A Data-Driven Unified Framework for Predicting Citation Dynamics

Abstract

With the rising interest in predicting the scientific output, various efforts have been made to predict a scientist's h-index or the citation trajectory of a publication. In this work, we employ a dynamic categorization for scientists to ensure at each stage of their careers a comparison amongst their peers and combine this grouping with predictive models to estimate a scientist's future impact, as expressed by citation counts. Moreover, we investigate a wide range of factors identifying their importance in determining the future of science for different performance and academic levels with particular emphasis on features describing a scholar's position in multi-layered collaboration and citation networks. The robustness of the approach is examined on a longitudinal dataset centered around 700,302 data points representing Computer Scientists in various time periods with their complete networks of over 18 million collaboration links and 36 million citations. Our results indicate up to 30 percent improvement in prediction performance compared to baseline methods along with an average R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> =0.96 for short term and R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> =0.91 for long term predictions.