Clickbait Detection in Indonesia Headline News Using BERT Ensemble Models

Abstract

This study explores the effectiveness of combining BERT (Bidirectional Encoder Representations from Transformers) with convolutional neural networks (CNN) and multilayer perceptrons (MLP) for a specific task. The results showcase promising performance of the BERT + CNN model, with precision, recall, and F1-score values of 0.91, 0.86, and 0.89, respectively. The BERT + MLP model exhibits consistent performance with an F1-score of 0.875. A comparative analysis against a previous study utilizing IndoBERT highlights the competitive edge of our BERT + CNN model, particularly in terms of recall and F1-score. Additionally, our proposed model demonstrates competitive performance against other state-of-the- art models such as RoBERTa and xlmRoBERTa. This study contributes valuable insights into the optimization of BERT -based models for specific tasks, emphasizing the efficacy of the BERT + CNN architecture.