Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evaluating the effect of viral news on social media engagement

Emanuele Sangiorgio, Niccolò Di Marco, Gabriele Etta, Matteo Cinelli, Roy Cerqueti, Walter Quattrociocchi

TL;DR

The paper addresses how viral posts affect user engagement on Facebook and YouTube, across four European languages, by applying a Bayesian structural time-series (BSTS) based comparative interrupted time-series (CITS) framework to post-viral engagement windows. It defines virality via a bivariate spreading and interactions metric, detects viral posts with platform-specific z-score thresholds, and evaluates impact with two research questions on growth and persistence. The key findings show that viral events are typically transient, often failing to produce sustained engagement, with a stronger boost when virality occurs as a sudden exogenous shock rather than after a prior growth phase; faster virality tends to fade sooner, while slower, sustained growth yields more persistent effects. The study highlights the elastic nature of collective attention and argues for continuous, steady engagement strategies over reliance on viral spikes, with implications for news outlets and social platforms alike.

Abstract

This study examines Facebook and YouTube content from over a thousand news outlets in four European languages from 2018 to 2023, using a Bayesian structural time-series model to evaluate the impact of viral posts. Our results show that most viral events do not significantly increase engagement and rarely lead to sustained growth. The virality effect usually depends on the engagement trend preceding the viral post, typically reversing it. When news emerges unexpectedly, viral events enhances users' engagement, reactivating the collective response process. In contrast, when virality manifests after a sustained growth phase, it represents the final burst of that growth process, followed by a decline in attention. Moreover, quick viral effects fade faster, while slower processes lead to more persistent growth. These findings highlight the transient effect of viral events and underscore the importance of consistent, steady attention-building strategies to establish a solid connection with the user base rather than relying on sudden visibility spikes.

Evaluating the effect of viral news on social media engagement

TL;DR

The paper addresses how viral posts affect user engagement on Facebook and YouTube, across four European languages, by applying a Bayesian structural time-series (BSTS) based comparative interrupted time-series (CITS) framework to post-viral engagement windows. It defines virality via a bivariate spreading and interactions metric, detects viral posts with platform-specific z-score thresholds, and evaluates impact with two research questions on growth and persistence. The key findings show that viral events are typically transient, often failing to produce sustained engagement, with a stronger boost when virality occurs as a sudden exogenous shock rather than after a prior growth phase; faster virality tends to fade sooner, while slower, sustained growth yields more persistent effects. The study highlights the elastic nature of collective attention and argues for continuous, steady engagement strategies over reliance on viral spikes, with implications for news outlets and social platforms alike.

Abstract

This study examines Facebook and YouTube content from over a thousand news outlets in four European languages from 2018 to 2023, using a Bayesian structural time-series model to evaluate the impact of viral posts. Our results show that most viral events do not significantly increase engagement and rarely lead to sustained growth. The virality effect usually depends on the engagement trend preceding the viral post, typically reversing it. When news emerges unexpectedly, viral events enhances users' engagement, reactivating the collective response process. In contrast, when virality manifests after a sustained growth phase, it represents the final burst of that growth process, followed by a decline in attention. Moreover, quick viral effects fade faster, while slower processes lead to more persistent growth. These findings highlight the transient effect of viral events and underscore the importance of consistent, steady attention-building strategies to establish a solid connection with the user base rather than relying on sudden visibility spikes.
Paper Structure (24 sections, 11 equations, 6 figures, 3 tables)

This paper contains 24 sections, 11 equations, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Literature review
  3. Attention economy
  4. Human dynamics on social media
  5. Virality on social media
  6. Research design
  7. Data
  8. Defining virality
  9. Spreading
  10. Interactions
  11. Detecting virality
  12. Evaluating virality impact
  13. Research Question 1: Does virality induce engagement growth?
  14. Research Question 2: Do the faster-manifesting effects persist longer?
  15. Results
  16. ...and 9 more sections

Figures (6)

  • Figure 1: Distributions of Spreading and Engagement $z-$scores for Facebook and YouTube. The inset plots of each chart display the subset of values exceeding the threshold for the given metric, showing its breakdown between viral contents (which exceed the threshold in both metrics) and non-viral ones (hence exceeding the threshold only in the displayed metric).
  • Figure 2: Distributions of viral posts per source on Facebook and YouTube. Facebook sample sizes: Pages = 613, Posts = 9514; YouTube sample sizes: Channels = 161, Videos = 3624.
  • Figure 3: Effects after virality with increasing time windows. The $y$-axis represents the percentage of Growth, No Effect, and Decrease cases, for each time scale on the $x$-axis. Links represent the effects flow between consecutive weeks, color-coded according to the first observed effect (i.e., in the 2-week window).
  • Figure 4: Density of the trend preceding the viral post and the average absolute effect on engagement for the 2-week time window. Trend Pre Virality is the $\beta1$ coefficient of the regression estimated by the BSTS on the weeks preceding virality. Given its previous trend, the average Absolute Effect is the average effect on the Engagement after the viral event. Only events with a statistically significant effect on Engagement are shown.
  • Figure 5: Persistency of the virality effect based on its time of emergence. Solid lines represent the estimated decay for the observed curves - from the 2nd week to the 5th week of emergence time - along with their observed values. The dotted lines represent the corresponding extrapolated decay curves for 0 and 1 week after virality as emergence time.
  • ...and 1 more figures