Exploring the Effectiveness of Bi-LSTM in Detecting Indonesian-Language Hoax News

Abstract

This study aims to develop and evaluate a hoax detection model based on Bidirectional Long Short-Term Memory (Bi-LSTM) using a Semi-Supervised Learning approach. In the context of the increasing spread of false information on online platforms, the model is designed to automatically classify news articles as hoaxes or non-hoaxes, even when labeled data is limited. The initial model was trained on a labeled minor dataset and then used to predict labels for an unlabeled major dataset. After combining both datasets, a retraining process was conducted to improve the models generalization to various linguistic styles and sentence structures. Evaluation results show that the model achieved an accuracy of 84%, recall of 76.9%, precision of 70%, and an F1-score of 73.3%. These findings demonstrate that the semi-supervised approach, which combines labeled and unlabeled data, can significantly enhance model performance in hoax detection tasks. This study contributes to the development of an effective and adaptable automated hoax detection system that addresses linguistic challenges in online news texts.