Source-linked AI summary
Collective emotions online and their influence on community life
Anna Chmiel, Julian Sienkiewicz, Mike Thelwall, Georgios Paltoglou, Kevan Buckley, Arvid Kappas, Janusz A. Hołyst
TL;DR
The paper examines whether emotions in online exchanges arise through intra-group interactions and influence e-community activity. Using sentiment analysis of over 4 million comments, it finds collective emotional clustering and longer BBC discussions following more emotional openings.
Problem
It is unclear whether emotions in e-communities primarily reflect members’ personalities or intra-group interactions, and whether they influence group activities.
Method
The study applies sentiment-analysis classifiers to over 4 million comments from blogs, BBC forums, and Digg, analyzing emotional clusters and discussion threads.
Results
Long same-valence clusters exceeded independent-message expectations, with preferential cluster-growth patterns and longer BBC discussions following more emotional first ten comments.
Takeaways & Limitations
Internet communication can create and modulate collective emotional states, while emotional expressiveness sustains some e-communities.
Takeaways & Limitations
The data-collection and sentiment-classifier methods contain errors that are only partly estimable, although random classifier errors are more likely to break than create clusters.
Abstract
from arXiv · showhide
E-communities, social groups interacting online, have recently become an object of interdisciplinary research. As with face-to-face meetings, Internet exchanges may not only include factual information but also emotional information - how participants feel about the subject discussed or other group members. Emotions are known to be important in affecting interaction partners in offline communication in many ways. Could emotions in Internet exchanges affect others and systematically influence quantitative and qualitative aspects of the trajectory of e-communities? The development of automatic sentiment analysis has made large scale emotion detection and analysis possible using text messages collected from the web. It is not clear if emotions in e-communities primarily derive from individual group members' personalities or if they result from intra-group interactions, and whether they influence group activities. We show the collective character of affective phenomena on a large scale as observed in 4 million posts downloaded from Blogs, Digg and BBC forums. To test whether the emotions of a community member may influence the emotions of others, posts were grouped into clusters of messages with similar emotional valences. The frequency of long clusters was much higher than it would be if emotions occurred at random. Distributions for cluster lengths can be explained by preferential processes because conditional probabilities for consecutive messages grow as a power law with cluster length. For BBC forum threads, average discussion lengths were higher for larger values of absolute average emotional valence in the first ten comments and the average amount of emotion in messages fell during discussions. Our results prove that collective emotional states can be created and modulated via Internet communication and that emotional expressiveness is the fuel that sustains some e-communities.
3 School of Humanities and Social Sciences, Jacobs University Bremen, Bremen, Germany
Using sentiment analysis of large online discussions, the paper finds that emotions cluster across consecutive posts and that emotional intensity can sustain some e-community discussions. These effects vary by community and emotional valence, while classifier and data-collection errors constrain measurement.
- Data and approach: The study analyzes over 4 million comments from blogs, BBC forums, and Digg using automatic sentiment analysis.Discussion threads were converted into message chains and analyzed for collective emotional patterns.
- Collective emotional clustering: Long same-valence clusters occur more frequently than expected under mutually independent messages.For BBC forums, 91 negative clusters of length 25 contrasted with an i.i.d. prediction of 6; Digg had 57 positive clusters of length 11 versus 2 predicted.
- Collective emotional clustering: Conditional probabilities of repeating an emotion increase with cluster length, supporting preferential processes rather than an i.i.d. process.The increase holds for n < 20 before saturation, and the characteristic exponent α measures the strength of the preferential process.
- Collective emotional clustering: Positive, negative, and neutral messages tend to provoke responses with the same valence across BBC forums, Digg, and Blogs06 blogs.The pattern appears in communities dominated by either negative or positive emotions, indicating collective effects across different e-community types.
- Valence and interaction strength: Larger α values indicate stronger collective behavior when the corresponding emotion is less frequent.The paper interprets α as an indirect measure of emotional interaction strength through the influence of the latest emotional cluster.
- Limitations: Data-collection and sentiment-classifier errors can only be partly estimated, although random classifier errors are expected to break clusters more often than create them.The authors therefore suggest that raw cluster statistics likely underestimate the strength of existing clustering.
- Discussion vitality: In BBC forums, higher emotional levels in the first ten comments correspond to longer discussions, whereas this effect is absent in Digg and Blogs06.The average emotional content of messages falls during discussions, and the initial emotional fuel is described as becoming exhausted when threads end.
- Implications: The study concludes that collective emotional states can be created and modulated online, with emotional expressiveness sustaining some e-communities.The findings support development of models of collective emotions and possible tools for monitoring discussion emotion levels.
Life – Supporting Information File S1
The supporting analysis models emotional cluster sizes and tests whether consecutive messages with the same valence exhibit preferential attraction rather than independence. It also examines how emotional sequences and discussion structure relate across online communities.
- Data structure: Chronological ordering simplifies forum-like BBC and Digg discussions for comparing emotional sequences across communities.Blogs posts were originally arranged differently, whereas BBC and Digg data were arranged chronologically.
- Cluster models: Under an i.i.d. process, cluster probabilities depend on the emotion frequency and decline geometrically with cluster size.The model assumes no dependence between consecutive events and a constant event distribution over time.
- Preferential attraction: Conditional probabilities of continuing the same emotion increase as a power law with cluster length, supporting preferential attraction across datasets.The scaling is observed for all datasets, although its range varies by community and emotional valence; BBC neutral clusters fit across the whole data range.
- Model limitations: The preferential-scaling approximation is bounded by assumptions on maximum cluster size and exponent, and produces spurious large-n behavior for Blogs06 neutral clusters.The normalization approximation requires nmax <40 or α<0.1; Blogs06 neutral clusters violate these conditions, and artificial scaling causes the spurious tail behavior.
- Preferential attraction: Preferential-attraction distributions fit the observed cumulative cluster sizes better than i.i.d. or Markov models, especially for large clusters.The comparison is reported in the supporting distribution analysis, while power-law fits are also better than linear fits despite a short scaling regime.