app.py

import os
from dotenv import load_dotenv
load_dotenv()

import requests
from bs4 import BeautifulSoup
from newsapi import NewsApiClient
import pandas as pd
import torch
import soundfile as sf
import gradio as gr
from transformers import (
    AutoModelForSequenceClassification, AutoTokenizer, pipeline,
    BartTokenizer, BartForConditionalGeneration,
    MarianMTModel, MarianTokenizer,
    BarkModel, AutoProcessor
)

# -------------------------
# Global Setup and Environment Variables
# -------------------------
NEWS_API_KEY = os.getenv("NEWS_API_KEY")  # Set this in your .env file

device = "cuda:0" if torch.cuda.is_available() else "cpu"

# -------------------------
# News Extraction Functions
# -------------------------
def fetch_and_scrape_news(company, api_key, count=11, output_file='news_articles.xlsx'):
    newsapi = NewsApiClient(api_key=api_key)
    all_articles = newsapi.get_everything(q=company, language='en', sort_by='relevancy', page_size=count)
    articles = all_articles.get('articles', [])
    
    scraped_data = []
    for article in articles:
        url = article.get('url')
        if url:
            scraped_article = scrape_news(url)
            if scraped_article:
                scraped_article['url'] = url
                scraped_data.append(scraped_article)
    
    df = pd.DataFrame(scraped_data)
    df.to_excel(output_file, index=False, header=True)
    print(f"News scraping complete. Data saved to {output_file}")
    return df

def scrape_news(url):
    headers = {"User-Agent": "Mozilla/5.0"}
    response = requests.get(url, headers=headers)
    if response.status_code != 200:
        print(f"Failed to fetch the page: {url}")
        return None
    soup = BeautifulSoup(response.text, "html.parser")
    headline = soup.find("h1").get_text(strip=True) if soup.find("h1") else "No headline found"
    paragraphs = soup.find_all("p")
    article_text = " ".join(p.get_text(strip=True) for p in paragraphs)
    return {"headline": headline, "content": article_text}

# -------------------------
# Sentiment Analysis Setup
# -------------------------
sentiment_model_name = "cross-encoder/nli-distilroberta-base"
sentiment_model = AutoModelForSequenceClassification.from_pretrained(
    sentiment_model_name, 
    torch_dtype=torch.float16,
    device_map="auto"
)
sentiment_tokenizer = AutoTokenizer.from_pretrained(sentiment_model_name)
classifier = pipeline("zero-shot-classification", model=sentiment_model, tokenizer=sentiment_tokenizer)
labels = ["positive", "negative", "neutral"]

# -------------------------
# Summarization Setup
# -------------------------
bart_tokenizer = BartTokenizer.from_pretrained('facebook/bart-large-cnn')
bart_model = BartForConditionalGeneration.from_pretrained('facebook/bart-large-cnn')

def split_into_chunks(text, tokenizer, max_tokens=1024):
    words = text.split()
    chunks = []
    current_chunk = []
    current_length = 0
    for word in words:
        tokenized_word = tokenizer.encode(word, add_special_tokens=False)
        if current_length + len(tokenized_word) <= max_tokens:
            current_chunk.append(word)
            current_length += len(tokenized_word)
        else:
            chunks.append(' '.join(current_chunk))
            current_chunk = [word]
            current_length = len(tokenized_word)
    if current_chunk:
        chunks.append(' '.join(current_chunk))
    return chunks

# -------------------------
# Translation Setup (English to Hindi)
# -------------------------
translation_model_name = 'Helsinki-NLP/opus-mt-en-hi'
trans_tokenizer = MarianTokenizer.from_pretrained(translation_model_name)
trans_model = MarianMTModel.from_pretrained(translation_model_name)

def translate_text(text):
    tokens = trans_tokenizer(text, return_tensors="pt", padding=True)
    translated = trans_model.generate(**tokens)
    return trans_tokenizer.decode(translated[0], skip_special_tokens=True)

# -------------------------
# Bark TTS Setup (Hindi)
# -------------------------
bark_model = BarkModel.from_pretrained("suno/bark-small").to(device)
processor = AutoProcessor.from_pretrained("suno/bark")

# -------------------------
# Main Pipeline Function
# -------------------------
def process_company(company):
    # Step 1: Fetch and scrape news
    fetch_and_scrape_news(company, NEWS_API_KEY)
    df = pd.read_excel('news_articles.xlsx')
    print("Scraped Articles:")
    print(df)
    
    articles_data = []
    for index, row in df.iterrows():
        article_text = row.get("content", "")
        title = row.get("headline", "No title")
        url = row.get("url", "")
        chunks = split_into_chunks(article_text, bart_tokenizer)
        chunk_summaries = []
        for chunk in chunks:
            inputs = bart_tokenizer([chunk], max_length=1024, return_tensors='pt', truncation=True)
            summary_ids = bart_model.generate(inputs.input_ids, num_beams=4, max_length=130, min_length=30, early_stopping=True)
            chunk_summary = bart_tokenizer.decode(summary_ids[0], skip_special_tokens=True)
            chunk_summaries.append(chunk_summary)
        final_summary = ' '.join(chunk_summaries)
        sentiment_result = classifier(final_summary, labels)
        sentiment = sentiment_result["labels"][0]
        
        articles_data.append({
            "Title": title,
            "Summary": final_summary,
            "Sentiment": sentiment,
            "URL": url
        })
    
    # Comparative Analysis: Build a simple sentiment distribution
    sentiment_distribution = {"Positive": 0, "Negative": 0, "Neutral": 0}
    for article in articles_data:
        key = article["Sentiment"].capitalize()
        sentiment_distribution[key] += 1
    
    # Step 2: Translate summaries and generate Hindi speech
    translated_summaries = [translate_text(article["Summary"]) for article in articles_data]
    final_translated_text = "\n\n".join(translated_summaries)
    inputs = processor(final_translated_text, return_tensors="pt").to(device)
    speech_output = bark_model.generate(**inputs)
    audio_path = "final_summary.wav"
    sf.write(audio_path, speech_output[0].cpu().numpy(), bark_model.generation_config.sample_rate)
    
    # Build final report
    report = {
        "Company": company,
        "Articles": articles_data,
        "Comparative Sentiment Score": {
            "Sentiment Distribution": sentiment_distribution,
            "Coverage Differences": "Detailed comparative analysis not implemented",
            "Topic Overlap": "Topic extraction not implemented"
        },
        "Final Sentiment Analysis": "Overall sentiment analysis not fully computed",
        "Audio": audio_path
    }
    return report, audio_path

# Gradio Interface Function
def gradio_interface(company):
    report, audio_path = process_company(company)
    return report, audio_path

# -------------------------
# Gradio UI Setup
# -------------------------
iface = gr.Interface(
    fn=gradio_interface,
    inputs=gr.Textbox(label="Enter Company Name"),
    outputs=[
        gr.JSON(label="News Sentiment Report"),
        gr.Audio(type="filepath", label="Hindi Summary Audio")
    ],
    title="News Summarization & Text-to-Speech",
    description="Enter a company name to fetch news articles, perform sentiment analysis, and listen to a Hindi TTS summary."
)

if __name__ == "__main__":
    iface.launch()