Ridderprint

The history, evolution, and future of big data and analytics 33 Cluster (N) Id First Author, Year Weighted Degree Closen ess Between ness 481 Hammer, 2012 19 .232 .000 625 Kleih, 2011 19 .232 .000 2.4 Study 2: Algorithmic Historiography The development of a field over time can be displayed by ordering the most important publications in a field in the sequence in which they appeared, along with the citation relations between these publications (Garfield, 2004; Van Eck & Waltman, 2014a). Such an evolutionary visualization of a field illustrates the history of science and scholarship and has been referred to as an algorithmic historiography (Garfield, 2001; Garfield, 2004). Like other bibliometric methods, a historiography considers the relationships between various primary papers. However, the direction rather than the weight of this relationship is of importance as relationships are binary–a primary paper either does or does not cite a second primary paper. A historiography helps understanding paradigm shifts, as the changes in the citation of key papers of a field demonstrate how basic concepts and the overall perception of paradigms have changed over time (Garfield, Pudovkin, & Istomin, 2003). We conducted the historiography in CitNetExplorer (Van Eck &Waltman, 2014a) on the earlier described full sample of secondary papers. CitNetExplorer is a software tool for visualizing and analysing citation networks of scientific publications. It is especially useful for analysing the development of a research field over time as it shows how publications build on each other. CitNetExplorer reduces this full citation network in two ways. First, it identifies the “ core ” publications through the concept of k-cores (Seidman, 1983), where publications are considered core when they have a certain minimum number of ingoing or outgoing citation relations with other core publications. Van Eck & Waltman (2014a) consider publications core if they have citation relations with at least ten other core publications whereas Garfield and colleagues (2003) propose to limit the core publications to approximately 5% of the total number of publications. For our dataset, such settings resulted in a rather incomprehensive network with less than 21 publications. Therefore, we decided to expand this set iteratively - balancing the network’s comprehensiveness and interpretability - which resulted in an optimal network of fifty core publications (approximately 15% of the total number of publications). Second, CitNetExplorer performed a so-called transitive reduction of the citation network. Here, the program distinguishes essential from non-essential citation relations in the network, and only the essential relations are retained (Van Eck & Waltman, 2014a). Citation relations are classified as essential if there are no other pathways (i.e. relations) connecting two publications. Removing all non-essential relations minimizes the edges in the network while ensuring that all previously connected publications still have a pathway connecting them. Next, CitNetExplorer draws the nodes in the resulting network