Using Personalisation to Tackle Misinformation and Disinformation on Social Media

Abstract

Social media�s ubiquity in contemporary society is undeniable, serving as a crucial platform for communication, information sharing, and community building. However, this digital landscape is fraught with misinformation�false or misleading information presented as fact�which poses significant risks to societal discourse, public health, and democratic processes. The proliferation of misinformation on social media platforms can distort public perception, influence elections, and under-mine trust in institutions. Traditionally, efforts to mitigate misinformation have primarily adopted a uniform strategy, assuming a homogenous audience. This `one-size-fits-all� approach overlooks the nuanced ways in which individuals consume and respond to information based on their unique characteristics, such as cognitive abilities, political beliefs, and personality traits. Emerging evidence suggests individual factors influence a person�s susceptibility to misinformation, yet the specific ways these traits shape responses to interventions require further investigation, particularly with a focus on developing personalised intervention strategies. The domain of persuasive technologies, which aims to change behaviours or attitudes through persuasion rather than coercion, illustrates the efficacy of personalised approaches. Per-sonalisation tailors� content and strategies to the individual, increasing relevance and effectiveness. Misinformation can be seen as a form of persuasion, so a compelling argument exists for applying personalised strategies to counteract its influence. This perspective identifies a critical research gap and sets the stage for investigating personalised interventions against misinformation. In response to the identified need, this thesis proposes a Personalised Adaptive Interven-tion (PAI) framework. This framework leverages user profiles�incorporating characteristics and be-haviours�to tailor misinformation intervention strategies. The goal is to enhance the efficacy of interventions by making them more relevant and transparent to the targeted users. This framework is intended for application by both social media platforms seeking to curb the spread of misinfor-mation and researchers dedicated to crafting more effective digital interventions. The thesis sets forth to achieve several key objectives: (1) to investigate misinformation trends on social media and analyse user characteristics and behaviours that contribute to vulnera-bility, (2) to review existing interventions targeting online misinformation, and (3) to develop a PAI framework. This framework is (4) iteratively refined and validated, based on empirical evidence, to ensure its effectiveness and adaptability in real-time misinformation mitigation. The final version of the PAI framework presented in this thesis represents the culmination of this research. A mixed-methods approach underpins this research, combining qualitative insights from expert review with quantitative data analysis. This dual methodology enables a deeper understanding of user engage-ment with misinformation and the nuanced preferences for personalised interventions. Quantitative analysis in two user experiments revealed statistically significant differences in user responses to interventions, influenced by their unique characteristics and behaviours. These findings were instrumental in refining the PAI framework, allowing for the development of more adaptable intervention strategies. Finally, a simulation of the PAI framework was conducted to vali-date the higher effectiveness of personalised interventions when compared to a baseline system. This thesis contributes significantly to the field of human computer interaction, misinfor-mation interventions and personalisation. It identifies user characteristics and behaviours that may increase individual�s susceptibility to online misinformation and how reactions to interventions vary among user profiles. This variability suggests that the nature of the intervention encountered can influence trust in interventions. Moreover, the thesis offers substantial practical contributions: the PAI framework is poised to assist social media companies and researchers in crafting and imple-menting personalised digital interventions. This research advances our understanding of digital in-terventions on social media and lays the groundwork for future innovations in digital intervention technologies.