Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,162 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from predict import run_prediction
|
2 |
+
from io import StringIO
|
3 |
+
import json
|
4 |
+
import gradio as gr
|
5 |
+
import spacy
|
6 |
+
from spacy import displacy
|
7 |
+
from transformers import AutoTokenizer, AutoModelForTokenClassification,RobertaTokenizer,pipeline
|
8 |
+
import torch
|
9 |
+
import nltk
|
10 |
+
from nltk.tokenize import sent_tokenize
|
11 |
+
from fin_readability_sustainability import BERTClass, do_predict
|
12 |
+
import pandas as pd
|
13 |
+
import en_core_web_sm
|
14 |
+
from fincat_utils import extract_context_words
|
15 |
+
from fincat_utils import bert_embedding_extract
|
16 |
+
import pickle
|
17 |
+
lr_clf = pickle.load(open("lr_clf_FiNCAT.pickle",'rb'))
|
18 |
+
|
19 |
+
nlp = en_core_web_sm.load()
|
20 |
+
nltk.download('punkt')
|
21 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
22 |
+
|
23 |
+
#SUSTAINABILITY STARTS
|
24 |
+
tokenizer_sus = RobertaTokenizer.from_pretrained('roberta-base')
|
25 |
+
model_sustain = BERTClass(2, "sustanability")
|
26 |
+
model_sustain.to(device)
|
27 |
+
model_sustain.load_state_dict(torch.load('sustainability_model.bin', map_location=device)['model_state_dict'])
|
28 |
+
|
29 |
+
def get_sustainability(text):
|
30 |
+
df = pd.DataFrame({'sentence':sent_tokenize(text)})
|
31 |
+
actual_predictions_sustainability = do_predict(model_sustain, tokenizer_sus, df)
|
32 |
+
highlight = []
|
33 |
+
for sent, prob in zip(df['sentence'].values, actual_predictions_sustainability[1]):
|
34 |
+
if prob>=4.384316:
|
35 |
+
highlight.append((sent, 'non-sustainable'))
|
36 |
+
elif prob<=1.423736:
|
37 |
+
highlight.append((sent, 'sustainable'))
|
38 |
+
else:
|
39 |
+
highlight.append((sent, '-'))
|
40 |
+
return highlight
|
41 |
+
#SUSTAINABILITY ENDS
|
42 |
+
|
43 |
+
#CLAIM STARTS
|
44 |
+
def score_fincat(txt):
|
45 |
+
li = []
|
46 |
+
highlight = []
|
47 |
+
txt = " " + txt + " "
|
48 |
+
k = ''
|
49 |
+
for word in txt.split():
|
50 |
+
if any(char.isdigit() for char in word):
|
51 |
+
if word[-1] in ['.', ',', ';', ":", "-", "!", "?", ")", '"', "'"]:
|
52 |
+
k = word[-1]
|
53 |
+
word = word[:-1]
|
54 |
+
st = txt.find(" " + word + k + " ")+1
|
55 |
+
k = ''
|
56 |
+
ed = st + len(word)
|
57 |
+
x = {'paragraph' : txt, 'offset_start':st, 'offset_end':ed}
|
58 |
+
context_text = extract_context_words(x)
|
59 |
+
features = bert_embedding_extract(context_text, word)
|
60 |
+
if(features[0]=='None'):
|
61 |
+
highlight.append(('None', ' '))
|
62 |
+
return highlight
|
63 |
+
prediction = lr_clf.predict(features.reshape(1, 768))
|
64 |
+
prediction_probability = '{:.4f}'.format(round(lr_clf.predict_proba(features.reshape(1, 768))[:,1][0], 4))
|
65 |
+
highlight.append((word, ' In-claim' if prediction==1 else 'Out-of-Claim'))
|
66 |
+
# li.append([word,' In-claim' if prediction==1 else 'Out-of-Claim', prediction_probability])
|
67 |
+
else:
|
68 |
+
highlight.append((word, ' '))
|
69 |
+
#headers = ['numeral', 'prediction', 'probability']
|
70 |
+
#dff = pd.DataFrame(li)
|
71 |
+
# dff.columns = headers
|
72 |
+
return highlight
|
73 |
+
|
74 |
+
|
75 |
+
##Summarization
|
76 |
+
summarizer = pipeline("summarization", model="knkarthick/MEETING_SUMMARY")
|
77 |
+
def summarize_text(text):
|
78 |
+
resp = summarizer(text)
|
79 |
+
stext = resp[0]['summary_text']
|
80 |
+
return stext
|
81 |
+
|
82 |
+
|
83 |
+
def split_in_sentences(text):
|
84 |
+
doc = nlp(text)
|
85 |
+
return [str(sent).strip() for sent in doc.sents]
|
86 |
+
def make_spans(text,results):
|
87 |
+
results_list = []
|
88 |
+
for i in range(len(results)):
|
89 |
+
results_list.append(results[i]['label'])
|
90 |
+
facts_spans = []
|
91 |
+
facts_spans = list(zip(split_in_sentences(text),results_list))
|
92 |
+
return facts_spans
|
93 |
+
##Forward Looking Statement
|
94 |
+
fls_model = pipeline("text-classification", model="yiyanghkust/finbert-fls", tokenizer="yiyanghkust/finbert-fls")
|
95 |
+
def fls(text):
|
96 |
+
results = fls_model(split_in_sentences(text))
|
97 |
+
return make_spans(text,results)
|
98 |
+
|
99 |
+
##Company Extraction
|
100 |
+
ner=pipeline('ner',model='Jean-Baptiste/camembert-ner-with-dates',tokenizer='Jean-Baptiste/camembert-ner-with-dates', aggregation_strategy="simple")
|
101 |
+
def fin_ner(text):
|
102 |
+
replaced_spans = ner(text)
|
103 |
+
new_spans=[]
|
104 |
+
for item in replaced_spans:
|
105 |
+
item['entity']=item['entity_group']
|
106 |
+
del item['entity_group']
|
107 |
+
new_spans.append(item)
|
108 |
+
return {"text": text, "entities": new_spans}
|
109 |
+
|
110 |
+
|
111 |
+
#CUAD STARTS
|
112 |
+
def load_questions():
|
113 |
+
questions = []
|
114 |
+
with open('questions.txt') as f:
|
115 |
+
questions = f.readlines()
|
116 |
+
return questions
|
117 |
+
|
118 |
+
|
119 |
+
def load_questions_short():
|
120 |
+
questions_short = []
|
121 |
+
with open('questionshort.txt') as f:
|
122 |
+
questions_short = f.readlines()
|
123 |
+
return questions_short
|
124 |
+
questions = load_questions()
|
125 |
+
questions_short = load_questions_short()
|
126 |
+
def quad(query,file):
|
127 |
+
with open(file.name) as f:
|
128 |
+
paragraph = f.read()
|
129 |
+
questions = load_questions()
|
130 |
+
questions_short = load_questions_short()
|
131 |
+
if (not len(paragraph)==0) and not (len(query)==0):
|
132 |
+
print('getting predictions')
|
133 |
+
predictions = run_prediction([query], paragraph, 'marshmellow77/roberta-base-cuad',n_best_size=5)
|
134 |
+
answer = ""
|
135 |
+
if predictions['0'] == "":
|
136 |
+
answer = 'No answer found in document'
|
137 |
+
else:
|
138 |
+
with open("nbest.json") as jf:
|
139 |
+
data = json.load(jf)
|
140 |
+
for i in range(1):
|
141 |
+
raw_answer=data['0'][i]['text']
|
142 |
+
answer += f"{data['0'][i]['text']} -- \n"
|
143 |
+
answer += f"Probability: {round(data['0'][i]['probability']*100,1)}%\n\n"
|
144 |
+
#summarizer = pipeline("summarization", model="knkarthick/MEETING_SUMMARY")
|
145 |
+
#resp = summarizer(answer)
|
146 |
+
#stext = resp[0]['summary_text']
|
147 |
+
|
148 |
+
# highlight,dff=score_fincat(answer)
|
149 |
+
return answer,summarize_text(answer),fin_ner(answer),score_fincat(answer),get_sustainability(answer),fls(answer)
|
150 |
+
|
151 |
+
|
152 |
+
# b6 = gr.Button("Get Sustainability")
|
153 |
+
#b6.click(get_sustainability, inputs = text, outputs = gr.HighlightedText())
|
154 |
+
|
155 |
+
|
156 |
+
#iface = gr.Interface(fn=get_sustainability, inputs="textbox", title="CONBERT",description="SUSTAINABILITY TOOL", outputs=gr.HighlightedText(), allow_flagging="never")
|
157 |
+
#iface.launch()
|
158 |
+
|
159 |
+
iface = gr.Interface(fn=quad, inputs=[gr.Dropdown(choices=questions,label='SEARCH QUERY'),gr.inputs.File(label='TXT FILE')], title="CONBERT",description="SUSTAINABILITY TOOL",article='Article', outputs=[gr.outputs.Textbox(label='Answer'),gr.outputs.Textbox(label='Summary'),gr.HighlightedText(label='NER'),gr.HighlightedText(label='CLAIM'),gr.HighlightedText(label='SUSTAINABILITY'),gr.HighlightedText(label='FLS')], allow_flagging="never")
|
160 |
+
|
161 |
+
|
162 |
+
iface.launch()
|