Modeling Large-Scale Collaboration on GitHub

Members of loosely-coupled teams are often unaware of the work done by other team members. As such, identifying and sharing relevant information with a user about their teammates’ activity can enhance collaborative success. Determining the relevance of information becomes more difficult as teams grow large and users dispersed. In this work, we introduce a model to predict information relevance in largescale collaboration that addresses these difficulties. Our model also captures a one-to-many relationship between users and projects and allows for multiple types of user interaction. Using data from the GitHub Archive, we demonstrate that our approach can predict both user-repository interaction sequences and the likelihood of success for a given repository. Specifically, we show that our model predicts all types of user interactions at 65% precision and active user interactions at above 80%. It also predicts a repository’s stars and forks, two important metrics of success, within 5.7 of the true value. These findings bode well for the creation of large-scale information sharing systems that facilitate successful teamwork at scale.