Detecting rising stars in dynamic collaborative networks

In today's complex academic environment the process of performance evaluation of scholars is becoming increasingly difficult. Evaluation committees often need to search in several repositories in order to deliver their evaluation summary report for an individual. However, it is extremely difficult to infer performance indicators that pertain to the evolution and the dynamics of a scholar. In this paper we propose a novel computational methodology based on unsupervised machine learning that can act as an important tool at the hands of evaluation committees of individual scholars. The suggested methodology compiles a list of several key performance indicators (features) for each scholar and monitors them over time. All these indicators are used in a clustering framework which groups the scholars into categories by automatically discovering the optimal number of clusters using clustering validity metrics. A profile of each scholar can then be inferred through the labeling of the clusters with the used performance indicators. These labels can ultimately act as the main profile characteristics of the individuals that belong to that cluster. Our empirical analysis gives emphasis on the “rising stars” who demonstrate the biggest improvement over time across all of the key performance indicators (KPIs), and can also be employed for the profiling of scholar groups.