Analysis of Sampling using Data Augmentation on Imbalance Fake News Dataset

During this global pandemic of COVID-19, the proliferation of fake news had devastating effects on the economy and public health. From the origin of the virus, spread, self-medication to hoaxes on vaccination, it created more panic than the fatality of the virus. W.H.O. has coined the term Infodemic describing this state of society. It has raised fear and anxiety among people, which has indirectly provoked sharing and consuming of any and every information related to healthcare at disposal. Thus, the spread of health-related information over digital media has increased manifolds causing an imbalance in the media ecosystem. For better preparedness and control of the situation, it is necessary to mitigate fear among people, disperse fake news, and dispel misinformation. Many contributions were made to the research community to identify fake news without considering the imbalanced nature of the data. This work highlights the imbalanced nature of this data and analyzes approaches to sample fake news using clustering and data augmentation techniques. For data augmentation, an intuitive approach of partial back translation is proposed over direct duplication in oversampling. For this study, a dataset of healthcare news is compiled and published. This work further opens the direction of back translation application for handling imbalanced data experimentation with multiple languages and taking into account syntax and semantics of the language.