Source-linked AI summary
Happiness is assortative in online social networks
Johan Bollen, Bruno Goncalves, Guangchen Ruan, Huina Mao
TL;DR
The paper addresses whether psychological-state assortativity occurs in online social networks whose ties are mediated without physical contact. It measures individual Twitter users’ SWB from six months of tweets and analyzes correlations across their network ties. SWB is positive and highly assortative, with stronger ties showing stronger assortativity, although the study does not determine whether homophily or contagion produces it.
Problem
It was unknown whether psychological states such as happiness exhibit assortative mixing in online social systems without physical contact, despite the importance of understanding sentiment across online ties.
Method
The study estimates individual SWB from six months of tweets and measures pairwise and neighborhood assortativity among connected Twitter users, including analyses with different edge weights.
Results
SWB is positive and highly assortative among Twitter users, with pairwise and neighborhood assortativity converging to approximately 0.750 as stronger edge-weight thresholds are used.
Takeaways & Limitations
Twitter users tend to connect with others sharing similar levels of general happiness, particularly through stronger social ties.
Takeaways & Limitations
The study does not determine whether the observed SWB assortativity arises from homophily, mood contagion, or both.
Abstract
from arXiv · showhide
Social networks tend to disproportionally favor connections between individuals with either similar or dissimilar characteristics. This propensity, referred to as assortative mixing or homophily, is expressed as the correlation between attribute values of nearest neighbour vertices in a graph. Recent results indicate that beyond demographic features such as age, sex and race, even psychological states such as "loneliness" can be assortative in a social network. In spite of the increasing societal importance of online social networks it is unknown whether assortative mixing of psychological states takes place in situations where social ties are mediated solely by online networking services in the absence of physical contact. Here, we show that general happiness or Subjective Well-Being (SWB) of Twitter users, as measured from a 6 month record of their individual tweets, is indeed assortative across the Twitter social network. To our knowledge this is the first result that shows assortative mixing in online networks at the level of SWB. Our results imply that online social networks may be equally subject to the social mechanisms that cause assortative mixing in real social networks and that such assortative mixing takes place at the level of SWB. Given the increasing prevalence of online social networks, their propensity to connect users with similar levels of SWB may be an important instrument in better understanding how both positive and negative sentiments spread through online social ties. Future research may focus on how event-specific mood states can propagate and influence user behavior in "real life".
1 Introduction
Assortative mixing describes networks’ tendency to connect individuals with similar characteristics, including psychological states. This paper asks whether happiness is similarly assortative in Twitter’s online-only social network and reports strong evidence that it is.
- Assortative mixing is a network tendency to favor connections between vertices with similar characteristics, while disassortative mixing favors dissimilar characteristics.Most networks exhibit both tendencies to some degree, depending on the characteristic examined.
- Psychological states such as loneliness can be assortative, with individuals preferentially relating to others reporting similarly elevated loneliness.This homophilic tendency increases over time.
- It remained unclear whether homophily or contagion operates in online social systems where ties need not arise from physical contact or in-person communication.The two proposed processes are homophilic attachment and emotional contagion.
- Twitter follower relations form directed, unweighted ties centered on interest in another user’s content rather than necessarily reciprocated friendship.This structure differs from offline friendship networks, where ties are generally symmetric and vary in strength.
- Online environments have already shown homophily in personal preferences and sentiment, increasing interest in how mood and sentiment relate across online social ties.Such questions matter for understanding online social behavior and responses to large-scale events.
- Using 129 million tweets from 102,009 users over 6 months, the study measures individual SWB and finds Twitter users’ overall SWB positive and highly assortative.Users are preferentially connected to others with similar general happiness, and tie strength significantly modulates SWB assortativity.
2 Data and methods
The study constructs a Twitter Friend network from reciprocal follows, filters users by activity, and measures individual SWB from six months of tweets. It then evaluates pairwise and neighborhood correlations between users’ SWB values.
- Data collection: 129 million tweets and follower histories for 4,844,430 users provide the underlying activity and social-connection data.Tweets were collected from November 28, 2008 to May 2009, supplemented by complete histories for over 4 million users.
- Creating a Twitter Friend network: Friend ties retain only reciprocal follower relations, so two users are connected when both follow each other.This converts the directed Follower graph into a network of reciprocal Friend connections.
- Creating a Twitter Friend network: 102,009 users and 2,361,547 edges remain in the largest Connected Component after activity filtering and extraction.Users posting more than one tweet per day on average were retained before extracting the largest Connected Component.
- Creating a Twitter Friend network: 97.9% of users in the Twitter Friend network are retained in the largest Connected Component, whose diameter is 14 despite low density.The resulting component is therefore highly connected across its users.
- User-level measurements of Subjective Well-Being: SWB is inferred from six months of tweets using OpinionFinder after requiring at least one tweet per day in the Connected Component.SWB is defined as the fractional difference between positive and negative tweets, with Np(u) and Nn(u) denoting their respective counts.
- Defining SWB assortativity: Pairwise assortativity correlates SWB values across connected user pairs, while neighborhood assortativity correlates each user’s SWB with the mean SWB of connected Friends.The pairwise measure uses source and target SWB vectors; the neighborhood measure uses user SWB and average neighborhood SWB vectors.
3 Results and discussion
Twitter users show statistically significant positive SWB assortativity at both pairwise and neighborhood levels. Assortativity strengthens when weakly weighted connections are excluded, converging near 0.75.
- 3.1 SWB distribution: SWB values are bimodally distributed, with peaks near [−0.1, 0.1] and [0.2, 0.4]; 50% of users have SWB ≤0.1.Excluding users with SWB=0, most users have positive SWB in [0.1, 0.4], peaking at 0.16; 95% have SWB ≤0.285.
- 3.2 Pairwise and neighborhood SWB assortativity: 0.443⋆⋆⋆ pairwise and 0.689⋆⋆⋆ neighborhood SWB assortativity were observed, both with p-values < 0.001.The estimates used 2,062,714 edges for pairwise assortativity and 102,009 nodes for neighborhood assortativity.
- 3.2 Pairwise and neighborhood SWB assortativity: Users with SWB values in the same range are preferentially connected, despite substantial scatter and the bimodal SWB distribution.The pairwise relation is not obviously linear, with clusters around [−0.05, 0.05] and [0.1, 0.3].
- 3.3 Edge weight and SWB assortativity: Increasing the edge threshold raises both assortativity measures nonlinearly; pairwise assortativity stabilizes near 0.750 after ε increases to 0.10.At ε=0.1, the edge count falls from 2,062,714 to 479,401, while neighborhood assortativity reaches approximately 0.760 for ε∈[0.10, 0.90].
- 3.4 Discussion: The results indicate that low-SWB users connect predominantly to low-SWB users and high-SWB users to high-SWB users.The authors interpret this pattern as consistent with either homophilic attachment or contagion, without distinguishing between them.
- 3.4 Discussion: Pairwise and neighborhood assortativity converge near 0.750 when weak ties are removed, indicating stronger SWB similarity among higher-weight connections.The authors suggest weakly weighted connections reduce pairwise assortativity because that measure includes all individual user-to-user links.
4 Conclusion
The study concludes that SWB is assortative in Twitter’s social network, with happy users tending to connect to happy users and unhappy users to unhappy users. It does not determine whether homophilic attachment, mood contagion, or both produce this pattern.
- 4 Conclusion: SWB is strongly related among users with reciprocal Twitter follower links, with happy users tending to connect to happy users.The conclusion also reports that unhappy users tend to be predominantly connected to unhappy users.
- 4 Conclusion: Increasing edge-weight thresholds make pairwise and neighborhood assortativity converge, suggesting strongest assortativity occurs among a limited number of strong social ties.The authors propose that weaker ties may fulfill a different social role.
- 4 Conclusion: The study does not establish whether homophilic attachment, mood contagion, or both mechanisms cause the observed SWB assortativity.Possible alternatives include preferentially forming similar-SWB connections, converging over time, or expressing SWB relative to friends.