Create process.py
Browse files- process.py +149 -0
process.py
ADDED
@@ -0,0 +1,149 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
import argparse
|
3 |
+
import sys
|
4 |
+
from collections import defaultdict
|
5 |
+
from transformers import AutoTokenizer
|
6 |
+
|
7 |
+
|
8 |
+
def read_conjunctive_sentences(args):
|
9 |
+
with open(args.conjunctions_file, 'r') as fin:
|
10 |
+
sent = True
|
11 |
+
sent2conj = defaultdict(list)
|
12 |
+
conj2sent = dict()
|
13 |
+
currentSentText = ''
|
14 |
+
for line in fin:
|
15 |
+
if line == '\n':
|
16 |
+
sent = True
|
17 |
+
continue
|
18 |
+
if sent:
|
19 |
+
currentSentText = line.replace('\n', '')
|
20 |
+
sent = False
|
21 |
+
else:
|
22 |
+
conj_sent = line.replace('\n', '')
|
23 |
+
sent2conj[currentSentText].append(conj_sent)
|
24 |
+
conj2sent[conj_sent] = currentSentText
|
25 |
+
|
26 |
+
return sent2conj
|
27 |
+
|
28 |
+
|
29 |
+
def get_conj_free_sentence_dicts(sentence, sent_to_conj, sent_id):
|
30 |
+
flat_extractions_list = []
|
31 |
+
sentence = sentence.replace('\n', '')
|
32 |
+
if sentence in list(sent_to_conj.keys()):
|
33 |
+
for s in sent_to_conj[sentence]:
|
34 |
+
sentence_and_extractions_dict = {
|
35 |
+
"sentence": s + " [unused1] [unused2] [unused3] [unused4] [unused5] [unused6]",
|
36 |
+
"sentId": sent_id, "entityMentions": list(),
|
37 |
+
"relationMentions": list(), "extractionMentions": list()}
|
38 |
+
flat_extractions_list.append(sentence_and_extractions_dict)
|
39 |
+
return flat_extractions_list
|
40 |
+
|
41 |
+
return [{
|
42 |
+
"sentence": sentence + " [unused1] [unused2] [unused3] [unused4] [unused5] [unused6]",
|
43 |
+
"sentId": sent_id, "entityMentions": list(),
|
44 |
+
"relationMentions": list(), "extractionMentions": list()}]
|
45 |
+
|
46 |
+
|
47 |
+
def add_joint_label(ext, ent_rel_id):
|
48 |
+
"""add_joint_label add joint labels for sentences
|
49 |
+
"""
|
50 |
+
|
51 |
+
none_id = ent_rel_id['None']
|
52 |
+
sentence_length = len(ext['sentText'].split(' '))
|
53 |
+
entity_label_matrix = [[none_id for j in range(sentence_length)] for i in range(sentence_length)]
|
54 |
+
relation_label_matrix = [[none_id for j in range(sentence_length)] for i in range(sentence_length)]
|
55 |
+
label_matrix = [[none_id for j in range(sentence_length)] for i in range(sentence_length)]
|
56 |
+
ent2offset = {}
|
57 |
+
for ent in ext['entityMentions']:
|
58 |
+
ent2offset[ent['emId']] = ent['span_ids']
|
59 |
+
try:
|
60 |
+
for i in ent['span_ids']:
|
61 |
+
for j in ent['span_ids']:
|
62 |
+
entity_label_matrix[i][j] = ent_rel_id[ent['label']]
|
63 |
+
except:
|
64 |
+
print("span ids: ", sentence_length, ent['span_ids'], ext)
|
65 |
+
sys.exit(1)
|
66 |
+
ext['entityLabelMatrix'] = entity_label_matrix
|
67 |
+
for rel in ext['relationMentions']:
|
68 |
+
arg1_span = ent2offset[rel['arg1']['emId']]
|
69 |
+
arg2_span = ent2offset[rel['arg2']['emId']]
|
70 |
+
|
71 |
+
for i in arg1_span:
|
72 |
+
for j in arg2_span:
|
73 |
+
# to be consistent with the linking model
|
74 |
+
relation_label_matrix[i][j] = ent_rel_id[rel['label']] - 2
|
75 |
+
relation_label_matrix[j][i] = ent_rel_id[rel['label']] - 2
|
76 |
+
label_matrix[i][j] = ent_rel_id[rel['label']]
|
77 |
+
label_matrix[j][i] = ent_rel_id[rel['label']]
|
78 |
+
ext['relationLabelMatrix'] = relation_label_matrix
|
79 |
+
ext['jointLabelMatrix'] = label_matrix
|
80 |
+
|
81 |
+
|
82 |
+
def tokenize_sentences(ext, tokenizer):
|
83 |
+
cls = tokenizer.cls_token
|
84 |
+
sep = tokenizer.sep_token
|
85 |
+
wordpiece_tokens = [cls]
|
86 |
+
|
87 |
+
wordpiece_tokens_index = []
|
88 |
+
cur_index = len(wordpiece_tokens)
|
89 |
+
# for token in ext['sentText'].split(' '):
|
90 |
+
for token in ext['sentence'].split(' '):
|
91 |
+
tokenized_token = list(tokenizer.tokenize(token))
|
92 |
+
wordpiece_tokens.extend(tokenized_token)
|
93 |
+
wordpiece_tokens_index.append([cur_index, cur_index + len(tokenized_token)])
|
94 |
+
cur_index += len(tokenized_token)
|
95 |
+
wordpiece_tokens.append(sep)
|
96 |
+
|
97 |
+
wordpiece_segment_ids = [1] * (len(wordpiece_tokens))
|
98 |
+
return {
|
99 |
+
'sentId': ext['sentId'],
|
100 |
+
'sentText': ext['sentence'],
|
101 |
+
'entityMentions': ext['entityMentions'],
|
102 |
+
'relationMentions': ext['relationMentions'],
|
103 |
+
'extractionMentions': ext['extractionMentions'],
|
104 |
+
'wordpieceSentText': " ".join(wordpiece_tokens),
|
105 |
+
'wordpieceTokensIndex': wordpiece_tokens_index,
|
106 |
+
'wordpieceSegmentIds': wordpiece_segment_ids
|
107 |
+
}
|
108 |
+
|
109 |
+
|
110 |
+
def write_dataset_to_file(dataset, dataset_path):
|
111 |
+
print("dataset: {}, size: {}".format(dataset_path, len(dataset)))
|
112 |
+
with open(dataset_path, 'w', encoding='utf-8') as fout:
|
113 |
+
for idx, ext in enumerate(dataset):
|
114 |
+
fout.write(json.dumps(ext))
|
115 |
+
if idx != len(dataset) - 1:
|
116 |
+
fout.write('\n')
|
117 |
+
|
118 |
+
|
119 |
+
def process(args, sent2conj):
|
120 |
+
extractions_list = []
|
121 |
+
auto_tokenizer = AutoTokenizer.from_pretrained(args.embedding_model)
|
122 |
+
print("Load {} tokenizer successfully.".format(args.embedding_model))
|
123 |
+
|
124 |
+
ent_rel_id = json.load(open(args.ent_rel_file, 'r', encoding='utf-8'))["id"]
|
125 |
+
sentId = 0
|
126 |
+
with open(args.source_file, 'r', encoding='utf-8') as fin, open(args.target_file, 'w', encoding='utf-8') as fout:
|
127 |
+
for line in fin:
|
128 |
+
sentId += 1
|
129 |
+
exts = get_conj_free_sentence_dicts(line, sent2conj, sentId)
|
130 |
+
for ext in exts:
|
131 |
+
# ext = ext.strip()
|
132 |
+
ext_dict = tokenize_sentences(ext, auto_tokenizer)
|
133 |
+
add_joint_label(ext_dict, ent_rel_id)
|
134 |
+
extractions_list.append(ext_dict)
|
135 |
+
fout.write(json.dumps(ext_dict))
|
136 |
+
fout.write('\n')
|
137 |
+
|
138 |
+
|
139 |
+
if __name__ == '__main__':
|
140 |
+
parser = argparse.ArgumentParser(description='Process sentences.')
|
141 |
+
parser.add_argument("--source_file", type=str, help='source file path')
|
142 |
+
parser.add_argument("--target_file", type=str, help='target file path')
|
143 |
+
parser.add_argument("--conjunctions_file", type=str, help='conjunctions file.')
|
144 |
+
parser.add_argument("--ent_rel_file", type=str, default="ent_rel_file.json", help='entity and relation file.')
|
145 |
+
parser.add_argument("--embedding_model", type=str, default="bert-base-uncased", help='embedding model.')
|
146 |
+
|
147 |
+
args = parser.parse_args()
|
148 |
+
sent2conj = read_conjunctive_sentences(args)
|
149 |
+
process(args, sent2conj)
|