A Hybrid Machine Learning Model for Clickbait Detection Using Large Language Models and Classical Machine Learning

Clickbait headlines, which are exaggerated or sensationalized to gain attention, are significant problems in the digital media sector since they take undue advantage of consumers. This work delivers a hybrid machine learning approach combining the strengths of Large Language Models (LLM), particularly BERT, with the traditional Machine Learning (ML) techniques for detecting clickbait. The model utilizes embeddings from the pre-trained BERT transformer to store the deep semantic features of the headlines. Then, it combines these with much more prevalent language features like punctuation, word count, and sentence length. The model trains and evaluates the performance of multiple classifiers, Logistic Regression (LogReg), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), K-Nearest Neighbors (KNN), and Multilayer Perceptron (MLP) for the detection of clickbait headlines. Experimental findings indicate that MLP outperforms the other models, with the highest accuracy being 98.29%, thus setting the precedence of transformer-based embeddings when coupled with classical classifiers. The combination dramatically enhances the ability of the model to classify clickbait and nonclickbait headlines, making it more resistant to linguistic variations. This work's innovation utilizes deep contextual embeddings and linguistic signals to offer a richer feature representation for clickbait classification.