Signed Social Networks: Methods and Applications to Disinformation and Conflict

• In Event: Preprocessing, Annotation, and Interpretation in Computational Social Science

Abstract

A great deal of research in computational political science relies on social interaction graph data to analyze online communities. In most studies, these interactions are treated as identical, even though they may suggest opposite intentions from one user to another. For instance, users commonly mention others not just to endorse their views, but also to criticize them, which can alter how we understand these interactions. It is thus crucial to meaningfully and accurately assign signs to edges in interaction graphs in order to more accurately represent the attitudes and opinions of social network users. Despite recent progress in sentiment analysis and related tools, it remains challenging to infer the stance of online communications with desired accuracy and scale for downstream research. In this project, we propose to empirically examine the existing approaches in order to identify their strengths and limitations (e.g., VADER, Darwish et al. 2019, GPT-3, etc.). We will combine several methods to develop a new technique that works well in most social interaction graph signing problems and produces significant progress in a diverse set of applications, including measuring political conflict and countering disinformation on Twitter during the 2022 US midterm election. Our primary goal is to determine how online debates on issues such as climate change or abortion evolve over time and are influenced by the extent of user interactions and the nature of their stances. Our second goal is to refine disinformation datasets that currently do not distinguish between people spreading vs. opposing disinformation. This should both increase our understanding of how disinformation spreads and improve existing detection methods. Ultimately, the tools and approaches developed here could be used for a variety of other applications to gain a substantially stronger insight into social media interactions, and in turn, to enhance the health of the social media ecosystem.

Authors

• Kellin Pelrine, McGill University

• André Blais, University of Montreal

• Jean-Francois Godbout, Universite de Montreal

• Reihaneh Rabbany, McGill University