Similarity indexes for scientometric research: A comparative analysis

Abstract

A significant number of papers in the field of scientometrics addressed the comparisons of various similarity indexes. However, there is still a debate on the appropriateness of an index compared to others, beacause of the assessment differences reported in the literature. The objective of this paper is to make a comparative analysis of the five most used similarity indexes for the three scientometric analysis types: co-word, co-citation and co-authorship. A total of 388 papers addressing similarity indexes in scientometric analysis over three decades were retrieved from the Web of Science and examined; of which 49 were retained as the most relevant according to selective criteria. The approach consisted of building cross matrices for the five indexes (Jaccard, Dice-Sorensson, Salton, Pearson, and Association Strength) for the three types of scientometric analysis. For each of these analyses, a distinction is made between papers according to their theoretical or empirical results. Furthermore, papers are classified according to the mathematical formula of the similarity index being used (vector vs non vector). In the 49 relevant papers being selected, the comparative analysis showed that there is still no consensus on the appropriateness of an index for co-word and co-authorship analyses, while for co-citation, Salton is the widely preferred one. The Association Strength is the less covered and compared to other indexes for the three analysis types. An open source computer program was developed as a tool to facilitate empirical comparative studies of indexes. It allows generating normalized matrix of any chosen index for the two mathematical variants.