Building a knowledge graph with automatically acquired publication classifications

Abstract

Research information systems represent an optimal data source for research analyses of universities. In order to guarantee these analysis qualities, it is necessary to provide a data structure that covers a large information content. An important part of this is the classification of researchers into research categories in order to build up research profiles. Since such a classification process is very time-consuming, it is necessary that such a classification takes place automatically. For automation, the research information system of Leipzig University (leuris) has available as text data source the research papers with titles and abstracts, as well as partly the full text papers. Using state of the art text analytical methods, we generate a Topic Model which can assign one or more of its Topics to the individual publications. By linking the publications with the authors, a knowledge graph can be built up, which provides a good structure for detailed search queries. One of the key features of the knowledge graph is the ability to extract research profiles of researchers. Leveraging these opportunities, it is possible to find researchers with similar research interests and promote collaboration, if they have not been aware that there are other researchers doing research in the same direction.