A Flask-based web app that predicts whether a news article headline is REAL or FAKE from a given URL. It scrapes the page headline, detects and translates non‑English text to English, vectorizes with TF‑IDF, and classifies using Logistic Regression. Authenticated users have their analyzed URLs saved to MySQL.
dataset.csv with columns news and label (0 = REAL, 1 = FAKE).TfidfVectorizer + LogisticRegression./ and sees a single-page interface.<h1> (BeautifulSoup).username and full_name into MySQL.headline, result (“REAL” or “FAKE”), language, and whether it was saved./ (GET): Renders the UI./api/me (GET): Returns current session user./api/signup (POST): Creates user; starts a session./api/signin (POST): Authenticates; starts a session./api/signout (POST): Clears session./detect (POST): Core inference endpoint.secret_key.requests to fetch the URL with a standard user-agent and timeout.BeautifulSoup to extract the first <h1> as the headline.langdetect to detect language of the headline.deep-translator (GoogleTranslator) to translate to English if not English.TfidfVectorizer for text feature extraction.LogisticRegression(max_iter=1000) for binary classification.dataset.csv and kept in memory.users table: stores username, full name, password hash.url_entries table: stores username, full name, and submitted URL.generate_password_hash.username and full_name.templates/index.html: Single-page app with modals and fetch-based API calls.static/styles.css: Modern glassmorphism styles.app.py: Flask app, ML pipeline, routes, DB init, and handlers.templates/index.html: Frontend page with forms, modals, and client JS.static/styles.css: UI styles.dataset.csv: Training data with news,label.requirements.txt: Python dependencies.venv/: (local virtual environment, optional)python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
set FLASK_SECRET_KEY=your_very_secret_key
app.py match your setup
app.py.final_project and required tables if they don’t exist.python app.py
http://127.0.0.1:4000dataset.csv is loaded and the model is trained./api/me
null.
{ "username": "john", "full_name": "John Doe" }
/api/signup (x-www-form-urlencoded or JSON)
username, full_name, password{ ok: true, username, full_name } or { error }/api/signin (x-www-form-urlencoded or JSON)
username, password{ ok: true, username, full_name } or { error }/api/signout
{ ok: true }/detect (x-www-form-urlencoded)
url{
"headline": "Detected or translated headline text",
"result": "REAL",
"language": "en",
"saved": true
}
{ "error": "URL fetch failed: ..." }
{ "error": "No headline found" }
{ "error": "Model prediction failed: ..." }
curl -X POST http://127.0.0.1:4000/detect ^
-H "Content-Type: application/x-www-form-urlencoded" ^
--data-urlencode "url=https://example.com/news-article"
dataset.csv with headers:
news (string): headline or news textlabel (int): 0 = REAL, 1 = FAKETfidfVectorizer → LogisticRegression(max_iter=1000)."REAL" for 0, "FAKE" for 1.app.py.final_project (auto-created if missing).users(id, username UNIQUE, full_name, password_hash, created_at)url_entries(id, username, full_name, url, created_at, INDEX(username))/) with:
/api/me, /api/signup, /api/signin, /api/signout, /detect.dataset.csv exists with columns news,label. Missing/invalid data will be logged on startup.<h1> element. Sites without <h1> may fail.app.py to match your environment.FLASK_SECRET_KEY in production.dataset.csv and restart the app to retrain the model.