A Survey on Automated Fact-Checking

Abstract

Fact-checking has become increasingly important due to the speed with which both information and misinformation can spread in the modern media ecosystem. Therefore, researchers have been exploring how fact-checking can be automated, using techniques based on natural language processing, machine learning, knowledge representation, and databases to automatically predict the veracity of claims. In this paper, we survey automated fact-checking stemming from natural language processing, and discuss its connections to related tasks and disciplines. In this process, we present an overview of existing datasets and models, aiming to unify the various definitions given and identify common concepts. Finally, we highlight challenges for future research.

Countering Misinformation: Evidence, Knowledge Gaps, and Implications of Current Interventions

“…Debunking can be fact-based (i.e., correcting a specific misperception) or logic-based, focusing on the epistemic quality of the misinformation or the manipulation techniques used to mislead (Cook et al, 2017;. Some technology companies, notably Meta (formerly Facebook), use debunking, drawing on both automated (Thorne & Vlachos, 2018;Guo et al, 2022) and human-centred methodologies to moderate content on their platforms. Debunking is almost synonymous with fact-checking, but not quite: one can fact-check a story and rate it as true, whereas debunking only pertains to misinformation.…”

“…Automated content labelling typically relies on machine learning and neural network models to automate the content moderation process (Alaphilippe et al, 2019;Gorwa et al, 2020). Although specific techniques vary considerably, the overall aim is to classify content into problem categories (e.g., probable misinformation) or to match uploads against a database of problem content (e.g., known cases of misinformation); see Thorne and Vlachos (2018) and Guo et al (2022) for an overview. Research into the effectiveness of content labels, such as smoking and alcohol warnings, has identified several necessary information processing steps (Conzola & Wogalter, 2001).…”

Overview of the CLEF–2022 CheckThat! Lab on Fighting the COVID-19 Infodemic and Fake News Detection

“…Major research attention has been paid to the social media [63,81]. Within the realm of misinformation and disinformation there are a number of research areas such as identifying the checkworthiness of a claim [74,85], claim detection [30,[35][36][37], fact-checked claims [32,83,96] etc. Shared tasks has also been organized in the last few years, which are similar to CheckThat!…”

On Dimensions of Plausibility for Narrative Information Access to Digital Libraries

“…In contrast to fact-checking, where the task refers to deciding whether some statement is indeed a fact (and thus true) or is rejected because it has no evidence [32,14], narrative structures may be composed out of several statements. Moreover, we do not require a universally valid decision, i.e., based on a given ground truth (a theory in physics for example), we can make a narrative plausible, whereas we cannot make the narrative plausible on a different ground truth.…”