Towards a Novel Benchmark for Automatic Generation of ClaimReview Markup

Abstract

The spreading of disinformation throughout the web has become a critical problem for a democratic society. The dissemination of fake news has become a profitable business and a common practice among politicians and content producers. On the other hand, journalists and fact-checkers work unceasingly to debunk misinformation and prevent it from further spreading. In 2015, a new web markup called ClaimReview has been introduced to grant access to the fact-checking article's meaning by search engines. It is an important initiative to fight fake news by promoting and highlighting fact-check articles among users. However, barely half of fact-checkers have adopted the ClaimReview markup so far, resulting in low findability of fact-check articles, especially in under-represented countries and languages. In this work, we investigate the viability of using Artificial Intelligence for generating ClaimReview automatically. Besides promoting fact-check articles, the automatic generating of ClaimReview is an important step towards the creation of updated multilingual knowledge base for fighting disinformation. Our experiments show noticeable results, which indicate a viable solution in a production environment. Furthermore, this work has created a benchmark that can be used in upcoming investigations in this domain.