app.py

import streamlit as st
import pandas as pd
import numpy as np
import re
import string
import joblib
import matplotlib.pyplot as plt
import seaborn as sns
from wordcloud import WordCloud
import nltk
from nltk.corpus import stopwords
from sklearn.metrics import silhouette_score

# Download stopwords if not available
nltk.download("stopwords")

# Load models and vectorizer
kmeans = joblib.load("kmeans_fake_news.pkl")
lda = joblib.load("lda_fake_news.pkl")
vectorizer = joblib.load("tfidf_vectorizer.pkl")

# Load dataset
DATASET_URL = "https://www.kaggle.com/datasets/mrisdal/fake-news"
fake_df = pd.read_csv("Fake.csv")

# Preprocessing
stop_words = set(stopwords.words("english"))

def clean_text(text):
    """Cleans the input text by removing punctuation, numbers, and stopwords."""
    text = text.lower()
    text = re.sub(f"[{string.punctuation}]", "", text)  # Remove punctuation
    text = re.sub(r"\d+", "", text)  # Remove numbers
    text = " ".join([word for word in text.split() if word not in stop_words])  # Remove stopwords
    return text

fake_df = fake_df[['title', 'text']].dropna()
fake_df['content'] = fake_df['title'] + " " + fake_df['text']
fake_df['clean_text'] = fake_df['content'].apply(clean_text)

# Transform text into TF-IDF features
X = vectorizer.transform(fake_df['clean_text'])
fake_df['cluster'] = kmeans.predict(X)

# Get top words for LDA topics
words = np.array(vectorizer.get_feature_names_out())  
top_words = [" ".join(words[np.argsort(topic)][-10:]) for topic in lda.components_]

# Sidebar Navigation
st.sidebar.title("Navigation")
page = st.sidebar.radio("Go to", ["Dataset", "Visualizations", "Model Info", "Model Metrics", "Predictor"])

# Model Information Page
if page == "Model Info":
    st.title("Model Information")
    
    st.write("### Machine Learning Models Used")
    st.markdown(
        """
        - **K-Means Clustering**: Used to group fake news articles into clusters based on their content similarity.
        - **Latent Dirichlet Allocation (LDA)**: Used for topic modeling to extract the main topics from fake news articles.
        - **TF-IDF Vectorizer**: Transforms the textual content into numerical features to be used by the models.
        """
    )

# Dataset Page
elif page == "Dataset":
    st.title("Fake News Topic Analyzer")
    
    st.write("### About the Dataset")
    st.markdown(
        """
        The dataset contains **fake news articles** collected from multiple sources. 
        It includes titles, article texts, and publishing dates. 
        We use this dataset for **unsupervised clustering and topic modeling**.
        """
    )
    st.write(f"📂 **Dataset Source:** [Kaggle: Fake News](<{DATASET_URL}>)")
    
    st.write("### Sample Data (Raw)")
    st.dataframe(fake_df[['title', 'text']].head())
    
    st.write("### Sample Data (Cleaned)")
    st.dataframe(fake_df[['clean_text']].head())

    st.write("### Word Cloud of Most Frequent Words")
    wordcloud = WordCloud(width=800, height=400, background_color="white").generate(" ".join(fake_df['clean_text']))
    fig, ax = plt.subplots()
    ax.imshow(wordcloud, interpolation="bilinear")
    ax.axis("off")
    st.pyplot(fig)

# Visualizations Page
elif page == "Visualizations":
    st.title("Fake News Clustering & Topic Modeling")
    
    st.write("### Cluster Distribution")
    fig, ax = plt.subplots()
    sns.countplot(x=fake_df['cluster'], ax=ax, palette="viridis")
    ax.set_xlabel("Cluster")
    ax.set_ylabel("Number of Articles")
    st.pyplot(fig)
    
    st.write("### Topic Words from LDA")
    for idx, words in enumerate(top_words):
        st.write(f"**Topic {idx}:** {words}")

# Model Metrics Page
elif page == "Model Metrics":
    st.title("Model Clustering Performance")
    
    sil_score = silhouette_score(X, fake_df['cluster'])
    st.write(f"### Silhouette Score (K-Means Clustering): **{sil_score:.4f}**")
    
    st.write("### Sample Articles per Cluster")
    for cluster_id in sorted(fake_df['cluster'].unique()):
        st.write(f"#### Cluster {cluster_id} Samples")
        st.dataframe(fake_df[fake_df['cluster'] == cluster_id][['title', 'text']].head(3))

# Predictor Page
elif page == "Predictor":
    st.title("Fake News Topic Analyzer")
    
    user_input = st.text_area("Enter news content:")
    
    if st.button("Analyze"):
        if user_input.strip():
            cleaned_input = clean_text(user_input)
            vectorized_input = vectorizer.transform([cleaned_input])
            cluster_pred = kmeans.predict(vectorized_input)[0]
            topic_pred = np.argmax(lda.transform(vectorized_input))
            
            st.write(f"### Predicted Cluster: {cluster_pred}")
            
            # Handle out-of-range topic index
            if topic_pred < len(top_words):
                st.write(f"### Predicted Topic: {topic_pred} - {top_words[topic_pred]}")
            else:
                st.write(f"### Predicted Topic: {topic_pred} (No keywords available)")