app.py

import streamlit as st
import torch
import newspaper
import json

# Load model directly
from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("sofzcc/distilbert-base-uncased-fake-news-checker")
model = AutoModelForSequenceClassification.from_pretrained("sofzcc/distilbert-base-uncased-fake-news-checker")

def extract_news_text(url):
    article = newspaper.Article(url=url, language='en')
    article.download()
    article.parse()

    article ={
        "title": str(article.title),
        "text": str(article.text),
        "published_date": str(article.publish_date),
        "keywords": article.keywords,
        "summary": str(article.summary)
    }

    return article['text']


# Function to predict if news is real or fake
def predict_news(news_text):
    inputs = tokenizer(news_text, return_tensors="pt", truncation=True, padding=True, max_length=512)
    with torch.no_grad():
        outputs = model(**inputs)
    logits = outputs.logits
    predictions = torch.argmax(logits, dim=-1).item()
    return "Real" if predictions == 1 else "Fake"

# Check if a GPU is available and move the model to the appropriate device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

# Define text classification function
def classify_text(text):
    # Tokenize the input text
    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=512)
    inputs = {key: value.to(device) for key, value in inputs.items()}

    # Ensure the model is in evaluation mode
    model.eval()

    # Perform the forward pass
    with torch.no_grad():
        outputs = model(**inputs)

    logits = outputs.logits

    # Convert logits to probabilities
    probabilities = torch.nn.functional.softmax(logits, dim=-1)

    # Get the predicted class
    predicted_class = torch.argmax(probabilities, dim=1).item()

    # Define the label mapping
    labels = ['False', 'True']

    # Return the predicted label and probabilities
    predicted_label = labels[predicted_class]
    probabilities = probabilities.cpu().numpy()

    return predicted_label, probabilities


# Streamlit App
st.title("Fake News Detector")

# Add disclaimer
with st.expander("Disclaimer"):
    st.markdown("""
    
    **Important Notice:**
    
    This model was trained exclusively on news articles from Reuters. As a result, the model may be biased towards considering news from Reuters as "True" and may not accurately classify news from other sources. 
    
    **Usage Warning:**
    
    - This model is intended for experimental and educational purposes only.
    - We do not take any responsibility for the outcomes or decisions made based on the results provided by this model.
    - The model should not be used for any critical or real-world applications, especially those that involve significant consequences or decision-making.
    - Users are encouraged to apply their own judgment and consult multiple sources when evaluating the credibility of news.
    
    **By using this model, you acknowledge and accept these terms and disclaimers.**
    """)

st.write("Enter a news article URL below to check if it's real or fake:")

news_url = st.text_area("News URL", height=100)


if st.button("Evaluate URL"):
    if news_url:
        try:
            news_text = extract_news_text(news_url)
            predicted_label, probabilities = classify_text(news_text)
            st.write(f"The news article is predicted to be: **{predicted_label}**")
        except:
            st.write("It wasn't possible to fetch the article text. Enter the news article text below to check if it's real or fake.")
    else:
        st.write("Please enter some news URL to evaluate.")

st.write("Enter a news article text below to check if it's real or fake:")
news_text = st.text_area("News Text", height=300)
if st.button("Evaluate Text"):
    if news_text:
        try:       
            predicted_label, probabilities = classify_text(news_text)
            st.write(f"The news article is predicted to be: **{predicted_label}**")
        except:
            st.write("It wasn't possible to asses the article text.")            
    else:
        st.write("Please enter some news URL to evaluate.")