Community Aliveness: Discovering Interaction Decay Patterns in Online Social Communities

摘要

Online Social Communities (OSCs) provide a medium for connecting people,sharing news, eliciting information, and finding jobs, among others. Thedynamics of the interaction among the members of OSCs is not always growthdynamics. Instead, a $\textit{decay}$ or $\textit{inactivity}$ dynamics oftenhappens, which makes an OSC obsolete. Understanding the behavior and thecharacteristics of the members of an inactive community help to sustain thegrowth dynamics of these communities and, possibly, prevents them from beingout of service. In this work, we provide two prediction models for predictingthe interaction decay of community members, namely: a Simple Threshold Model(STM) and a supervised machine learning classification framework. We conductedevaluation experiments for our prediction models supported by a $\textit{groundtruth}$ of decayed communities extracted from the StackExchange platform. Theresults of the experiments revealed that it is possible, with satisfactoryprediction performance in terms of the F1-score and the accuracy, to predictthe decay of the activity of the members of these communities usingnetwork-based attributes and network-exogenous attributes of the members. Theupper bound of the prediction performance of the methods we used is $0.91$ and$0.83$ for the F1-score and the accuracy, respectively. These results indicatethat network-based attributes are correlated with the activity of the membersand that we can find decay patterns in terms of these attributes. The resultsalso showed that the structure of the decayed communities can be used tosupport the alive communities by discovering inactive members.