[论文解读] Disinformation and Misinformation on Twitter during the Novel Coronavirus Outbreak

Binxuan Huang, Kathleen M. Carley|arXiv (Cornell University)|Jun 7, 2020

Misinformation and Its Impacts参考文献 3被引用 52

一句话总结

该研究分析了来自 12 million 用户的 67.4 million COVID-19 相关推文（Jan 29–Mar 4, 2020），以描绘谁在传播错误信息、讨论什么，以及跨国信息流如何。

ABSTRACT

As the novel coronavirus spread globally, a growing public panic was expressed over the internet. We examine the public discussion concerning COVID-19 on Twitter. We use a dataset of 67 million tweets from 12 million users collected between January 29, 2020 and March 4, 2020. We categorize users based on their home countries, social identities, and political orientation. We find that news media, government officials, and individual news reporters posted a majority of influential tweets, while the most influential ones are still written by regular users. Tweets mentioning "fake news" URLs and disinformation story-lines are also more likely to be spread by regular users. Unlike real news and normal tweets, tweets containing URLs pointing to "fake news" sites are most likely to be retweeted within the source country and so are less likely to spread internationally.

研究动机与目标

• Understand which user types (by home country, social identity, and political orientation) are most influential in Twitter discussions about COVID-19.
• Identify who discusses disinformation stories and fake news URLs during the early pandemic period.
• Assess the geographic distribution and international spread of discussions involving low-credibility information.
• Characterize the global network through which disinformation and misinformation spread across countries.

提出的方法

• Collect and filter a 67.4 million tweet dataset (Jan 29–Mar 4, 2020) using COVID-19 related keywords.
• Predict user home country with a high-accuracy location model (92.96% for country; 95.4% identity; 87.4% political orientation).
• Classify user types into seven identities (news media, news reporter, celebrity, government official, sport, company, regular user).
• Curate fake news site lists (black, red, orange) and real news sites; map five disinformation story-lines (bio-weapon, garlic, sesame oil, bleach, chlorine dioxide).
• Analyze tweets containing fake-news URLs and misinfo mentions to assess who tweets them and bot-like behavior (using a bot-detection method with 60% cutoff).
• Examine geographic distributions and inter-country retweet flows, using geotags and country-level location predictions; measure divergence (KL) and entropy across country distributions.

实验结果

研究问题

• RQ1 What types of users send influential (highly retweeted) tweets during the global health emergency?
• RQ2 Who discusses disinformation stories and fake-news URLs, and what are their attributes?
• RQ3 Where in the world do users who discuss low-credibility information come from, and how is this distribution shaped?
• RQ4 What is the global network for information flow about low credibility content across countries?

主要发现

• Regular users (including bot-like ones) produce the majority of influential tweets, with news agencies and government officials contributing substantial but not majority shares.
• About 90% of tweets containing fake-news URLs and misinfo mentions are initiated by regular users, with lower credibility sites driving more bot-like activity.
• Users linking to black/red/orange fake-news sites are disproportionately bots, especially for black sites (e.g., 58.74% bots among regulars citing black URLs).
• The bio-weapon conspiracy is the most widely spread story-line; bleach and chlorine-dioxide claims are also prominent, with many bleach tweets appearing as satire.
• Discussions of fake-news URLs and misinformation are heavily US-centric, with higher normalized probabilities and greater KL-divergence from underlying English-speaking populations than real-news discussions.
• Disinformation topics tend to remain within countries, whereas real-news discussions show higher international spread; WHO and global health authorities helped counter disinformation.
• Geographic distribution and entropy analyses show that misinfo discussions have lower country diversity than real-news discussions, and flows are concentrated from the US to other countries.