-
Notifications
You must be signed in to change notification settings - Fork 29
/
preprocess.py
517 lines (448 loc) · 20.7 KB
/
preprocess.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
import csv
import re
import os
from typing import List
from tqdm import tqdm
import logging
import pandas as pd
import torch
import copy
import numpy as np
import torch
raw_data_path = './'
# 合并相似的title 和 text
def concat(title, content):
if type(title) is float or str(title) in str(content):
return str(content)
else:
return str(title) + " " + str(content)
def get_jaccard_similar(title, text):
if type(title) is float:
return 0
if type(text) is float:
text = str(text)
char_list_1 = []
char_list_2 = []
for char in title:
char_list_1.append(char)
for char in text:
char_list_2.append(char)
char_list_both = [char for char in char_list_1 if char in char_list_2]
return len(char_list_both) / (len(char_list_1) + len(char_list_2) - len(char_list_both))
# return len(char_list_1) + 0.0001 / len(char_list_both) + 0.0001
# return len(char_list_both) / (len(char_list_1) if len(char_list_1) < len(char_list_2) else len(char_list_2))
def process_title(title, text, similar):
if title == text or similar > 0.75:
return ""
else:
return title
def cos_similar(vector1, vector2):
dot_product = 0.0
normA = 0.0
normB = 0.0
for a, b in zip(vector1, vector2):
dot_product += a * b
normA += a ** 2
normB += b ** 2
if normA == 0.0 or normB == 0.0:
return 0
else:
return round(dot_product / ((normA ** 0.5) * (normB ** 0.5)), 2)
train_data = pd.read_csv(os.path.join(raw_data_path,'Round2_train.csv'), encoding='utf-8')
test_data = pd.read_csv(os.path.join(raw_data_path,'round2_test.csv'), encoding='utf-8')
train_data["similar"] = train_data.apply(lambda x: get_jaccard_similar(x["title"], x["text"]), axis=1)
test_data["similar"] = test_data.apply(lambda x: get_jaccard_similar(x["title"], x["text"]), axis=1)
train_data["title"] = train_data.apply(lambda x: process_title(x["title"], x["text"], x["similar"]), axis=1)
# train_data["text"] = train_data.apply(lambda x: process_text(x["title"], x["text"], x["similar"]), axis=1)
test_data["title"] = test_data.apply(lambda x: process_title(x["title"], x["text"], x["similar"]), axis=1)
# test_data["text"] = test_data.apply(lambda x: process_text(x["title"], x["text"], x["similar"]), axis=1)
del train_data["similar"]
del test_data["similar"]
train_data["text"] = train_data.apply(lambda x: concat(x['title'], x['text']), axis=1)
test_data["text"] = test_data.apply(lambda x: concat(x['title'], x['text']), axis=1)
train_data["title"] = ""
test_data["title"] = ""
train_data.to_csv(os.path.join(raw_data_path,'Train_Data_round2.csv'), encoding='utf-8', index=False)
test_data.to_csv(os.path.join(raw_data_path,'Test_Data_round2.csv'), encoding='utf-8', index=False)
# 更改路径
train_data = pd.read_csv(os.path.join(raw_data_path,'Train_Data_round2.csv'))
test_data = pd.read_csv(os.path.join(raw_data_path,'Test_Data_round2.csv'))
test_entity_null_id = test_data[test_data['entity'].isnull()]['id']
test_data['entity'] = test_data['entity'].fillna(' ')
test_data['title'] = test_data['title'].fillna('')
test_data['text'] = test_data['text'].fillna('')
test_data['text']= test_data.apply(lambda x:x['title']+' '+x['text']if x['title']!=x['text'] else x['text'],axis=1)
test_data['text'] = test_data.apply(lambda x:x['text'].strip(),axis=1)
train_data['entity'] = train_data['entity'].fillna(' ')
train_data['title'] = train_data['title'].fillna('')
train_data['text'] = train_data.apply(lambda x: x['title']+' '+x['text'] if x['title'] != x['text'] else x['text'],axis=1)
train_data['text'] = train_data.apply(lambda x:x['text'].strip(),axis=1)
def process_chaohua_and_jing(text: str, entity_list: List[str]) -> str:
new_entity_list = []
for i in entity_list:
temp = i.strip()
temp = re.sub('\?*','',temp)
new_entity_list.append(temp)
entity_list = new_entity_list
pattern = "#[^#]*#"
final_text = []
while True:
res = re.search(pattern, text)
if res is None:
final_text.append(text)
break
chunk_with_ent = res.group()
curr_begin_idx, curr_end_idx = res.span()
text_part1 = text[0: curr_begin_idx]
text_part2 = text[curr_end_idx:]
flag = True
for ent in entity_list:
if ent in chunk_with_ent:
flag = False
break
final_text.append(text_part1)
final_text.append("" if flag else chunk_with_ent)
text = text_part2
#final_text = re.sub('#|\[超话\]', "", "".join(final_text))
final_text = re.sub('\[超话\]', "", "".join(final_text))
return final_text
def process_at(text: str, entity_list: List[str]) -> str:
pattern = "@[^(@|：| |,|，|。)]*( |，|,|：|:|。)|(//|回复)@[^(@|:)]*:"
final_text = []
while True:
res = re.search(pattern, text)
if res is None:
final_text.append(text)
break
chunk_with_ent = res.group()
curr_begin_idx, curr_end_idx = res.span()
text_part1 = text[0: curr_begin_idx]
text_part2 = text[curr_end_idx:]
flag = True
for ent in entity_list:
if ent in chunk_with_ent:
flag = False
break
final_text.append(text_part1)
final_text.append("" if flag else chunk_with_ent)
text = text_part2
final_text = re.sub('@|//', "", "".join(final_text))
return final_text
def process_laiyuan(text: str, entity_list: List[str]) -> str:
pattern = "(文章|本文)?来源[：\:]\s?[^(，|。|\?|\||：|/||\s)]*[，。\?：/\|\s]"
final_text = []
while True:
res = re.search(pattern, text)
if res is None:
final_text.append(text)
break
chunk_with_ent = res.group()
curr_begin_idx, curr_end_idx = res.span()
text_part1 = text[0: curr_begin_idx]
text_part2 = text[curr_end_idx:]
flag = True
for ent in entity_list:
if ent in chunk_with_ent:
flag = False
break
final_text.append(text_part1)
final_text.append("" if flag else chunk_with_ent)
text = text_part2
return "".join(final_text)
def eliminate_special_str(text: str, entity_list: List[str]) -> str:
# 去除特殊字符
regex1 = "①|②|③|④|⑤|⑥|⑦|⑧|⑨|⑩"
text = re.sub(regex1, "", text)
regex1 = "▽|▼|▲|■|█|▎|▌|★|►|▶|▼|—|\*|↑|→|°|·|ˇ|¤|é|§|●|…|☆|─|↓|�|✬|◆|▍|△|√|〖|〗|⊙|■|◆"
text = re.sub(regex1, " ", text)
# 去除微博@的数据
# text = process_at(text, entity_list)
# 处理##与超话
text = re.sub('\[超话\]', " ", text)
text = process_chaohua_and_jing(text, entity_list)
# 去除js代码
regex2 = "[a-z0-9'():/?\"=_\->]*[a-z'():/?\"=_\->][.:;=\"]+[a-z'():/?\"=_\->][a-z*0-9*'():/._?\"=>;\"]*"
text = re.sub(regex2, " ", text)
# 去掉js function的代码
regex2 = "\(function\(\)\{.*\}?\(?\)?"
text = re.sub(regex2, " ", text)
# 去掉html标签
regex2 = "<[^>]*>"
text = re.sub(regex2, " ", text)
# 去掉url
regex2 = "(http|ftp|https):\/\/[\w\-_]+(\.[\w\-_]+)+([\w\-\.,@?^=%&:/~\+#]*[\w\-\@?^=%&/~\+#])?"
text = re.sub(regex2, " ", text)
# 去除 IMG
regex3 = "\{IMG[^\}]*\}"
text = re.sub(regex3, " ", text)
# 去除记者 文章来源 责任编辑
regex4 = "（[^（^）]*(记者|微信号|作者|通讯员)：?[^）^（]*）|【?编者按】?|新华社成都6月24日电"
text = re.sub(regex4, " ", text)
# 去除ID 以及编号
regex5 = "（ID[^）]*）|\[[a-zA-Z0-9]*\]"
text = re.sub(regex5, " ", text)
# 去除 [doge] [cp] 等
regex5 = "\[[a-zA-Z0-9]*\]"
text5 = re.sub(regex5, " ", text)
# 去除 点击上方蓝字
regex6 = "【字号[^】]*】|点击上方蓝字|点击上方|【图】|回复使用道具举报|查看更多 |当前位置[：\:]|全文："
text = re.sub(regex6, " ", text)
regex6 = "收藏本站|收藏\([^\)]*\)|[0-9]*收藏|点击排行 |评论（[^）]*）|分享到腾讯微博|分享到新浪微博|赞[0-9]*|看全文 |(详细内容请)?点击："
text = re.sub(regex6, " ", text)
regex6 = "下载附件保存到相册|上传赞赏|支持([0-9]*人打赏)?分享|微信图片_[0-9]*|IMG_[0-9]*"
text = re.sub(regex6, " ", text)
regex6 = "资讯 |(网络)?电影 |电视剧 |综艺 |VIP(会员)? |首页 |导航 |娱乐 |片花 |脱口秀 |动漫 |游戏(视频|中心)? |搞笑 |微信 "
text = re.sub(regex6, " ", text)
regex6 = "体育 |教育 |儿童 |母婴 |生活 |健康 |军事 |汽车 |公益 |纪录片 |文学 |漫画 |热点 |风云榜 |全网影视 |应用商店 "
text = re.sub(regex6, " ", text)
regex6 = "大头 |爱奇艺号 |泡泡广场 |会员精选 |VR |泡泡 |旅游 |音乐 |时尚 |原创 |拍客 |科技 |奇秀直播 |直播中心 |商城 "
text = re.sub(regex6, " ", text)
regex6 = "网友评论 |最新评论 |暂无评论 |热度排行 |评论排行 |推荐 视频"
text = re.sub(regex6, " ", text)
regex7 = "\([^\(]*公众号：[^\)]*\)|(:\d*上传)?赞赏支持(\d*人)?(打赏)?分享:赞\d*\|收藏\(\d*\)"
text = re.sub(regex7, " ", text)
regex7 = "QQ空间 |微信 |朋友圈 |扫描二维码关注|[0-9]*(个回答|人关注)|主页 |企业动态 |正文 |舆情监测 |登录 |注册 |下载 |获取更多机会 |(顶|踩)一下 "
text = re.sub(regex7, " ", text)
regex7 = "\.(png|jpg)\s?\([^\)]*\)"
text = re.sub(regex7, " ", text)
text = process_laiyuan(text, entity_list)
regex7 = "[\(（【\[][^(\(|（|【|\[)]*(编辑)：[^(\)|）)]*[\)）】\]]"
text = re.sub(regex7, " ", text)
regex7 = "(责任)?编辑[：\:]\s?[^(，|。|\?|\||：|/|\s)]*[，。\?：/\|\s]"
text = re.sub(regex7, " ", text)
# 去掉单独的数字,以及带括号的数字(124) （12435）
regex_number = "[0-9]+ |[\(（\{][0-9a-zA-Z]+[\)）\}]"
text = re.sub(regex_number, " ", text)
# 去掉超长字符串
regex_super_long = "[a-zA-Z0-9]{50,}"
text = re.sub(regex_super_long, " ", text)
# 去除股票代码
regex8 = "[\(（][0-9]*[\.\-][a-zA-Z]*[\)）]|[\(（][a-zA-Z]*[:：][a-zA-Z]*[\)）]"
text = re.sub(regex8, " ", text)
# 去除  "
regex9 = " |>|&ldquo|&lsquo|&rsquo|""
text = re.sub(regex9, " ", text)
# 去除 去除两个（时间）的情况
regex11 = "\(?\d*/\d*/\d*\)?|\(?\d*-\d*-\d*\)?|\(?\d*\.\d*\.\d*\)?|\(?\d*年\d*月\d*日\)?|\(?\d*:\d*:\d*\)?"
text = re.sub(regex11, " ", text)
#去除 空格
regex10 = " |" + " " + "| " + "| " + "|\s+|\s+"
text = re.sub(regex10, " ", text)
return text
# 生成预处理后的文件
# 修改路径
for index, item in train_data.iterrows():
text = item['text']
entity_list = item['entity'].split(';')
processed_text = eliminate_special_str(text,entity_list)
train_data.iloc[index,2] = processed_text
train_data.to_csv(os.path.join(raw_data_path,'Train_Data_Title_processed_final.csv'),index=False)
for index, item in test_data.iterrows():
text = item['text']
entity_list = item['entity'].split(';')
test_data.iloc[index,2] = processed_text
test_data.to_csv(os.path.join(raw_data_path,'Test_Data_Title_processed_final.csv'),index=False)
# 准备生成二分类的五折数据
train_data = pd.read_csv(os.path.join(raw_data_path,'Train_Data_Title_processed_final.csv'))
test_data = pd.read_csv(os.path.join(raw_data_path,'Test_Data_Title_processed_final.csv'))
test_entity_null_id = test_data[test_data['entity'].isnull()]['id']
test_data['entity'] = test_data['entity'].fillna(' ')
test_data['title'] = test_data['title'].fillna('')
test_data['text'] = test_data['text'].fillna('')
test_data['text']= test_data.apply(lambda x:x['title']+' '+x['text']if x['title']!=x['text'] else x['text'],axis=1)
test_data['text'] = test_data.apply(lambda x:x['text'].strip(),axis=1)
train_data['entity'] = train_data['entity'].fillna(' ')
train_data['title'] = train_data['title'].fillna('')
train_data['text'] = train_data.apply(lambda x: x['title']+' '+x['text'] if x['title'] != x['text'] else x['text'],axis=1)
train_data['text'] = train_data.apply(lambda x:x['text'].strip(),axis=1)
# 去除空entity
train_data = train_data[train_data['entity'].map(lambda x : len(x)>1)]
train_data.shape
test_data_no_entity = test_data[test_data['entity'].map(lambda x : len(x)<=1)]
test_data = test_data[test_data['entity'].map(lambda x : len(x)>1)]
test_data.shape
test_data_no_entity.shape
def select_test(context,title,entity_list,key_entity_list):
return entity_list
def select_by_key_entity(context,title,entity_list,key_entity_list):
new_list = [i for i in entity_list]
new_list = sorted(new_list,key= lambda x:len(x),reverse=True)
final_list = []
for i in new_list:
flag = True
for j in final_list:
if i in j and i and i not in key_entity_list :
flag = False
break
if flag:
final_list.append(i)
return final_list
train_entity = pd.DataFrame(columns=['id','text','entity','negative'])
test_entity = pd.DataFrame(columns=['id','text','entity'])
for index,item in train_data.iterrows():
if item['negative'] == 0:
entity_list = item['entity'].split(';')
for i in entity_list:
train_entity = train_entity.append(pd.Series({'id':item['id'],'text':item['text'],'entity':i,'negative':0}),ignore_index=True)
else:
entity_list = item['entity'].split(';')
key_entity_list = item['key_entity'].split(';')
select_entity_list = select_by_key_entity(item['text'],'',entity_list, key_entity_list)
for i in select_entity_list:
if i not in key_entity_list:
train_entity = train_entity.append(pd.Series({'id':item['id'],'text':item['text'],'entity':i,'negative':0}),ignore_index=True)
else:
train_entity = train_entity.append(pd.Series({'id':item['id'],'text':item['text'],'entity':i,'negative':1}),ignore_index=True)
for index,item in test_data.iterrows():
entity_list = item['entity'].split(';')
select_entity_list = select_test(item['text'],'',entity_list,key_entity_list)
if len(select_entity_list) == 0:
print(entity_list)
for i in select_entity_list:
test_entity = test_entity.append(pd.Series({'id':item['id'],'text':item['text'],'entity':i}),ignore_index=True)
# 生成五折数据
# 修改路径
num = torch.randperm(len(train_entity)).tolist()
pd_list = []
for i in list(range(5)):
left = i* int(len(num)/5)
right = left+int(len(num)/5)
temp_pd = train_entity.iloc[num[left:right]]
pd_list.append(temp_pd)
for i in pd_list:
print (len(i))
for i in range(5):
temp_dev_pd = pd_list[i]
train_cv = copy.deepcopy(train_entity)
train_cv = train_cv.append(temp_dev_pd,ignore_index=True).drop_duplicates(keep=False)
print('--',temp_dev_pd.shape,'--',train_cv.shape)
path = os.path.join(os.path.join(raw_data_path,'fusai_cv_data','cv_'+str(i)))
if not os.path.exists(path):
os.makedirs(path)
dev = pd.DataFrame({'index':temp_dev_pd['id'],'question':temp_dev_pd['entity'],
'sentence':temp_dev_pd['text'],'label':temp_dev_pd['negative']})
dev_id = pd.DataFrame({'sentence':temp_dev_pd['text'],'id':temp_dev_pd['id']})
train = pd.DataFrame({'index':train_cv['id'],'question':train_cv['entity'],
'sentence':train_cv['text'],'label':train_cv['negative']})
test = pd.DataFrame({'index':test_entity['id'],'question':test_entity['entity'],'sentence':test_entity['text']})
dev.to_csv(os.path.join(path,'dev.tsv'),sep='\t',index=False)
dev_id.to_csv(os.path.join(path,'dev_id.tsv'),sep='\t',index=False)
train.to_csv(os.path.join(path,'train.tsv'),sep='\t',index=False)
test.to_csv(os.path.join(path,'test.tsv'),sep='\t',index=False)
# change length from cv file
def find_all(sub,s):
index_list = []
index = s.find(sub)
while index != -1:
index_list.append(index)
index = s.find(sub,index+1)
if len(index_list) > 0:
return index_list
else:
return -1
def get_span(loc,text,length):
left = loc-length
if loc+length < len(text):
right= loc+length
else:
right = len(text)
return text[left:right]
for i in range(5):
path= os.path.join(raw_data_path,'fusai_cv_data','cv_'+str(i))
data=['train','dev','test']
save_path = os.path.join(raw_data_path,'fusai_cv_data_max512')
if not os.path.exists(save_path):
os.mkdir(save_path)
save_path_cv =os.path.join(save_path,'cv_'+str(i))
#save_path_cv =save_path+'cv_'+str(i)
if not os.path.exists(save_path_cv):
os.mkdir(save_path_cv)
for j in data:
data_path = os.path.join(path,j+'.tsv')
#data_path = path+j+'.tsv'
train=pd.read_csv(data_path,sep='\t')
print(data_path)
num = 0
for index,item in train.iterrows():
entity = item['question']
text = item['sentence']
if len(text) > 512:
if pd.isna(entity):
continue
loc = text.find(entity)
if loc == -1:
train.iloc[index,2]=text[0:512]
else:
# print(loc,entity,'\n',text,[m.start() for m in re.finditer(entity, text)])
last_loc = find_all(entity,text)[-1]
if last_loc<512:
train.iloc[index,2]=text[0:512]
else:
head = text[0:100]
start_list = find_all(entity,text)[0:5]
if len(start_list)==0:
continue
span_length = (512-100)//len(start_list)
content = [get_span(i,text,span_length)for i in start_list]
text_target = head+' '.join(content)
train.iloc[index,2]=text_target
# print(item['index'],'\t',entity,'\t',len(start_list))
num +=1
train.to_csv(os.path.join(save_path,'cv_'+str(i),j+'.tsv'),sep='\t',index=False)
#train.to_csv(save_path+'cv_'+str(i)+'/'+j+'.tsv',sep='\t',index=False)
print(num)
# change length from cv file
## add <>
nan_num =0
for i in range(5):
path= os.path.join(raw_data_path,'fusai_cv_data_max512','cv_'+str(i))
data=['test','train','dev']
save_path = os.path.join(raw_data_path,'fusai_cv_data_max512_span_fc')
if not os.path.exists(save_path):
os.mkdir(save_path)
save_path_cv =os.path.join(save_path,'cv_'+str(i))
#save_path_cv =save_path+'cv_'+str(i)
if not os.path.exists(save_path_cv):
os.mkdir(save_path_cv)
for j in data:
data_path = os.path.join(path,j+'.tsv')
#data_path = path+j+'.tsv'
train=pd.read_csv(data_path,sep='\t')
print(data_path)
num = 0
for index,item in train.iterrows():
entity = item['question']
text = item['sentence']
if j != 'test':
entity_all = train_data[train_data['id']==item['index']]['entity'].item().split(';')
else:
entity_all = test_data[test_data['id']==item['index']]['entity'].item().split(';')
long = None
try:
for x in entity_all:
if type(x) is not float:
if entity in x and len(x) != len(entity):
long = x
break
except :
long = None
print('nan entity: ',entity)
continue
if long is None:
if type(entity) is not float and len(entity)>1 and '?'not in entity and '(' not in entity and '（' not in entity and '*'not in entity:
print(entity)
text = re.sub(entity,'['+entity+']',text)
train.iloc[index,2]=text
num +=1
else:
if type(entity) is not float and len(entity)>1 and '?'not in entity and '(' not in entity and '（' not in entity and '*'not in entity:
if type(long) is not float and len(long)>1 and '?'not in long and '(' not in long and '（' not in long and '*'not in long:
print(entity)
text = re.sub(long,'@@@@@@@',text)
text = re.sub(entity,'['+entity+']',text)
text = re.sub('@@@@@@@',long,text)
train.iloc[index,2]=text
num +=1
train.to_csv(os.path.join(save_path,'cv_'+str(i),j+'.tsv'),sep='\t',index=False)
#train.to_csv(save_path+'cv_'+str(i)+'/'+j+'.tsv',sep='\t',index=False)
print(num)