Mendeley readership as a filtering tool to identify highly cited publications

TL;DR

This study demonstrates that Mendeley readership scores are more effective than journal citation scores in identifying highly cited publications across disciplines and can serve as early impact indicators for recent research.

Contribution

It provides large-scale evidence that Mendeley readership scores outperform journal citation scores in predicting highly cited papers and highlights their potential as early impact indicators.

Findings

Readership scores outperform citation scores in identifying highly cited papers.

86.5% of publications have at least one Mendeley reader.

Readership scores are more prevalent for recent publications.

Abstract

This study presents a large scale analysis of the distribution and presence of Mendeley readership scores over time and across disciplines. We study whether Mendeley readership scores (RS) can identify highly cited publications more effectively than journal citation scores (JCS). Web of Science (WoS) publications with DOIs published during the period 2004-2013 and across 5 major scientific fields have been analyzed. The main result of this study shows that readership scores are more effective (in terms of precision/recall values) than journal citation scores to identify highly cited publications across all fields of science and publication years. The findings also show that 86.5% of all the publications are covered by Mendeley and have at least one reader. Also the share of publications with Mendeley readership scores is increasing from 84% in 2004 to 89% in 2009, and decreasing from 88% in 2010 to 82% in 2013. However, it is noted that publications from 2010 onwards exhibit on average a higher density of readership vs. citation scores. This indicates that compared to citation scores, readership scores are more prevalent for recent publications and hence they could work as an early indicator of research impact. These findings highlight the potential and value of Mendeley as a tool for scientometric purposes and particularly as a relevant tool to identify highly cited publications.