app.py

import os
import feedparser
import urllib.parse
import newspaper
import functools
from transformers import pipeline, BartForConditionalGeneration, BartTokenizer
from sentence_transformers import SentenceTransformer, util
from datetime import datetime
from concurrent.futures import ThreadPoolExecutor
import pandas as pd
import time
import sys
import gradio as gr
from flask import Flask, send_file

# Define sentiment analysis pipeline
sentiment_analysis = pipeline("sentiment-analysis", model="ProsusAI/finbert")

# Load Sentence Transformer model
sentence_model = SentenceTransformer("all-MiniLM-L6-v2")

# Load BART model and tokenizer for detailed news summary
bart_model_name = "facebook/bart-large-cnn"
bart_model = BartForConditionalGeneration.from_pretrained(bart_model_name)
bart_tokenizer = BartTokenizer.from_pretrained(bart_model_name)

# Cache for storing fetched articles
article_cache = {}

def fetch_article(url):
    """Fetch article text from URL."""
    if url not in article_cache:
        article = newspaper.Article(url)
        article.download()
        article.parse()
        article_cache[url] = article.text
    return article_cache[url]

def fetch_and_analyze_news_entry(entry, company_name, company_ticker, location):
    """Fetch and analyze sentiment for a single news entry."""
    title = entry.title
    url = entry.link
    domain = urllib.parse.urlparse(url).netloc  # Extract domain from URL
    publishing_date = entry.published_parsed  # Extract publishing date
    
    # Analyze sentiment regardless of article text availability
    try:
        label, score = analyze_sentiment(title)
        sentiment_label = "Positive" if label == "positive" else "Negative" if label == "negative" else "Neutral"
    except Exception as e:
        print(f"Error analyzing sentiment for title: {title}. Error: {e}")
        sentiment_label = "Unknown"
    
    try:
        # Fetch article text using caching
        article_text = fetch_article(url)
    except Exception as e:
        print(f"Error fetching article at URL: {url}. Skipping article.")
        return {
            "title": title,
            "url": url,
            "domain": domain,  # Include domain in the result
            "location": location,  # Include location in the result
            "publishing_date": datetime.fromtimestamp(time.mktime(publishing_date)).strftime("%Y-%m-%d %H:%M:%S"),  # Convert to normal date format
            "sentiment": sentiment_label,
            "detailed_summary": "Paywall Detected",
            "similarity_score": calculate_similarity(company_name, company_ticker, title)  # Calculate similarity based on title
        }
    
    # Generate detailed news summary using BART model
    detailed_summary = news_detailed(article_text)
    
    # Calculate sentence similarity
    similarity_score = calculate_similarity(company_name, company_ticker, title)
    
    return {
        "title": title,
        "url": url,
        "domain": domain,  # Include domain in the result
        "location": location,  # Include location in the result
        "publishing_date": datetime.fromtimestamp(time.mktime(publishing_date)).strftime("%Y-%m-%d %H:%M:%S"),  # Convert to normal date format
        "sentiment": sentiment_label,
        "detailed_summary": detailed_summary,
        "similarity_score": similarity_score
    }

def fetch_and_analyze_news(company_name, company_ticker, event_name, start_date=None, end_date=None, location=None, num_news=5, include_domains=None, exclude_domains=None):
    """Fetch and analyze news entries."""
    # Constructing the Google News RSS feed URL
    query_name = f"{company_name} {event_name} {location}"
    
    # Add date range to the query if start_date and end_date are provided
    if start_date and end_date:
        query_name += f" after:{start_date} before:{end_date}"
    
    # Add domain suggestions and exclusions to the query
    if include_domains:
        include_domains_query = " OR ".join(f"site:{domain.strip()}" for domain in include_domains)
        query_name += f" {include_domains_query}"
    
    if exclude_domains:
        exclude_domains_query = " ".join(f"-site:{domain.strip()}" for domain in exclude_domains)
        query_name += f" {exclude_domains_query}"
    
    encoded_query_name = urllib.parse.quote(query_name)
    rss_url_name = f"https://news.google.com/rss/search?q={encoded_query_name}"
    
    # Parsing the RSS feed for company name
    feed_name = feedparser.parse(rss_url_name)
    news_entries_name = feed_name.entries[:num_news]
    
    analyzed_news_name = []

    # Fetch and analyze news entries for company name in parallel
    with ThreadPoolExecutor() as executor:
        analyze_news_entry_func = functools.partial(fetch_and_analyze_news_entry, company_name=company_name, company_ticker=company_ticker, location=location)
        analyzed_news_name = list(executor.map(analyze_news_entry_func, news_entries_name))
    
    return analyzed_news_name

def news_detailed(article_text, max_length=250):
    """Generate detailed news summary using BART model."""
    inputs = bart_tokenizer([article_text], max_length=max_length, truncation=True, return_tensors="pt")
    summary_ids = bart_model.generate(inputs["input_ids"], num_beams=4, max_length=max_length, length_penalty=2.0, early_stopping=True)
    detailed_summary = bart_tokenizer.decode(summary_ids[0], skip_special_tokens=True)
    return detailed_summary

def calculate_similarity(company_name, company_ticker, title, threshold=0.4):
    """Calculate sentence similarity."""
    company_name_prefix = f"News Regarding {company_name}"
    
    embeddings_company_name = sentence_model.encode([company_name_prefix], convert_to_tensor=True)
    embeddings_title = sentence_model.encode([title], convert_to_tensor=True)

    similarity_score_company_name = util.pytorch_cos_sim(embeddings_company_name, embeddings_title).item()

    weighted_similarity_score = similarity_score_company_name

    return weighted_similarity_score

def analyze_sentiment(title):
    print("Analyzing sentiment...")
    # Perform sentiment analysis on the input title
    result = sentiment_analysis(title)
    # Extract sentiment label and score from the result
    labels = result[0]['label']
    scores = result[0]['score']
    print("Sentiment analyzed successfully.")
    return labels, scores

def find_similar_news_titles(above_threshold_news, similarity_threshold=0.9):
    similar_news = []
    titles = [news['title'] for news in above_threshold_news]
    embeddings = sentence_model.encode(titles, convert_to_tensor=True)
    cosine_scores = util.pytorch_cos_sim(embeddings, embeddings)
    
    for i in range(len(titles)):
        for j in range(i + 1, len(titles)):
            if cosine_scores[i][j] >= similarity_threshold:
                similar_news.append({
                    "title_1": titles[i],
                    "title_2": titles[j],
                    "similarity_score": cosine_scores[i][j].item()
                })
    return similar_news

def fetch_news(company_name, company_ticker, event_name, start_date, end_date, location, num_news, include_domains, exclude_domains):
    analyzed_news_name = fetch_and_analyze_news(company_name, company_ticker, event_name, start_date, end_date, location, num_news, include_domains, exclude_domains)
    
    above_threshold_news = [news for news in analyzed_news_name if news is not None and news['similarity_score'] >= 0.3]
    below_threshold_news = [news for news in analyzed_news_name if news is not None and news['similarity_score'] < 0.3]
    
    similar_news = find_similar_news_titles(above_threshold_news, similarity_threshold=0.9)
    
    above_threshold_df = pd.DataFrame(above_threshold_news)
    below_threshold_df = pd.DataFrame(below_threshold_news)
    similar_news_df = pd.DataFrame(similar_news)
    
    file_name = f"{company_name}_News_Data.xlsx"
    file_path = os.path.join("/tmp", file_name)

    with pd.ExcelWriter(file_path) as writer:
        above_threshold_df.to_excel(writer, sheet_name='Above_Threshold', index=False)
        below_threshold_df.to_excel(writer, sheet_name='Below_Threshold', index=False)
        similar_news_df.to_excel(writer, sheet_name='Similar_News', index=False)
    
    # Return file path to be used in the download link
    return file_path

# Gradio interface setup
inputs = [
    gr.Textbox(label="Company Name"),
    gr.Textbox(label="Company Ticker"),
    gr.Textbox(label="Event Name"),
    gr.Textbox(label="Start Date (optional)"),
    gr.Textbox(label="End Date (optional)"),
    gr.Textbox(label="Location (optional)"),
    gr.Number(label="Number of News to Fetch"),
    gr.Textbox(label="Include Domains (comma-separated)", placeholder="e.g., example.com,example.org"),
    gr.Textbox(label="Exclude Domains (comma-separated)", placeholder="e.g., example.net,example.info")
]

outputs = gr.Markdown(label="Output Message")

interface = gr.Interface(
    fn=fetch_news,
    inputs=inputs,
    outputs=outputs,
    title="Company News Analyzer",
    description="Fetch and analyze news articles based on company and event details."
)

# Flask setup
app = Flask(__name__)

# Route for handling file download
@app.route('/download/<file_name>')
def download(file_name):
    file_path = os.path.join("/tmp", file_name)
    return send_file(file_path, as_attachment=True)

# Route for serving Gradio interface
@app.route('/')
def serve_interface():
    return interface.launch()

if __name__ == "__main__":
    app.run(debug=True)