Master's Thesis Ivana Hacajová

Fact-checking has been an open problem in the field of computer science for some time now. In the recent years, the need to be able to verify information has become more important due to the online space being filled with misinformation. There are many fact checking initiatives with people manually verifying claims, but the volume and speed in which new claims emerge, call for an automated solution.

The standard solutions to this problem used to involve BERT-like language models and using Wikipedia as a source of evidence to accept or refute a claim. The most recent development in NLP allow for utilising Large Language Models. Another source of evidence are Knowledge Basis, which can be searched to retrieve relevant information. In this work we explore the ways in which LLMs can improve the performance together with utilising knowledge basis as an additional source of evidence.

Furthermore, most of the recent papers focus only on general, non-domain specific datasets. In this work we compare performance among claims from multiple domains, namely medical and climate related ones. We employ a medical knowledge base to support the evidence.