A comparison of machine learning approaches for predicting ClaimReview markup attributes from fact-checking websites

Abstract

The spreading of fake news is a reality within modern times. However, in the daily fight against disinformation, the fact-checking agencies are one of the strongest allies. Some techniques have been in place to help in this battle, and one of them is the ClaimReview web markup, which had been introduced to grant access to fact-checking articles meaning by search engines. Despite its importance within this context, barely half of the fact-checkers have adopted it. Therefore, in this work, we provide a starting point for the automatic generation of ClaimReview markup, investigating means to predict Claim- Review’s attributes using machine learning models. By experimenting and comparing the baseline approach, Support Vector Machine, with the state-of-the-art (BERT) we have achieved noticeable results, creating a benchmark for upcoming researches in this domain.