Social network dynamics in a conference setting
(disclaimer: This was not peer reviewed and is not serious at all :)
To study the dynamics in social network topology we decided evaluate how some nodes (also called humans) interact in defined experimental conditions. We used the scientific meeting setting that we think can serve as a model for this type of studies. We observed human-human interactions during the meeting breaks by taking snapshots and calculating inter-human distances. We defined an arbitrary cut-off to determine the binary interactions between all the humans present in the study.
The first analysis we preformed was under the so call "conference breaks" model where our nodes are allowed to interact for brief time intervals after being subjected lengthy lectures.
We observed an interesting clustered network topology that can be described with a power law distribution. Most nodes in the network have few interactions while a small fractions of humans was found to consistently interact with a large number of other nodes. We found also some nodes that did not show any interactions in our studies even when several "conference breaks" were preformed. We believe that these could be pseudo-humans that were included in our study by mistake. These pseudo-humans might be on the way to extinction from the humeone.
Having built this network of human-human interaction on a large scale we decided to investigate what human properties might be correlated with human hubs. We used previous large-scale studies of human properties like height, gender and number of papers published to test this.
We show here that although gender shows a significant correlation with human hubness, the best predictor for hubs in the conference breaks networks is actually number of papers published. We tried to refine this further by introducing a new human measurement we call "hypeness". Hypeness of a human was calculated as a modification of the number of papers published weighted by the impact factor of the journals where the papers were published and also the number of times cited in popular media articles. We show here that hypeness does significant better at predicting hub nodes in this network.
Given that networks are dynamic we set out to map the changes in network structure with time. To simulate this we perturbed the gathering using a small-compound (EtOH) that we administered in liquid form. With time we observed a noticeable change in the network. Although the overall topological properties were maintained, the nature of the hubs changed dramatically. In this new network state that we call the "drunk" state, the best predictor for the highly connected hubs is clearly gender. We believe this clearly proves that social networks in conference settings are very dynamic with time.
To prove that gender was indeed the best indicator of hubness and not some strange artifact we used deletions studies. Random female nodes where struck with a sudden case of "sleepiness" and the perturbed network was observed. We show here that random female deletion leads to a rapid collapse of the network. The same is not observed with random deletion of the hypest nodes, proving our initial proposition.