Networks All the Way Down: Assessing Modeling Choices for Political Conversation. Workshop by Sarah Shugars, September 22, 2021.
Political conversations, whether online or in person, are networked along multiple dimensions: people come into contact with each other through social networks, they spread messages and ideas using semantic networks, and conversational interactions themselves form a network of back-and-forth exchange. Each of these networked dimensions can be valuable in understanding the political implications of discourse and for developing appropriate interventions around the spread of misinformation and toxic speech. Yet it is rarely practical or meaningful to consider all of these networks simultaneously. Indeed, most studies focus on a single type of social, semantic, or conversational network and make explicit choices about the content of interest and the types of relationships examined. Research on Twitter, for example, may consider social networks formed by follower relationships, semantic networks formed by hashtag co-occurrence, or conversational networks of replies and interactions. Each of these networks is meaningful in its own right, but only captures a piece of the larger public discourse. This paper therefore examines the network modeling choices researchers must make when studying political conversations. Using diverse corpora including Twitter exchanges, Reddit threads, and U.S. Congressional debates, we present a framework for modeling the social, semantic, and conversational networks of political discourse in a range of contexts. We illustrate what can and cannot be inferred from individual network models, and assess the sensitivity of findings to various modeling choices. Ultimately, this paper presents a roadmap to assist researchers in identifying the network models most appropriate for different research questions related to political discourse.
Sarah Shugars is a computational political scientist, studying American political behavior and developing new methods in natural language processing, network analysis, and machine learning.
The workshop materials are available here.