Challenges in managing semantic annotations in harvested research objects in a national CRIS context

Abstract

Harvested metadata on research objects can include links between the primary domain objects such as organizational identifiers associated with dataset, persons identified with ORCIDs linked to publications and publications connected through ISSNs to publishing channels. This kind of linkage is the bread-and-butter of the CRIS systems and usually comprehensively maintained. When it comes to the more subjective description of a domain object, such as keywords, themes, or subject headings, the issues related to data management and modeling become prominent with challenges such as flexibility of free text keywords as opposed to authoritative, but rigid classification systems. Many CRIS objects also already contain an extensive description of the content, just meant for human consumption, in the form of an abstract or similar summary text. With the help of automated data mining and annotation tools, these textual representations can be processed into structured data. This paper presents the processing pipelines implemented as part of the research.fi portal for automatic linking of different research inputs based on automatically extracted ontology concepts and discusses the implications of utilizing them as part of the research.fi platform. But more than simply discussing the annotation of research objects and the creation of word clusters for representation of the semantic content of research objects, we also discuss challenges related to maintaining the automatically produced metadata, as the utilized ontologies evolve, annotation algorithms develop, connections between research objects and mined word clusters change over time.