Update app.py
Browse files
app.py
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
import streamlit as st
|
| 2 |
from transformers import pipeline
|
| 3 |
import re
|
| 4 |
-
|
| 5 |
import requests
|
| 6 |
|
| 7 |
API_URL = "https://api-inference.huggingface.co/models/microsoft/prophetnet-large-uncased-squad-qg"
|
|
@@ -81,6 +81,7 @@ if evidence_text:
|
|
| 81 |
|
| 82 |
|
| 83 |
import pandas as pd
|
|
|
|
| 84 |
import numpy as np
|
| 85 |
from allennlp.predictors.predictor import Predictor
|
| 86 |
import allennlp_models.tagging
|
|
@@ -89,11 +90,14 @@ predictor = Predictor.from_path("structured-prediction-srl-bert.tar.gz")
|
|
| 89 |
#---------------------------------------------------------------
|
| 90 |
def claim(text):
|
| 91 |
import re
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
|
|
|
|
|
|
|
|
|
| 97 |
def srl_allennlp(sent):
|
| 98 |
try:
|
| 99 |
#result = predictor.predict(sentence=sent)['verbs'][0]['description']
|
|
@@ -160,11 +164,16 @@ def claim(text):
|
|
| 160 |
else:
|
| 161 |
substr = substr + val_string[i]
|
| 162 |
else:
|
| 163 |
-
|
| 164 |
-
|
| 165 |
-
|
| 166 |
-
|
| 167 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 168 |
#----------FOR COLUMN "WHAT"------------#
|
| 169 |
df['what'] = ''
|
| 170 |
for j in range(len(df['modified'])):
|
|
@@ -187,10 +196,15 @@ def claim(text):
|
|
| 187 |
else:
|
| 188 |
substr = substr + val_string[i]
|
| 189 |
else:
|
| 190 |
-
|
| 191 |
-
|
| 192 |
-
|
| 193 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 194 |
|
| 195 |
#----------FOR COLUMN "WHY"------------#
|
| 196 |
df['why'] = ''
|
|
@@ -214,10 +228,15 @@ def claim(text):
|
|
| 214 |
else:
|
| 215 |
substr = substr + val_string[i]
|
| 216 |
else:
|
| 217 |
-
|
| 218 |
-
|
|
|
|
|
|
|
|
|
|
| 219 |
|
| 220 |
-
|
|
|
|
|
|
|
| 221 |
|
| 222 |
#----------FOR COLUMN "WHEN"------------#
|
| 223 |
df['when'] = ''
|
|
@@ -241,10 +260,15 @@ def claim(text):
|
|
| 241 |
else:
|
| 242 |
substr = substr + val_string[i]
|
| 243 |
else:
|
| 244 |
-
|
| 245 |
-
|
|
|
|
|
|
|
|
|
|
| 246 |
|
| 247 |
-
|
|
|
|
|
|
|
| 248 |
|
| 249 |
|
| 250 |
#----------FOR COLUMN "WHERE"------------#
|
|
@@ -269,163 +293,337 @@ def claim(text):
|
|
| 269 |
else:
|
| 270 |
substr = substr + val_string[i]
|
| 271 |
else:
|
| 272 |
-
|
| 273 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 274 |
|
| 275 |
-
df['where'][j] = where
|
| 276 |
|
| 277 |
data=df[["claim","who","what","why","when","where"]].copy()
|
| 278 |
-
import re
|
| 279 |
-
def remove_trail_comma(text):
|
| 280 |
-
x = re.sub(",\s*$", "", text)
|
| 281 |
-
return x
|
| 282 |
-
|
| 283 |
-
|
| 284 |
-
data['claim']=data['claim'].apply(lambda x: str(x).replace('\'','').replace('\'',''))
|
| 285 |
-
data['claim']=data['claim'].apply(lambda x: str(x).replace('[','').replace(']',''))
|
| 286 |
-
|
| 287 |
-
|
| 288 |
-
|
| 289 |
-
data['who']=data['who'].apply(lambda x: str(x).replace(" 's","'s"))
|
| 290 |
-
data['who']=data['who'].apply(lambda x: str(x).replace("s ’","s’"))
|
| 291 |
-
data['who']=data['who'].apply(lambda x: str(x).replace(" - ","-"))
|
| 292 |
-
data['who']=data['who'].apply(lambda x: str(x).replace('\'','').replace('\'',''))
|
| 293 |
-
# data['who']=data['who'].apply(lambda x: str(x).replace('"','').replace('"',''))
|
| 294 |
-
data['who']=data['who'].apply(lambda x: str(x).replace('[','').replace(']',''))
|
| 295 |
-
data['who']=data['who'].apply(lambda x: str(x).rstrip(','))
|
| 296 |
-
data['who']=data['who'].apply(lambda x: str(x).lstrip(','))
|
| 297 |
-
data['who']=data['who'].apply(lambda x: str(x).replace('None,','').replace('None',''))
|
| 298 |
-
data['who']=data['who'].apply(remove_trail_comma)
|
| 299 |
-
|
| 300 |
-
|
| 301 |
-
|
| 302 |
-
data['what']=data['what'].apply(lambda x: str(x).replace(" 's","'s"))
|
| 303 |
-
data['what']=data['what'].apply(lambda x: str(x).replace("s ’","s’"))
|
| 304 |
-
data['what']=data['what'].apply(lambda x: str(x).replace(" - ","-"))
|
| 305 |
-
data['what']=data['what'].apply(lambda x: str(x).replace('\'','').replace('\'',''))
|
| 306 |
-
# data['what']=data['what'].apply(lambda x: str(x).replace('"','').replace('"',''))
|
| 307 |
-
data['what']=data['what'].apply(lambda x: str(x).replace('[','').replace(']',''))
|
| 308 |
-
data['what']=data['what'].apply(lambda x: str(x).rstrip(','))
|
| 309 |
-
data['what']=data['what'].apply(lambda x: str(x).lstrip(','))
|
| 310 |
-
data['what']=data['what'].apply(lambda x: str(x).replace('None,','').replace('None',''))
|
| 311 |
-
data['what']=data['what'].apply(remove_trail_comma)
|
| 312 |
-
|
| 313 |
-
data['why']=data['why'].apply(lambda x: str(x).replace(" 's","'s"))
|
| 314 |
-
data['why']=data['why'].apply(lambda x: str(x).replace("s ’","s’"))
|
| 315 |
-
data['why']=data['why'].apply(lambda x: str(x).replace(" - ","-"))
|
| 316 |
-
data['why']=data['why'].apply(lambda x: str(x).replace('\'','').replace('\'',''))
|
| 317 |
-
# data['why']=data['why'].apply(lambda x: str(x).replace('"','').replace('"',''))
|
| 318 |
-
data['why']=data['why'].apply(lambda x: str(x).replace('[','').replace(']',''))
|
| 319 |
-
data['why']=data['why'].apply(lambda x: str(x).rstrip(','))
|
| 320 |
-
data['why']=data['why'].apply(lambda x: str(x).lstrip(','))
|
| 321 |
-
data['why']=data['why'].apply(lambda x: str(x).replace('None,','').replace('None',''))
|
| 322 |
-
data['why']=data['why'].apply(remove_trail_comma)
|
| 323 |
-
|
| 324 |
-
data['when']=data['when'].apply(lambda x: str(x).replace(" 's","'s"))
|
| 325 |
-
data['when']=data['when'].apply(lambda x: str(x).replace("s ’","s’"))
|
| 326 |
-
data['when']=data['when'].apply(lambda x: str(x).replace(" - ","-"))
|
| 327 |
-
data['when']=data['when'].apply(lambda x: str(x).replace('\'','').replace('\'',''))
|
| 328 |
-
# data['when']=data['when'].apply(lambda x: str(x).replace('"','').replace('"',''))
|
| 329 |
-
data['when']=data['when'].apply(lambda x: str(x).replace('[','').replace(']',''))
|
| 330 |
-
data['when']=data['when'].apply(lambda x: str(x).rstrip(','))
|
| 331 |
-
data['when']=data['when'].apply(lambda x: str(x).lstrip(','))
|
| 332 |
-
data['when']=data['when'].apply(lambda x: str(x).replace('None,','').replace('None',''))
|
| 333 |
-
data['when']=data['when'].apply(remove_trail_comma)
|
| 334 |
-
|
| 335 |
-
data['where']=data['where'].apply(lambda x: str(x).replace(" 's","'s"))
|
| 336 |
-
data['where']=data['where'].apply(lambda x: str(x).replace("s ’","s’"))
|
| 337 |
-
data['where']=data['where'].apply(lambda x: str(x).replace(" - ","-"))
|
| 338 |
-
data['where']=data['where'].apply(lambda x: str(x).replace('\'','').replace('\'',''))
|
| 339 |
-
# data['where']=data['where'].apply(lambda x: str(x).replace('"','').replace('"',''))
|
| 340 |
-
data['where']=data['where'].apply(lambda x: str(x).replace('[','').replace(']',''))
|
| 341 |
-
data['where']=data['where'].apply(lambda x: str(x).rstrip(','))
|
| 342 |
-
data['where']=data['where'].apply(lambda x: str(x).lstrip(','))
|
| 343 |
-
data['where']=data['where'].apply(lambda x: str(x).replace('None,','').replace('None',''))
|
| 344 |
-
data['where']=data['where'].apply(remove_trail_comma)
|
| 345 |
return data
|
| 346 |
#-------------------------------------------------------------------------
|
| 347 |
-
def split_ws(input_list):
|
| 348 |
-
import re
|
| 349 |
output_list = []
|
| 350 |
for item in input_list:
|
| 351 |
-
split_item =
|
| 352 |
-
|
| 353 |
-
|
| 354 |
-
|
|
|
|
|
|
|
| 355 |
|
| 356 |
#--------------------------------------------------------------------------
|
| 357 |
-
def
|
| 358 |
-
|
| 359 |
-
|
| 360 |
-
|
| 361 |
-
|
| 362 |
-
|
| 363 |
-
|
| 364 |
-
|
| 365 |
-
|
| 366 |
-
|
| 367 |
-
|
| 368 |
-
|
| 369 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 370 |
try:
|
| 371 |
-
|
| 372 |
-
|
| 373 |
-
|
| 374 |
-
|
| 375 |
-
|
| 376 |
-
|
| 377 |
-
|
| 378 |
-
|
| 379 |
-
|
| 380 |
-
|
| 381 |
-
|
| 382 |
-
|
| 383 |
-
|
| 384 |
-
|
| 385 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 386 |
except:
|
| 387 |
pass
|
| 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 |
|
|
@@ -433,10 +631,13 @@ if claim_text:
|
|
| 433 |
if evidence_text:
|
| 434 |
df=claim(claim_text)
|
| 435 |
df["evidence"]=evidence_text
|
| 436 |
-
|
| 437 |
-
|
| 438 |
-
|
| 439 |
-
|
|
|
|
|
|
|
|
|
|
| 440 |
# a,b=qa_evidence(final_data)
|
| 441 |
# qa_evidence(final_data)
|
| 442 |
# st.json(qa_evidence(final_data))
|
|
|
|
| 1 |
import streamlit as st
|
| 2 |
from transformers import pipeline
|
| 3 |
import re
|
| 4 |
+
import time
|
| 5 |
import requests
|
| 6 |
|
| 7 |
API_URL = "https://api-inference.huggingface.co/models/microsoft/prophetnet-large-uncased-squad-qg"
|
|
|
|
| 81 |
|
| 82 |
|
| 83 |
import pandas as pd
|
| 84 |
+
from rouge_score import rouge_scorer
|
| 85 |
import numpy as np
|
| 86 |
from allennlp.predictors.predictor import Predictor
|
| 87 |
import allennlp_models.tagging
|
|
|
|
| 90 |
#---------------------------------------------------------------
|
| 91 |
def claim(text):
|
| 92 |
import re
|
| 93 |
+
def remove_special_chars(text):
|
| 94 |
+
# Remove special characters that are not in between numbers
|
| 95 |
+
text = re.sub(r'(?<!\d)[^\w\s]+(?!\d)', '', text)
|
| 96 |
+
|
| 97 |
+
return text
|
| 98 |
+
|
| 99 |
+
df = pd.DataFrame({'claim' : remove_special_chars(text)},index=[0])
|
| 100 |
+
|
| 101 |
def srl_allennlp(sent):
|
| 102 |
try:
|
| 103 |
#result = predictor.predict(sentence=sent)['verbs'][0]['description']
|
|
|
|
| 164 |
else:
|
| 165 |
substr = substr + val_string[i]
|
| 166 |
else:
|
| 167 |
+
pass
|
| 168 |
+
if len(substr)!= 0:
|
| 169 |
+
who.append(substr)
|
| 170 |
+
else:
|
| 171 |
+
pass
|
| 172 |
+
|
| 173 |
+
|
| 174 |
+
df['who'][j] = "<sep>".join(who)
|
| 175 |
+
# else:
|
| 176 |
+
# continue
|
| 177 |
#----------FOR COLUMN "WHAT"------------#
|
| 178 |
df['what'] = ''
|
| 179 |
for j in range(len(df['modified'])):
|
|
|
|
| 196 |
else:
|
| 197 |
substr = substr + val_string[i]
|
| 198 |
else:
|
| 199 |
+
pass
|
| 200 |
+
if len(substr)!= 0:
|
| 201 |
+
what.append(substr)
|
| 202 |
+
else:
|
| 203 |
+
pass
|
| 204 |
+
|
| 205 |
+
df['what'][j] = "<sep>".join(what)
|
| 206 |
+
# else:
|
| 207 |
+
# continue
|
| 208 |
|
| 209 |
#----------FOR COLUMN "WHY"------------#
|
| 210 |
df['why'] = ''
|
|
|
|
| 228 |
else:
|
| 229 |
substr = substr + val_string[i]
|
| 230 |
else:
|
| 231 |
+
pass
|
| 232 |
+
if len(substr)!= 0:
|
| 233 |
+
why.append(substr)
|
| 234 |
+
else:
|
| 235 |
+
pass
|
| 236 |
|
| 237 |
+
df['why'][j] = "<sep>".join(why)
|
| 238 |
+
# else:
|
| 239 |
+
# continue
|
| 240 |
|
| 241 |
#----------FOR COLUMN "WHEN"------------#
|
| 242 |
df['when'] = ''
|
|
|
|
| 260 |
else:
|
| 261 |
substr = substr + val_string[i]
|
| 262 |
else:
|
| 263 |
+
pass
|
| 264 |
+
if len(substr)!= 0:
|
| 265 |
+
when.append(substr)
|
| 266 |
+
else:
|
| 267 |
+
pass
|
| 268 |
|
| 269 |
+
df['when'][j] = "<sep>".join(when)
|
| 270 |
+
# else:
|
| 271 |
+
# continue
|
| 272 |
|
| 273 |
|
| 274 |
#----------FOR COLUMN "WHERE"------------#
|
|
|
|
| 293 |
else:
|
| 294 |
substr = substr + val_string[i]
|
| 295 |
else:
|
| 296 |
+
pass
|
| 297 |
+
if len(substr)!= 0:
|
| 298 |
+
where.append(substr)
|
| 299 |
+
else:
|
| 300 |
+
pass
|
| 301 |
+
|
| 302 |
+
df['where'][j] = "<sep>".join(where)
|
| 303 |
|
|
|
|
| 304 |
|
| 305 |
data=df[["claim","who","what","why","when","where"]].copy()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 306 |
return data
|
| 307 |
#-------------------------------------------------------------------------
|
| 308 |
+
def split_ws(input_list, delimiter="<sep>"):
|
|
|
|
| 309 |
output_list = []
|
| 310 |
for item in input_list:
|
| 311 |
+
split_item = item.split(delimiter)
|
| 312 |
+
for sub_item in split_item:
|
| 313 |
+
sub_item = sub_item.strip()
|
| 314 |
+
if sub_item:
|
| 315 |
+
output_list.append(sub_item)
|
| 316 |
+
return output_list
|
| 317 |
|
| 318 |
#--------------------------------------------------------------------------
|
| 319 |
+
def calc_rouge_l_score(list_of_evidence, list_of_ans):
|
| 320 |
+
scorer = rouge_scorer.RougeScorer(['rougeL'], use_stemmer=True)
|
| 321 |
+
scores = scorer.score(' '.join(list_of_evidence), ' '.join(list_of_ans))
|
| 322 |
+
return scores['rougeL'].fmeasure
|
| 323 |
+
#-------------------------------------------------------------------------
|
| 324 |
+
|
| 325 |
+
|
| 326 |
+
def rephrase_question_who(question):
|
| 327 |
+
if not question.lower().startswith("who"):
|
| 328 |
+
words = question.split()
|
| 329 |
+
words[0] = "Who"
|
| 330 |
+
return " ".join(words)
|
| 331 |
+
else:
|
| 332 |
+
return question
|
| 333 |
+
#------------------------------------------------------------------------
|
| 334 |
+
def gen_qa_who(df):
|
| 335 |
+
list_of_ques_who=[]
|
| 336 |
+
list_of_ans_who=[]
|
| 337 |
+
list_of_evidence_answer_who=[]
|
| 338 |
+
rouge_l_scores=[]
|
| 339 |
+
for i,row in df.iterrows():
|
| 340 |
+
srl=df["who"][i]
|
| 341 |
+
claim=df['claim'][i]
|
| 342 |
+
answer= split_ws(df["who"])
|
| 343 |
+
evidence=df["evidence"][i]
|
| 344 |
+
#time.sleep(10)
|
| 345 |
+
if srl!="":
|
| 346 |
try:
|
| 347 |
+
for j in range(0,len(answer)):
|
| 348 |
+
FACT_TO_GENERATE_QUESTION_FROM = f"""generate_the_question_from_context_using_the_next_answer:{answer[j]} [SEP] context:{claim}"""
|
| 349 |
+
#FACT_TO_GENERATE_QUESTION_FROM = f"""generate_who_based_question_from_context_using_the_next_answer:{answer[j]} [SEP] context:{claim}"""
|
| 350 |
+
#time.sleep(10)
|
| 351 |
+
question_ids = query({"inputs":FACT_TO_GENERATE_QUESTION_FROM,
|
| 352 |
+
"num_beams":5,
|
| 353 |
+
"early_stopping":True,
|
| 354 |
+
"min_length": 100,"wait_for_model":True})[0]['generated_text'].capitalize()
|
| 355 |
+
question_ids = rephrase_question_who(question_ids)
|
| 356 |
+
list_of_ques_who.append(f"""Q{j+1}:{question_ids}""")
|
| 357 |
+
list_of_ans_who.append(f"""Ans{j+1}:{answer[j]}""")
|
| 358 |
+
input_evidence = f"answer_the_next_question_from_context: {question_ids} context: {evidence}"
|
| 359 |
+
#time.sleep(10)
|
| 360 |
+
answer_evidence = query_evidence({"inputs":input_evidence,"truncation":True,"wait_for_model":True})[0]['generated_text']
|
| 361 |
+
if answer_evidence.lower() in evidence.lower():
|
| 362 |
+
list_of_evidence_answer_who.append(f"""Evidence{j+1}:{answer_evidence}""")
|
| 363 |
+
else:
|
| 364 |
+
answer_evidence=""
|
| 365 |
+
list_of_evidence_answer_who.append(f"""No mention of 'who'in any related documents.""")
|
| 366 |
+
threshold = 0.2
|
| 367 |
+
list_of_pairs = [(answer_evidence, answer[j])]
|
| 368 |
+
rouge_l_score = calc_rouge_l_score(answer_evidence, answer[j])
|
| 369 |
+
if rouge_l_score >= threshold:
|
| 370 |
+
verification_status = 'Verified Valid'
|
| 371 |
+
elif rouge_l_score == 0:
|
| 372 |
+
verification_status = 'Not verifiable'
|
| 373 |
+
else:
|
| 374 |
+
verification_status = 'Verified False'
|
| 375 |
+
rouge_l_scores.append(verification_status)
|
| 376 |
except:
|
| 377 |
pass
|
| 378 |
+
else:
|
| 379 |
+
list_of_ques_who="No claims"
|
| 380 |
+
list_of_ans_who=""
|
| 381 |
+
list_of_evidence_answer_who="No mention of 'who'in any related documents."
|
| 382 |
+
rouge_l_scores="Not verifiable"
|
| 383 |
+
return list_of_ques_who,list_of_ans_who,rouge_l_scores,list_of_evidence_answer_who
|
| 384 |
#------------------------------------------------------------
|
| 385 |
+
|
| 386 |
+
def rephrase_question_what(question):
|
| 387 |
+
if not question.lower().startswith("what"):
|
| 388 |
+
words = question.split()
|
| 389 |
+
words[0] = "What"
|
| 390 |
+
return " ".join(words)
|
| 391 |
+
else:
|
| 392 |
+
return question
|
| 393 |
+
#----------------------------------------------------------
|
| 394 |
+
def gen_qa_what(df):
|
| 395 |
+
list_of_ques_what=[]
|
| 396 |
+
list_of_ans_what=[]
|
| 397 |
+
list_of_evidence_answer_what=[]
|
| 398 |
+
rouge_l_scores=[]
|
| 399 |
+
for i,row in df.iterrows():
|
| 400 |
+
srl=df["what"][i]
|
| 401 |
+
claim=df['claim'][i]
|
| 402 |
+
answer= split_ws(df["what"])
|
| 403 |
+
evidence=df["evidence"][i]
|
| 404 |
+
#time.sleep(10)
|
| 405 |
+
if srl!="":
|
| 406 |
+
try:
|
| 407 |
+
for j in range(0,len(answer)):
|
| 408 |
+
FACT_TO_GENERATE_QUESTION_FROM = f"""generate_the_question_from_context_using_the_next_answer:{answer[j]} [SEP] context:{claim}"""
|
| 409 |
+
#time.sleep(10)
|
| 410 |
+
question_ids = query({"inputs":FACT_TO_GENERATE_QUESTION_FROM,
|
| 411 |
+
"num_beams":5,
|
| 412 |
+
"early_stopping":True,
|
| 413 |
+
"min_length": 100,"wait_for_model":True})[0]['generated_text'].capitalize()
|
| 414 |
+
question_ids = rephrase_question_what(question_ids)
|
| 415 |
+
list_of_ques_what.append(f"""Q{j+1}:{question_ids}""")
|
| 416 |
+
list_of_ans_what.append(f"""Ans{j+1}:{answer[j]}""")
|
| 417 |
+
input_evidence = f"answer_the_next_question_from_context: {question_ids} context: {evidence}"
|
| 418 |
+
#time.sleep(10)
|
| 419 |
+
answer_evidence = query_evidence({"inputs":input_evidence,"truncation":True,"wait_for_model":True})[0]['generated_text']
|
| 420 |
+
if answer_evidence.lower() in evidence.lower():
|
| 421 |
+
list_of_evidence_answer_what.append(f"""Evidence{j+1}:{answer_evidence}""")
|
| 422 |
+
|
| 423 |
+
else:
|
| 424 |
+
answer_evidence=""
|
| 425 |
+
list_of_evidence_answer_what.append(f"""No mention of 'what'in any related documents.""")
|
| 426 |
+
threshold = 0.2
|
| 427 |
+
list_of_pairs = [(answer_evidence, answer[j])]
|
| 428 |
+
rouge_l_score = calc_rouge_l_score(answer_evidence, answer[j])
|
| 429 |
+
if rouge_l_score >= threshold:
|
| 430 |
+
verification_status = 'Verified Valid'
|
| 431 |
+
elif rouge_l_score == 0:
|
| 432 |
+
verification_status = 'Not verifiable'
|
| 433 |
+
else:
|
| 434 |
+
verification_status = 'Verified False'
|
| 435 |
+
rouge_l_scores.append(verification_status)
|
| 436 |
+
except:
|
| 437 |
+
pass
|
| 438 |
+
else:
|
| 439 |
+
list_of_ques_what="No claims"
|
| 440 |
+
list_of_ans_what=""
|
| 441 |
+
list_of_evidence_answer_what="No mention of 'what'in any related documents."
|
| 442 |
+
rouge_l_scores="Not verifiable"
|
| 443 |
+
return list_of_ques_what,list_of_ans_what,rouge_l_scores,list_of_evidence_answer_what
|
| 444 |
+
#----------------------------------------------------------
|
| 445 |
+
|
| 446 |
+
def rephrase_question_why(question):
|
| 447 |
+
if not question.lower().startswith("why"):
|
| 448 |
+
words = question.split()
|
| 449 |
+
words[0] = "Why"
|
| 450 |
+
return " ".join(words)
|
| 451 |
+
else:
|
| 452 |
+
return question
|
| 453 |
+
|
| 454 |
+
#---------------------------------------------------------
|
| 455 |
+
def gen_qa_why(df):
|
| 456 |
+
list_of_ques_why=[]
|
| 457 |
+
list_of_ans_why=[]
|
| 458 |
+
list_of_evidence_answer_why=[]
|
| 459 |
+
rouge_l_scores=[]
|
| 460 |
+
for i,row in df.iterrows():
|
| 461 |
+
srl=df["why"][i]
|
| 462 |
+
claim=df['claim'][i]
|
| 463 |
+
answer= split_ws(df["why"])
|
| 464 |
+
evidence=df["evidence"][i]
|
| 465 |
+
#time.sleep(10)
|
| 466 |
+
if srl!="":
|
| 467 |
+
try:
|
| 468 |
+
for j in range(0,len(answer)):
|
| 469 |
+
FACT_TO_GENERATE_QUESTION_FROM = f"""generate_the_question_from_context_using_the_next_answer:{answer[j]} [SEP] context:{claim}"""
|
| 470 |
+
#time.sleep(10)
|
| 471 |
+
question_ids = query({"inputs":FACT_TO_GENERATE_QUESTION_FROM,
|
| 472 |
+
"num_beams":5,
|
| 473 |
+
"early_stopping":True,
|
| 474 |
+
"min_length": 100,"wait_for_model":True})[0]['generated_text'].capitalize()
|
| 475 |
+
question_ids = rephrase_question_why(question_ids)
|
| 476 |
+
list_of_ques_why.append(f"""Q{j+1}:{question_ids}""")
|
| 477 |
+
list_of_ans_why.append(f"""Ans{j+1}:{answer[j]}""")
|
| 478 |
+
input_evidence = f"answer_the_next_question_from_context: {question_ids} context: {evidence}"
|
| 479 |
+
#time.sleep(10)
|
| 480 |
+
answer_evidence = query_evidence({"inputs":input_evidence,"truncation":True,"wait_for_model":True})[0]['generated_text']
|
| 481 |
+
if answer_evidence.lower() in evidence.lower():
|
| 482 |
+
list_of_evidence_answer_why.append(f"""Evidence{j+1}:{answer_evidence}""")
|
| 483 |
+
else:
|
| 484 |
+
answer_evidence=""
|
| 485 |
+
list_of_evidence_answer_why.append(f"""No mention of 'why'in any related documents.""")
|
| 486 |
+
threshold = 0.2
|
| 487 |
+
list_of_pairs = [(answer_evidence, answer[j])]
|
| 488 |
+
rouge_l_score = calc_rouge_l_score(answer_evidence, answer[j])
|
| 489 |
+
if rouge_l_score >= threshold:
|
| 490 |
+
verification_status = 'Verified Valid'
|
| 491 |
+
elif rouge_l_score == 0:
|
| 492 |
+
verification_status = 'Not verifiable'
|
| 493 |
+
else:
|
| 494 |
+
verification_status = 'Verified False'
|
| 495 |
+
rouge_l_scores.append(verification_status)
|
| 496 |
+
except:
|
| 497 |
+
pass
|
| 498 |
+
else:
|
| 499 |
+
list_of_ques_why="No claims"
|
| 500 |
+
list_of_ans_why=""
|
| 501 |
+
list_of_evidence_answer_why="No mention of 'why'in any related documents."
|
| 502 |
+
rouge_l_scores="Not verifiable"
|
| 503 |
+
return list_of_ques_why,list_of_ans_why,rouge_l_scores,list_of_evidence_answer_why
|
| 504 |
+
|
| 505 |
+
#---------------------------------------------------------
|
| 506 |
+
|
| 507 |
+
def rephrase_question_when(question):
|
| 508 |
+
if not question.lower().startswith("when"):
|
| 509 |
+
words = question.split()
|
| 510 |
+
words[0] = "When"
|
| 511 |
+
return " ".join(words)
|
| 512 |
+
else:
|
| 513 |
+
return question
|
| 514 |
+
#---------------------------------------------------------
|
| 515 |
+
def gen_qa_when(df):
|
| 516 |
+
list_of_ques_when=[]
|
| 517 |
+
list_of_ans_when=[]
|
| 518 |
+
list_of_evidence_answer_when=[]
|
| 519 |
+
rouge_l_scores=[]
|
| 520 |
+
for i,row in df.iterrows():
|
| 521 |
+
srl=df["when"][i]
|
| 522 |
+
claim=df['claim'][i]
|
| 523 |
+
answer= split_ws(df["when"])
|
| 524 |
+
evidence=df["evidence"][i]
|
| 525 |
+
#time.sleep(10)
|
| 526 |
+
if srl!="":
|
| 527 |
+
try:
|
| 528 |
+
for j in range(0,len(answer)):
|
| 529 |
+
FACT_TO_GENERATE_QUESTION_FROM = f"""generate_the_question_from_context_using_the_next_answer:{answer[j]} [SEP] context:{claim}"""
|
| 530 |
+
#time.sleep(10)
|
| 531 |
+
question_ids = query({"inputs":FACT_TO_GENERATE_QUESTION_FROM,
|
| 532 |
+
"num_beams":5,
|
| 533 |
+
"early_stopping":True,
|
| 534 |
+
"min_length": 100,"wait_for_model":True})[0]['generated_text'].capitalize()
|
| 535 |
+
question_ids = rephrase_question_when(question_ids)
|
| 536 |
+
list_of_ques_when.append(f"""Q{j+1}:{question_ids}""")
|
| 537 |
+
list_of_ans_when.append(f"""Ans{j+1}:{answer[j]}""")
|
| 538 |
+
input_evidence = f"answer_the_next_question_from_context: {question_ids} context: {evidence}"
|
| 539 |
+
#time.sleep(10)
|
| 540 |
+
answer_evidence = query_evidence({"inputs":input_evidence,"truncation":True,"wait_for_model":True})[0]['generated_text']
|
| 541 |
+
if answer_evidence.lower() in evidence.lower():
|
| 542 |
+
list_of_evidence_answer_when.append(f"""Evidence{j+1}:{answer_evidence}""")
|
| 543 |
+
else:
|
| 544 |
+
answer_evidence=""
|
| 545 |
+
list_of_evidence_answer_when.append(f"""No mention of 'when'in any related documents.""")
|
| 546 |
+
threshold = 0.2
|
| 547 |
+
list_of_pairs = [(answer_evidence, answer[j])]
|
| 548 |
+
rouge_l_score = calc_rouge_l_score(answer_evidence, answer[j])
|
| 549 |
+
if rouge_l_score >= threshold:
|
| 550 |
+
verification_status = 'Verified Valid'
|
| 551 |
+
elif rouge_l_score == 0:
|
| 552 |
+
verification_status = 'Not verifiable'
|
| 553 |
+
else:
|
| 554 |
+
verification_status = 'Verified False'
|
| 555 |
+
rouge_l_scores.append(verification_status)
|
| 556 |
+
except:
|
| 557 |
+
pass
|
| 558 |
+
else:
|
| 559 |
+
list_of_ques_when="No claims"
|
| 560 |
+
list_of_ans_when=""
|
| 561 |
+
list_of_evidence_answer_when="No mention of 'when'in any related documents."
|
| 562 |
+
rouge_l_scores="Not verifiable"
|
| 563 |
+
return list_of_ques_when,list_of_ans_when,rouge_l_scores,list_of_evidence_answer_when
|
| 564 |
+
|
| 565 |
+
#------------------------------------------------------
|
| 566 |
+
|
| 567 |
+
def rephrase_question_where(question):
|
| 568 |
+
if not question.lower().startswith("where"):
|
| 569 |
+
words = question.split()
|
| 570 |
+
words[0] = "Where"
|
| 571 |
+
return " ".join(words)
|
| 572 |
+
else:
|
| 573 |
+
return question
|
| 574 |
+
#------------------------------------------------------
|
| 575 |
+
def gen_qa_where(df):
|
| 576 |
+
list_of_ques_where=[]
|
| 577 |
+
list_of_ans_where=[]
|
| 578 |
+
list_of_evidence_answer_where=[]
|
| 579 |
+
rouge_l_scores=[]
|
| 580 |
+
for i,row in df.iterrows():
|
| 581 |
+
srl=df["where"][i]
|
| 582 |
+
claim=df['claim'][i]
|
| 583 |
+
answer= split_ws(df["where"])
|
| 584 |
+
evidence=df["evidence"][i]
|
| 585 |
+
#time.sleep(10)
|
| 586 |
+
if srl!="":
|
| 587 |
+
try:
|
| 588 |
+
for j in range(0,len(answer)):
|
| 589 |
+
FACT_TO_GENERATE_QUESTION_FROM = f"""generate_the_question_from_context_using_the_next_answer:{answer[j]} [SEP] context:{claim}"""
|
| 590 |
+
#time.sleep(10)
|
| 591 |
+
question_ids = query({"inputs":FACT_TO_GENERATE_QUESTION_FROM,
|
| 592 |
+
"num_beams":5,
|
| 593 |
+
"early_stopping":True,
|
| 594 |
+
"min_length": 100,"wait_for_model":True})[0]['generated_text'].capitalize()
|
| 595 |
+
question_ids = rephrase_question_where(question_ids)
|
| 596 |
+
list_of_ques_where.append(f"""Q{j+1}:{question_ids}""")
|
| 597 |
+
list_of_ans_where.append(f"""Ans{j+1}:{answer[j]}""")
|
| 598 |
+
input_evidence = f"answer_the_next_question_from_context: {question_ids} context: {evidence}"
|
| 599 |
+
#time.sleep(10)
|
| 600 |
+
answer_evidence = query_evidence({"inputs":input_evidence,"truncation":True,"wait_for_model":True})[0]['generated_text']
|
| 601 |
+
if answer_evidence.lower() in evidence.lower():
|
| 602 |
+
list_of_evidence_answer_where.append(f"""Evidence{j+1}:{answer_evidence}""")
|
| 603 |
+
else:
|
| 604 |
+
answer_evidence=""
|
| 605 |
+
list_of_evidence_answer_where.append(f"""No mention of 'where'in any related documents.""")
|
| 606 |
+
threshold = 0.2
|
| 607 |
+
list_of_pairs = [(answer_evidence, answer[j])]
|
| 608 |
+
rouge_l_score = calc_rouge_l_score(answer_evidence, answer[j])
|
| 609 |
+
if rouge_l_score >= threshold:
|
| 610 |
+
verification_status = 'Verified Valid'
|
| 611 |
+
elif rouge_l_score == 0:
|
| 612 |
+
verification_status = 'Not verifiable'
|
| 613 |
+
else:
|
| 614 |
+
verification_status = 'Verified False'
|
| 615 |
+
rouge_l_scores.append(verification_status)
|
| 616 |
+
except:
|
| 617 |
+
pass
|
| 618 |
+
else:
|
| 619 |
+
list_of_ques_where="No claims"
|
| 620 |
+
list_of_ans_where=""
|
| 621 |
+
list_of_evidence_answer_where="No mention of 'where'in any related documents."
|
| 622 |
+
rouge_l_scores="Not verifiable"
|
| 623 |
+
return list_of_ques_where,list_of_ans_where,rouge_l_scores,list_of_evidence_answer_where
|
| 624 |
+
|
| 625 |
+
#------------------------------------------------------
|
| 626 |
+
|
| 627 |
|
| 628 |
#------------------------------------------------------------
|
| 629 |
|
|
|
|
| 631 |
if evidence_text:
|
| 632 |
df=claim(claim_text)
|
| 633 |
df["evidence"]=evidence_text
|
| 634 |
+
final_df = pd.DataFrame(columns=['Who Claims', 'What Claims', 'When Claims', 'Where Claims', 'Why Claims'])
|
| 635 |
+
final_df["Who Claims"]=gen_qa_who(df)
|
| 636 |
+
final_df["What Claims"]=gen_qa_what(df)
|
| 637 |
+
final_df["When Claims"]=gen_qa_when(df)
|
| 638 |
+
final_df["Where Claims"]=gen_qa_where(df)
|
| 639 |
+
final_df["Why Claims"]=gen_qa_why(df)
|
| 640 |
+
st.dataframe(final_df)
|
| 641 |
# a,b=qa_evidence(final_data)
|
| 642 |
# qa_evidence(final_data)
|
| 643 |
# st.json(qa_evidence(final_data))
|