Cross-Domain Health Misinformation Detection on Indonesian Social Media

Abstract

Indonesia is among the world's most prolific coun- tries in terms of internet and social media usage. Social media serves as a primary platform for disseminating and accessing all types of information, including health-related data. However, much of the content generated on these platforms is unverified and often falls into the category of misinformation, which poses risks to public health. It is essential to ensure the credibility of the information available to social media users, thereby helping them make informed decisions and reducing the risks associated with health misinformation. Previous research on health misinformation detection has predominantly focused on English-language data or has been limited to specific health crises, such as COVID-19. Consequently, there is a need for a more comprehensive approach which not only focus on single issue or domain. This study proposes the development of a new corpus that encompasses various health topics from Indonesian social media. Each piece of content within this corpus will be manually annotated by expert to label a social media post as either misinformation or fact. Additionally, this research involves experimenting with machine learning models, including traditional and deep learning models. Our finding shows that the new cross-domain dataset is able to achieve better performance compared to those trained on the COVID dataset, highlighting the importance of diverse and representative training data for building robust health misinformation detection system.