Spaces:
Running
Running
File size: 2,696 Bytes
98cc895 9c53030 98cc895 9c53030 98cc895 9c53030 98cc895 9c53030 98cc895 9c53030 98cc895 9c53030 98cc895 9c53030 |
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 |
import nltk
nltk.download('punkt')
import pandas as pd
import gradio as gr
from nltk import sent_tokenize
from transformers import pipeline
from gradio.themes.utils import red, green
detector = pipeline(task='text-classification', model='SJTU-CL/RoBERTa-large-ArguGPT-sent')
color_map = {
'0%': green.c400,
'10%': green.c300,
'20%': green.c200,
'30%': green.c100,
'40%': green.c50,
'50%': None,
'60%': red.c50,
'70%': red.c100,
'80%': red.c200,
'90%': red.c300,
'100%': red.c400
}
def predict_doc(doc):
sents = sent_tokenize(doc)
data = {'sentence': [], 'label': [], 'score': []}
res = []
for sent in sents:
prob = predict_one_sent(sent)
data['sentence'].append(sent)
data['score'].append(round(prob, 4))
if prob < 0.1: label = '0%'
elif prob < 0.2: label = '10%'
elif prob < 0.3: label = '20%',
elif prob < 0.4: label = '30%',
elif prob < 0.5: label = '40%',
elif prob < 0.6: label = '50%',
elif prob < 0.7: label = '60%',
elif prob < 0.8: label = '70%',
elif prob < 0.9: label = '80%',
elif prob < 1: label = '90%'
else: label = '100%'
res.append((sent, label))
if prob <= 0.5:
data['label'].append('Human')
else: data['label'].append('Machine')
df = pd.DataFrame(data)
df.to_csv('result.csv')
overall_score = df.score.mean()
sum_str = ''
if overall_score <= 0.5: overall_label = 'Human'
else: overall_label = 'Machine'
sum_str = f'The essay is probably written by {overall_label}. The probability of being generated by AI is {overall_score}'
# print(sum_str)
return sum_str, res, df, 'result.csv'
def predict_one_sent(sent):
'''
convert to prob
LABEL_1, 0.66 -> 0.66
LABEL_0, 0.66 -> 0.34
'''
res = detector(sent)[0]
org_label, prob = res['label'], res['score']
if org_label == 'LABEL_0': prob = 1 - prob
return prob
with gr.Blocks() as demo:
text_in = gr.Textbox(
lines=5,
label='Essay input',
info='Please enter the essay in the textbox'
)
button = gr.Button('Predict who writes this essay!')
summary = gr.Textbox(lines=1, label='Result summary')
sent_res = gr.HighlightedText(
label = 'Labeled Result'
).style(color_map=color_map)
tab = gr.DataFrame(
label='Table with Probability Score',
max_rows=10
)
csv_f = gr.File(
label='CSV file storing data with all sentences.'
)
button.click(predict_doc, inputs=[text_in], outputs=[summary, sent_res, tab, csv_f], api_name='predict_doc')
demo.launch()
|