ABM for COVID-19 Vaccination Information's Spread on Facebook

  Agent-Based Modeling Applications 

Article 1 - ENSEMBLE OF OPINION DYNAMICS MODELS TO UNDERSTAND 

THE ROLE OF THE UNDECIDED IN THE VACCINATION DEBATE 

This article analyzes opinion dynamics regarding the COVID-19 vaccination campaign, in  terms of pro-vaccine, anti-vaccine, and neutral populations. Leveraging Facebook data, the team  experimented with three different models to better understand the movement of undecided  people into either the pro-vaccine or anti-vaccine groups in terms of the information shared  with them on Facebook. The first model, called the SIS model, is one in which undecided people  are known to be indifferent about the issue, and can become “infected” by either the pro/anti- vaccine factions and they can “recover” if they lose interest in the debate and go back to  neutrality. The second model, called a Voters model, studies a third set of neutral Facebook  pages representing a centrist position. The third is a Bilingual model, which describes a context  where neutral individuals are “in agreement with both pro and anti-vax factions”. The goal is to  better understand the effects of opinion dynamics on the initially static network, modeling the  behavior of those undecided in a news spread (regardless of whether the news is factual or  “fake news”-based). 

    In terms of the models themselves, the researchers use nodes to model the initial groupings in terms of pro, anti, or neutral to vaccination. The agent-based model then analyzes  the spread of information in terms of a neutral node’s awareness of either pro/anti-vaccine information from a Facebook page and how that spread of information tends to “infect” the  neutral population, both based on the spread of information awareness on Facebook as well as  the depth of information available for the pro/anti-vaccine factions on Facebook. This news transmission rate is specifically what is modeled on an individual node-level to better  understand the overall impact on the collective opinion dynamic network in turn. What was  generally found was that the pro-vaccine-related content on Facebook tended to be too far  from the core network of neutral opinions, therefore not impacting as many nodes as quickly to  change their opinions to pro-vaccine as the anti-vaccine-related content was able to. This could  make some sense because those who are neutral about the vaccine may have some underlying  skepticisms about injecting something seemingly new or unproven into their bodies, which may  allow for skeptical anti-vaccine-related content to better resonate with the core of their network  versus pro-vaccine content which might come off as too complicated/gimmicky. 

    There are many advantages to leveraging agent-based models to analyze the spread of  information on Facebook regarding the COVID-19 vaccine. For one, as illustrated by the nodes  showing individuals’ movements from neutral to pro/anti-vaccine, agent-based models excel at  modeling individual behavior changes in the context of the larger collective. As seen in the  figure below from this article, one could theoretically understand and track the different paths  that individuals within the group take starting from neutrality, as well as easily see which type of  information is infecting the middle/majority core of the neutral group itself. This model type  also allows the researchers to effectively model these social interactions and the influence of  communication on changing social behavior as compared to other models. Thus, agent-based  models may be more advantageous than other statistical network models or frameworks such  as game theory because they can incorporate diverse agent aTributes specific to those  individuals (such as their specific beliefs and social connections, which is more representative of  true human opinions and social behaviors), as well as the flexibility to incorporate other 

contextual factors such as communication social norms that arise from analysis in order to  better fine-tune the models dynamically. All-in-all, agent-based models are a really interesting  way to model such a complex and interconnected social opinion dynamic in such a way that is  realistic while still being relatively interpretable from visualizations/summarizations of results.