Source-linked AI summary
Measuring Emotional Contagion in Social Media
Emilio Ferrara, Zeyao Yang
TL;DR
The paper asks whether emotions spread through online interactions without non-verbal cues and without ethically concerning content manipulation. Using Twitter sentiment analysis and a reshuffled null model, it finds emotional contagion patterns in both negative and positive content, including a strong stimulus-response relationship and differing user susceptibility. The study cautions that observational data cannot separate contagion from homophily.
Problem
The study asks whether emotions spread through online interactions despite the absence of in-person non-verbal cues, while prior manipulation experiments raise ethical concerns.
Method
The authors analyze Twitter exposure and responses using sentiment scores, reconstructed preceding stimuli, and a reshuffled null model that discounts emotional contagion and other correlational biases.
Results
The results suggest emotional contagion for both negative and positive sentiment, with a very strong linear relationship between stimulus and response valence and distinct susceptibility patterns across users.
Takeaways & Limitations
The findings provide quantitative evidence for emotional contagion on Twitter while avoiding direct interaction or content manipulation.
Takeaways & Limitations
Observational data cannot separate emotional contagion from homophily, so further work is needed to understand how the two dynamics intertwine.
Abstract
from arXiv · showhide
Social media are used as main discussion channels by millions of individuals every day. The content individuals produce in daily social-media-based micro-communications, and the emotions therein expressed, may impact the emotional states of others. A recent experiment performed on Facebook hypothesized that emotions spread online, even in absence of non-verbal cues typical of in-person interactions, and that individuals are more likely to adopt positive or negative emotions if these are over-expressed in their social network. Experiments of this type, however, raise ethical concerns, as they require massive-scale content manipulation with unknown consequences for the individuals therein involved. Here, we study the dynamics of emotional contagion using Twitter. Rather than manipulating content, we devise a null model that discounts some confounding factors (including the effect of emotional contagion). We measure the emotional valence of content the users are exposed to before posting their own tweets. We determine that on average a negative post follows an over-exposure to 4.34% more negative content than baseline, while positive posts occur after an average over-exposure to 4.50% more positive contents. We highlight the presence of a linear relationship between the average emotional valence of the stimuli users are exposed to, and that of the responses they produce. We also identify two different classes of individuals: highly and scarcely susceptible to emotional contagion. Highly susceptible users are significantly less inclined to adopt negative emotions than the scarcely susceptible ones, but equally likely to adopt positive emotions. In general, the likelihood of adopting positive emotions is much greater than that of negative emotions.
INTRODUCTION
This study examines whether emotions spread through online interactions and seeks quantitative evidence without manipulating users’ social-media content. It uses Twitter to study emotional contagion while addressing ethical concerns associated with large-scale content manipulation.
- INTRODUCTION: Online emotional contagion may occur even without the non-verbal cues typical of in-person interactions.
- INTRODUCTION: Facebook-based content manipulation experiments raise ethical concerns because the consequences of massive-scale manipulation are unknown.
- INTRODUCTION: The study observes Twitter activity without filtering, prioritizing, ranking, or otherwise manipulating users’ information.
- INTRODUCTION: A null model discounts emotional contagion and other correlational biases while reconstructing the stimuli users encountered before tweeting.
- INTRODUCTION: The work provides quantitative evidence about emotional contagion while avoiding the ethically problematic consequences of prior experimental manipulation.
MATERIALS AND METHODS
The study quantifies tweet sentiment and reconstructs the content users encountered shortly before posting. It uses Twitter users, their followees’ recent tweets, sentiment classifications, and a reshuffled baseline to analyze emotional exposure.
- MATERIALS AND METHODS: SentiStrength assigns each tweet separate positive and negative sentiment scores designed for short, informal social-media text.
- MATERIALS AND METHODS: The polarity score S ranges from -4 to +4, with S = 0 denoting neutral sentiment when positive and negative scores are equal.
- MATERIALS AND METHODS: The dataset begins with a random sample of 3,800 users who posted at least one English tweet during the last week of September 2014.
- MATERIALS AND METHODS: For each user tweet, the study collects followees’ tweets from the preceding one-hour window.
- MATERIALS AND METHODS: The analysis retains windows containing at least 20 tweets and restricts tweets to English text without URLs or media content.
- MATERIALS AND METHODS: Tweets are classified as negative for S ≤−1, neutral for S = 0, and positive for S ≥1.
RESULTS
Using an observational Twitter analysis and a reshuffled null model, the study measures emotional contagion without controlled content manipulation. Stimulus valence predicts response valence, while susceptibility varies substantially across users and is generally more associated with positive than negative adoption.
- Analysis design: The analysis used a reshuffled null model that discounts exposure effects and the possibility of emotional contagion.For each tweet, the model sampled tweets with replacement from a sentiment bucket in a quantity matching the observed one-hour exposure history.
- Emotional contagion: 4.34% more negative tweets preceded negative posts, while 4.50% more positive tweets preceded positive posts than the null-model baseline.Negative posts also followed 1.09% less positive content, and positive posts followed 1.29% less negative content.
- Stimulus-response valence: R2 = 0.975 indicates a very strong linear relationship between stimulus and response valence.A valence -1 stimulus is followed by response valence about -0.8, while valence +1 is followed by response valence around +0.6.
- User susceptibility: About 80% of users had up to 50% of their tweets affected by emotional contagion, whereas 20% exceeded 50%.The study characterized each user by the fraction of tweets affected by emotional contagion.
- User susceptibility: Highly and scarcely susceptible users were defined as the top and bottom 15% of the susceptibility distribution, respectively.Their positive and negative susceptibility fractions were then compared independently.
- User susceptibility: Positive emotional contagion was 1.6 times more likely than negative contagion among low-susceptibility users and 3.96 times more likely among high-susceptibility users.Highly susceptible users were less inclined to adopt negative emotions than scarcely susceptible users but equally likely to adopt positive emotions.
DISCUSSION
An observational Twitter analysis uses a null model to estimate emotional contagion without manipulating users’ timelines. It finds aligned stimulus-response valence patterns, while noting that contagion cannot be separated from homophily in observational data.
- A null model discounts confounding factors, including contagion, while tracking stimuli and responses for 3,800 users over one week.
- 4.34% more negative stimuli precede negative tweets on average, while 4.50% more positive tweets precede positive ones.
- A strong linear relation between stimulus and response valence suggests a common mechanism regulating negative and positive emotions.
- Observational analysis cannot separate emotional contagion from homophily, so further work is needed to disentangle their effects.