Curbing the Spread: An Exploratory Study of the COVID-19 Infodemic in the United States
Author
Windfeld, Andreas Christian
Term
4. term
Publication year
2020
Submitted on
2020-10-07
Pages
80
Abstract
This thesis examines the COVID-19 infodemic — the surge of information, including rumors and false claims — in the United States. We study how people searched for misinformation from January to August 2020 by analyzing queries from Microsoft Bing's Coronavirus intent dataset (2020). We assess how often misinformation-related searches appeared overall and by state, which themes were most common, and what factors might influence these patterns. The analysis combines manual content coding (researchers label queries by hand), supervised machine learning (a model trained on labeled examples to classify text), and exploratory data analysis. To check the robustness of the findings, we compare patterns with Google Trends. Top misinformation-related searches included QAnon, herd immunity, hydroxychloroquine coronavirus, Bill Gates coronavirus, and malaria drugs for coronavirus. Early in the period, searches focused mainly on the origin of the virus. Over time, interest shifted toward miracle cures, alternative treatments, and conspiracy theories. Wyoming showed a notably higher level of misinformation-related searching, both relative to the state's total queries and to its population size. We also observed a recent shift in activity from states with the largest US cities toward more rural states. At the state level, we found no association between misinformation levels and implemented COVID-19 policies or political orientation. Overall, the study maps how misinformation changed during the first months of the pandemic. Future work could test causal relationships, use topic modeling and other unsupervised methods to refine misinformation types, and apply this approach to other countries.
Denne afhandling undersøger COVID-19-infodemi — den store mængde information, herunder rygter og falske påstande — i USA. Vi ser på, hvordan folk søgte efter misinformation fra januar til august 2020 ved at analysere søgninger i Microsoft Bing Coronavirus intent-datasættet (2020). Vi kortlægger, hvor ofte misinformationrelaterede søgninger forekom samlet og på delstatsniveau, hvilke temaer der var mest udbredte, og hvilke faktorer der kan have påvirket mønstrene. Analysen kombinerer manuel indholdskodning (forskere mærker søgninger i hånden), overvåget maskinlæring (en model trænes på mærkede eksempler til at klassificere tekst) og udforskende dataanalyse. For at teste robustheden sammenlignes resultaterne med mønstre i Google Trends. De hyppigste misinformationrelaterede søgninger omfattede QAnon, herd immunity (flokimmunitet), hydroxychloroquine coronavirus, Bill Gates coronavirus og malaria drugs for coronavirus. Tidligt i perioden handlede søgningerne især om virussets oprindelse. Senere flyttede interessen sig mod mirakelmidler, alternative behandlinger og konspirationsteorier. Wyoming havde et markant højere niveau af misinformationrelaterede søgninger både i forhold til statens samlede søgninger og til befolkningsstørrelsen. Vi så også en nylig bevægelse i aktiviteten fra delstater med de største storbyer mod mere landlige delstater. På delstatsniveau fandt vi ingen sammenhæng mellem niveauet af misinformation og gennemførte COVID-19-tiltag eller politisk orientering. Samlet giver studiet et overblik over, hvordan misinformation ændrede sig i de første måneder af pandemien. Fremtidig forskning kan undersøge årsagsforhold, bruge emnemodellering og anden ikke-overvåget læring til at finpudse typer af misinformation og anvende samme tilgang i andre lande.
[This apstract has been rewritten with the help of AI based on the project's original abstract]