Paying Attention to Each Other in Visible Work Communities: Modeling Bursty Systems of Multiple Activity Streams

Abstract

Online work projects, from open source to wikipedia, have emerged as an important phenomenon. These communities offer exciting opportunities to investigate social processes because they leave traces of their activity over time. We argue that the rapid visibility of others' work afforded by the information systems used by these projects reaches out and attracts the attention of others who are peripherally aware of the group's online space, prompting them to begin or intensify their participation, binding separate individual streams of activity into a social entity. Previous work has suggested that for certain types of bursty social behavior (e.g. email), the frequency of the behavior is not homogeneously distributed but rather can be divided into two generative mechanisms: active sessions and passive background participation. We extend this work for the case of multiple conditionally independent streams of behavior, where each stream is characterized by these two generative mechanisms. Our model can characterized by a double-chain hidden markov model, allowing efficient inference using expectation-maximization. We apply this model to visible work communities by modeling each participant as a single stream of behavior, assessing transition probabilities between active sessions of different participants. This allows us to examine the extent to which the various members of the community are influenced by the active participation of others. Our results indicate that an active session by a participant at least triples the likelihood of another participant beginning an active session.