A Taxonomy of COVID-19 Disinformation
A Taxonomy of COVID-19 Disinformation is a project created during the 2019–2020 Assembly Student Fellowship at the Berkman Klein Center at Harvard University. One of three tracks in the Assembly Program, the Assembly Student Fellowship convenes Harvard students from across a variety of schools and disciplines to tackle the spread and consumption of disinformation. Assembly Student Fellows conducted their work independently with light advisory guidance from faculty advisors and staff.
The taxonomy and this post were authored by Jayshree Sarathy (GSAS ‘24), Sandhira Wijayaratne (HMS ‘20), and Michael Wornow (College ‘20).
With backgrounds in computer science, statistics, and medicine, our Assembly Student Fellow group considered a number of potential project ideas at the intersection of disinformation, technology, and health. However, with the rise of the COVID-19 pandemic and the infodemic surrounding it, we pivoted toward studying coronavirus-related disinformation. As COVID-19’s global footprint increased in size, we felt overwhelmed by the confusing flood of misinformation left in its wake. What content was deemed inauthentic, and who were the authenticators? What were the motivations behind the various disinformation campaigns around the virus? How were stakeholders keeping up with an infodemic growing as fast as the pandemic on which it was based? Our project aimed to explore these questions.
After initial discussions with project advisors, including Dr. Joan Donovan, we decided to create a taxonomy of COVID-19 disinformation to better disaggregate the complexity of the coronavirus infodemic. In collecting content for this project, we also came across various efforts to mitigate COVID-19 disinformation, and we developed our findings into a catalog intended to complement our taxonomy.
Classifying Disinformation Related to COVID-19
The central goal of our project was to identify and classify the various threads of disinformation campaigns surrounding COVID-19 in March and April 2020, as well as the types of responses that stakeholders have pursued to combat this disinformation.
We created two infographics. The first illustrates the current landscape of COVID-19 disinformation and shows how different themes are clustered together; the second details how different actors affected by disinformation have attempted to remedy this issue.
Infographic 1: Landscape of COVID-19 Disinformation
We began by compiling a list of sources covering instances of COVID-19 disinformation in March and April 2020. These were found by mining Google searches, Wikipedia indexes, memes and articles shared on social media, FirstDraft (a database for COVID-19 info), and from Professors Donovan and Jonathan Zittrain, and David O’Brien, Assistant Research Director at the Berkman Klein Center. For each instance of inauthentic content, we noted both the original source and type of the disinformation, actors involved, apparent motivation, and extent of impact. As we compiled pieces of news, we began identifying broadly shared themes and divided them into relevant sub-categories. After analyzing our dataset, we identified two main axes across which disinformation could be classified — the target of the attack and the motivation of the agent behind the attack.
Targets ranged from individuals to communities to nation-states. This categorization was fairly straightforward based on the content of the news articles that we had gathered, which usually identified the target of the disinformation on which they reported.
Motivations were ranked on a spectrum defined by the following two primary goals — (1) monetary or (2) political gain. We categorized the motivations behind each disinformation campaign based on the apparent or most obvious outcome that would result from a target believing the disinformation (which, again, was often identified by the news article itself).
We recognize it is at times difficult, and subjective, to attribute motivations to specific pieces of disinformation. And, we recognize that our dataset is by nature incomplete. However, after inspecting our dataset, we do believe that it’s often possible to glean the central goal of a piece of disinformation based on its context and content, and that uncovering these apparent motives is useful in disaggregating patterns of inauthentic content.
One of the predominant approaches for analyzing disinformation, the ABC framework, was developed by Camille François of Graphika and the Berkman Klein Center. By primarily considering the “Actor,” “Behavior,” and “Content,” this framework does not make motivation a central component of its analysis. However, we believe that motivation is an under-explored piece of the disinformation puzzle. If we can better understand why bad actors are launching information operations and how these varying operations flow through the information ecosystem, then we may be able to better combat them by more accurately targeting their root cause. Our taxonomy is an initial offering; while we started with targets and motivation, further axes of categorization certainly exist to build upon this work.
Finally, in order to give the reader an additional dimension for viewing how the data were clustered, we also labeled the content of each piece of disinformation in our plot.
Using D3.js, a data tool to create interactive visualizations, we made an interactive 2D plot of the news articles that we had collected, and embedded the plot on a website.
Infographic 2: Responses to COVID-19 Disinformation
Once we had created a taxonomy of COVID-19 disinformation, we wanted to explore how this disinformation was being addressed by various stakeholders and communities. We compiled a list of mitigation efforts from March to April 2020, again using online searches, Wikipedia articles, and pointers from our advisors, noting the type of mitigation as well as the key actors. When sifting through these responses, we were not surprised that most of the pushback against the COVID-19 infodemic largely came from professional spheres that fell into four sectors: social media platforms, governments, academia, and the press. While acknowledging these groups bear tight connections to the responsibility of limiting online misinformation, along with the resources to adequately do so, we still hoped to see more representation among civil society groups and individuals — though we acknowledge that our data collection measures may have missed these actors. Sifting through the actions themselves, we noted many coalesced around some form of policy change limiting inauthentic content, greater access to accurate information, or support for journalistic efforts and fact-checking.
In our infographic, created through Visme, we document what we believed to be the most notable of the mitigation efforts in each of these sectors. We found that, generally, while platforms were attempting to limit the spread of “bad” information, academics were trying to improve access to “good information.” And, broadly, while Western governments attempted to utilize administrative tools to provide accurate COVID-19 information (such as the US CDC creating clear videos on the importance of social distancing), some Southeast Asian nations instead pushed punitive laws against platforms that didn’t comply with taking down malicious disinformation. We saw that the media, in addition to covering the threat of COVID-19 disinformation, also aided in fact-checking online material and rumors and databanking this material for use by journalists.
Though our analysis focused on these four main sectors connected to COVID-19 disinformation (social media platforms, governments, academia, and the press), there are other relevant actors, such as industry sellers like Amazon, who fell outside of the scope of our analysis.
Conclusions
While more efforts are needed to understand the COVID-19 infodemic, we hope our project helps contribute to the body of knowledge seeking to analyze the disinformation enveloping this pandemic. This initial taxonomy attempts to categorize COVID-19 disinformation along two critical axes of targets and motivations. At the time we did this preliminary exercise, in March and April 2020, it illustrated to us that most inauthentic content on the scope and origin of COVID-19 clustered around political motives and targeted states, while most medical disinformation was profit-motivated and targeted individuals. When mapping the responses to COVID-19 disinformation, we observed two types of approaches: one approach focused on limiting access to disinformation and increasing access to accurate information, versus an approach with more directive actions, such as restricting access or criminalizing types of content. Examining which approaches were most beneficial to containing the disinformation (and where) will be critical next steps in learning from COVID-19. We hope this project aids in conceptualizing this infodemic in its current stages, and supports thinking on how best to handle future health-related disinformation — and, specifically, how such conceptualizations and approaches may differ from those of political disinformation. As COVID-19 continues to complicate the world’s lived and digital lives, we hope the global community takes collective efforts to halt this pandemic.
For more information on the Taxonomy of COVID-19 Disinformation project, visit the team’s website. Learn more about the Assembly: Disinformation program at www.bkmla.org.
