Temporal trends in the interaction network formed using the public Twitter stream

Abstract

Interaction networks from social networks such as Twitter reveal how users communicate with each other. Here we generate sampled interaction networks of users from the public Twitter stream starting from 2016, and measured their structure through maximum matching. We observed bimodality in the distribution of the unmatched node fraction. The bimodality reflects the presence of two distinct structures obtained from the interaction network when sampling, showing the sensitivity of the network to window sampling. With time, said bimodality started to occur at shorter sampling window sizes and hints at changes in interaction within the platform. It remains to be seen if the change occurs due to changes in user behavior, or is due to users responding to Twitter's algorithmic changes.