Predicting Most Influential Paper Award Using Citation Count

Abstract

The early identification of the influential papers is of great significance for assessing the scientific achievements of researchers and institutions as it can help in addressing the processes in an academic and scientific field, such as promotions, recruitment decisions, and funding allocation. This work evaluates features for predicting the most influential paper award that is given by several renowned conferences, ten years subsequent to their publication. The data of five renowned conferences, i.e., ICSE, ICFP, POPL, PLDI, and OOPSLA is used to predict the long-term citations to identify the most influential paper of the respective conference. GD boost model is considered to be better performing among the five different machine learning algorithms. The results show that a three to five years of the time window is good enough to evaluate the most influential paper award. Additionally, the assessment of time window and the citation trajectory of awarded and non awarded papers shows that the citation trajectory of the awarded paper vary from the Citation gain patterns of non-awarded paper.