Carnegie Mellon University
Browse

Information Diffusion Over Diverse Social Media Platforms and the Simulated Cross-Platform Impacts of Interventions

Download (42.63 MB)
thesis
posted on 2024-05-24, 18:50 authored by Isabel E. Murdock

 Misinformation spread over social media deepens social and political divisions within American society and countries worldwide. While social media platforms give voice to traditionally underrepresented groups and facilitate social connections, they simultaneously allow conspiracy theory promoters to reach new audiences and malicious actors to spread harmful content and conduct influence operations. Thus, misinformation propagation, whether it occurs inadvertently or results from deliberate disinformation campaigns, is a significant threat to healthy societies and global stability. Prior work has explored various approaches for modeling the spread of misinformation across social media, ranging from epidemiological models to stochastic agent-based ones. However, much of it fails to address an important aspect of the current social media environment. Namely, many users view and share content over multiple social media platforms, each with different social structures, posting mechanisms, and moderation policies. 

By tackling the subject of information diffusion over social media within the context of multi-platform networks, this work aims to 1) provide novel and more realistic models of misinformation spread and 2) reveal effective mechanisms for curbing its propagation. This is accomplished by first providing an empirical analysis of multi-platform content spread over Twitter, Facebook, and Reddit during two U.S. elections and a major U.S. Supreme Court deci?sion. The insights and motivation garnered through this analysis are subsequently used to build and validate two agent-based models reflecting user behavior and interactions on diverse social media platforms. The first model simulates interactions based on Reddit and incorporates the platform’s community-based network structure and decentralized moderation system. The other facilitates information diffusion based on Twitter’s followership network structure and post-sharing mechanisms. 

The thesis culminates by combining the individual platform models into a multi-layer, agent-based model that simulates multi-platform communications and information diffusion. Virtual experiments are performed using the model to study the potential cross-platform impacts of interventions implemented on each platform, including content removal, user banning, and prebunking. These models and their insights contribute to ongoing efforts to develop effective misinformation mitigation strategies and design systems that result in more resilient societies. 

History

Date

2024-05-03

Degree Type

  • Dissertation

Department

  • Electrical and Computer Engineering

Degree Name

  • Doctor of Philosophy (PhD)

Advisor(s)

Osman Yagan Kathleen M. Carley

Usage metrics

    Licence

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC