File size: 7,542 Bytes
ea0ff68
 
 
 
 
 
 
 
 
 
 
 
 
7d555a4
4bd2beb
ea0ff68
7d555a4
4bd2beb
 
1623114
a17012f
 
1623114
 
 
ea0ff68
9f940ae
f7beae7
ae7fad7
9f940ae
ea0ff68
fe34c7c
ea0ff68
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
12ce985
16ae58d
12ce985
9f940ae
 
f804b80
a17012f
445f48d
4bd2beb
31b9770
4bd2beb
1623114
0b4e51e
458520f
ea0ff68
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a17012f
4a74247
4bd2beb
 
 
31b9770
4bd2beb
1623114
0b4e51e
458520f
1623114
445f48d
4bd2beb
a17012f
 
 
 
 
 
 
 
4a74247
4201d5b
4bd2beb
31b9770
4bd2beb
 
 
31b9770
4bd2beb
ea0ff68
74c7d7a
ea0ff68
1623114
ea0ff68
 
 
1623114
ea0ff68
1623114
7c811b0
f733e13
02ff828
adb8993
458520f
9f940ae
 
 
f733e13
02ff828
adb8993
458520f
23755c7
9f940ae
ea0ff68
 
 
 
 
 
 
 
 
 
633c00d
ea0ff68
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
import json
import os

import streamlit as st
import pickle

from transformers import AutoTokenizer, BertForSequenceClassification, pipeline
from sklearn.feature_extraction.text import TfidfVectorizer


def load_models():
    st.session_state.loaded = True

    with open('models/tfidf_vectorizer_untrue_inform_detection_tfidf_bg_0.96_F1_score_3Y_N_Q1_082023.pkl', 'rb') as f:
        st.session_state.tfidf_vectorizer_untrue_inf = pickle.load(f)

    with open('models/SVM_model_untrue_inform_detection_tfidf_bg_0.96_F1_score_3Y_N_Q1_082023.pkl', 'rb') as f:
        st.session_state.untrue_detector = pickle.load(f)

    st.session_state.bert_disinfo = pipeline(task="text-classification",
                                     model=BertForSequenceClassification.from_pretrained("usmiva/bert-desinform-bg", num_labels=2),
                                     tokenizer=AutoTokenizer.from_pretrained("usmiva/bert-desinform-bg"))
    st.session_state.bert_gpt = pipeline(task="text-classification",
                                     model=BertForSequenceClassification.from_pretrained("usmiva/bert-deepfake-bg", num_labels=2),
                                     tokenizer=AutoTokenizer.from_pretrained("usmiva/bert-deepfake-bg"))

    st.session_state.emotions = pipeline(task="text-classification",
                                     model=BertForSequenceClassification.from_pretrained("TRACES/emotions", use_auth_token=os.environ['ACCESS_TOKEN2'],  num_labels=11),
                                     tokenizer=AutoTokenizer.from_pretrained("usmiva/bert-web-bg"))



def load_content():
    with open('resource/page_content.json', encoding='utf8') as json_file:
        return json.load(json_file)


def switch_lang(lang):
    if 'lang' in st.session_state:
        if lang == 'bg':
            st.session_state.lang = 'bg'
        else:
            st.session_state.lang = 'en'


if 'lang' not in st.session_state:
    st.session_state.lang = 'bg'

if all([
    'bert_gpt_result' not in st.session_state,
    'untrue_detector_result' not in st.session_state,
    'bert_disinfo_result' not in st.session_state,
    'emotions_result' not in st.session_state
    ]):
    st.session_state.bert_gpt_result = [{'label': '', 'score': 1}]

    st.session_state.untrue_detector_result = ''
    st.session_state.untrue_detector_probability = 1
    
    st.session_state.bert_disinfo_result = [{'label': '', 'score': 1}]
        
    st.session_state.emotions_result = [{'label': '', 'score': 1}]    

content = load_content()
if 'loaded' not in st.session_state:
    load_models()

#######################################################################################################################

st.title(content['title'][st.session_state.lang])

col1, col2, col3 = st.columns([1, 1, 10])
with col1:
    st.button(
        label='EN',
        key='en',
        on_click=switch_lang,
        args=['en']
    )
with col2:
    st.button(
        label='BG',
        key='bg',
        on_click=switch_lang,
        args=['bg']
    )

if 'agree' not in st.session_state:
    st.session_state.agree = False

if st.session_state.agree:
    tab_tool, tab_terms = st.tabs([content['tab_tool'][st.session_state.lang], content['tab_terms'][st.session_state.lang]])

    with tab_tool:
        user_input = st.text_area(content['textbox_title'][st.session_state.lang],
                                  content['text_placeholder'][st.session_state.lang]).strip('\n')
    
        if st.button(content['analyze_button'][st.session_state.lang]):
            st.session_state.bert_gpt_result = st.session_state.bert_gpt(user_input)
            
            user_tfidf_untrue_inf = st.session_state.tfidf_vectorizer_untrue_inf.transform([user_input])
            st.session_state.untrue_detector_result = st.session_state.untrue_detector.predict(user_tfidf_untrue_inf)[0]
            st.session_state.untrue_detector_probability = st.session_state.untrue_detector.predict_proba(user_tfidf_untrue_inf)[0]
            st.session_state.untrue_detector_probability = max(st.session_state.untrue_detector_probability[0], st.session_state.untrue_detector_probability[1]) 

            st.session_state.bert_disinfo_result = st.session_state.bert_disinfo(user_input)
            
            st.session_state.emotions_result = st.session_state.emotions(user_input)

            

        if st.session_state.bert_gpt_result[0]['label'] == 'LABEL_1':
            st.warning(content['bert_gpt'][st.session_state.lang] +
                       str(round(st.session_state.bert_gpt_result[0]['score'] * 100, 2)) +
                       content['bert_gpt_prob'][st.session_state.lang], icon = "⚠️")
        else:
            st.success(content['bert_human'][st.session_state.lang] +
                       str(round(st.session_state.bert_gpt_result[0]['score'] * 100, 2)) +
                       content['bert_human_prob'][st.session_state.lang], icon="✅")
        
        if st.session_state.untrue_detector_result == 0:
            st.warning(content['untrue_getect_yes'][st.session_state.lang] +
                       str(round(st.session_state.untrue_detector_probability * 100, 2)) +
                       content['untrue_yes_proba'][st.session_state.lang], icon="⚠️")
        else:
            st.success(content['untrue_getect_no'][st.session_state.lang] +
                       str(round(st.session_state.untrue_detector_probability * 100, 2)) +
                       content['untrue_no_proba'][st.session_state.lang], icon="✅")
    
        if st.session_state.bert_disinfo_result[0]['label'] == 'LABEL_1':
            st.warning(content['bert_yes_1'][st.session_state.lang] +
                       str(round(st.session_state.bert_disinfo_result[0]['score'] * 100, 2)) +
                       content['bert_yes_2'][st.session_state.lang], icon = "⚠️")
        else:
            st.success(content['bert_no_1'][st.session_state.lang] +
                       str(round(st.session_state.bert_disinfo_result[0]['score'] * 100, 2)) +
                       content['bert_no_2'][st.session_state.lang], icon="✅")

        if st.session_state.emotions_result[0]['score'] < 0.97:
            st.warning(content['emotions_label_1'][st.session_state.lang] + 
                       str(st.session_state.emotions_result[0]['label']) + 
                       content['emotions_label_2'][st.session_state.lang] +
                       str(round(st.session_state.emotions_result[0]['score'] * 100, 2)) + 
                       content['emotions_label_3'][st.session_state.lang] +
                       content['emotions_label_4'][st.session_state.lang], icon = "⚠️")
        else:
            st.info(content['emotions_label_1'][st.session_state.lang] + 
                       str(st.session_state.emotions_result[0]['label']) + 
                       content['emotions_label_2'][st.session_state.lang] +
                       str(round(st.session_state.emotions_result[0]['score'] * 100, 2)) + 
                       content['emotions_label_3'][st.session_state.lang])

    
        st.info(content['disinformation_definition'][st.session_state.lang], icon="ℹ️")

    with tab_terms:
        st.write(content['disclaimer'][st.session_state.lang])

else:
    st.write(content['disclaimer_title'][st.session_state.lang])
    st.write(content['disclaimer'][st.session_state.lang])
    if st.button(content['disclaimer_agree_text'][st.session_state.lang]):
        st.session_state.agree = True
        st.experimental_rerun()