Ninghan CHEN will defend her PhD thesis Decoding the Real World : Tackling Virtual Ethnographic Challenges through Data-Driven Methods on Tuesday, 27 June 2023, 14h00-17h00, Campus Belval in room MNO.1010.
As the Internet has become an integral part of our daily lives, the virtual world, particularly online social networks (OSNs), have evolved into crucial platforms for learning and idea exchange. These online interactions generate a considerable amount of data, reflecting real-world human behaviours and experiences. This leads to the question: how can this wealth of data from the virtual world be utilised to investigate real-world phenomena and challenges? One solution lies in virtual ethnography, an ethnographic approach enhanced by computational tools.
In the past three years, the COVID-19 pandemic has brought to light a unique problem emanating from this abundance of data: the infodemic. This term, referring to an overabundance of information, accurate or not, has compounded the challenges presented by the pandemic. Misinformation and fake news have inundated OSNs, fostering confusion, fear, and harmful behaviours among individuals in the real world.
To combat the infodemic, governments and healthcare bodies have deployed interventions on OSNs. These interventions aim to amplify trustworthy information, control the spread of misinformation and fake news, and understand public sentiment and policy reactions. This thesis provides an exhaustive examination of the infodemic, employing Social Network Analysis (SNA) as the computational tool. It underscores the significance of three SNA applications—social characteristics, information diffusion, and sentiment analysis—in addressing the aforementioned OSN-based interventions. Concerning social characteristics, our objective is to identify users who genuinely contribute to information diffusion. We introduce two novel measures to evaluate the actual performance of individual users and user subgroups in diffusing COVID-19 information. We also shed light on the heightened mental distress experienced by influential users during the COVID-19 pandemic.
In terms of information diffusion, we propose two prediction models. These models consider the content of messages, users’ susceptibility, and influence to estimate a message’s eventual reach and identify users likely to disseminate the message. We validate our models through experiments, and the results indicate that our models surpass existing methods.
For sentiment analysis, we devise a Graph Neural Network-based text classification framework to extract vaccine attitudes from text posts on social media. We use the vaccine attitudes of users’ friends as contextual information to minimise the interference of linguistic nuances like sarcasm. Lastly, to confirm the consistency between virtual world data and real-world phenomena, we conduct a comprehensive cross-validation. This involves comparing virtual and real-world data concerning COVID-19 vaccine hesitancy across different regions and time periods.