app.py

import base64
import streamlit as st
import numpy as np
import re
import emoji
import os

import nltk
from nltk.tokenize import word_tokenize
from nltk.corpus import stopwords
from nltk.stem import WordNetLemmatizer

# Download necessary resources
nltk.download('punkt_tab')
nltk.download('stopwords')
nltk.download('wordnet')

import tensorflow
import keras
from keras.utils import pad_sequences

import pickle

# Set Streamlit page configuration
st.set_page_config(page_title="News Category Classifier", page_icon="📰", layout="centered")


# Function to set background image
def set_background(image_path):
    if not os.path.exists(image_path):
        st.error(f"❌ Background image not found: {image_path}")
        return

    with open(image_path, "rb") as img_file:
        encoded_img = base64.b64encode(img_file.read()).decode()

    bg_image_style = f"""
    <style>
        body {{
            background-image: url("data:image/jpg;base64,{encoded_img}");
            background-size: cover;
            background-repeat: no-repeat;
            background-position: center;
            background-attachment: fixed;
        }}
    </style>
    """
    st.markdown(bg_image_style, unsafe_allow_html=True)

# Set background image
set_background("Images/News image.jpg")


# Initialize stopwords and lemmatizer
stop_words = set(stopwords.words('english')).union({"pm"})
lemmatizer = WordNetLemmatizer()


# Preprocessing function
def pre_process(x):
    x = x.lower()
    x = re.sub("<.*?>", "", x)  # Remove HTML tags
    x = re.sub("http[s]?://\S+", "", x)  # Remove URLs
    x = re.sub("[@#]\S+", "", x)  # Remove mentions and hashtags
    x = re.sub(r"\_+", " ", x)  # Replace underscores with space
    x = emoji.demojize(x)  # Convert emojis to text
    x = re.sub(":.*?:", "", x)  # Remove emoji text
    x = re.sub("[^a-zA-Z0-9\s_]", "", x)  # Remove special characters
    words = word_tokenize(x)
    words = [word for word in words if word not in stop_words]
    x = " ".join([lemmatizer.lemmatize(word) for word in words])
    return x


# Cache model loading to improve performance
@st.cache_resource
def load_model():
    model_path = "news_model.keras"
    vectorizer_path = "news_tv_model.keras"
    label_encoder_path = "label_encoder.pkl"

    model = keras.models.load_model(model_path)
    vectorizer = keras.models.load_model(vectorizer_path)
    
    with open(label_encoder_path, 'rb') as file:
        label_encoder = pickle.load(file)

    return model, vectorizer, label_encoder


# Load the models
model, vectorizer, label_encoder = load_model()


# Prediction function
def predict_category(text):
    processed_text = [pre_process(text)]
    text_vectorized = pad_sequences(vectorizer(processed_text).numpy().tolist(), padding='pre', maxlen=82)
    prediction = model.predict(text_vectorized)
    category_idx = np.argmax(prediction, axis=1)[0]
    return label_encoder.inverse_transform([category_idx])[0]


# Streamlit UI
st.markdown(
    """
    <style>
    .title {
        color: #ffffff;
        font-size: 2.4em;
        text-align: center;
        font-weight: 700;
        text-transform: uppercase;
        text-shadow: 2px 2px 8px rgba(0, 0, 0, 1.0);
        padding: 10px;
    }
    .subtitle {
        color: #ffff;
        font-size: 1.3em;
        text-align: center;
        font-weight: 600;
        text-shadow: 1px 1px 6px rgba(0, 0, 0, 1.0);
        padding: 5px;
    }
    .classify-button {
        background-color: #3498db;
        color: white;
        font-size: 1.2em;
        padding: 12px 24px;
        border: none;
        border-radius: 8px;
        cursor: pointer;
        display: block;
        margin: 20px auto;
        transition: 0.3s;
    }
    .classify-button:hover {
        background-color: #2980b9;
    }
    .result-box {
        background: linear-gradient(135deg, #6284FF 30%, #FF0000 70%);
        padding: 20px;
        border-radius: 10px;
        text-align: center;
        margin-top: 30px;
        position: relative;
        overflow: hidden;
        border: 2px solid transparent;
        background-clip: padding-box, border-box;
        border-image: linear-gradient(135deg, #6284FF 30%, #FF0000 70%);
        border-image-slice: 0;
        transition: transform 0.3s ease-in-out, box-shadow 0.3s ease-in-out;
    }
    .result-box:hover {
        transform: scale(1.05);
        box-shadow: 0px 10px 30px rgba(98, 132, 255, 0.8),
                    0px 10px 30px rgba(255, 0, 0, 0.8);
    }
    .result-text {
        font-size: 1.8em;
        color: #ffffff;
        font-weight: 900;
        text-shadow: 3px 3px 10px rgba(0, 0, 0, 0.5);
        animation: fadeIn 0.8s ease-in-out;
    }
    </style>
    """,
    unsafe_allow_html=True
)

# Page title
st.markdown("<div class='title'>📰 News Classifier</div>", unsafe_allow_html=True)
st.markdown("<div class='subtitle'>Enter a news headline or article snippet to analyze its category.</div>", unsafe_allow_html=True)

# User input
user_input = st.text_area("Enter text here:", height=150, placeholder="Type your news text here...")

# Button to analyze
if st.button("Analyze 🍿", key="analyze_button"): 
    if user_input.strip():
        category = predict_category(user_input)
        st.markdown(f"<div class='result-box'><span class='result-text'>🗂️ Predicted Category: <strong>{category}</strong></span></div>", unsafe_allow_html=True)
    else:
        st.warning("⚠️ Please enter some text to analyze.")