Tackling Clickbait with Machine Learning: A Comparative Study of Binary Classification Models for YouTube Title

Abstract

Clickbait is a form of internet content whose function is to attract attention and entice users to click on a link. Its main goal generally is to generate more advertisement revenue for the creator. Clickbait is featured heavily on all kinds of social media, especially YouTube where there is a strong financial incentive to do so. Clickbait content constitutes 47.56% of content from mainstream broadcast media and US companies spent an average of 9.8% of their advertising budget on clickbait contents. Clickbait classification is the first and most important step in resolving the proliferation of clickbait content. Contributing to this, we aim to detect YouTube clickbait videos by building several binary classification machine learning models trained on an open-sourced dataset of 31.987 English YouTube video titles from GitHubGist to differentiate between clickbait or non-clickbait YouTube titles. The machine learning models are based on Na¨ıve-Bayes, Support Vector Machine (SVM), and Long Short-Term Memory (LSTM) Network, with the final objective to compare each model's resulting effectiveness. The best-performing resulting model from this study is a kernel TF-IDF SVM model scoring 98.53% on accuracy, precision, recall, and f1-score which outperforms the past experiments that is using the same models.