Distributional characteristics of Dimensions concepts: An Empirical Analysis using Zipf’s law

Abstract

The massive growth in scholarly outputs during the last few decades has resulted into the creation of several scholarly databases to index the outputs. These scholarly databases index publication records and provide different metadata fields for different kinds of usage ranging from retrieval and research evaluation to various scientometric analysis. The ‘author keywords’ is one such important metadata field provided by many databases and used for different text-based and thematic structure analysis. The Dimensions database, however, does not provide ‘author keywords’ metadata field, instead it provides automatically generated terms from the article full texts, called ‘concepts’. Therefore, it is not clear whether different text-based analysis can be done with data provided by Dimensions database. Therefore, this article explores the distributional characteristics of Dimensions concepts. The Dimensions concept data obtained for a sufficiently large sample of scholarly articles is analysed through rank frequency distribution plots in the log–log space. Existence of Zipfian distribution is explored. The results indicate that Dimensions concepts adhere to the Zipfian properties which in turn indicates that Dimensions concepts have similar distributional characteristics as author keywords and hence they may have the same expressive power as that of author or index keywords for scientometric exercises. The study is novel as it is the first study to explore the distributional characteristics of the Dimensions concepts, particularly with respect to Zipfian properties, which provide the statistical foundation for understanding the Dimensions concepts and help to model and analyse them.