YouTube Clickbait Project

CS121 Spring 2019 Webpage: https://clickbait-prediction.herokuapp.com/

Python script

• clickbait_test_real.py script: We used YouTube API and combined both models to make prediction. Provide the script with the youtube url to analyze. The script will print the model prediction: a number between 1 and 0, 1 is the video is 100% clickbait, 0 means the video is 0% clickbait.

Usage:

clickbait_test_real.py --url URL

• The predict.py script: If you don't want to install YouTube API related dependencies, you can use this script instead. Note that you will need to provide all the video information by yourself.

Usage:

predict.py [-h] --title TITLE [--views VIEWS] [--likes LIKES]
           [--dislikes DISLIKES] [--comments COMMENTS] [--imagepath IMAGEPATH]

JSON files

• credential_sample.json: This file allows the script to work without retrieving the authentication code every time.
• client_secret_3.json: The client_secrets.json file format is a JSON formatted file containing the client ID, client secret, and other OAuth 2.0 parameters. Here is an example client_secrets.json file for a web application:

Models

• Metada model: saved as svm. It's a support vector machine model based on alessiovierti/youtube-clickbait-detector
• Thumbnail model: saved as models/clickbait-model-2.pthR. It's a ResNet34 model, trained with fastai library.

Web app

• app.py: A Flask application to connect front-end and back-end.
• clickbaitHome.html: HTML of the home web page.