Source-linked AI summary
Anger is More Influential Than Joy: Sentiment Correlation in Weibo
Rui Fan, Jichang Zhao, Yan Chen, Ke Xu
TL;DR
The paper addresses limited evidence on how local social structure shapes emotional correlation and propagation in online networks. It classifies Weibo sentiment and constructs an interaction network to measure emotion correlations among connected users. Anger shows the strongest correlation, while sadness is low; stronger interactions and higher node degree are associated with greater sentiment influence.
Problem
How local network structure affects emotion correlation in online social networks remains insufficiently studied, limiting understanding of sentiment influence and contagion.
Method
The study classifies Weibo tweets into anger, joy, sadness, and disgust, then constructs a weighted interaction network to measure correlations among connected users.
Results
Anger has the strongest emotion correlation, sadness has low correlation, and sentiment correlation increases with interaction strength and node degree.
Takeaways & Limitations
The findings provide insights for modeling sentiment influence and propagation in online social networks.
Abstract
from arXiv · showhide
Recent years have witnessed the tremendous growth of the online social media. In China, Weibo, a Twitter-like service, has attracted more than 500 million users in less than four years. Connected by online social ties, different users influence each other emotionally. We find the correlation of anger among users is significantly higher than that of joy, which indicates that angry emotion could spread more quickly and broadly in the network. While the correlation of sadness is surprisingly low and highly fluctuated. Moreover, there is a stronger sentiment correlation between a pair of users if they share more interactions. And users with larger number of friends posses more significant sentiment influence to their neighborhoods. Our findings could provide insights for modeling sentiment influence and propagation in online social networks.
1. Introduction
Online social networks provide large-scale behavioral data, including users’ emotional states, but how local network structure shapes emotion correlation remains insufficiently studied. This paper addresses that gap by examining four sentiments among connected Weibo users.
- 1. Introduction: Online social networks provide large-scale records for investigating human behavior and emotional communication.Weibo had more than 500 million registered users in less than four years and over 100 million Chinese tweets were published daily.
- 1. Introduction: Homophily extends beyond demographics to psychological states such as loneliness and happiness.Prior research found happiness assortative in Twitter and average happiness scores positively correlated across the network.
- 1. Introduction: Online social media records convey users’ sentiments alongside factual information about social events.Computer-mediated emotional communication is described as comparable to face-to-face communication in emotionality and personal character.
- 1. Introduction: How local network structure affects emotion correlation remains insufficiently studied, despite its importance for understanding sentiment influence and contagion.The paper identifies this as a central gap in prior research.
- 1. Introduction: The study classifies sentiment into anger, joy, sadness, and disgust and investigates correlations between connected Weibo users.It also reports that anger correlates more strongly than joy, sadness has trivial correlation, and local interaction structure boosts correlation.
2. Related works
Prior work mines sentiment from social-media text using lexicons or machine-learning classifiers, while emotion states help characterize users, groups, and real-world events.
- 2. Related works: Lexicon-based sentiment analysis determines a tweet’s sentiment by counting positive and negative sentimental words.Prior work also assigned happiness scores to words and measured affect in large Twitter datasets.
- 2. Related works: Machine-learning sentiment classification uses terms, smileys, and emoticons as features.Reported experiments found machine-learning classifiers outperform a human-word-list baseline, while another framework classified four emotions without manually labelled training tweets.
- 2. Related works: Emotion states are important for understanding user behavior from both individual and group perspectives.Users’ mood states are also described as significantly affected by real-world events.
3. Methods
The study constructs a Weibo interaction network, classifies tweets into four emotions, and quantifies sentiment correlation between users at different network distances. It uses Pearson and Spearman correlations to examine emotion relationships and local network structure.
- Methods: The methodology collects Weibo tweets, constructs an interaction network, classifies tweet sentiments, and defines emotion correlations.The network is built from retweets or mentions exceeding threshold T, while correlations are computed for connected users.
- Methods: The interaction network is an undirected weighted graph whose links represent retweet or mention interactions above threshold T.Users are retained when they posted more than one tweet every two days on average during the six-month period.
- Methods: Tweets are assigned to anger, sadness, joy, or disgust using a Bayesian classifier trained from emoticon-labeled Weibo tweets.About 3.5 million tweets with valid emoticons provide the labeled training corpus, while roughly 95% of Weibo tweets lack emoticons.
- Methods: The analyses fix T = 30 to obtain a sufficiently large network with convincing interaction strength and compare correlations as hop distance varies.The resulting comparisons use both Pearson and Spearman correlation measures.
- Methods: For each user, an emotion vector records the fractions of tweets expressing anger, joy, sadness, and disgust.These four fractions represent each user’s sentiment status.
- Methods: Pairwise sentiment correlation groups user pairs by hop distance and compares source and target emotion sequences using Pearson or Spearman correlation.Spearman correlation uses rank differences between corresponding source and target sequence values.
4. Results
The results show that sentiment correlation varies by emotion and social structure. Anger is most strongly correlated, while tie strength and node degree mainly increase influence among close or highly connected users.
- Anger has the strongest sentiment correlation, indicating faster and broader propagation than other emotions.Both Pearson and Spearman correlations identify anger as unusually strong, with influence extending across roughly three hops.
- Sadness and disgust show weak correlations, with sadness below 0.15 at h = 1 and correlations becoming weak beyond h > 3.For anger and joy, correlations fluctuate around 0 when h > 5, showing that social distance limits influence.
- Shuffling emotion sequences removes correlation for every sentiment, supporting the significance of the observed correlations between connected users.The shuffled results indicate no emotion homophily among randomly paired users.
- Higher interaction thresholds increase correlations within two hops, with anger’s Pearson correlation reaching around 0.52 while sadness and disgust remain below 0.25.The effect of tie strength is concentrated among close neighbors, whereas social distance remains the primary constraint on influence.
- Sentiment correlation generally increases with node degree, especially for anger and joy, while sadness and disgust fluctuate around 0.2 or lower.At degree 30, anger’s correlation reaches 0.85; the network is small, so behavior beyond that degree remains uncertain.
- Overall, tie strength and node degree enhance sentiment influence mainly for anger and joy, whereas their effects on sadness and disgust are limited.The paper identifies social distance as a major constraint and notes that larger-degree behavior may change beyond the observed range.
5. Empirical Explanation
The paper examines retweeted keywords and topics to explain why anger spreads strongly in Weibo while sadness spreads weakly. Angry posts cluster around domestic social problems and diplomatic conflicts, whose discussion and retweeting accelerate information diffusion.
- Keyword and topic analysis: Retweeted keywords and topics are mined to interpret the differing spread of emotions in Weibo.The analysis uses emotion-specific retweeted tweets and their associated keywords or topics to summarize real-world events and social issues.
- Anger-related events: Anger is associated with domestic social problems such as food security, government bribery, and demolition for resettlement.Keywords including “government,” “bribery,” and “demolition” recur around these events.
- Keyword and topic analysis: Figure 7 presents example Chinese keywords extracted separately for anger and sadness, with the top 20 keywords translated into English.The translated keyword list is provided through the link identified in the caption.
- Anger-related events: Diplomatic conflicts, including China–United States, China–South Korea, and China–Japan tensions, also trigger angry moods and patriotic responses.Examples include the Yellow Sea drill and the China–Japan ship collision, alongside keywords such as “Diaoyu Island” and “Philippines.”
- Propagation mechanism: Social problems and diplomatic issues spread rapidly as users express anger through posting, retweeting, and criticism.The paper links these practices to faster dissemination of related news and the formation of public opinion and collective behavior.
6. Discussion and Conclusion
The study examines emotional correlation in Weibo using four sentiment categories and connects emotion spread to local network structure. It reports stronger influence for anger than joy, weak sadness correlation, and structural effects that inform sentiment-propagation modeling.
- Discussion and conclusion: The study separates sentiment into anger, joy, sadness, and disgust to analyze their correlations in Weibo.The dataset is publicly available to the research community.
- Discussion and conclusion: Anger is more influential than joy, while sadness has low correlation; angry tweets can therefore spread quickly and broadly in the network.The paper connects angry public opinion about social problems and diplomatic issues with rapid information propagation.
- Discussion and conclusion: The paper conjectures that anger contributes to large-scale propagation of negative social news and online collective behavior.Examples include food security and demolition-for-resettlement controversies.
- Local network structure: Emotion correlation strengthens when two users interact more, and a node’s degree enhances sentiment influence on its neighborhood, especially for anger and joy.These findings are presented as guidance for modeling sentiment influence and spread in social networks.