Analysis of Scientific Collaboration Networks

In scientometrics, the quantitative study of science, network analysis has become a prominent tool. The kinds of networks most frequently examined have been citation networks (mapping links between publications based on references) and collaboration networks (mapping the collaborative relationships between researchers based on joint publications) [Mil04][New01]. Collaboration networks are at the center of this work, too. Based on the extensive DBLP publication database, relations between authors and publications are compiled as graphs, mapping almost the entire field of computer science. This allows us to examine and quantify the collaborative relations between researchers using graph-based methods. The focus on collaboration is supported by empirical results showing that research increasingly happens in the form of teamwork, and publications produced by teams of authors typically reach a greater impact [WJU07][Sol09]. In the course of this work, the collaboration network is explored from several angles. A variety of techniques for the analysis of complex networks, especially social network analysis, yield insights into its composition. Of particular interest here is whether the natural components of the network correspond to separate fields with regard to research content. These analyses give us a general picture of the network and lead up to the central point of this work, the question how the network is shaped by social events in the academic realm: The Dagstuhl seminars assemble researchers with the goal of fostering (collaborative) work in cutting-edge areas of computer science. We examine whether such events leave a track in the structure of the network. Furthermore, the significance of centrality in the collaborative network is explored, both for authors and for publications. The structure of the work can be outlined briefly as follows: After some preliminary definitions (Section 1.2), Chapter 2 describes in detail the preparation of source data, leading to the modeling of the collaboration network as different graphs. Chapter 3 briefly highlights some technical aspects of the implementation. In Chapter 4, standard concepts from network analysis are applied in order to report general properties of the resulting network, including connectedness, degree distribution and k-core decomposition. Chapter 5 is concerned with clustering, i.e. partitioning the graph into internally dense subgraphs. After calculating a clustering of the collaboration network, we look into the question whether authors form collaborative clusters according to the topical similarity of their work. As a core element of this work, we then explore the question whether participation in seminars designed to foster collaboration, the …