-
Notifications
You must be signed in to change notification settings - Fork 38
/
structgpt_for_webqsp.py
1218 lines (1093 loc) · 52 KB
/
structgpt_for_webqsp.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
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
import argparse
import json
import logging
import os
import pickle
import re
import openai
from KnowledgeBase.KG_api import KnowledgeGraph
# from KnowledgeBase.sparql_executor import *
import multiprocessing as mp
class ChatGPT:
def __init__(self, args, prompt_path, prompt_name, max_tokens):
self.args = args
self.history_messages = []
self.history_contents = []
self.max_tokens = max_tokens
self.prompt = self.load_prompt_template(prompt_path, prompt_name)
self.idx_mapping = {"0": "first", "1": "second", "2": "third", "3": "fourth", "4": "fifth", "5": "sixth",
"6": "seventh",
"7": "eighth", "8": "ninth", "9": "tenth"}
def get_response(self, input_text, turn_type, tpe_name=None):
if self.args.debug:
message = self.create_message(input_text, turn_type, tpe_name)
self.history_messages.append(message)
self.history_contents.append(message['content'])
print("query API to get message:\n%s" % message['content'])
# message = self.query_API_to_get_message(self.history)
# self.history.append(message)
# response = self.parse_result(message)
response = input("input the returned response:")
else:
message = self.create_message(input_text, turn_type, tpe_name)
self.history_messages.append(message)
self.history_contents.append(message['content'])
message = self.query_API_to_get_message(self.history_messages)
self.history_messages.append(message)
self.history_contents.append(message['content'])
response = self.parse_result(message, turn_type)
return response
def get_response_v1(self, input_text, turn_type, tpe_name=None):
if self.args.debug:
message = self.create_message_v1(input_text, turn_type)
self.history_messages.append(message)
self.history_contents.append(message['content'])
print("query API to get message:\n%s" % message['content'])
# message = self.query_API_to_get_message(self.history)
# self.history.append(message)
# response = self.parse_result(message)
response = input("input the returned response:")
else:
message = self.create_message_v1(input_text, turn_type)
self.history_messages.append(message)
self.history_contents.append(message['content'])
message = self.query_API_to_get_message(self.history_messages)
self.history_messages.append(message)
self.history_contents.append(message['content'])
response = self.parse_result_v1(message, turn_type)
return response
def create_message(self, input_text, turn_type, tpe_name):
if turn_type == "initial": # the initial query
instruction = self.prompt[turn_type]['instruction']
template = self.prompt[turn_type]['init_template']
self.question = input_text
input_text = instruction + template.format(question=input_text, tpe=tpe_name)
elif turn_type == "continue_template":
input_text = self.prompt[turn_type]
elif turn_type == "question_template":
template = self.prompt[turn_type]
input_text = template.format(idx=self.idx_mapping[input_text])
elif turn_type == "answer_template":
template = self.prompt[turn_type]
if len(input_text) > 0:
input_text = template["valid"].format(facts=input_text)
else:
input_text = template["invalid"]
elif turn_type == "final_query_template":
template = self.prompt[turn_type]
input_text = template.format(question=self.question)
else:
raise NotImplementedError
message = {'role': 'user', 'content': input_text}
return message
def create_message_v1(self, input_text, turn_type):
if turn_type == "instruction": # the initial query
instruction = self.prompt['instruction']
input_text = instruction
elif turn_type == "init_relation_rerank":
template = self.prompt['init_relation_rerank']
question, tpe, can_rels = input_text
input_text = template.format(question=question, tpe=tpe, relations=can_rels)
elif turn_type == "ask_question":
template = self.prompt['ask_question']
idx, relations = input_text
idx = self.idx_mapping[idx]
input_text = template.format(idx=idx, relations=relations)
elif turn_type == "ask_answer":
facts = input_text
template = self.prompt['ask_answer']
input_text = template.format(facts=facts)
elif turn_type == "ask_final_answer_or_next_question":
question, serialized_facts = input_text
template = self.prompt['ask_final_answer_or_next_question']
input_text = template.format(facts=serialized_facts, question=question)
elif turn_type == "condition":
input_text = self.prompt['continue_template']['condition']
elif turn_type == "continue":
input_text = self.prompt['continue_template']['continue']
elif turn_type == "stop":
input_text = self.prompt['continue_template']['stop']
elif turn_type == 'relation_rerank':
template = self.prompt['relation_rerank']
question, can_rels = input_text
input_text = template.format(question=question, relations=can_rels)
else:
raise NotImplementedError
message = {'role': 'user', 'content': input_text}
return message
def query_API_to_get_message(self, messages):
while True:
try:
res = openai.ChatCompletion.create(
model="gpt-3.5-turbo",
messages=messages,
temperature=0,
max_tokens=self.max_tokens,
top_p=1,
frequency_penalty=0,
presence_penalty=0,
)
return res['choices'][0]['message']
except openai.error.RateLimitError:
print('openai.error.RateLimitError\nRetrying...')
time.sleep(30)
except openai.error.ServiceUnavailableError:
print('openai.error.ServiceUnavailableError\nRetrying...')
time.sleep(20)
except openai.error.Timeout:
print('openai.error.Timeout\nRetrying...')
time.sleep(20)
except openai.error.APIError:
print('openai.error.APIError\nRetrying...')
time.sleep(20)
except openai.error.APIConnectionError:
print('openai.error.APIConnectionError\nRetrying...')
time.sleep(20)
# except openai.error.InvalidRequestError:
# print('openai.error.InvalidRequestError\nRetrying...')
def parse_result(self, result, turn_type):
content = result['content'].strip()
if turn_type in ["initial", "question_template"]:
if "should be" in content:
content = content.split("should be")[1].strip()
if content.startswith('"') and content.endswith('"'):
content = content[1:-1]
else:
matchObj = re.search(r'"(.*?)"', content)
if matchObj is not None:
content = matchObj.group()
content = content[1:-1]
else:
content = content.strip().strip('"')
print("Not exactly parse, we directly use content: %s" % content)
return content
def parse_result_v1(self, result, turn_type):
content = result['content'].strip()
if turn_type in ["ask_question", "continue"]:
if "the simple question:" in content:
content = content.split("the simple question:")[1].strip()
if content.startswith('"') and content.endswith('"'):
content = content[1:-1]
else:
matchObj = re.search(r'"(.*?)"', content)
if matchObj is not None:
content = matchObj.group()
content = content[1:-1]
else:
content = content.strip().strip('"')
print("Not exactly parse, we directly use content: %s" % content)
return content
def parse_result_v2(self, result, turn_type):
content = result['content'].strip()
return content
def reset_history(self):
self.history_messages = []
self.history_contents = []
def reset_history_messages(self):
self.history_messages = []
def reset_history_contents(self):
self.history_contents = []
def load_prompt_template(self, prompt_path, prompt_name):
if prompt_path.endswith(".json"):
with open(prompt_path, "rb") as f:
prompt = json.load(f)
return prompt[prompt_name]
def get_response_v2(self, input_text, turn_type):
message = self.create_message_v2(input_text, turn_type)
self.history_messages.append(message)
self.history_contents.append(message['content'])
message = self.query_API_to_get_message(self.history_messages)
self.history_messages.append(message)
self.history_contents.append(message['content'])
response = message['content'].strip()
return response
def create_message_v2(self, input_text, turn_type):
if turn_type == "instruction": # the initial query
instruction = self.prompt['instruction']
input_text = instruction
# ykm
# elif turn_type == "init_relation_rerank":
# template = self.prompt['init_relation_rerank']
# can_rels, question, tpe, hop = input_text
# if hop == 1:
# hop = "first"
# elif hop == 2:
# hop = "second"
# elif hop == 3:
# hop = "third"
# input_text = template.format(question=question, tpe=tpe, relations=can_rels, hop=hop)
elif turn_type == "init_relation_rerank":
template = self.prompt['init_relation_rerank']
can_rels, question, tpe = input_text
input_text = template.format(question=question, tpe=tpe, relations=can_rels)
elif turn_type == "constraints_flag":
template = self.prompt['constraints_flag']
question, tpe, selected_relations = input_text
if len(selected_relations) > 1:
selected_relations = "are " + ", ".join(selected_relations)
else:
selected_relations = "is " + ", ".join(selected_relations)
input_text = template.format(question=question, tpe=tpe, selected_relations=selected_relations)
elif turn_type == "ask_final_answer_or_next_question":
question, serialized_facts = input_text
template = self.prompt['ask_final_answer_or_next_question']
input_text = template.format(facts=serialized_facts, question=question)
elif turn_type == "choose_constraints":
question, relation_tails, tpe_name = input_text
template = self.prompt['choose_constraints']
input_text = template.format(question=question, relation_tails=relation_tails, tpe=tpe_name)
elif turn_type == "final_query_template":
template = self.prompt['final_query_template']
input_text = template.format(question=input_text)
elif turn_type == 'relation_rerank':
template = self.prompt['relation_rerank']
can_rels, question, tpe, selected_relations = input_text
# 暂时注释掉
# if len(selected_relations) > 1:
# selected_relations = "are " + ", ".join(selected_relations)
# else:
# selected_relations = "is " + ", ".join(selected_relations)
selected_relations = "".join(selected_relations)
input_text = template.format(question=question, relations=can_rels, tpe=tpe,
selected_relations=selected_relations)
elif turn_type == 'relation_rerank_2hop':
template = self.prompt['relation_rerank_2hop']
can_rels, question, tpe, sub_question, selected_relations = input_text
sub_question = ", ".join(sub_question)
selected_relations = ", ".join(selected_relations)
input_text = template.format(question=question, relations=can_rels, tpe=tpe,
first_sub_question=sub_question, first_relation=selected_relations)
elif turn_type == 'relation_rerank_3hop':
template = self.prompt['relation_rerank_3hop']
can_rels, question, tpe, sub_question, selected_relations = input_text
first_sub_question = sub_question[0]
second_sub_question = sub_question[1]
fisrt_relation = selected_relations[0]
second_relation = selected_relations[1]
input_text = template.format(question=question, relations=can_rels, tpe=tpe,
first_sub_question=first_sub_question, first_relation = fisrt_relation,
second_sub_question=second_sub_question, second_relation=second_relation)
elif turn_type == 'direct_ask_final_answer':
template = self.prompt['direct_ask_final_answer']
question = input_text
input_text = template.format(question=question)
elif turn_type == 'final_answer_organize':
template = self.prompt['final_answer_organize']
input_text = template
else:
raise NotImplementedError
message = {'role': 'user', 'content': input_text}
return message
class Retriever:
def __init__(self, args):
self.args = args
# self.initialize_PLM(args)
self.initialize_KG(args)
def get_retrieval_information(self, first_flag=False, gold_relations=None):
triples_per_hop, tails = self.KG.get_facts_1hop(self.cur_ents, self.args.max_triples_per_relation,
first_flag, gold_relations)
self.reset_cur_ents(tails)
# self.reset_last_ents(self.cur_ents)
return triples_per_hop
# 直接获得triples
def get_retrieval_information_direct(self, response, tpe, first_flag=False, gold_relations=None):
triples, tails = self.KG.get_facts_1hop_direct(response, tpe, self.cur_ents, self.tokenizer, self.retriever,
self.args.topk,
self.args.filter_score, self.args.max_triples_per_relation,
first_flag, gold_relations)
self.reset_cur_ents(tails)
# self.reset_last_ents(self.cur_ents)
return triples
def get_retrieval_relations(self, first_flag=False):
rels = self.KG.get_rels_1hop(self.cur_ents, first_flag)
return rels
def initialize_KG(self, args):
self.KG = KnowledgeGraph(args.kg_source_path, args.ent_type_path, args.ent2id_path, args.rel2id_path)
def initialize_PLM(self, args):
tokenizer = AutoTokenizer.from_pretrained(args.model_path)
self.tokenizer = tokenizer
model = AutoModel.from_pretrained(args.model_path)
# self.retriever = model.cuda("cuda:" + str(args.device))
self.retriever = None
def reset_cur_ents(self, entity_list):
self.cur_ents = entity_list
# print("Current entity num: ", len(self.cur_ents))
def update_cur_ents(self, filtered_triples_per_hop):
new_tails = set()
for tri in filtered_triples_per_hop[1]:
h, r, t = tri
try:
t_id = self.KG.ent2id[t]
new_tails.add(t_id)
except Exception as e:
logging.exception(e)
print("Entity string: %s not in ent2id dict" % t)
continue
new_tails = list(new_tails)
self.reset_cur_ents(new_tails)
def extract_facts(self, facts, response):
response = response.lower().strip()
# if response.startswith("the relevant relations:"):
# response = response.replace("the relevant relations:", "")
# response = response.strip()
# nor_rels = response.split(",")
# nor_rels = [rel.strip() for rel in nor_rels]
# else:
# nor_rels = response
filtered_facts = []
for tri in facts:
h, r, t = tri
if self.filter_relation(r):
continue
nor_r = self.normalize_relation(r)
if nor_r in response:
filtered_facts.append(tri)
return filtered_facts
def filter_relation(self, rel):
# same criteria as GraftNet
relation = rel
if relation == "common.topic.notable_types": return False
if relation == "base.kwebbase.kwtopic.has_sentences": return False
domain = relation.split(".")[0]
if domain == "type" or domain == "common": return True
return False
def should_ignore(self, rel):
if self.filter_relation(rel):
return True
return False
def normalize_relation(self, rel):
# e.g. <fb:film.film.other_crew>
rel_surface = rel
# replace '.' and '_' with ' '
rel_surface = rel_surface.replace('.', ' ')
# only keep the last two words
rel_surface = ' '.join(rel_surface.split(' ')[-2:])
rel_surface = rel_surface.replace('_', ' ')
return rel_surface
def get_one_hop_cand_rels(self, question):
pass
def get_tails_list(self, cur_ents):
tails = [self.KG.id2ent[ent] for ent in cur_ents]
return tails
class Solver:
def __init__(self, args):
self.args = args
self.LLM = ChatGPT(args=args, prompt_path=args.prompt_path, prompt_name=args.prompt_name,
max_tokens=args.max_tokens)
self.SLM = Retriever(args)
self.max_serialization_tokens = args.max_llm_input_tokens
self.load_ent2name(args.ent2name_path)
self.log = []
self.selected_relations = []
# 暂时添加一个selected_sub_questions = []来存放解析的子问题
self.selected_sub_questions = []
def forward_v2(self, question, tpe_str, tpe_id):
self.LLM.reset_history()
self.SLM.reset_cur_ents([tpe_id])
self.reset_history()
iterative_step = 0
# start_response = self.LLM.get_response_v2("", "instruction")
# self.log.append(start_response)
while True:
# select
all_rel_one_hop = self.SLM.get_retrieval_relations(first_flag=iterative_step == 0)
if len(all_rel_one_hop) == 0:
final_answers = self.LLM.get_response_v2(question, "final_query_template")
break
serialized_rels = self.extract_can_rels(all_rel_one_hop, normalize_rel=False)
if args.debug:
print("Step-%d: serialized_rels:%s" % (iterative_step, serialized_rels))
if iterative_step == 0:
llm_selected_rels = self.LLM.get_response_v2((serialized_rels, question, tpe_str),
"init_relation_rerank")
else:
llm_selected_rels = self.LLM.get_response_v2(
(serialized_rels, question, tpe_str, self.selected_relations), "relation_rerank")
self.LLM.reset_history_messages()
if args.debug:
print("Step-%d: llm_selected_rels:%s" % (iterative_step, llm_selected_rels))
selected_relations_list = self.parse_llm_selected_relations(llm_selected_rels, all_rel_one_hop)
if args.debug:
print("Step-%d: selected_relations_list:%s" % (iterative_step, selected_relations_list))
if len(selected_relations_list) == 0:
final_answers = self.LLM.get_response_v2(question, "final_query_template")
break
self.selected_relations.extend(selected_relations_list)
if args.debug:
print("Step-%d: self.selected_relations:%s" % (iterative_step, self.selected_relations))
filtered_triples_per_hop = self.SLM.get_retrieval_information(first_flag=iterative_step == 0,
gold_relations=selected_relations_list)
cvt_triples, mid_triples, entstr_triples = self.classify_triples(filtered_triples_per_hop)
if len(cvt_triples) > 0:
# constraint
if args.debug:
print("Step-%d: Constraints" % iterative_step)
constraints_candidate = self.serialize_constraints(cvt_triples)
if args.debug:
print("Step-%d: constraints_candidate:%s" % (iterative_step, constraints_candidate))
constraint_response = self.LLM.get_response_v2((question, constraints_candidate, tpe_str), "choose_constraints")
self.log.append(constraint_response)
if args.debug:
print("Step-%d: constraint_response:%s" % (iterative_step, constraint_response))
if self.has_constraints(constraint_response):
filtered_triples_per_hop = self.filter_triples(filtered_triples_per_hop, cvt_triples,
constraint_response)
self.SLM.update_cur_ents(filtered_triples_per_hop)
if args.debug:
print("Step-%d: filtered_triples_per_hop:%s" % (iterative_step, filtered_triples_per_hop))
if args.debug:
print("Step-%d: self.SLM.cur_ents:%s" % (iterative_step, self.SLM.cur_ents))
serialized_facts = self.serialize_facts(filtered_triples_per_hop)
self.log.append(serialized_facts)
if args.debug:
print("Step-%d: serialized_facts:%s" % (iterative_step, serialized_facts))
final_ans_or_next_que = self.LLM.get_response_v2((question, serialized_facts),
"ask_final_answer_or_next_question")
self.log.append(final_ans_or_next_que)
# 新加的
final_answers = self.parse_result(final_ans_or_next_que, "final_answer")
self.log.append(final_answers)
break
return final_answers, self.LLM.history_contents, self.log
def reset_selected_list(self):
self.selected_sub_questions = []
self.selected_relations = []
def is_end(self, response, iterative_step):
if "no" in response.lower() or iterative_step > 8:
return True
else:
return False
def load_ent2name(self, ent2name_path):
with open(ent2name_path, "rb") as f:
self.cvt_flag_dict, self.mid_mapping_dict = pickle.load(f)
def convert_hyper_facts_to_text(self, facts):
subj, rels, objs = facts
if self.is_cvt(subj):
return None
elif subj in self.mid_mapping_dict:
subj_surface = self.mid_mapping_dict[subj]
elif self.is_ent(subj):
# print("head entity %s doesn't have name, we skip this triple." % subj)
return None
else:
subj_surface = subj
flat_facts = []
for rel, obj in zip(rels, objs):
if self.should_ignore(rel):
continue
else:
nor_rel = self.normalize_relation(rel)
if self.is_cvt(obj):
continue
elif obj in self.mid_mapping_dict:
obj_surface = self.mid_mapping_dict[obj]
elif self.is_ent(obj):
# print("tail entity %s doesn't have name, we skip this triple." % obj)
continue
else:
obj_surface = obj
flat_facts.append((subj_surface, nor_rel, obj_surface))
return flat_facts
def convert_fact_to_text(self, fact, normalize_rel=False):
subj, rel, obj = fact
if self.should_ignore(rel):
return None
if rel.endswith(".from"):
rel = rel.rstrip(".from")
rel = rel + ".start_time"
if rel.endswith(".to"):
rel = rel.rstrip(".to")
rel = rel + ".end_time"
rel_surface = self.normalize_relation(rel) if normalize_rel else rel
# subject
if subj.startswith("CVT"):
subj_surface = subj
elif subj in self.mid_mapping_dict:
subj_surface = self.mid_mapping_dict[subj]
elif subj.startswith("m.") or subj.startswith('g.'):
# print("head entity %s doesn't have name, we skip this triple." % subj)
return None
else:
subj_surface = subj
# object
if obj.startswith("CVT"):
obj_surface = obj
elif obj in self.mid_mapping_dict:
obj_surface = self.mid_mapping_dict[obj]
elif obj.startswith("m.") or obj.startswith('g.'):
# print("tail entity %s doesn't have name, we skip this triple." % obj)
return None
else:
obj_surface = obj
return (subj_surface, rel_surface, obj_surface)
def extract_can_rels(self, all_rel_one_hop, normalize_rel=True):
rel_prompt = '"{relation}"'
nor_rels_set = []
for rel in all_rel_one_hop:
if self.filter_relation(rel):
continue
nor_r = self.normalize_relation(rel) if normalize_rel else rel
if nor_r not in nor_rels_set:
nor_rels_set.append(rel_prompt.format(relation=nor_r))
rel_candidate = ", ".join(nor_rels_set)
return rel_candidate
def serialize_rels(self, rels, normalize_rel=True):
nor_rels_set = []
for rel in rels:
if self.filter_relation(rel):
continue
nor_r = self.normalize_relation(rel) if normalize_rel else rel
if nor_r not in nor_rels_set:
nor_rels_set.append(nor_r)
# rel_candidate = ", ".join(nor_rels_set)
rel_candidate = ";\n ".join(nor_rels_set)
return rel_candidate
# 直接拼接
def serialize_facts_direct(self, facts):
# 拼接triples
facts_str_for_one_tail_ent = ["(" + ", ".join(fact) + ")" for fact in facts]
serialized_facts = ""
for fact in facts_str_for_one_tail_ent:
serialized_facts_tmp = serialized_facts + fact + "; "
serialized_facts = serialized_facts_tmp
return serialized_facts
def serialize_facts(self, facts_per_hop):
h_r_t = defaultdict(lambda: defaultdict(set))
visited_flag = {}
name2cvt_tmp = {}
cvt_count = 0
all_facts = []
for hop, facts in facts_per_hop.items():
if len(facts) > 0:
for fact in facts:
h, r, t = fact
if self.is_cvt(h):
if h not in name2cvt_tmp:
cvt = "CVT_" + str(cvt_count)
cvt_count += 1
name2cvt_tmp[h] = cvt
h = name2cvt_tmp[h]
if self.is_cvt(t):
if t not in name2cvt_tmp:
cvt = "CVT_" + str(cvt_count)
cvt_count += 1
name2cvt_tmp[t] = cvt
t = name2cvt_tmp[t]
fact = (h, r, t)
all_facts.append(fact)
visited_flag[fact] = False
h_r_t[h][r].add(t)
if len(all_facts) > 0:
all_facts_str = []
for tri in all_facts:
facts_str_for_one_tail_ent = []
if not visited_flag[tri]:
h, r, t = tri
if t.startswith("CVT") and len(h_r_t[t]) == 0:
continue
if h.startswith("CVT"):
# print("Qid:[%s] has single cvt head entities." % qid)
# logger.info(triples_per_hop)
continue
elif t.startswith("CVT"):
st = self.convert_fact_to_text(tri, normalize_rel=False)
facts_str_for_one_tail_ent.append(st)
one_hop_triples = h_r_t[t]
if len(one_hop_triples) > 0:
for key_r, value_ts in one_hop_triples.items():
for t_ in value_ts:
visit_tri = (t, key_r, t_)
if not visited_flag[visit_tri]:
visited_flag[visit_tri] = True
st = self.convert_fact_to_text(visit_tri, normalize_rel=False)
if st is not None:
assert len(st) == 3
facts_str_for_one_tail_ent.append(st)
# h_new = t
# r_new = []
# t_new = []
# for key_r, value_ts in one_hop_triples.items():
# for t_ in value_ts:
# visit_tri = (t, key_r, t_)
# if not visited_flag[visit_tri]:
# r_new.append(key_r)
# t_new.append(t_)
# visited_flag[visit_tri] = True
# tri_new = (t, r_new, t_new)
# if len(r_new) == len(t_new) > 0:
# str_tri_list = self.convert_hyper_facts_to_text(tri_new)
# if str_tri_list is not None:
# for st in str_tri_list:
# assert len(st) == 3
# if st not in facts_str:
# facts_str.append(st)
else:
st = self.convert_fact_to_text(tri, normalize_rel=False)
if st is not None:
assert len(st) == 3
if st not in facts_str_for_one_tail_ent:
facts_str_for_one_tail_ent.append(st)
facts_str_for_one_tail_ent = ["(" + ", ".join(fact) + ")" for fact in facts_str_for_one_tail_ent]
facts_str = ", ".join(facts_str_for_one_tail_ent)
all_facts_str.append(facts_str)
# facts_str = ["(" + ", ".join(fact) + ")" for fact in facts_str]
serialized_facts = ""
for fact in all_facts_str:
serialized_facts_tmp = serialized_facts + fact + "; "
if len(serialized_facts_tmp.split()) > self.max_serialization_tokens:
break
else:
serialized_facts = serialized_facts_tmp
serialized_facts = serialized_facts.strip("; ")
else:
serialized_facts = ""
return serialized_facts
def serialize_facts_v1(self, facts):
if len(facts) > 0:
h_r_t = defaultdict(lambda: defaultdict(set))
visited_flag = {}
for fact in facts:
h, r, t = fact
visited_flag[tuple(fact)] = False
h_r_t[h][r].add(t)
facts_str = []
for tri in facts:
if not visited_flag[tuple(tri)]:
h, r, t = tri
if self.is_cvt(t) and len(h_r_t[t]) == 0:
continue
if self.is_cvt(h):
# print("Qid:[%s] has single cvt head entities." % qid)
# logger.info(triples_per_hop)
continue
elif self.is_cvt(t):
one_hop_triples = h_r_t[t]
if len(one_hop_triples) > 0:
h_new = t
r_new = []
t_new = []
for key_r, value_ts in one_hop_triples.items():
for t_ in value_ts:
visit_tri = (t, key_r, t_)
if not visited_flag[visit_tri]:
r_new.append(key_r)
t_new.append(t_)
visited_flag[visit_tri] = True
tri_new = (h, r_new, t_new)
if len(r_new) == len(t_new) > 0:
str_tri_list = self.convert_hyper_facts_to_text(tri_new)
if str_tri_list is not None:
for st in str_tri_list:
assert len(st) == 3
if st not in facts_str:
facts_str.append(st)
else:
st = self.convert_fact_to_text(tri)
if st is not None:
assert len(st) == 3
if st not in facts_str:
facts_str.append(st)
facts_str = ["(" + ", ".join(fact) + ")" for fact in facts_str]
serialized_facts = ""
for fact in facts_str:
serialized_facts_tmp = serialized_facts + fact + "; "
if len(serialized_facts_tmp.split()) > self.max_serialization_tokens:
break
else:
serialized_facts = serialized_facts_tmp
# serialized_facts = "; ".join(facts_str)
serialized_facts = serialized_facts.strip("; ")
else:
serialized_facts = ""
return serialized_facts
def is_cvt(self, entity):
if self.cvt_flag_dict[entity]:
return True
else:
return False
def is_ent(self, ent_str):
if type(ent_str) is not bool and (ent_str.startswith("m.") or ent_str.startswith("g.")):
return True
else:
return False
def filter_relation(self, rel):
# same criteria as GraftNet
relation = rel
if relation == "common.topic.notable_types": return False
if relation == "base.kwebbase.kwtopic.has_sentences": return False
domain = relation.split(".")[0]
if domain == "type" or domain == "common": return True
return False
def should_ignore(self, rel):
if self.filter_relation(rel):
return True
return False
def normalize_relation(self, rel):
# e.g. <fb:film.film.other_crew>
rel_surface = rel
# replace '.' and '_' with ' '
rel_surface = rel_surface.replace('.', ' ')
# only keep the last two words
rel_surface = ' '.join(rel_surface.split(' ')[-2:])
rel_surface = rel_surface.replace('_', ' ')
return rel_surface
def parse_llm_selected_relations(self, llm_sel_rels_str, can_rels):
# llm_sel_rels = llm_sel_rels_str.strip(" ;.|,<>`[]'")
# llm_sel_rels = llm_sel_rels.split(',')
# llm_sel_rels = [rel.strip(" ;.|,<>`[]'").strip(" ;.|,<>`[]'") for rel in llm_sel_rels]
# llm_sel_rel_list = []
# for rel in llm_sel_rels:
# if rel in can_rels:
# llm_sel_rel_list.append(rel)
# else:
# print(rel)
# if len(llm_sel_rel_list) == 0:
# for rel in can_rels:
# if rel in llm_sel_rels_str:
# llm_sel_rel_list.append(rel)
# print("-----llm_ser_rels:\n%s\ndoesn't match the predefined format" % llm_sel_rels)
llm_sel_rel_list = []
for rel in can_rels:
if rel in llm_sel_rels_str:
llm_sel_rel_list.append(rel)
return llm_sel_rel_list
def parse_result(self, response, parse_type):
response = response.lower()
if parse_type == "next_question":
if "the next question:" in response:
next_question = response.split("the next question:")[1].strip()
elif ":" in response:
next_question = response.split(":")[1].strip()
else:
next_question = response
print("Not parse the next question exactly, directly use the response: ", response)
return next_question
elif parse_type == "final_answer":
if "the final answers:" in response:
final_answer = response.split("the final answers:")[1].strip()
# 暂时注释掉
elif ":" in response:
final_answer = response.split(":")[1].strip()
# 新添加的用于解析direct query
else:
final_answer = response
# 暂时注释掉
# print("Not parse the final answer exactly, directly use the response: ", response)
return final_answer
def classify_triples(self, filtered_triples_per_hop):
cvt_triples, mid_triples, entstr_triples = set(), set(), set()
if 0 in filtered_triples_per_hop:
triples_0 = filtered_triples_per_hop[0]
else:
triples_0 = []
if 1 in filtered_triples_per_hop:
triples_1 = filtered_triples_per_hop[1]
else:
triples_1 = []
if len(triples_1) == 0:
for tri in triples_0:
if self.is_ent(tri[2]):
mid_triples.add(tuple(tri))
else:
entstr_triples.add(tuple(tri))
else:
for tri in triples_1:
cvt_triples.add(tuple(tri))
return cvt_triples, mid_triples, entstr_triples
def serialize_constraints(self, cvt_triples):
r2t_set = defaultdict(set)
for tri in cvt_triples:
subj, rel, obj = tri
if self.should_ignore(rel):
continue
if rel.endswith(".from"):
rel = rel.rstrip(".from")
rel = rel + ".start_time"
if rel.endswith(".to"):
rel = rel.rstrip(".to")
rel = rel + ".end_time"
rel_surface = rel
# object
if obj in self.mid_mapping_dict:
obj_surface = self.mid_mapping_dict[obj]
elif obj.startswith("m.") or obj.startswith('g.'):
# print("tail entity %s doesn't have name, we skip this triple." % obj)
continue
else:
obj_surface = obj
if obj_surface == "To" or "has_no_value" in rel:
continue
r2t_set[rel_surface].add(obj_surface)
constraints = []
for r, t_set in r2t_set.items():
t_set = ['"' + t + '"' for t in t_set]
constraints.append('"' + r + '"' + ": [" + ", ".join(t_set) + "]")
# constraints = constraints.rstrip("\n")
constraints = "\n".join(constraints)
return constraints
def has_constraints(self, constraint_response):
if "no" in constraint_response.lower():
return False
else:
return True
def filter_triples(self, filtered_triples_per_hop, cvt_triples, constraint_response):
valid_cvt_nodes = set()
h_r_t = defaultdict(list)
for tri in cvt_triples:
h, r, t = tri
h_r_t[h].append((r, t))
for cvt, r_ts in h_r_t.items():
flag = True
at_leat_one_flag = False
for r_t in r_ts:
rel, obj = r_t
if rel.endswith(".from"):
rel = rel.rstrip(".from")
rel = rel + ".start_time"
if rel.endswith(".to"):
rel = rel.rstrip(".to")
rel = rel + ".end_time"
rel_surface = rel
# object
if obj in self.mid_mapping_dict:
obj_surface = self.mid_mapping_dict[obj]
elif obj.startswith("m.") or obj.startswith('g.'):
# print("tail entity %s doesn't have name, we skip this triple." % obj)
continue
else:
obj_surface = obj
if rel_surface.lower() in constraint_response.lower():
at_leat_one_flag = True
if obj_surface.lower() not in constraint_response.lower():
flag = False
break
if flag and at_leat_one_flag:
valid_cvt_nodes.add(cvt)
# 添加软约束条件，解析cvt结点的rel，若有两部分在response中则选中
if len(valid_cvt_nodes) == 0:
for cvt, r_ts in h_r_t.items():
flag = True
at_leat_one_flag = False
for r_t in r_ts:
rel, obj = r_t
if rel.endswith(".from"):
rel = rel.rstrip(".from")
rel = rel + ".start_time"
if rel.endswith(".to"):
rel = rel.rstrip(".to")
rel = rel + ".end_time"
rel_surface = rel
# object
if obj in self.mid_mapping_dict:
obj_surface = self.mid_mapping_dict[obj]
elif obj.startswith("m.") or obj.startswith('g.'):
# print("tail entity %s doesn't have name, we skip this triple." % obj)
continue
else:
obj_surface = obj
rel_surface_list = rel_surface.split(".")
for rel in rel_surface_list:
if rel.lower() in constraint_response.lower():
at_leat_one_flag = True
if obj_surface.lower() not in constraint_response.lower():
flag = False
break
else:
flag = True
if flag and at_leat_one_flag:
valid_cvt_nodes.add(cvt)
break