Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,397 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# example 1
|
2 |
+
from textwrap3 import wrap
|
3 |
+
import torch
|
4 |
+
import random
|
5 |
+
import numpy as np
|
6 |
+
import nltk
|
7 |
+
nltk.download('punkt')
|
8 |
+
nltk.download('brown')
|
9 |
+
nltk.download('wordnet')
|
10 |
+
from nltk.corpus import wordnet as wn
|
11 |
+
from nltk.tokenize import sent_tokenize
|
12 |
+
nltk.download('stopwords')
|
13 |
+
from nltk.corpus import stopwords
|
14 |
+
import string
|
15 |
+
import pke
|
16 |
+
import traceback
|
17 |
+
from flashtext import KeywordProcessor
|
18 |
+
from similarity.normalized_levenshtein import NormalizedLevenshtein
|
19 |
+
normalized_levenshtein = NormalizedLevenshtein()
|
20 |
+
from collections import OrderedDict
|
21 |
+
from sklearn.metrics.pairwise import cosine_similarity
|
22 |
+
import nltk
|
23 |
+
nltk.download('omw-1.4')
|
24 |
+
import gradio as gr
|
25 |
+
question_model = T5ForConditionalGeneration.from_pretrained('ramsrigouthamg/t5_squad_v1')
|
26 |
+
question_tokenizer = T5Tokenizer.from_pretrained('ramsrigouthamg/t5_squad_v1')
|
27 |
+
question_model = question_model.to(device)
|
28 |
+
|
29 |
+
# filter keywords
|
30 |
+
!wget https://github.com/explosion/sense2vec/releases/download/v1.0.0/s2v_reddit_2015_md.tar.gz
|
31 |
+
!tar -xvf s2v_reddit_2015_md.tar.gz
|
32 |
+
import numpy as np
|
33 |
+
from sense2vec import Sense2Vec
|
34 |
+
s2v = Sense2Vec().from_disk('s2v_old')
|
35 |
+
from sentence_transformers import SentenceTransformer
|
36 |
+
|
37 |
+
|
38 |
+
text = """Elon Musk has shown again he can influence the digital currency market with just his tweets. After saying that his electric vehicle-making company
|
39 |
+
Tesla will not accept payments in Bitcoin because of environmental concerns, he tweeted that he was working with developers of Dogecoin to improve
|
40 |
+
system transaction efficiency. Following the two distinct statements from him, the world's largest cryptocurrency hit a two-month low, while Dogecoin
|
41 |
+
rallied by about 20 percent. The SpaceX CEO has in recent months often tweeted in support of Dogecoin, but rarely for Bitcoin. In a recent tweet,
|
42 |
+
Musk put out a statement from Tesla that it was “concerned” about the rapidly increasing use of fossil fuels for Bitcoin (price in India) mining and
|
43 |
+
transaction, and hence was suspending vehicle purchases using the cryptocurrency. A day later he again tweeted saying, “To be clear, I strongly
|
44 |
+
believe in crypto, but it can't drive a massive increase in fossil fuel use, especially coal”. It triggered a downward spiral for Bitcoin value but
|
45 |
+
the cryptocurrency has stabilised since. A number of Twitter users welcomed Musk's statement. One of them said it's time people started realising
|
46 |
+
that Dogecoin “is here to stay” and another referred to Musk's previous assertion that crypto could become the world's future currency."""
|
47 |
+
|
48 |
+
for wrp in wrap(text, 150):
|
49 |
+
print (wrp)
|
50 |
+
print ("\n")
|
51 |
+
|
52 |
+
|
53 |
+
# summerization with t5
|
54 |
+
from transformers import T5ForConditionalGeneration,T5Tokenizer
|
55 |
+
summary_model = T5ForConditionalGeneration.from_pretrained('t5-base')
|
56 |
+
summary_tokenizer = T5Tokenizer.from_pretrained('t5-base')
|
57 |
+
|
58 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
59 |
+
summary_model = summary_model.to(device)
|
60 |
+
|
61 |
+
|
62 |
+
|
63 |
+
def set_seed(seed: int):
|
64 |
+
random.seed(seed)
|
65 |
+
np.random.seed(seed)
|
66 |
+
torch.manual_seed(seed)
|
67 |
+
torch.cuda.manual_seed_all(seed)
|
68 |
+
|
69 |
+
set_seed(42)
|
70 |
+
|
71 |
+
|
72 |
+
|
73 |
+
def postprocesstext (content):
|
74 |
+
final=""
|
75 |
+
for sent in sent_tokenize(content):
|
76 |
+
sent = sent.capitalize()
|
77 |
+
final = final +" "+sent
|
78 |
+
return final
|
79 |
+
|
80 |
+
|
81 |
+
def summarizer(text,model,tokenizer):
|
82 |
+
text = text.strip().replace("\n"," ")
|
83 |
+
text = "summarize: "+text
|
84 |
+
# print (text)
|
85 |
+
max_len = 512
|
86 |
+
encoding = tokenizer.encode_plus(text,max_length=max_len, pad_to_max_length=False,truncation=True, return_tensors="pt").to(device)
|
87 |
+
|
88 |
+
input_ids, attention_mask = encoding["input_ids"], encoding["attention_mask"]
|
89 |
+
|
90 |
+
outs = model.generate(input_ids=input_ids,
|
91 |
+
attention_mask=attention_mask,
|
92 |
+
early_stopping=True,
|
93 |
+
num_beams=3,
|
94 |
+
num_return_sequences=1,
|
95 |
+
no_repeat_ngram_size=2,
|
96 |
+
min_length = 75,
|
97 |
+
max_length=300)
|
98 |
+
|
99 |
+
|
100 |
+
dec = [tokenizer.decode(ids,skip_special_tokens=True) for ids in outs]
|
101 |
+
summary = dec[0]
|
102 |
+
summary = postprocesstext(summary)
|
103 |
+
summary= summary.strip()
|
104 |
+
|
105 |
+
return summary
|
106 |
+
|
107 |
+
|
108 |
+
summarized_text = summarizer(text,summary_model,summary_tokenizer)
|
109 |
+
|
110 |
+
|
111 |
+
print ("\noriginal Text >>")
|
112 |
+
for wrp in wrap(text, 150):
|
113 |
+
print (wrp)
|
114 |
+
print ("\n")
|
115 |
+
print ("Summarized Text >>")
|
116 |
+
for wrp in wrap(summarized_text, 150):
|
117 |
+
print (wrp)
|
118 |
+
print ("\n")
|
119 |
+
|
120 |
+
|
121 |
+
|
122 |
+
# answer span extraction
|
123 |
+
|
124 |
+
|
125 |
+
def get_nouns_multipartite(content):
|
126 |
+
out=[]
|
127 |
+
try:
|
128 |
+
extractor = pke.unsupervised.MultipartiteRank()
|
129 |
+
extractor.load_document(input=content,language='en')
|
130 |
+
# not contain punctuation marks or stopwords as candidates.
|
131 |
+
pos = {'PROPN','NOUN'}
|
132 |
+
#pos = {'PROPN','NOUN'}
|
133 |
+
stoplist = list(string.punctuation)
|
134 |
+
stoplist += ['-lrb-', '-rrb-', '-lcb-', '-rcb-', '-lsb-', '-rsb-']
|
135 |
+
stoplist += stopwords.words('english')
|
136 |
+
# extractor.candidate_selection(pos=pos, stoplist=stoplist)
|
137 |
+
extractor.candidate_selection(pos=pos)
|
138 |
+
# 4. build the Multipartite graph and rank candidates using random walk,
|
139 |
+
# alpha controls the weight adjustment mechanism, see TopicRank for
|
140 |
+
# threshold/method parameters.
|
141 |
+
extractor.candidate_weighting(alpha=1.1,
|
142 |
+
threshold=0.75,
|
143 |
+
method='average')
|
144 |
+
keyphrases = extractor.get_n_best(n=15)
|
145 |
+
|
146 |
+
|
147 |
+
for val in keyphrases:
|
148 |
+
out.append(val[0])
|
149 |
+
except:
|
150 |
+
out = []
|
151 |
+
traceback.print_exc()
|
152 |
+
|
153 |
+
return out
|
154 |
+
|
155 |
+
|
156 |
+
|
157 |
+
def get_keywords(originaltext,summarytext):
|
158 |
+
keywords = get_nouns_multipartite(originaltext)
|
159 |
+
print ("keywords unsummarized: ",keywords)
|
160 |
+
keyword_processor = KeywordProcessor()
|
161 |
+
for keyword in keywords:
|
162 |
+
keyword_processor.add_keyword(keyword)
|
163 |
+
|
164 |
+
keywords_found = keyword_processor.extract_keywords(summarytext)
|
165 |
+
keywords_found = list(set(keywords_found))
|
166 |
+
print ("keywords_found in summarized: ",keywords_found)
|
167 |
+
|
168 |
+
important_keywords =[]
|
169 |
+
for keyword in keywords:
|
170 |
+
if keyword in keywords_found:
|
171 |
+
important_keywords.append(keyword)
|
172 |
+
|
173 |
+
return important_keywords[:10]
|
174 |
+
|
175 |
+
|
176 |
+
imp_keywords = get_keywords(text,summarized_text)
|
177 |
+
print (imp_keywords)
|
178 |
+
|
179 |
+
|
180 |
+
|
181 |
+
def get_question(context,answer,model,tokenizer):
|
182 |
+
text = "context: {} answer: {}".format(context,answer)
|
183 |
+
encoding = tokenizer.encode_plus(text,max_length=384, pad_to_max_length=False,truncation=True, return_tensors="pt").to(device)
|
184 |
+
input_ids, attention_mask = encoding["input_ids"], encoding["attention_mask"]
|
185 |
+
|
186 |
+
outs = model.generate(input_ids=input_ids,
|
187 |
+
attention_mask=attention_mask,
|
188 |
+
early_stopping=True,
|
189 |
+
num_beams=5,
|
190 |
+
num_return_sequences=1,
|
191 |
+
no_repeat_ngram_size=2,
|
192 |
+
max_length=72)
|
193 |
+
|
194 |
+
|
195 |
+
dec = [tokenizer.decode(ids,skip_special_tokens=True) for ids in outs]
|
196 |
+
|
197 |
+
|
198 |
+
Question = dec[0].replace("question:","")
|
199 |
+
Question= Question.strip()
|
200 |
+
return Question
|
201 |
+
|
202 |
+
|
203 |
+
|
204 |
+
for wrp in wrap(summarized_text, 150):
|
205 |
+
print (wrp)
|
206 |
+
print ("\n")
|
207 |
+
|
208 |
+
for answer in imp_keywords:
|
209 |
+
ques = get_question(summarized_text,answer,question_model,question_tokenizer)
|
210 |
+
print (ques)
|
211 |
+
print (answer.capitalize())
|
212 |
+
print ("\n")
|
213 |
+
|
214 |
+
|
215 |
+
|
216 |
+
|
217 |
+
# filter keywords
|
218 |
+
|
219 |
+
# paraphrase-distilroberta-base-v1
|
220 |
+
sentence_transformer_model = SentenceTransformer('msmarco-distilbert-base-v3')
|
221 |
+
|
222 |
+
|
223 |
+
|
224 |
+
|
225 |
+
|
226 |
+
def filter_same_sense_words(original,wordlist):
|
227 |
+
filtered_words=[]
|
228 |
+
base_sense =original.split('|')[1]
|
229 |
+
print (base_sense)
|
230 |
+
for eachword in wordlist:
|
231 |
+
if eachword[0].split('|')[1] == base_sense:
|
232 |
+
filtered_words.append(eachword[0].split('|')[0].replace("_", " ").title().strip())
|
233 |
+
return filtered_words
|
234 |
+
|
235 |
+
def get_highest_similarity_score(wordlist,wrd):
|
236 |
+
score=[]
|
237 |
+
for each in wordlist:
|
238 |
+
score.append(normalized_levenshtein.similarity(each.lower(),wrd.lower()))
|
239 |
+
return max(score)
|
240 |
+
|
241 |
+
def sense2vec_get_words(word,s2v,topn,question):
|
242 |
+
output = []
|
243 |
+
print ("word ",word)
|
244 |
+
try:
|
245 |
+
sense = s2v.get_best_sense(word, senses= ["NOUN", "PERSON","PRODUCT","LOC","ORG","EVENT","NORP","WORK OF ART","FAC","GPE","NUM","FACILITY"])
|
246 |
+
most_similar = s2v.most_similar(sense, n=topn)
|
247 |
+
# print (most_similar)
|
248 |
+
output = filter_same_sense_words(sense,most_similar)
|
249 |
+
print ("Similar ",output)
|
250 |
+
except:
|
251 |
+
output =[]
|
252 |
+
|
253 |
+
threshold = 0.6
|
254 |
+
final=[word]
|
255 |
+
checklist =question.split()
|
256 |
+
for x in output:
|
257 |
+
if get_highest_similarity_score(final,x)<threshold and x not in final and x not in checklist:
|
258 |
+
final.append(x)
|
259 |
+
|
260 |
+
return final[1:]
|
261 |
+
|
262 |
+
def mmr(doc_embedding, word_embeddings, words, top_n, lambda_param):
|
263 |
+
|
264 |
+
# Extract similarity within words, and between words and the document
|
265 |
+
word_doc_similarity = cosine_similarity(word_embeddings, doc_embedding)
|
266 |
+
word_similarity = cosine_similarity(word_embeddings)
|
267 |
+
|
268 |
+
# Initialize candidates and already choose best keyword/keyphrase
|
269 |
+
keywords_idx = [np.argmax(word_doc_similarity)]
|
270 |
+
candidates_idx = [i for i in range(len(words)) if i != keywords_idx[0]]
|
271 |
+
|
272 |
+
for _ in range(top_n - 1):
|
273 |
+
# Extract similarities within candidates and
|
274 |
+
# between candidates and selected keywords/phrases
|
275 |
+
candidate_similarities = word_doc_similarity[candidates_idx, :]
|
276 |
+
target_similarities = np.max(word_similarity[candidates_idx][:, keywords_idx], axis=1)
|
277 |
+
|
278 |
+
# Calculate MMR
|
279 |
+
mmr = (lambda_param) * candidate_similarities - (1-lambda_param) * target_similarities.reshape(-1, 1)
|
280 |
+
mmr_idx = candidates_idx[np.argmax(mmr)]
|
281 |
+
|
282 |
+
# Update keywords & candidates
|
283 |
+
keywords_idx.append(mmr_idx)
|
284 |
+
candidates_idx.remove(mmr_idx)
|
285 |
+
|
286 |
+
return [words[idx] for idx in keywords_idx]
|
287 |
+
|
288 |
+
def get_distractors_wordnet(word):
|
289 |
+
distractors=[]
|
290 |
+
try:
|
291 |
+
syn = wn.synsets(word,'n')[0]
|
292 |
+
|
293 |
+
word= word.lower()
|
294 |
+
orig_word = word
|
295 |
+
if len(word.split())>0:
|
296 |
+
word = word.replace(" ","_")
|
297 |
+
hypernym = syn.hypernyms()
|
298 |
+
if len(hypernym) == 0:
|
299 |
+
return distractors
|
300 |
+
for item in hypernym[0].hyponyms():
|
301 |
+
name = item.lemmas()[0].name()
|
302 |
+
#print ("name ",name, " word",orig_word)
|
303 |
+
if name == orig_word:
|
304 |
+
continue
|
305 |
+
name = name.replace("_"," ")
|
306 |
+
name = " ".join(w.capitalize() for w in name.split())
|
307 |
+
if name is not None and name not in distractors:
|
308 |
+
distractors.append(name)
|
309 |
+
except:
|
310 |
+
print ("Wordnet distractors not found")
|
311 |
+
return distractors
|
312 |
+
|
313 |
+
def get_distractors (word,origsentence,sense2vecmodel,sentencemodel,top_n,lambdaval):
|
314 |
+
distractors = sense2vec_get_words(word,sense2vecmodel,top_n,origsentence)
|
315 |
+
print ("distractors ",distractors)
|
316 |
+
if len(distractors) ==0:
|
317 |
+
return distractors
|
318 |
+
distractors_new = [word.capitalize()]
|
319 |
+
distractors_new.extend(distractors)
|
320 |
+
# print ("distractors_new .. ",distractors_new)
|
321 |
+
|
322 |
+
embedding_sentence = origsentence+ " "+word.capitalize()
|
323 |
+
# embedding_sentence = word
|
324 |
+
keyword_embedding = sentencemodel.encode([embedding_sentence])
|
325 |
+
distractor_embeddings = sentencemodel.encode(distractors_new)
|
326 |
+
|
327 |
+
# filtered_keywords = mmr(keyword_embedding, distractor_embeddings,distractors,4,0.7)
|
328 |
+
max_keywords = min(len(distractors_new),5)
|
329 |
+
filtered_keywords = mmr(keyword_embedding, distractor_embeddings,distractors_new,max_keywords,lambdaval)
|
330 |
+
# filtered_keywords = filtered_keywords[1:]
|
331 |
+
final = [word.capitalize()]
|
332 |
+
for wrd in filtered_keywords:
|
333 |
+
if wrd.lower() !=word.lower():
|
334 |
+
final.append(wrd.capitalize())
|
335 |
+
final = final[1:]
|
336 |
+
return final
|
337 |
+
|
338 |
+
sent = "What cryptocurrency did Musk rarely tweet about?"
|
339 |
+
keyword = "Bitcoin"
|
340 |
+
|
341 |
+
# sent = "What did Musk say he was working with to improve system transaction efficiency?"
|
342 |
+
# keyword= "Dogecoin"
|
343 |
+
|
344 |
+
|
345 |
+
# sent = "What company did Musk say would not accept bitcoin payments?"
|
346 |
+
# keyword= "Tesla"
|
347 |
+
|
348 |
+
|
349 |
+
# sent = "What has Musk often tweeted in support of?"
|
350 |
+
# keyword = "Cryptocurrency"
|
351 |
+
|
352 |
+
print (get_distractors(keyword,sent,s2v,sentence_transformer_model,40,0.2))
|
353 |
+
|
354 |
+
|
355 |
+
|
356 |
+
|
357 |
+
context = gr.inputs.Textbox(lines=10, placeholder="Enter paragraph/content here...")
|
358 |
+
output = gr.outputs.HTML( label="Question and Answers")
|
359 |
+
radiobutton = gr.inputs.Radio(["Wordnet", "Sense2Vec"])
|
360 |
+
|
361 |
+
def generate_question(context,radiobutton):
|
362 |
+
summary_text = summarizer(context,summary_model,summary_tokenizer)
|
363 |
+
for wrp in wrap(summary_text, 100):
|
364 |
+
print (wrp)
|
365 |
+
# np = getnounphrases(summary_text,sentence_transformer_model,3)
|
366 |
+
np = get_keywords(context,summary_text)
|
367 |
+
print ("\n\nNoun phrases",np)
|
368 |
+
output=""
|
369 |
+
for answer in np:
|
370 |
+
ques = get_question(summary_text,answer,question_model,question_tokenizer)
|
371 |
+
if radiobutton=="Wordnet":
|
372 |
+
distractors = get_distractors_wordnet(answer)
|
373 |
+
else:
|
374 |
+
distractors = get_distractors(answer.capitalize(),ques,s2v,sentence_transformer_model,40,0.2)
|
375 |
+
# output= output + ques + "\n" + "Ans: "+answer.capitalize() + "\n\n"
|
376 |
+
output = output + "<b style='color:blue;'>" + ques + "</b>"
|
377 |
+
output = output + "<br>"
|
378 |
+
output = output + "<b style='color:green;'>" + "Ans: " +answer.capitalize()+ "</b>"+"<br>"
|
379 |
+
if len(distractors)>0:
|
380 |
+
for distractor in distractors[:4]:
|
381 |
+
output = output + "<b style='color:brown;'>" + distractor+ "</b>"+"<br>"
|
382 |
+
output = output + "<br>"
|
383 |
+
|
384 |
+
summary ="Summary: "+ summary_text
|
385 |
+
for answer in np:
|
386 |
+
summary = summary.replace(answer,"<b>"+answer+"</b>" + "<br>")
|
387 |
+
summary = summary.replace(answer.capitalize(),"<b>"+answer.capitalize()+"</b>")
|
388 |
+
output = output + "<p>"+summary+"</p>"
|
389 |
+
output = output + "<br>"
|
390 |
+
return output
|
391 |
+
|
392 |
+
|
393 |
+
iface = gr.Interface(
|
394 |
+
fn=generate_question,
|
395 |
+
inputs=[context,radiobutton],
|
396 |
+
outputs=output)
|
397 |
+
iface.launch(debug=True)
|