Scientific impact of an author and role of self-citations

• Published in: Scientometrics Vol. 122; no. 2; pp. 915 - 932
• Main Authors: Amjad, Tehmina, Rehmat, Yusra, Daud, Ali, Abbasi, Rabeeh Ayaz
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.02.2020; Springer Nature B.V
• Subjects: Bayesian analysis; Bibliometrics; Citation analysis; Computer Science; Hirsch index; Information Storage and Retrieval; Library Science; Prediction models; Regression analysis; Scientometrics; Self citation; Trends; Self-citations; H-index; Predictive models; Author impact factor; Self-citations trends
• ISSN: 0138-9130; 1588-2861
• DOI: 10.1007/s11192-019-03334-2

In bibliometric and scientometric research, the quantitative assessment of scientific impact has boomed over the past few decades. Citations, being playing a major role in enhancing the impact of researchers, have become a very significant part of a plethora of new techniques for measuring scientific impact. Self-citations, though can be used genuinely to credit someone’s own work, can play a significant role in artificial manipulation of scientific impact. In this research, we study the impact of self-citations on enhancing the scientific impact of an author using a dataset retrieved from AMiner ranging from 1936 to 2014 from the computer science domain. We investigated the relations among trends of self-citation and their influence on scientific impact. We also studied its influence on ranking metrics including author impact factor and H-Index. By analyzing self-citations over time, we discover five basic self-citation trends, which are early, middle, later, multi and none. Distinctly different patterns were observed in self-citations trends. The results show that self-citations, if totally removed from total received citations, negatively influence the AIF and H-Index values and hence can be used to artificially boost the scientific impact. We used regression-based prediction models to predict the influence of self-citations on future H-Index. Classifiers including Logistic Regression, Naïve Bayes and K-NN were used with an accuracy of 93%, 73% and 60% respectively.