Text Classification for Clickbait Detection: A Model-Driven Approach Using CountVectorizer and ML Classifiers

Abstract

In both digital and mainstream media, clickbait headlines are rampant forms of receiving and distributing attention, designed to maximize click through rates which in turn delegitimize content and help spread misinformation. In this paper we propose an efficient and interpretable machine learning framework for the binary classification of clickbait vs non-clickbait headlines using traditional models. Our pipeline involves the preprocessing step, followed by n-gram feature extraction via CountVectorizer and the classification done using Multinomial Naive Bayes, Logistic Regression and XGBoost. The models were trained and evaluated on accuracy, precision, recall, F1 score and ROC AUC, using a publicly available dataset. Our results indicate that Naive Bayes and Logistic Regression models enjoyed better performance with an accuracy of 95.88% and an F1-score of 95.88% and an AUC of 0.99, performing better than the more complex XGBoost classifier. Confirming the ability of lightweight models for real time clickbait detection, we further show that traditional machine learning is also interpretable and scalable.