-
Notifications
You must be signed in to change notification settings - Fork 0
/
data_disambiguation.py
222 lines (174 loc) · 9.54 KB
/
data_disambiguation.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
import random
import json
import re
import torch
from torch.utils.data import Dataset, DataLoader
class AttentionDataset(Dataset):
def __init__(self, mentions, kb, tokenizer, args, istest):
super(AttentionDataset, self).__init__()
self.mentions = mentions
self.kb = kb
self.all_candidates = list(kb.keys())
self.tokenizer = tokenizer
self.info_token_num = args.info_token_num
self.cand_num = args.cand_num
self.max_ent_len = args.max_ent_len - 1 - args.info_token_num
self.max_text_len = args.max_text_len
self.max_length = args.max_len
self.or_token = "[text]"
self.istest = istest
def __len__(self):
return len(self.mentions)
def __getitem__(self, index):
if self.istest:
return self.eval_dataset(index)
else:
return self.train_dataset(index)
def pad_values(self, tokens, value, max_len):
return (tokens + [value] * max_len)[:max_len]
def train_dataset(self, index):
data = self.mentions[index]
info_token = [f"[info{i}]" for i in range(self.info_token_num)]
text = data["text"]
mention_data = data["mention_data"]
splited_text = text.split(" ")
mention_start, mention_end = splited_text.index("[E1]"), splited_text.index("[\E1]")
mention = " ".join(splited_text[mention_start + 1:mention_end])
mention_token = self.tokenizer.tokenize(mention)
kb_id = mention_data["kb_id"]
candidates = mention_data["candidates"][:self.cand_num]
if kb_id not in candidates:
candidates = candidates[:self.cand_num - 1] + [kb_id]
if len(candidates) < self.cand_num:
sim_neg = random.sample(list(self.all_candidates), k=self.cand_num - len(candidates))
candidates += sim_neg
assert len(candidates) == self.cand_num
random.shuffle(candidates)
labels = [kb_id == candidate for candidate in candidates]
assert sum(labels) > 0
max_half_text = (self.max_text_len - len(mention_token)) // 2 - 1
text_tokens = self.tokenizer.tokenize(text)
# pattern_tokens = self.tokenizer.tokenize(pattern)
men_start = text_tokens.index("[E1]")
men_end = text_tokens.index("[\E1]")
text_tokens = text_tokens[max(0, men_start - max_half_text):men_end + max_half_text][:self.max_text_len - 2]
text_tokens = [self.tokenizer.cls_token] + text_tokens + [self.tokenizer.sep_token]
text_input_ids = self.tokenizer.convert_tokens_to_ids(text_tokens)
mention_pos = [0, text_tokens.index("[E1]"), text_tokens.index("[\E1]")]
text_input_ids = self.pad_values(text_input_ids, self.tokenizer.pad_token_id, self.max_text_len)
text_attention_mask = [1] * len(text_tokens)
text_attention_mask = self.pad_values(text_attention_mask, 0, self.max_text_len)
assert self.tokenizer.sep_token_id in text_input_ids
candiates_input_ids = []
candidates_attention_masks = []
for i in range(len(labels)):
entity_id = candidates[i]
entity_names = self.kb[entity_id]
entity_text = remove_punctuation(entity_names["text"])
entity_text = " ".join((entity_text.split(" "))[:self.max_ent_len])
entity_text_tokens = self.tokenizer.tokenize(entity_text)
entity_name_tokens = info_token + entity_text_tokens + [self.tokenizer.sep_token] + \
text_tokens[1:]
entity_name_ids = self.tokenizer.convert_tokens_to_ids(entity_name_tokens)
entity_attention_mask = [1] * len(entity_name_tokens)
entity_name_ids = self.pad_values(entity_name_ids, self.tokenizer.pad_token_id, self.max_length)
entity_attention_mask = self.pad_values(entity_attention_mask, 0, self.max_length)
candiates_input_ids.append(entity_name_ids)
candidates_attention_masks.append(entity_attention_mask)
labels = [[i] * self.info_token_num for i in labels]
text_input_ids = torch.tensor(text_input_ids).long()
text_attention_mask = torch.tensor(text_attention_mask).long()
candidates_input_ids = torch.tensor(candiates_input_ids).long()
candidates_attention_masks = torch.tensor(candidates_attention_masks).long()
mention_pos = torch.tensor(mention_pos).long()
labels = torch.tensor(labels).view(-1).float()
return text_input_ids, text_attention_mask, candidates_input_ids, candidates_attention_masks, \
mention_pos, labels
def eval_dataset(self, index):
data = self.mentions[index]
info_token = [f"[info{i}]" for i in range(self.info_token_num)]
text = data["text"]
mention_data = data["mention_data"]
kb_id = mention_data["kb_id"]
splited_text = text.split(" ")
mention_start, mention_end = splited_text.index("[E1]"), splited_text.index("[\E1]")
mention = " ".join(splited_text[mention_start + 1:mention_end])
mention_token = self.tokenizer.tokenize(mention)
candidates = mention_data["candidates"]
if not candidates:
candidates = random.sample(self.all_candidates, k=self.cand_num)
labels = [candidate == kb_id for candidate in candidates]
assert sum(labels) > 0
max_half_text = (self.max_text_len - len(mention_token)) // 2 - 1
text_tokens = self.tokenizer.tokenize(text)
# pattern_tokens = self.tokenizer.tokenize(pattern)
men_start = text_tokens.index("[E1]")
men_end = text_tokens.index("[\E1]")
text_tokens = text_tokens[max(0, men_start - max_half_text):men_end + max_half_text][:self.max_text_len - 2]
text_tokens = [self.tokenizer.cls_token] + text_tokens + [self.tokenizer.sep_token]
text_input_ids = self.tokenizer.convert_tokens_to_ids(text_tokens)
mention_pos = [0, text_tokens.index("[E1]"), text_tokens.index("[\E1]")]
text_input_ids = self.pad_values(text_input_ids, self.tokenizer.pad_token_id, self.max_text_len)
text_attention_mask = [1] * len(text_tokens)
text_attention_mask = self.pad_values(text_attention_mask, 0, self.max_text_len)
assert self.tokenizer.sep_token_id in text_input_ids
candiates_input_ids = []
candidates_attention_masks = []
for i in range(len(labels)):
entity_id = candidates[i]
entity_names = self.kb[entity_id]
title = entity_names["title"]
entity_text = remove_punctuation(entity_names["text"])
entity_text = " ".join(entity_text.split(" ")[:self.max_ent_len])
entity_text_tokens = self.tokenizer.tokenize(entity_text)
entity_name_tokens = info_token + entity_text_tokens + [self.tokenizer.sep_token] + \
text_tokens[1:]
entity_name_ids = self.tokenizer.convert_tokens_to_ids(entity_name_tokens)
entity_attention_mask = [1] * len(entity_name_tokens)
entity_name_ids = self.pad_values(entity_name_ids, self.tokenizer.pad_token_id, self.max_length)
entity_attention_mask = self.pad_values(entity_attention_mask, 0, self.max_length)
candiates_input_ids.append(entity_name_ids)
candidates_attention_masks.append(entity_attention_mask)
labels = [[i] * self.info_token_num for i in labels]
text_input_ids = torch.tensor(text_input_ids).long()
text_attention_mask = torch.tensor(text_attention_mask).long()
candiates_input_ids = torch.tensor(candiates_input_ids).long()
candidates_attention_masks = torch.tensor(candidates_attention_masks).long()
mention_pos = torch.tensor(mention_pos).long()
labels = torch.tensor(labels).view(-1).float()
return text_input_ids, text_attention_mask, candiates_input_ids, candidates_attention_masks, \
mention_pos, labels
def generate_samples(batch):
input_ids, attention_masks, labels = [], [], []
for b in batch:
input_ids += b["input_ids"]
attention_masks += b["attention_masks"]
labels += b["labels"]
input_ids = torch.tensor(input_ids).long()
attention_masks = torch.tensor(attention_masks).long()
labels = torch.tensor(labels).float()
return input_ids, attention_masks, labels
def load_data(data_path):
with open(data_path, encoding="utf-8") as f:
data = [json.loads(line) for line in f]
return data
def load_entities(data_path):
with open(data_path, encoding="utf-8") as f:
data = json.loads(f.read())
return data
def make_single_loader(data_set, bsz, shuffle, coll_fn=None):
if coll_fn is not None:
loader = DataLoader(data_set, bsz, shuffle=shuffle, collate_fn=coll_fn)
else:
loader = DataLoader(data_set, bsz, shuffle=shuffle)
return loader
def remove_punctuation(sentence):
remove_chars = '[’!"#$%&\'()*+,-.:;<=>?@,。?★、…【】《》?“”‘’!\\^_`{|}~]+'
result = re.sub(remove_chars, ' ', sentence)
result = ' '.join(result.split())
return result
def get_attention_mention_loader(samples, kb, tokenizer, shuffle, is_test, args):
samples_set = AttentionDataset(samples, kb, tokenizer, args, is_test)
if is_test:
return make_single_loader(samples_set, 1, False)
return make_single_loader(samples_set, args.batch, shuffle)