Research and Analysis of IndoBERT Hyperparameter Tuning in Fake News Detection

Abstract

The rapid advancement of communication technology has transformed how information is shared, but it has also brought concerns about the proliferation of false information. A recent report by the Ministry of Communication and Informatics in Indonesia revealed that around 800,000 websites were involved in spreading false information, underscoring the seriousness of the problem. To combat this issue, researchers have focused on developing techniques to detect and combat fake news. This research centers on using IndoBERT-base-p1 for fake news detection and aims to enhance its performance through three methods to tune the hyperparameter value of the model namely: Bayesian optimization, grid search, and random search. After comparing the outcomes of the three hyperparameter tuning methods, Bayesian Optimization emerged as the most effective approach. Achieving a precision of 88.79%, recall of 94.5%, and F1-score of 91.56% for the “fake” label, Bayesian Optimization outperformed the other hyperparameter tuning methods as well as the model using the fine-tuning hyperparameter value. These findings emphasize the importance of hyperparameter tuning in improving the accuracy of fake news detection models. Utilizing Bayesian Optimization and optimizing the specified hyperparameters, the model demonstrated superior performance in accurately identifying instances of fake news, providing a valuable tool in the ongoing battle against disinformation in the digital realm.

Keywords

• fake news
• bert
• indobert
• hyperparameter tuning
• natural language processing