Algorithmic models and automated fact-checking. A systematic literature review

Abstract

Automated fact-checking consists of automatically determining the veracity of a claim by applying existing artificial intelligence technologies to classify it into one of the categories commonly used by human fact-checkers (true, misleading, false, etc.). This paper presents the first systematic literature review in Spanish on the evolution of research on this topic. It also aims to analyze the level of accuracy of algorithmic solutions and the impact of published work, using descriptive and inferential statistical treatments (chi-square and Kruskal-Wallis tests). According to our results, the highest volume of contributions was concentrated in the last three years, mainly from the Asian region and United States. Papers proposing integrated algorithmic methods or systems predominate. Studies on linguistic models, which still have several limitations and below-average effectiveness, are in the majority. There is little attention to models based on image analysis, and the presence of fake audio detection algorithms is practically nonexistent. In line with previous work, our study concludes that there are no statistically significant differences in the level of accuracy of the diverse algorithmic models proposed, despite their different degrees of technical complexity.