Leveraging Computational Argumentation to Transparently Predict the Quality of Online Information

From anti-vaccination campaigns to political propaganda, misinformation and disinformation are spreading online, causing harm to society.

Contrasting mis- and disinformation is a challenging problem that can (and should) be tackled through human-in-the-loop approaches. Human intelligence and expertise can be employed to decompose the diverse layers of the problem and address them. For example, journalists evaluate the veracity of information through fact-checking, which is a thorough manual activity. However, not all information items require proper fact-checking: for example, opinions should not be treated like statements to be verified, yet their legitimacy, quality, and credibility might be assessed as well. Moreover, whereas statements can be individually fact-checked,  documents resulting from combining several statements should be evaluated as well. Statements might be linked in tendentious ways or be incompletely reported. Because of these considerations, it is crucial that humans take part in the process of quality evaluation of the information online and the problem is decomposed into multiple tasks at different levels of abstraction. When this is not the case, the effectiveness of fact-checking in contrasting misinformation can be limited or questionable (Barrera Rodriguez et al. 2017, Nyhan 2021).

At the same time, the speed and scale of online information call for an automated approach: information is produced at a high speed, and thus requires a fast evaluation. The fact-checking practice mentioned above is usually time-consuming. Similarly, the amount of information to be evaluated exceeds the available human computation resources.

Computational methods can help in automating and scaling the evaluation process. However, because of the sensitivity of the subject, it is important to enhance the transparency and explainability of these computational methods: winning user trust is fundamental to ensure the usefulness of the resulting assessments.

Computational argumentation is the field of research that focuses on mining and reasoning on the arguments in textual items. In this talk, I will outline the link I foresee between computational argumentation and information quality. I plan to seek such a connection in the context of the transparent and explainable use of AI. I will investigate the overall aim of identifying linguistic and logical features in text items that hint at their quality through transparent semi-automated AI pipelines, aiming at explaining multiple aspects of information quality.

Programme