Chuai & Zhao (2020), arXiv:2004.10399 (v1 22 April 2020, v3 27 August 2020; cs.SI).
Key takeaways
From the reading, NOTE-852.
- Data. 22,479 fake posts tagged by Weibo’s fact-checking committee after wide spread, and 10,000 true posts from verified accounts, 2011–2016. Emotion is the share of a post’s emotional terms in a hand-labelled five-emotion lexicon of 6,155 terms (87.1% coverage).
- Anger up, joy down. Mean anger share is 0.165 in fake and 0.112 in true news, joy 0.443 against 0.609; the gaps are larger between highly retweeted fake and lowly retweeted true posts, and hold under two classifiers and a rank measure. Disgust, sadness and fear are inconsistent.
- Retweets. In linear models of retweets within 48 hours with controls, anger share has coefficient 23.96 and joy −29.56 among fake posts (R² 0.084), 22.28 and −35.98 among all posts (R² 0.134). The paper’s “5.8 more retweets … in fake news” is the all-news figure; within fake news it is about 5.4.
- Elsewhere. On unlabelled English COVID-19 and election tweets anger predicts reaching 10 retweets (logit 0.215); the English fake/true gaps are small (0.123 against 0.110) or rest on 478 tweets.
- Survey. 1,291 responses to 15 posts; anger-dominated posts (two of them) drew slightly more anxiety-management and information-sharing motive than joy-dominated ones, at p ≈ 0.02–0.04.
- Intervention. Flagging posts above 20% anger would catch 46% of highly retweeted fake news and 22% of highly retweeted true news, on the assumption that a flagged post is never retweeted.
- Design. Observational throughout; “causally” in the text means “with controls in a regression”.
Standing in the record
Seeded on 2026-10-01 from the owner’s reading feed, where it was opened on three separate days, as part of the sweep tracked in the anthology’s issue 180 (https://github.com/dmarx/anthology-of-the-sota/issues/180), and routed here at the owner’s direction: its subject, how emotion goes with the spread of false news among people, is one no anthology topic holds. Read in full on 2026-10-14 in NOTE-852.
The reading supports one account the record lacked, filed as THEORY-329 at the strength the paper earns: an association between anger words and both falsity and retweets on Weibo, with the causal claim marked as not shown. It sits beside the record’s other work on information operations, LIT-083 (sharing disinformation incidents as threat intelligence), which covers the response side where this one covers the spread.