Using Graph Visualization to enhance representation and evaluation of work clusters

Abstract

The German National Library (DNB) uses the platform Culturegraph (www.culturegraph.org) to aggregate metadata of library holdings of German and Austrian library networks. In this data pool of more than 171 million items we perform work clustering. By offering an aggregated view of all publications representing a work, e.g. different editions and translations, the collection appears much more structured and searching is easier. In this paper we would like to show the use of graph visualization for display, analysis and evaluation of work clusters. Graph visualization enables users to obtain a more transparent view of the connections underlying the structure of a work cluster. Work clustering is achieved by creating and matching keys which combine different metadata elements of a bibliographic record. Applying a breadth-first search, publications with identical matchkeys are grouped together. We use different keys to represent a publication, so each publication can obtain more than one key. Thus, one matchkey establishes a connection between publications. If one of the publications shares a different matchkey with even more publications the network representing a work grows. Moving beyond visualization we can also gain statistical indicators from the graph to evaluate a work cluster. Degree, average path length or centrality measures can offer information about the internal structure of a work cluster. Particularly with large clusters information about the degree of connections between the members and the existence of more closely related subclusters is important for evaluating clusters. Using graph visualization thus not only assists in grasping the internal structure of a work cluster more clearly but also helps managing and evaluating large datasets, ultimately leading to better clustering results to support data representation and findability.