Characterization of Emotional Contagion in Collaborative Decision Support Systems

This research is focused on the analysis of how emotional aspects of people participating in decision-making through group discussions in online platforms manifest themselves. A particular emphasis is on the toxicity levels of discussions on various issues. Three platforms have been chosen for a comparative study: the first is a Reddit-like online forum, the second is Polis, where participants may vote on issues and is touted as a forum for computational democracy, and the third is called the Deliberatorium, developed at the Center for Collective Intelligence at the Massachusetts Institute of Technology. These platforms have their own specific structures that aid in group discussions on various issues. An important question that has been examined in this paper is how to characterize the contagion of toxicity in the three platforms and the extent to which their structures inhibit it. Our approach has been to first extract two sets of actual data from the three platforms: a control group and an experimental group where in the latter, an intervention is used by inserting an empathy statement to inhibit toxicity. Toxicity levels (determined using Google Perspective API) were compared using sample data from the three platforms. We use generative models that extend the sample datasets by adding synthetic nodes (artificial entries in group discussions) from the three platforms based on parameter values of popularity, novelty, root-bias, and reciprocity among participants. Two additional parameters of node type and toxicity score were added to these synthetic nodes. The generative models help in extending existing datasets to determine the contagion of toxicity in the platform. This is achieved by developing an index for toxicity contagion, EC tox expressed as a percentage of the total number of nodes. The idea of developing this index has been adapted from work in ecological studies where contagion of a certain type of land patch is examined in the presence of several different patch types. We use this metric to assess toxicity contagion in both the control and experimental groups of the three platforms. With the intervention of the empathy statement, Deliberatorium had an improvement in EC tox from 1.309% to 1.297%. In Forum, however, EC tox actually increased from 1.711% to 1.951%. Finally, in Polis, there was an improvement in EC tox from 1.750% to 1.663%. Thus, for Deliberatorium, the flow of toxicity is the least. We infer that there are inherent structural constructs in the design of the Deliberatorium, and that may naturally inhibit toxicity in group discussions.