Abstract
Due to the increasing amount of information that is available on the Internet (and sometimes, only on the Internet), especially on social media sites, it can even be difficult for experts to differentiate between what is true, false, or deliberately falsified. False claim detection has therefore become a vast research area, with extensive work on finding a concrete and robust solution to the deliberate deception of others with incomplete and/or fabricated information. While manual fact-checking is clearly possible, it is extremely time-consuming and unmanageable on a large scale. The automated analysis of online news and other digital content is important for many reasons, leading to calls for automated fact-checking systems that can verify the truthfulness of statements.Previous work in this area has largely focused on analysing the characteristics of the natural language used, for example, writing style and grammar rules. In contrast, this study looks at automated reasoning, with statements in online texts being checked for consistency with a knowledge base. Natural language processing is used to support this. The author considers that such analysis is best achieved by modelling false claims as claims that are inconsistent with a trusted knowledge base.
Consequently, the fact automated consistency testing (FACT) approach was introduced as a system of checking the consistency of facts in information that is not explicitly mentioned in an extracted text corpus. This approach is based on the following elements: information extraction (from natural language texts); checking claims against a trusted knowledge base, including against relations inferred via a logic system; the incremental building of a trusted knowledge base via a continuous learning technique, and the generation of explanations to facilitate users’ understanding of the system’s decisions. Consistency is checked using probabilistic soft logic (PSL) and a Markov logical network (MLN) for the purpose of comparison.
The results of this study indicate that automatic fact-checking algorithms offer a means of facilitating the detection of false claims, including what is commonly referred to as ‘fake news’. Here, FACT contributes to new knowledge through its ability to check the consistency of claims that are not explicitly mentioned in a text corpus. Thus, FACT permits the veracity of information found on the Web to be checked. The approach was demonstrated by checking facts about family-tree relationships against a corpus of Web resources, relating to the UK royal family and political relations.
| Date of Award | 12 May 2022 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Nello Cristianini (Supervisor) & Jonathan Lawry (Supervisor) |