Utilizing Transformer Models To Detect Vietnamese Fake News on Social Media Platforms

Abstract

The spread of fake news on social media has become a serious issue, leading to misinformation and causing harm to society. This research aims to develop a system for analyzing and classifying Vietnamese fake news using transformer models, with a particular focus on PhoBERT - a version of BERT optimized for Vietnamese. To address this issue, we collected a dataset consisting of Vietnamese posts on the Facebook social media platform and several articles from Vietnamese news sources, covering topics such as lifestyle news, current affairs, and politics. Nevertheless, there are still challenges due to data imbalance between the number of true and false news. The posts were labeled as real or fake, then underwent data preprocessing and were trained using transformer models and PhoBERT for Vietnamese. We also incorporated the TF-IDF and Word2Vec word embedding techniques to optimize the model's performance. To evaluate the performance of the models, we used various evaluation metrics such as Accuracy, Precision, Recall, F1 Score, and AUC. Our results indicate that PhoBERT outperforms other transformer models in detecting Vietnamese fake news, achieving high accuracy and reliability. This paper outlines the background, objectives, methodology, and future research directions, providing a comprehensive overview of the research and its contributions to the field of fake news detection.