CSH Talk by Max Pellert: “Emotional expression dynamics in social media”

The lecture by Max Pellert will take place at the Complexity Science Hub Vienna in room 201.

If you are interested in participating, please email to office@csh.ac.at

Abstract

A dataset of general social media text-based expression was analyzed: 22 million status updates were donated by 153727 Facebook users. By applying a lexicon-based approach, we quantified emotions in each post in terms of valence, the degree of pleasure associated with an emotional experience, and arousal, the level of activity induced by the emotional experience.

We further replicated our analysis with the unsupervised VADER method and supervised valence and arousal methods specific for Facebook. We analyzed how the emotions of a post predict the emotional content of the next post by the same user as a function of time between posts ∆t. It can be observed that emotional expression relaxes quickly but not instantly, as changes in short timescales are directed towards the baseline but do not fully converge. We captured this dynamics through a dynamic equation model following an agent-based modelling framework for emotions in online interaction. This way we modelled emotion eigendynamics as an example of an Ornstein–Uhlenbeck process.

We fitted the solution of this model against both datasets through nonlinear regression and came up with three main results: i) the presence of emotional expression indicates emotion regulation, evidencing a correction of emotional intensity as soon as a message is written, ii) valence and arousal decay exponentially towards their baselines, and iii) the baseline of valence is above its midpoint but the baseline of arousal is slightly below its midpoint (on a 1-9 scale). The above results are in line with well-established dynamics of emotions from previous research in psychology, but our analysis has several advantages. We observe a much longer period (for Facebook on average 534.36 days) over a larger population than the largest previous works with self-reports. Additionally, we capture general emotion dynamics rather than changes due to emotion labelling: This enables us to quantify individual dynamics that inform future computational models and analyses of emotional expression in social media.