The comparison of various similarity measurement approaches on interdisciplinary indicators

How to measure the interdisciplinary is a crucial topic in Interdisciplinary research (IDR), and the integrated indicators (e.g., Rao-Stirling) that combine three distinct components (variety, balance and disparity) has become one of the most promising attempts. Among the three components, variety and balance play relatively straightforward roles in diversity assessment but to what extent the (dis)similarity measuring approaches may affect the interdisciplinarity indicators is seldom discussed in the literature. In this paper, we compare various similarity measurement approaches from (1) different subject classification systems, (2) different normalization of (dis)similarity measure, (3) different (dis)similarity matrices of subjects, (4) different time windows; and (5) different levels of aggregations, using the academic publications labeled “Article” in eight selected journals published during the period 2009–2018 were selected as the sample dataset. Our results corroborate the following findings: First, a finer classification system with more subject categories increases the possibility that one cites sources from different subject categories. Second, different normalization approaches may lead to obviously different interdisciplinarity results, and such a finding is supported by the relatively low correlations between the interdisciplinarities calculated by Salton’s Cosine and Ochiai’s Cosine. Third, on the basis of Salton's cosine normalization, the interdisciplinary values obtained by different settings are highly correlated, especially in terms of different citation similarity matrices (cited, citing and cross-citation) and, in general, with different time windows. Fourth, results based on an aggregated dataset tend to overly expand the 'interdisciplinarity' degree of a journal, especially when the focused journal is actually 'multidisciplinary'.