NoNewsLite

NoNewsLite: Fake News Detection Model

Overview

NoNewsLite is a deep learning model designed to classify news articles as either real or fake. It uses a Bidirectional LSTM with Attention architecture trained on the ISOT Fake News Dataset. The model analyzes text from news titles and content to predict authenticity with high accuracy.

This repository provides:

• A pre-trained model for inference
• Complete code to run predictions locally
• An interactive web interface using Gradio for easy testing

---

How It Works

The model processes input text through these steps:

1. Text Preprocessing: Clean and tokenize input (remove URLs, HTML, stopwords)
2. Embedding: Convert words to numerical vectors
3. Bi-LSTM: Capture sequential patterns in both directions
4. Attention: Highlight important words that influence the prediction
5. Classification: Output probability of being real or fake news

The attention mechanism allows the model to explain its decisions by showing which words contributed most to the final prediction.

---

Requirements

To use this model, install the following packages:

pip install torch nltk gradio scikit-learn pandas numpy

# Run this script to start the demo
import gradio as gr
import torch
import json
import re
from nltk.corpus import stopwords
from nltk.tokenize import word_tokenize

# Load vocabulary
with open("vocab.json", "r") as f:
    word2idx = json.load(f)

# Define model architecture (see full implementation above)
class BiLSTM_Attention(nn.Module):
    # ... (model definition as provided earlier)
    pass

# Load model weights
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = BiLSTM_Attention(vocab_size=len(word2idx)).to(device)
model.load_state_dict(torch.load("pytorch_model.bin", map_location=device))
model.eval()

# Text preprocessing functions
def clean_and_preprocess_text(title, text):
    combined = f"{title} {text}".lower()
    combined = re.sub(r'https?://\S+|www\.\S+', '', combined)
    combined = re.sub(r'<.*?>', '', combined)
    combined = re.sub(r'[^a-zA-Z\s]', '', combined)
    combined = re.sub(r'\s+', ' ', combined).strip()
    tokens = word_tokenize(combined)
    tokens = [word for word in tokens if word not in stopwords.words('english')]
    return ' '.join(tokens)

def text_to_sequence(text, word2idx, max_len=512):
    tokens = text.split()
    seq = [word2idx.get(word, word2idx["<UNK>"]) for word in tokens]
    if len(seq) > max_len:
        seq = seq[:max_len]
    else:
        pad_token = word2idx["<PAD>"]
        seq += [pad_token] * (max_len - len(seq))
    return torch.tensor(seq, dtype=torch.long).unsqueeze(0)

def predict(title, text):
    cleaned = clean_and_preprocess_text(title, text)
    seq = text_to_sequence(cleaned, word2idx)
    seq = seq.to(device)

    with torch.no_grad():
        output, _ = model(seq)
        prob = output.item()
        prediction = "Real News" if prob > 0.5 else "Fake News"
        confidence = prob if prob > 0.5 else 1 - prob

    return f"**Prediction:** {prediction}\n\n**Confidence:** {confidence:.2%}"

# Launch Gradio interface
demo = gr.Interface(
    fn=predict,
    inputs=[
        gr.Textbox(placeholder="Enter news title...", label="Title"),
        gr.Textbox(placeholder="Enter news content...", label="Content", lines=6)
    ],
    outputs=gr.Markdown(label="Result"),
    title="🔍 NoNewsLite: Fake News Detection",
    description="Enter a news article's title and content to detect if it's real or fake using a Bi-LSTM with Attention.",
 
    examples=[
        ["Donald Trump Dies in Car Crash", "Breaking news: Former President Donald Trump has died in a tragic accident."],
        ["U.S. Republicans Flip Fiscal Script", "The Republican party has reversed its long-standing opposition to deficit spending."]
    ],
    theme="default"
)

demo.launch(share=True)  # Generates a public link

 or local inference with

# Example usage
title = "Scientists Confirm Climate Change Is Real"
text = "New study shows global temperatures rising due to greenhouse gas emissions."

result = predict(title, text)
print(result)