Source-linked AI summary
Good Friends, Bad News - Affect and Virality in Twitter
Lars Kai Hansen, Adam Arvidsson, Finn Årup Nielsen, Elanor Colleoni, Michael Etter
TL;DR
The paper examines whether affect influences Twitter virality differently in social and news communication. It analyzes classified news and non-news tweets using sentiment measures and retweet models, finding that negative sentiment supports news retweeting but not generic non-news diffusion.
Problem
Prior work linked positive affect with sharing, whereas classic news-diffusion theory emphasizes negative affect, leaving their applicability across Twitter’s social and news segments unresolved.
Method
The study analyzes COP15, random Twitter, and Brown Corpus data using a Naive Bayes news classifier, sentiment scoring, and generalized linear retweet models.
Results
Negative sentiment enhances retweeting in news tweets, while generic tweets show no such promotion and aroused non-news tweets favor positive content.
Takeaways & Limitations
Affect and virality have a content-dependent relationship: positive sentiment supports social diffusion, whereas negativity supports news propagation.
Abstract
from arXiv · showhide
The link between affect, defined as the capacity for sentimental arousal on the part of a message, and virality, defined as the probability that it be sent along, is of significant theoretical and practical importance, e.g. for viral marketing. A quantitative study of emailing of articles from the NY Times finds a strong link between positive affect and virality, and, based on psychological theories it is concluded that this relation is universally valid. The conclusion appears to be in contrast with classic theory of diffusion in news media emphasizing negative affect as promoting propagation. In this paper we explore the apparent paradox in a quantitative analysis of information diffusion on Twitter. Twitter is interesting in this context as it has been shown to present both the characteristics social and news media. The basic measure of virality in Twitter is the probability of retweet. Twitter is different from email in that retweeting does not depend on pre-existing social relations, but often occur among strangers, thus in this respect Twitter may be more similar to traditional news media. We therefore hypothesize that negative news content is more likely to be retweeted, while for non-news tweets positive sentiments support virality. To test the hypothesis we analyze three corpora: A complete sample of tweets about the COP15 climate summit, a random sample of tweets, and a general text corpus including news. The latter allows us to train a classifier that can distinguish tweets that carry news and non-news information. We present evidence that negative sentiment enhances virality in the news segment, but not in the non-news segment. We conclude that the relation between affect and virality is more complex than expected based on the findings of Berger and Milkman (2010), in short 'if you want to be cited: Sweet talk your friends or serve bad news to the public'.
1 Introduction
The paper frames Twitter virality as a tension between positive-affect sharing among social contacts and negative-affect propagation in news media. Because Twitter combines social-network and information-sharing features, it asks whether newsness changes how sentiment influences retweeting.
- Prior research: Berger and Milkman report that positive content is more viral in New York Times sharing, while some anger or anxiety also increases sharing.Their findings were based on 6,956 articles and controlled for surprise, interest, usefulness, and article prominence.
- Motivation: Classic news-diffusion theory instead emphasizes negative affect as promoting the selection and propagation of news.This creates the paper’s central apparent paradox about affect and virality.
- Twitter: Retweeting is treated as Twitter’s mechanism for quoting users and propagating memes and opinions.The paper distinguishes Twitter’s explicit RT and via notations from quotation-based meme tracking elsewhere.
- Twitter: Twitter combines social-network and news-medium characteristics, so retweeting may depend on both content type and intended audience.The paper notes that retweets can proliferate beyond reciprocal relationships through an interest graph.
- Research questions: The study asks how accurately tweets can be classified as news, how much Twitter is news, and whether sentiment affects retweeting differently in news and social communication.These questions are formalized as Q1–Q4.
2 Methods
The study classifies tweets as news, estimates sentiment, and models retweet probability across a news-event sample and random tweets. It combines a Brown Corpus classifier with generalized linear models using tweet-level, manipulable features.
- Data and design: The study uses Brown Corpus training data, complete COP15 event tweets, and a random Twitter sample to classify news and model retweeting.COP15 contains 207,782 tweets, while RANDOM contains 348,862 tweets.
- News classification: A Naive Bayes classifier trained on labeled Brown Corpus sentences estimates each tweet’s probability of being news.The model uses a bag-of-terms representation and a 75% training, 25% testing split.
- News classification: The study defines news tweets as those with p(news|w) > 0.5 and applies the trained classifier to RANDOM and COP15.This produces a news-tweet fraction for each Twitter sample.
- Sentiment scoring: Sentiment scoring uses a Twitter-specific English word list assigning 1,446 words valences from −5 to +5, then sums valence and absolute valence into tweet-level scores.The resulting scores represent valence and arousal.
- Retweet modeling: A binomial generalized linear model estimates retweet probability from extracted tweet features, with coefficients tested through likelihood-based statistics.The model focuses on features manipulable at tweeting time, including hashtags, mentions, and URLs.
- Retweet modeling: The analysis adds a news-by-negative-sentiment interaction feature to test whether negative sentiment matters differently for news and non-news tweets.This feature is used to examine the interaction between newsiness and negativity.
- Results setup: Table 1 reports t-values for negative sentiment and other binary covariates in logistic-retweet models for RANDOM and COP15.The covariates are negative sentiment, hashtag, mention, and URL.
3 Results
The results show that news classification separates tweet types, while sentiment has different retweet associations across samples and their interaction with newsiness.
- 84% ± 1% accuracy was achieved by the NLTK naive Bayes classifier on the Brown corpus test data.
- 23% of RANDOM tweets and 31% of COP15 tweets were classified as news using a probability threshold above 0.5.The RANDOM estimate could be a slight underestimate because the Brown corpus has about 0.1 news items a priori.
- In RANDOM, negative sentiment had no statistically detectable effect on retweeting, whereas in COP15 it significantly promoted retweeting.
- Among aroused RANDOM tweets, positive content strongly promoted retweets.Aroused tweets are those with non-zero positive or negative content.
- The interaction between negative sentiment and probability of news promoted retweeting in both COP15 and RANDOM.The interaction effect persisted when analysis was restricted to aroused tweets.
4 Discussion
The discussion reconciles competing accounts of affect and diffusion by distinguishing social or non-news communication from news-oriented Twitter content.
- Twitter combines social-network and news-medium characteristics, motivating analysis of how content type and audience relate to retweeting.
- About one quarter of RANDOM tweets and nearly one third of COP15 tweets were classified as news.
- Negative content was more strongly associated with retweeting in the news-driven COP15 sample than in the generic sample.
- Negative news was more retweeted than positive news, while positive content increased diffusion probability in the non-news segment.The discussion links the non-news pattern to sharing among friends and the news pattern to classic news-diffusion theory.
5 Conclusion
The paper concludes that affect and virality depend on whether tweets carry news, with negative sentiment associated with news diffusion and positive sentiment with generic diffusion.
- A simple Naive Bayes classifier reliably detected news in short communication.
- 23% of random tweets and 33% of COP15 tweets were identified as news.
- Negative sentiment was detrimental to retweeting in generic tweets, while negative news content propagated better.
- The conclusion is summarized as sweet-talking friends or serving bad news to the public.