A scientist's impact over time: The predictive power of clustering with peers

Abstract

The identification of latent patterns in big scholarly data that concern the performance of researchers is a significant task because it can potentially impact scien-tific careers since they are based in funding and promo-tion. This article investigates the temporal evolution of a scientist's impact. Instead of taking a detailed, microscopic view that examines the citation curves of every scientist's article, the article develops a scalable, macroscopicmethodology that uses the articles' citation profiles to build a more abstract and high-level profile that characterizes a scientist. This profile is utilized to cluster scientists in a set of 'performance' clusters. To this end, established techniques such as Principal Com-ponent Analysis and Self-OrganizingMap clustering are employed as well as a set of proposed heuristics. The effectiveness of the proposed methodology is examined by comparing the resulting rankings with the outcomes of the peer-review procedures that resulted in the E. F. Codd and the Turing awards. The good match be-tween the outcomes of computerized and peer-review procedures provides solid evidence that the proposed techniques constitute a promising analysis method for big scholarly data. © ACM 2016.