Source-linked AI summary
Assessing the risks of "infodemics" in response to COVID-19 epidemics
Riccardo Gallotti, Francesco Valle, Nicola Castaldo, Pierluigi Sacco, Manlio De Domenico
TL;DR
Assessing how infodemic risk interacted with epidemic dynamics during COVID-19 called for rigorous evidence. The study analyzed COVID-19 tweets and classified sources with a Harm Score, finding that misinformation trajectories around epidemic onset differed between Italy and the United States.
Problem
The real-time interaction between infodemic and epidemic dynamics during COVID-19 required rigorous, evidence-based assessment.
Source: p. 5Method
The authors monitored COVID-19 tweets and classified sources by their potential contribution to infodemic manipulation and misinformation using a Harm Score.
Source: Data collection · p. 20, CONSPIRACY/JUNK SCI · p. 22Results
Italy’s misinformation production collapsed after the epidemic outbreak, while misinformation in the United States remained basically constant.
Source: p. 12Takeaways & Limitations
The findings frame epidemic and infodemic dimensions as two sides of the same coin, with reliable information at all scales a key resource for mitigating adverse effects.
Source: p. 14The exposure measure is a lower bound because it does not track higher-order transmission pathways or users who read content without resharing it.
Source: p. 7
Abstract
from arXiv · showhide
Our society is built on a complex web of interdependencies whose effects become manifest during extraordinary events such as the COVID-19 pandemic, with shocks in one system propagating to the others to an exceptional extent. We analyzed more than 100 millions Twitter messages posted worldwide in 64 languages during the epidemic emergency due to SARS-CoV-2 and classified the reliability of news diffused. We found that waves of unreliable and low-quality information anticipate the epidemic ones, exposing entire countries to irrational social behavior and serious threats for public health. When the epidemics hit the same area, reliable information is quickly inoculated, like antibodies, and the system shifts focus towards certified informational sources. Contrary to mainstream beliefs, we show that human response to falsehood exhibits early-warning signals that might be mitigated with adequate communication strategies.
Methods
The study collected COVID-19 tweets worldwide, classified accounts as human or non-human, and mapped fact-checking databases’ domain labels into a unified reliability scheme.
Data collection
- 112.6 million of an estimated 161.2 million tweets posted through 10 March 2020 were collected using Twitter’s Filter and Search APIs and COVID-19-related terms.Collection began with Search API data on 21 January and continued in real time with the Filter API from 24 January; coverage exceeded 99.5% through 24 February, before message-volume restrictions limited collection.Source: Data collection · pp. 19–20
Human vs non-human classification
- The deep-learning method classified users as human or non-human, with reported accuracy above 90% and bot-identification precision above 95%.The method was described as stable when classifying broadcaster accounts; verified status was distinguished from human/non-human status.Source: Human vs non-human classification · p. 20
Fact Checking
- Manually checked domains from multiple public databases were mapped to a unified classification scheme, including a coarse reliable-versus-unreliable distinction.The source lists included scientific and journalistic databases, such as Washington Post, BuzzFeed News, PolitiFact, Bufale.net, and MediaBiasFactCheck; their differing labeling schemes motivated the mapping.Source: Fact Checking · pp. 20–21
MAINSTREAM MEDIA
The section classifies linked domains by reliability, distinguishing professionally accountable sources from unreliable content categories and domains that remain unknown.
MAINSTREAM MEDIA
- Reliable domains generally provide professionally fact-checked content and abide by media-accountability rules.Source: MAINSTREAM MEDIA · p. 21
- Satire, clickbait, political, and fake/hoax domains are classified as unreliable for humorous distortion, attention-seeking misrepresentation, partisan framing, or fabricated manipulation.Satire is explicitly intended as humor or social critique; fake/hoax content aims to mislead public opinion and provoke inflammatory responses.Source: MAINSTREAM MEDIA · pp. 21–22
- Other domains with hard-to-classify general content and unsuccessfully unshortened URL-shortening domains are categorized as unknown.YouTube videos are an example of general content in the “other” category.Source: MAINSTREAM MEDIA · p. 21
CONSPIRACY/JUNK SCI
The section defines conspiracy/junk science as the Harm Score’s most severe category for manipulative content opposing tested knowledge, and describes the source-classification and filtering process.
CONSPIRACY/JUNK SCI
- 4,988 domains became 3,892 final entries after duplicate resolution, expert harm-score classification, and filtering of poorly defined or unclassifiable domains.Hard duplicates reduced the set to 4,417; resolving soft duplicates by retaining the classification with the highest Harm Score reduced it to 3,920 before the final filtering step.Source: CONSPIRACY/JUNK SCI · pp. 22–23
- The Harm Score rises with systematic, intentional knowledge manipulation: science and mainstream media rank lowest, while fake/hoax and conspiracy/junk science rank highest.Satire is generally identifiable as unreliable; clickbait ranks higher because it presents fabricated or distorted information as true for commercial attention, while shadow and political content rank above general-purpose other content.Source: CONSPIRACY/JUNK SCI · pp. 22–23
- Conspiracy/junk science receives the highest Harm Score for ideological, inflammatory content that opposes tested knowledge and builds self-referential bubbles; it can mimic scientific reasoning and target alleged conspirators.The category describes sources that legitimize implausible accounts through arguments that coarsely imitate scientific reasoning.Source: CONSPIRACY/JUNK SCI · pp. 22–23
Data availability
- Aggregated, privacy-compliant information is publicly available through the Infodemics Observatory and a permanent repository; full datasets are available from the corresponding author on reasonable request.The passage says the Zenodo address/DOI will be provided with publication of the manuscript.Source: Data availability · p. 24