Fake Job Advertisement Detection System

A small Python application that estimates whether a job advertisement may be fraudulent. It uses a scikit-learn text classifier and saves submitted advertisements and results to MongoDB.

Requirements

• Python 3.10 or newer
• MongoDB running locally, or a MongoDB connection URI

Setup

From the project directory, create and activate a virtual environment, then install dependencies:

python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -r requirements.txt

Train the model using the included demonstration dataset:

python train_model.py

Start MongoDB in Docker (the named volume keeps data between container restarts), then run the web app:

docker run -d --name fake-job-detector-mongodb --restart unless-stopped -p 127.0.0.1:27017:27017 -v fake-job-detector-mongo-data:/data/db mongo:6.0
python app.py

If the MongoDB container already exists but is stopped, start it with docker start fake-job-detector-mongodb. Open http://127.0.0.1:5000 in a browser. Use this Flask address rather than VS Code Live Server: Flask serves the page and its prediction endpoint together. The checker accepts pasted text, PNG/JPG/JPEG/WebP images (drag and drop or browse, up to 8 MB), or public HTTP/HTTPS job-post URLs. Selecting an image immediately hides the input dashboard, shows a local image preview, and starts OCR analysis; the image is not uploaded to permanent storage. Results use Critical (55%+ estimated fraud risk), Average (30–54.9%), or Legit (under 30%). These are model risk bands, not a determination of whether an employer is genuinely verified; even the Legit result requires independent verification. URL checks support readable HTML and plain text only; pages that require sign-in, redirect, or return other file types are not supported. For safety, URLs resolving to private/local addresses are blocked, non-standard ports and redirects are not followed, and responses are limited to 2 MB. Extracted OCR/page text and analysis results are saved to MongoDB and can be reviewed in the result. A URL whose hostname does not resolve in public DNS cannot be fetched; check the spelling and public reachability of the address. Install the Tesseract OCR engine on Windows if it is not already installed:

winget install --id UB-Mannheim.TesseractOCR -e

Restart the app after installing Tesseract. If it is installed in a non-standard location, set the TESSERACT_CMD environment variable to its executable path. MongoDB defaults to mongodb://localhost:27017 and database fake_job_detector. Configure MONGODB_URI and MONGODB_DATABASE environment variables to use a different MongoDB instance or database. If saving fails, the app shows an error in the page and logs the underlying MongoDB error to the Flask console.

Training with your own data

Provide a UTF-8 CSV with text and label columns. Labels must be legitimate or fraudulent, and both classes must be present.

python train_model.py --dataset path\to\jobs.csv --output models\fake_job_model.pkl

The included sample is synthetic and intended only to demonstrate the workflow. It is not a representative dataset and should not be used to assess real-world accuracy. Train and evaluate with a sufficiently large, representative, labeled dataset before relying on the classifier.

Tests

python -m unittest discover -s tests