A multidisciplinary approach to combat disinformation in online discourse

To address the challenge of combating online disinformation, this thesis proposes an interdisciplinary approach combining network science and computer science methodologies. The contributions of this thesis are the results of parallel stud- ies. One research focus was understanding the dynamics be- hind the online spread of low-quality or unreliable content during major socio-political events. By analyzing social data, we found that factors such as polarized political communi- ties, echo chambers, and offline elements such as specific vot- ing systems are key indicators of significant flows of unreli- able information. Building on these insights, we developed strategies to coun- teract the spread of unreliable content. In this context, we developed (i) a novel methodology to detect echo chambers based on social interactions by modelling the concept of echo chambers with network science models and (ii) an automated system to estimate the trustworthiness of news publishers considering in turn (a) automating experts-designed reliabil- ity criteria using large language models (LLMs), (b) only users interactions within social media discussion and last, (c) an analysis of articles’ textual content to ascertain publisher trust- worthiness. This thesis presents a comprehensive framework that utilizes advanced techniques for analysing online social media dis- course to combat online disinformation. A prototype soft- ware tool, currently under development, aims to implement this framework by providing support to various facets of so- cial media analysis, including early assessments of news pub- lishers’ trustworthiness, evaluation of user behaviors related to the propensity to disseminate untrustworthy content, identification of participants in shared discourse or echo cham- bers, and augmentation of human expert efforts to classify the trustworthiness of previously unclassified news publish- ers.