid_to_vids.json")
dblp_aid_to_vids = utils.load_json(author_dir, "dblp_aid_to_vids.json")
aminer_aid_to_coauthors = utils.load_json(author_dir, "aminer_aid_to_coauthors.json")
dblp_aid_to_coauthors = utils.load_json(author_dir, "dblp_aid_to_coauthors.json")
coauthor_dict_cache = {**aminer_aid_to_coauthors, **dblp_aid_to_coauthors}
ego_attr = {**aminer_ego_author_attr, **dblp_ego_author_attr}
aid_to_pids_ego = {x: ego_attr[x].get("pubs", []) for x in ego_attr}
aid_to_pids = {**aid_to_pids_ego, **aminer_aid_to_sorted_pubs, **dblp_aid_to_sorted_pids}
aid_to_vids = {**aminer_aid_to_vids, **dblp_aid_to_vids}
for i, pair in enumerate(pairs_train + pairs_test):
if i < len(labels_train):
cur_adjs = adjs
cur_vertex_ids = vertex_ids
cur_vertex_types = vertex_types
else:
cur_adjs = adjs_test
cur_vertex_ids = vertex_ids_test
cur_vertex_types = vertex_types_test
node_set = set()
node_list = []
node_type_list = []
n_authors = 0
n_venues = 0
n_papers = 0
aaid, daid = pair['aminer'], pair['dblp']
aaid_dec = '{}-aa'.format(aaid)
daid_dec = '{}-ad'.format(daid)
n_authors += 2
# focal_node1 = aaid
# focal_node2 = daid
node_set.update({aaid, daid})
node_list += [aaid_dec, daid_dec]
node_type_list += [settings.AUTHOR_TYPE, settings.AUTHOR_TYPE]
# 1-ego venues
a_venues = aminer_aid_to_vids.get(aaid, [])[:10]
d_venues = dblp_aid_to_vids.get(daid, [])[:10]
venues_1_ego = [v['id'] for v in a_venues + d_venues]
for v in venues_1_ego:
if v not in node_set:
node_set.add(v)
v_dec = '{}-vv'.format(v)
n_venues += 1
node_list.append(v_dec)
node_type_list.append(settings.VENUE_TYPE)
# 1-ego papers
a_pubs = aminer_ego_author_attr.get(aaid, {}).get('pubs', [])[:20]
d_pubs = dblp_ego_author_attr.get(daid, {}).get('pubs', [])[:20]
pubs_1_ego = {item for item in a_pubs + d_pubs if item not in node_set}
node_set.update(pubs_1_ego)
node_list += ['{}-pp'.format(item) for item in pubs_1_ego]
node_type_list += [settings.PAPER_TYPE] * len(pubs_1_ego)
n_papers += len(pubs_1_ego)
# 1-ego coauthors
a_coauthors = aminer_aid_to_coauthors.get(aaid, [])[:10]
for cur_aid in a_coauthors:
if cur_aid not in node_set:
node_set.add(cur_aid)
node_list.append('{}-aa'.format(cur_aid))
node_type_list.append(settings.AUTHOR_TYPE)
n_authors += 1
d_coauthors = dblp_aid_to_coauthors.get(daid, [])[:10]
d_coauthors = {a for a in d_coauthors if a not in node_set}
node_set.update(d_coauthors)
node_list += ['{}-ad'.format(a) for a in d_coauthors]
node_type_list += [settings.AUTHOR_TYPE] * len(d_coauthors)
n_authors += len(d_coauthors)
# 2-ego
for a in a_coauthors:
cur_venues = aminer_aid_to_vids.get(a, [])[:5]
cur_vids = {v['id'] for v in cur_venues if v['id'] not in node_set}
node_set.update(cur_vids)
node_list += ['{}-vv'.format(v) for v in cur_vids]
node_type_list += [settings.VENUE_TYPE] * len(cur_vids)
n_venues += len(cur_vids)
for a in d_coauthors:
cur_venues = dblp_aid_to_vids.get(a, [])[:5]
cur_vids = {v['id'] for v in cur_venues if v['id'] not in node_set}
node_set.update(cur_vids)
node_list += ['{}-vv'.format(v) for v in cur_vids]
node_type_list += [settings.VENUE_TYPE] * len(cur_vids)
n_venues += len(cur_vids)
a_pubs_2 = []
for a in a_coauthors:
cur_pubs = aminer_aid_to_sorted_pubs.get(a, [])[:10]
a_pubs_2 += cur_pubs
d_pubs_2 = []
for a in d_coauthors:
cur_pubs = dblp_aid_to_sorted_pids.get(a, [])[:10]
d_pubs_2 += cur_pubs
pubs_2_ego = {item for item in a_pubs_2 + d_pubs_2 if item not in node_set}
node_set.update(pubs_2_ego)
node_list += ['{}-pp'.format(p) for p in pubs_2_ego]
node_type_list += [settings.PAPER_TYPE] * len(pubs_2_ego)
n_papers += len(pubs_2_ego)
cur_n_nodes_real = len(node_list)
assert len(node_set) == len(node_list) == len(node_type_list)
assert n_authors <= 22 and n_venues <= 120 and n_papers <= 240
# padding
for v_idx in range(len(node_set), n_nodes_ego):
node_list.append('-1')
node_type_list.append(settings.AUTHOR_TYPE)
assert len(node_list) == n_nodes_ego
cur_vertex_ids.append(node_list)
cur_vertex_types.append(node_type_list)
node_to_idx = {eid[:-3]: i for i, eid in enumerate(node_list)}
adj = np.zeros((n_nodes_ego, n_nodes_ego), dtype=np.bool_)
nnz = 0
for ii in range(cur_n_nodes_real):
v1 = node_list[ii][:-3]
t1 = node_type_list[ii]
if t1 == settings.AUTHOR_TYPE:
v_nbrs = aid_to_vids.get(v1, [])
v_nbrs = {item['id'] for item in v_nbrs}
v_nbrs_filtered = {item for item in v_nbrs if item in node_set}
nbrs_set = v_nbrs_filtered
p_nbrs = aid_to_pids.get(v1, [])
p_nbrs_filtered = {item for item in p_nbrs if item in node_set}
nbrs_set.update(p_nbrs_filtered)
a_nbrs = coauthor_dict_cache.get(v1, [])
a_nbrs_filtered = {item for item in a_nbrs if item in node_set}
nbrs_set.update(a_nbrs_filtered)
for nbr in nbrs_set:
nbr_idx_map = node_to_idx[nbr]
adj[ii, nbr_idx_map] = True
adj[nbr_idx_map, ii] = True
nnz += 1
cur_adjs.append(adj)
if i % 100 == 0:
logger.info('***********iter %d nnz %d', i, nnz)
logger.info('n_nodes real %d, n_authors %d, n_venues %d, n_papers %d.',
cur_n_nodes_real, n_authors, n_venues, n_papers)
out_dir = join(settings.OUT_DIR, "author", "hgat")
os.makedirs(out_dir, exist_ok=True)
np.save(join(out_dir, 'vertex_id_train.npy'), np.array(vertex_ids))
np.save(join(out_dir, 'vertex_types_train.npy'), np.array(vertex_types))
np.save(join(out_dir, 'adjacency_matrix_train.npy'), np.array(adjs))
np.save(join(out_dir, 'label_train.npy'), np.array(labels_train))
np.save(join(out_dir, 'vertex_id_test.npy'), np.array(vertex_ids_test))
np.save(join(out_dir, 'vertex_types_test.npy'), np.array(vertex_types_test))
np.save(join(out_dir, 'adjacency_matrix_test.npy'), np.array(adjs_test))
np.save(join(out_dir, 'label_test.npy'), np.array(labels_test))
if __name__ == "__main__":
# for svm
# gen_short_texts_sim_stat_features(entity_type="aff")
# gen_short_texts_sim_stat_features(entity_type="venue")
# gen_author_sim_struct_features()
# for rnn-based matching
# process_rnn_match_pair(entity_type="aff", max_seq1_len=10, max_seq2_len=5)
# process_rnn_match_pair(entity_type="venue", max_seq1_len=10, max_seq2_len=5)
# process_rnn_author_match_pair(max_seq1_len=10, max_seq2_len=5)
# for cnn-based matching
# process_cnn_match_pair(entity_type="aff")
# process_cnn_match_pair(entity_type="venue")
# process_cnn_author_match_pair()
# for hgat-based matching
gen_author_paired_subgraphs()
logger.info("done")