Networks and Information Filtering in Science

Abstract

During their scientific career, researchers make numerous choices about who they work with and what previously established knowledge they apply to create new findings. These decisions are influenced by, and in turn influence, the social and information networks scientists are embedded in. With an ever-increasing number of scientific publications, amounting to a body of knowledge far beyond what a single human mind can comprehend, there is a desire and need to reduce the available information. For example, to aid in identifying the most significant knowledge, statistical summaries, or indicators, are generated from citation data. Unfortunately, this form of information filtering can be problematic due to the different mechanisms that contribute to the accrual of citations. In this thesis, I empirically study information and social dynamics in science by analyzing the networks that emerge through collaborations and citations of millions of authors, extracted from a publication record database that covers most scientific disciplines and provides decades of publication metadata. The dissertation consists of four projects which together form the conceptual and methodological basis for a citation indicator, the s-index. Each of these projects implements a different aspect of an information system that can be used to explore the role of an author's network when evaluating scientific performance. First, I show that citation counts are strongly related to network distance. Successful authors are likely to be embedded in dense communities with short distances to potentially citing publications. Thus, I formulate an index that ranks authors according to the citation potential of their network position and then controls citation counts by comparing authors only to others with a similar network potential. Second, since even the evaluation of a single author requires full network information and also scores of all other authors, data needs to be prepared accordingly. Thus, I design an author name disambiguation method that can extract the required information from extensive databases. Third, to supplement the index with additional author-specific information, I develop a large-scale network layout algorithm to visualize the complete scientific community from the perspective of an individual scientist. Finally, to support transparency and reproducibility for such data-intensive computational research, I propose a systems design for an automatic reproduction of all generated data, that can also be applied to any scientific publication that is based on computational results.