Speaker: Frank Schweitzer (ETH Zürich)

Title: Success and failure - A complex network perspective

Abstract: The last 1.5 decades have brought important insights into the structure and the dynamics of complex networks. However, suprisingly little is known about the impact of such features on the performance of a system, on the global level, or its constituents (agents), on the local level. Hence, we need to link structure and dynamics with performance in a quantitative, empirically testable way. This means that for the global perspective we have to specify how network resilience or systemic risk depend on the structural properties of the system and the dynamics of the agents. For the local perspective, we have to understand the feedback between the network position of an agent and its future success or failure as an invidual. Most importantly, these insights have to be developed to a degree that allows us to predict success and failure, both on the systemic and the agent level.
In the talk, I will provide examples of large-scale analyses of different social networks that allow us to obtain the insights required. Specifically, for the local perspective, I will show that we can predict if a bug report of a user is valid or faulty, i.e., worth to be dealt with, before actually resolving the bug - a result that saves a lot of resources in open source projects. With respect to success in science, I will demonstrate to what extent the social position of an author in a co-authorship network is indicative of getting her paper into the top ten percent of her field, five years later. For the global perspective, I will present an analysis of the success and failure of online social platforms that allows us to estimate cascades of users leaving, which may eventually lead to the collapse of the platform. From such analyses, we learn that neither the structure nor the dynamics of the networks alone can predict a systemic failure or success. Instead, we need to develop a better understanding of the internal dynamics of their individual constituents.