Source-linked AI summary
Interplay between Emotional Dynamics and Network Structure on the Social Media Vent
Yuina Takahashi, Sho Tsugawa
TL;DR
Prior work leaves gaps in fine-grained, cross-category emotional analysis and in understanding individual alignment with surrounding emotions and network structure. This study analyzes Vent users’ pre-posting timelines and subsequent labels, finding both same- and cross-category associations and that highly aligned users tend to be close in the network.
Problem
Prior social-media studies often use coarse emotion categories, give limited attention to cross-category associations, and leave user alignment with surrounding emotions and network structure unclear.
Source: §1 · p. 2Method
The study uses Vent’s explicit emotion labels to analyze associations between pre-posting timeline emotions and subsequent labels, and to examine users’ alignment and network proximity.
Source: §1 · p. 2Results
Same-category emotions were overrepresented before posts across analyzed categories, several cross-category associations appeared, and highly aligned users tended to be close in the network.
Source: §5 · p. 7Takeaways & Limitations
The findings provide large-scale observational evidence linking emotional expression on Vent with short-term timeline context and network structure.
Source: §5 · p. 7
Abstract
from arXiv · showhide
The sharing and contagion of emotions on social media influence online interactions and users' psychological states. This study analyzes emotional dynamics on Vent, an emotion-sharing social media platform where users explicitly assign emotional labels to their own posts. Using a large-scale dataset, we examine how emotions in users' pre-posting timelines are associated with their subsequent emotional labels and how such associations vary across users. Our analysis reveals three key findings. First, users' subsequent emotional labels are associated with the emotional composition of their pre-posting timelines, with same-category emotions being overrepresented before posts in all analyzed categories. Second, several cross-category associations are observed; for example, Surprise was overrepresented before Fear posts; Affection and Happiness were overrepresented before Anger posts; and Affection was overrepresented before Sadness posts. Third, users differ in their degree of alignment with timeline emotional fluctuations, and highly aligned users tend to be located close to one another in the network. These findings provide large-scale observational evidence that emotional expression on Vent is associated with both short-term timeline context and network structure.
1 Introduction
Research has linked emotional content on social media to subsequent expression, but important gaps remain in fine-grained and cross-category analysis and in understanding individual differences in network context. This paper uses Vent to study pre-posting timeline emotions and users’ subsequent labels, including how users’ alignment varies across the network.
Prior work
Research gaps
- Prior studies often use coarse emotion labels, insufficiently explore cross-category associations, and leave unclear how user alignment with emotional fluctuations relates to network structure.Understanding these patterns may inform healthier social media design.Source: §1 · p. 2
Research questions and contribution
- Using Vent’s explicit emotion labels, the study examines pre-posting timeline associations and whether users highly aligned with timeline fluctuations are close to one another in the network.The analysis addresses both within-category and cross-category associations.Source: §1 · p. 2
2 The Vent Dataset
Vent lets users label their own posts with their current emotion, enabling analysis of explicitly stated emotions. The study filters its large public dataset to active users and seven target categories.
The Vent platform
- Vent users assign an emotional label, such as Sad or Happy, to their own post at posting time.This permits analysis of explicitly stated subjective emotions without estimating sentiment from text.Source: §2 · pp. 2–3
Target categories
- Table 1 presents the seven target categories, representative labels, and post distributions by number and percentage among active users.The table covers Sadness, Fear, Anger, Surprise, Positivity, Affection, and Happiness.Source: Table 1 · p. 3, §2 · p. 3
Dataset scale
- The public dataset contains 934,095 users, 33,623,414 posts, and 13,605,522 following edges, with 705 labels organized into 63 categories.Each post is linked to an emotional label.Source: §2 · p. 3
Filtering and target population
- The analysis retained seven categories and defined active users as those posting at least 30 times in them, yielding 18,681,803 posts by 137,822 users and 4,103,692 following edges.Feelings was excluded because it groups heterogeneous labels such as Tired, Meh, and Sleepy.Source: §2 · p. 3
3 Analysis of Timeline Emotional Associations
The analysis compares emotions in users’ recent followee timelines with their subsequent labels across seven categories. Same-category emotions were overrepresented before posts, and several cross-category associations were also observed.
3 Analysis of Timeline Emotional Associations
- The analysis considers seven categories and distinguishes within-category associations, where the same emotion appears in timeline and post, from cross-category associations involving different categories.Figure 1 compares pre-posting timeline distributions with the baseline and shows their ratio.Source: §3 · p. 3, §3.2 · p. 4, Figure 1 · p. 4
3.1 Quantification of Pre-posting Emotional Exposure
- The pre-posting timeline comprises followee posts from the three hours before a post; analyses required more than 20 timeline posts, yielding 1,035,304 posts from 29,454 users.The baseline was estimated by randomly sampling timelines of equal size from the pooled analyzed timelines and averaging their emotional distributions.Source: §3.1 · pp. 3–4
3.2 Within- and Cross-Category Emotional Associations
- Same-category timeline emotions were overrepresented before posts in every analyzed category; Positivity showed the largest increase, 1.607 times the baseline.Surprise showed the second-highest increase, 1.283 times baseline; the authors suggest its high arousal may contribute to its transmissibility.Source: §3.2 · pp. 4–5, Figure 1 · p. 4
- Before Fear posts, Surprise increased by a factor of 1.067; before Anger, Affection and Happiness increased by factors of 1.015 and 1.030; before Sadness, Affection was 1.026 times higher.The authors interpret these as observational associations, not causal mechanisms.Source: §3.2 · p. 5
4 Analysis of Susceptibility
The section quantifies how users’ posted emotions align with recent timeline fluctuations, then examines whether users with stronger alignment are connected in the network. It reports a right-skewed matching-rate distribution and greater network proximity among highly susceptible users than under a degree-preserving null model.
4.1 Quantification and Distribution of Susceptibility
- A user’s susceptibility index is the frequency with which their posted emotion matches the emotion with the largest recent increase on their timeline.The analysis included 6,908 users with at least 20 qualifying posts and more than 20 preceding timeline posts within three hours.Source: §4.1 · pp. 5–6
- Matching rates were right-skewed, with a small number of users showing very high alignment with the emotion most increased on their timelines.The authors note that this subgroup may excessively synchronize with surrounding negative emotional expressions, potentially increasing psychological distress.Source: §4.1 · p. 6, Figure 2 · p. 6
4.2 Homophily in Susceptibility
- The network analysis defined high-susceptibility users as the top 15% by index and compared their shortest paths with a degree-preserving edge-rewiring null model.The null-model values averaged 10 trials.Source: §4.2 · p. 6
- Susceptibility among users was moderately positively correlated with that of their followees (r = 0.437).The authors interpret this as users with similar susceptibility tending to be connected in the network.Source: §4.2 · p. 6
- High-susceptibility user pairs averaged 2.35 hops versus 3.29 in the null model, and 99.7% versus 63.0% were within three hops.Direct-following pairs were also approximately 8.3 times more numerous than in the null model.Source: §4.2 · p. 7, Table 2 · p. 7
- The clustering of highly susceptible users may create conditions for local reinforcement, but the analysis does not establish an echo-chamber effect.Source: §4.2 · p. 7
5 Conclusion
The study finds that emotional labels on Vent are associated with both recent timeline emotions and network structure. These observational patterns are bounded by potential exposure and alternative explanations, and the authors call for causal and cross-platform follow-up.
5 Conclusion
- On Vent, users’ emotional labels were associated with pre-posting timeline composition, including within- and cross-category patterns, while highly aligned users tended to be closer in the network than under a degree-preserving null model.Source: §5 · p. 7
- Because the analysis used followee posts in a preceding time window, it captured potential rather than actual exposure; stable user profiles, emotional homophily, or common events may also explain the associations.Source: §5 · p. 7
- The authors identify causal identification and generalizability to other social media platforms as directions for future work.Source: §5 · p. 7