Source-linked AI summary
Quantifying the Effect of Sentiment on Information Diffusion in Social Media
Emilio Ferrara, Zeyao Yang
TL;DR
The paper addresses limited understanding of how sentiment affects information diffusion and conversation dynamics on social media. Using sentiment analysis and temporal characterization of Twitter discussions, it finds that negative tweets spread faster while positive tweets reach broader audiences, with sentiment patterns varying by event dynamics. These findings support communication strategies that account for emotional valence.
Problem
The paper addresses limited quantitative evidence about how sentiment affects information diffusion and the sentiment patterns associated with different temporal conversation dynamics.
Method
The study applies SentiStrength to Twitter content and analyzes tweet-level diffusion alongside temporal classes and sentiment evolution of entire conversations.
Results
Negative tweets spread faster than positive ones, positive tweets reach broader audiences and collect more favorites, and anticipated versus unexpected events show distinct sentiment patterns.
Takeaways & Limitations
Understanding sentiment and diffusion dynamics can inform communication policies and strategies in marketing, public policy, and emergency management.
Abstract
from arXiv · showhide
Social media have become the main vehicle of information production and consumption online. Millions of users every day log on their Facebook or Twitter accounts to get updates and news, read about their topics of interest, and become exposed to new opportunities and interactions. Although recent studies suggest that the contents users produce will affect the emotions of their readers, we still lack a rigorous understanding of the role and effects of contents sentiment on the dynamics of information diffusion. This work aims at quantifying the effect of sentiment on information diffusion, to understand: (i) whether positive conversations spread faster and/or broader than negative ones (or vice-versa); (ii) what kind of emotions are more typical of popular conversations on social media; and, (iii) what type of sentiment is expressed in conversations characterized by different temporal dynamics. Our findings show that, at the level of contents, negative messages spread faster than positive ones, but positive ones reach larger audiences, suggesting that people are more inclined to share and favorite positive contents, the so-called positive bias. As for the entire conversations, we highlight how different temporal dynamics exhibit different sentiment patterns: for example, positive sentiment builds up for highly-anticipated events, while unexpected events are mainly characterized by negative sentiment. Our contribution is a milestone to understand how the emotions expressed in short texts affect their spreading in online social ecosystems, and may help to craft effective policies and strategies for content generation and diffusion.
INTRODUCTION
This study examines how sentiment shapes information diffusion on Twitter, addressing limited evidence about whether emotional valence affects spreading and discussion dynamics. It investigates content-level diffusion and sentiment patterns across conversations with different temporal evolution.
- Computational social science studies technologically mediated communication and information diffusion to understand effects on society, social awareness, and influence.
- Little work has quantified how sentiment affects information diffusion, despite evidence that online emotions transfer and content characteristics influence sharing.
- The study analyzes Twitter content spreading by examining how sentiment affects diffusion speed and content popularity.
- It also categorizes entire conversations by temporal evolution to examine how different discussion dynamics exhibit different sentiment patterns.
MATERIALS AND METHODS
The study quantifies tweet sentiment with SentiStrength and analyzes a large September 2014 Twitter sample, relating polarity to diffusion outcomes.
- Sentiment analysis: SentiStrength assigns each tweet positive and negative sentiment scores, each ranging from 1 for neutral to 5 for strongly positive or negative.
- Sentiment analysis: Polarity score S(t) is computed as the difference between positive and negative sentiment scores.
- Sentiment analysis: The polarity score ranges from -4 to +4, with zero representing tweets whose positive and negative scores are equal.
- Data: The dataset contains English public tweets from September 2014, excluding URLs and media content because their sentiment is difficult to capture computationally.
- Diffusion measures: Figure 2 measures average retweets, average favorites, and seconds until the first retweet as functions of tweet polarity.
- Data: 19,766,112 tweets from 8,130,481 distinct users were processed with SentiStrength and assigned sentiment scores.
RESULTS
The study finds that sentiment relates differently to the speed and breadth of Twitter diffusion, while conversation-level sentiment patterns vary across temporal dynamics. Four discussion classes capture distinct popularity evolutions, with anticipatory conversations tending positive and unexpected events more negative.
- Content diffusion: Positive tweets spread broader than neutral ones and collect more favorites, whereas negative posts are not more or less widely spread or favored than neutral posts.This pattern suggests a positivity bias in what people share and favorite.
- Content diffusion: Negative contents spread much faster than positive ones, although not significantly faster than neutral ones.Positive tweets require more time to receive rebroadcasts, while negative or neutral posts generally achieve their first retweet twice as fast.
- Conversation dynamics: Anticipatory discussions build before the peak and fade quickly afterward, unexpected events peak suddenly and decay rapidly, symmetric discussions remain balanced, and transient events are bursty but short.The classes were learned from proportions of tweets before, during, and after each conversation’s peak.
- Conversation dynamics: 1,522 active and exclusive conversations were classified into 64 anticipatory, 156 unexpected, 56 symmetric, and 1,246 transient discussions.Conversations were represented in seven-day windows centered on the peak day.
- Sentiment evolution: Anticipatory conversations show rising positive sentiment and below-average negative content, while unexpected events maintain negative sentiment around the dataset average.Across the dataset, neutral, positive, and negative tweets average 42.46%, 35.95%, and 21.59%, respectively.
- Sentiment evolution: Symmetric discussions show declining negative sentiment, falling from about 23% at inception to around 12% toward the end.These values are compared with the dataset’s 21.59% average negative sentiment.
DISCUSSION
Computational annotation of short texts enables analysis of how sentiment relates to information diffusion. Negative tweets spread faster, while positive content reaches larger audiences, and conversation sentiment varies with temporal dynamics, informing communication strategies.
- Method and scope: Computationally annotating emotional value in tweets enabled investigation of how sentiment relates to information diffusion.The study examines emotions expressed in short social-media texts at scale.
- Content diffusion: Negative tweets spread faster than neutral and positive tweets, with positive tweets taking almost twice as long as negative tweets to receive a first retweet.The comparison concerns the interval between original publication and first retweet.
- Content diffusion: Positive tweets were favored up to five times more and retweeted up to 2.5 times more than negative or neutral tweets.These results indicate a divergence between diffusion speed and audience reach.
- Conversation dynamics: Highly anticipated events generally exhibit positive sentiment, unexpected events negative sentiment, and transient events no particular emotional valence.Transient events are described as the norm on Twitter because their duration is very brief.
- Implications: Understanding sentiment and diffusion dynamics is relevant to communication policy across advertising, marketing, public policy, and emergency management.The paper frames these applications as having economic and social impact.