Fact-checking ecosystem: from modelling to semi-automated stakeholder-based detection of false information

Humanity has suffered from false, misleading, and malicious information, in short, problematic information, throughout history. Nevertheless, the advances in digital technologies, including the Internet, social media, and smart devices, accelerated and facilitated the dissemination of problematic information. As a complex phenomenon with several aspects to be studied, problematic information remains within many disciplines’ areas of interest. Nevertheless, the lack of a unified understanding of the phenomenon complicates the development of more effective solutions that benefit from multidisciplinary literature to detect problematic information. Focusing on a common type of problematic information, false information, this thesis aims to contribute to filling this gap by presenting a more comprehensive approach to understanding and combating false information. To this end, a topical analysis of the relevant literature was conducted through topic modelling over research publications. This was complemented by the generation of an entity-relationship model of the ecosystem to develop a better conceptual understanding of the landscape and to produce a tool that can be used to guide conceptual analysis of false information-related work and scenarios. Taking a closer look at the proposed model, then, user attitudes and behaviours towards fact-checking tools were examined through an online user survey designed with the help of a new taxonomy of fact-checking tools to collect some empirical evidence for the design of more trustworthy solutions. Most importantly, the conducted survey revealed that semi-automated fact-checking systems, combining human intelligence with automation speed and scalability, can build a higher level of trust among users, compared to manual and fully automated solutions. Therefore, based on the derived conceptual understanding and supported by the obtained empirical evidence, a semi-automated fact-checking tool that is entirely based on expert opinion discovery called aedFaCT was designed, implemented and tested. The presented research in this thesis indicates the potential of semi-automated approaches for the development of more trustworthy and usable fact-checking solutions. The impact of the presented work can be further enhanced by addressing its limitations and incorporating new technological advancements, including large language models (LLMs) and explainable artificial intelligence (xAI).