app.py

import gradio as gr
import joblib
import pickle
from huggingface_hub import hf_hub_download
from tensorflow.keras.preprocessing.sequence import pad_sequences

# Load the Tokenizer, TfidfVectorizer, and Logistic Regression model
vectorizer_path = hf_hub_download(repo_id="sanaa-11/arabic-fake-news-classification", filename="tfidf_vectorizer (1).joblib")
tokenizer_path = hf_hub_download(repo_id="sanaa-11/arabic-fake-news-classification", filename="tokenizer (1).pkl")
model_path = hf_hub_download(repo_id="sanaa-11/arabic-fake-news-classification", filename="Logistic_Regression_model.joblib")

vectorization = joblib.load(vectorizer_path)
with open(tokenizer_path, "rb") as f:
    tokenizer = pickle.load(f)
model = joblib.load(model_path)

# Define the prediction function
def predict_news(news_text):
    # Preprocess the input text directly using TfidfVectorizer
    tfidf_vector = vectorization.transform([news_text])

    # Predict with the model
    prediction = model.predict(tfidf_vector)
    return "Fake News" if prediction[0] == 0 else "Real News"


# Create Gradio interface
interface = gr.Interface(
    fn=predict_news,
    inputs=gr.Textbox(label="Enter Arabic News Text"),
    outputs=gr.Label(label="Prediction"),
    title="Arabic Fake News Classifier",
    description="Classifies Arabic news articles as Fake or Real."
)

# Launch the app
interface.launch()