tl;dr: How do you identify influencers in a community? How do they rise to, stay in, and fall from power? How does power in one community affect their behaviour across others?
The model of threaded conversation is ubiquitous on the social media platforms that make up our information ecosystem. One user makes a comment/post, which acts as a seed of conversation, drawing in other engaged users, going back-and-forth in a dance that contributes to individual understanding and collective sense-making. In this paper, we formalize this dance as a networked model of interaction: users reply to and engage with other users, with each interaction a weighted, directed edge from one user to another.
The formalization above lends itself to several useful metaphors: misinformation as contagion; polarization as clique formation; adjacent communities as overlapping epistemic networks with shared members. In addition to theoretical advantages pertaining to influence and neighborhoods, it also provides a suite of methodological tools from Network Science to measure the dynamics of these phenomena. Additionally, it provides an empirical setup to apply Kevin Zollman’s model of Network Epistemology for epistemic communities, situating individuals as agents assisting and influencing each other, contributing to the sense-making process.
In this project, by leveraging the topology of interactions, we use the iterative HITS centrality as a measure for user influence. It assigns two scores for each node in the network: an authority score which measures how much attention they receive, and a hub score which measures how good they are at routing attention to high-authority users. Using this, we identify and study ‘power’ users on Reddit for each month from 2016-19, ranked among the top 1% of their subreddits by influence. These users wield disproportionate influence within their large communities, and their identification allows us to examine questions of framing effects led by elites. We measure how the structure of their contributions across multiple subreddits change before they attain power and after. Additionally, to contextualize our findings, we examine the linguistic and semantic differences between their use of language, versus that of their immediate neighbors and other matched users.
In our preliminary results, we find that such power-users exhibit lower entropy in their contributions across subreddits. They become more focused in their contributions once they attain their status, focusing on the communities where their influence has increased and catapulted them to power. That said, the entropy distribution is fat-tailed, with a high number of users who continue to vary their contributions across subreddits. I also build community-level language models to measure the linguistic differences between the user and the community average. In conclusion, I talk about the value of these secondary user-level longitudinal measurements and describe how the data enables further study of the psychosocial phenomena of our interest.