Bridging the Language Gap: Transformer-Based BERT for Fake News Detection in Low-Resource Settings

Global rise in internet usage and access has caused fake news to become an ever spreading phenomenon. Since fake news is intended to influence public opinion, it has a significant impact on the world. False information spreading has the potential to be extremely harmful and can manipulate the public in many ways. Numerous methods for spotting fake news have been developed to stop its spread. However, it should be noted that this challenge is not only confined to widely spoken languages. Low resource languages face challenges in combating the spread of fake news due to limited semantic and computational constraints. This study demonstrates the use of various BERT models which are considered to be state-of-the-art to classify news as real or fake for a low resource language like Malay, closing the knowledge gap in fake news detection of languages with insufficient linguistics and computational resources. With the help of these experiments, we achieve maximum F1 scores of 86% for mBERT, 80% for XLnet, 87% for IndoBert, 88% for MalayBERT and 84% for mT5 respectively, suggesting their potential in addressing the challenges posed by fake news in Malay. The results of these models are compared with each other to study and draw inferences on their performances. These findings hold significant implications for the development of more robust and language-specific fake news detection systems, contributing to the overall effort to curb the spread of misinformation.