Dynamics of social interaction: Modeling the genesis of scientific collaboration

Collaboration plays a key role in physics and in the broader scientific enterprise. Here we develop a mathematical model for predicting new collaborations. We demonstrate that a simple ordinary differential equation model is a good fit to a data set that tracks collaborations resulting from four series of annual conferences on diverse scientific topics, 12 conferences in total over a period of five years. The model, inspired by the physics of catalysis, attempts to quantify the time-varying probability that any pair of individuals will initiate a new collaboration. It takes as input the pair's prior familiarity with one another as well as their pattern of interaction over time, and incorporates the effect of temporally decaying memory. This model accurately reproduces the collaborations formed across all first-year conferences in the four series and outperforms seven other candidate models. We also find evidence that prescribed interaction can lead to novel team formation, with observed collaboration probabilities increased by almost an order of magnitude. These results suggest that encounters among individual researchers at conferences, including encounters engineered by organizers, play an important role in shaping the future of science.