|
import os |
|
import gradio as gr |
|
from transformers import pipeline |
|
import spacy |
|
import subprocess |
|
import nltk |
|
from nltk.corpus import wordnet |
|
from spellchecker import SpellChecker |
|
from fastapi import FastAPI |
|
from pydantic import BaseModel |
|
import uvicorn |
|
|
|
|
|
api_app = FastAPI() |
|
|
|
|
|
pipeline_en = pipeline(task="text-classification", model="Hello-SimpleAI/chatgpt-detector-roberta") |
|
|
|
|
|
spell = SpellChecker() |
|
|
|
|
|
nltk.download('wordnet') |
|
nltk.download('omw-1.4') |
|
|
|
|
|
try: |
|
nlp = spacy.load("en_core_web_sm") |
|
except OSError: |
|
subprocess.run(["python", "-m", "spacy", "download", "en_core_web_sm"]) |
|
nlp = spacy.load("en_core_web_sm") |
|
|
|
|
|
class TextRequest(BaseModel): |
|
text: str |
|
|
|
|
|
def predict_en(text): |
|
res = pipeline_en(text)[0] |
|
return res['label'], res['score'] |
|
|
|
|
|
def remove_redundant_words(text): |
|
doc = nlp(text) |
|
meaningless_words = {"actually", "basically", "literally", "really", "very", "just"} |
|
filtered_text = [token.text for token in doc if token.text.lower() not in meaningless_words] |
|
return ' '.join(filtered_text) |
|
|
|
|
|
def capitalize_sentences_and_nouns(text): |
|
doc = nlp(text) |
|
corrected_text = [] |
|
|
|
for sent in doc.sents: |
|
sentence = [] |
|
for token in sent: |
|
if token.i == sent.start: |
|
sentence.append(token.text.capitalize()) |
|
elif token.pos_ == "PROPN": |
|
sentence.append(token.text.capitalize()) |
|
else: |
|
sentence.append(token.text) |
|
corrected_text.append(' '.join(sentence)) |
|
|
|
return '\n'.join(corrected_text) |
|
|
|
|
|
def force_first_letter_capital(text): |
|
sentences = text.split(". ") |
|
capitalized_sentences = [sentence[0].capitalize() + sentence[1:] if sentence else "" for sentence in sentences] |
|
return ". ".join(capitalized_sentences) |
|
|
|
|
|
def correct_tense_errors(text): |
|
doc = nlp(text) |
|
corrected_text = [] |
|
for token in doc: |
|
if token.pos_ == "VERB" and token.dep_ in {"aux", "auxpass"}: |
|
lemma = wordnet.morphy(token.text, wordnet.VERB) or token.text |
|
corrected_text.append(lemma) |
|
else: |
|
corrected_text.append(token.text) |
|
return ' '.join(corrected_text) |
|
|
|
|
|
def correct_singular_plural_errors(text): |
|
doc = nlp(text) |
|
corrected_text = [] |
|
|
|
for token in doc: |
|
if token.pos_ == "NOUN": |
|
if token.tag_ == "NN": |
|
if any(child.text.lower() in ['many', 'several', 'few'] for child in token.head.children): |
|
corrected_text.append(token.lemma_ + 's') |
|
else: |
|
corrected_text.append(token.text) |
|
elif token.tag_ == "NNS": |
|
if any(child.text.lower() in ['a', 'one'] for child in token.head.children): |
|
corrected_text.append(token.lemma_) |
|
else: |
|
corrected_text.append(token.text) |
|
else: |
|
corrected_text.append(token.text) |
|
|
|
return ' '.join(corrected_text) |
|
|
|
|
|
def correct_spelling(text): |
|
words = text.split() |
|
corrected_words = [] |
|
for word in words: |
|
corrected_word = spell.correction(word) |
|
corrected_words.append(corrected_word) |
|
return ' '.join(corrected_words) |
|
|
|
|
|
def rephrase_with_synonyms(text): |
|
doc = nlp(text) |
|
rephrased_text = [] |
|
|
|
for token in doc: |
|
pos_tag = None |
|
if token.pos_ == "NOUN": |
|
pos_tag = wordnet.NOUN |
|
elif token.pos_ == "VERB": |
|
pos_tag = wordnet.VERB |
|
elif token.pos_ == "ADJ": |
|
pos_tag = wordnet.ADJ |
|
elif token.pos_ == "ADV": |
|
pos_tag = wordnet.ADV |
|
|
|
if pos_tag: |
|
synonyms = wordnet.synsets(token.lemma_, pos=pos_tag) |
|
if synonyms: |
|
synonym = synonyms[0].lemmas()[0].name() |
|
if token.pos_ == "VERB": |
|
if token.tag_ == "VBG": |
|
synonym = synonym + 'ing' |
|
elif token.tag_ in {"VBD", "VBN"}: |
|
synonym = synonym + 'ed' |
|
elif token.tag_ == "VBZ": |
|
synonym = synonym + 's' |
|
rephrased_text.append(synonym) |
|
else: |
|
rephrased_text.append(token.text) |
|
else: |
|
rephrased_text.append(token.text) |
|
|
|
return ' '.join(rephrased_text) |
|
|
|
|
|
def paraphrase_and_correct(text): |
|
|
|
cleaned_text = remove_redundant_words(text) |
|
|
|
|
|
paraphrased_text = capitalize_sentences_and_nouns(cleaned_text) |
|
|
|
|
|
paraphrased_text = force_first_letter_capital(paraphrased_text) |
|
|
|
|
|
paraphrased_text = correct_singular_plural_errors(paraphrased_text) |
|
paraphrased_text = correct_tense_errors(paraphrased_text) |
|
|
|
|
|
paraphrased_text = rephrase_with_synonyms(paraphrased_text) |
|
|
|
|
|
paraphrased_text = correct_spelling(paraphrased_text) |
|
|
|
return paraphrased_text |
|
|
|
|
|
@api_app.post("/ai-detection") |
|
async def ai_detection(request: TextRequest): |
|
label, score = predict_en(request.text) |
|
return {"label": label, "score": score} |
|
|
|
|
|
@api_app.post("/paraphrase") |
|
async def paraphrase(request: TextRequest): |
|
corrected_text = paraphrase_and_correct(request.text) |
|
return {"corrected_text": corrected_text} |
|
|
|
|
|
with gr.Blocks() as demo: |
|
with gr.Tab("AI Detection"): |
|
t1 = gr.Textbox(lines=5, label='Text') |
|
button1 = gr.Button("🤖 Predict!") |
|
label1 = gr.Textbox(lines=1, label='Predicted Label 🎃') |
|
score1 = gr.Textbox(lines=1, label='Prob') |
|
|
|
|
|
button1.click(fn=predict_en, inputs=t1, outputs=[label1, score1]) |
|
|
|
with gr.Tab("Paraphrasing & Grammar Correction"): |
|
t2 = gr.Textbox(lines=5, label='Enter text for paraphrasing and grammar correction') |
|
button2 = gr.Button("🔄 Paraphrase and Correct") |
|
result2 = gr.Textbox(lines=10, label='Corrected Text', placeholder="The corrected text will appear here...") |
|
|
|
|
|
button2.click(fn=paraphrase_and_correct, inputs=t2, outputs=result2) |
|
|
|
|
|
demo.launch(share=True) |
|
|
|
|
|
if __name__ == "__main__": |
|
uvicorn.run(api_app, host="0.0.0.0", port=8000) |
|
|