Brinkeys is a demonstrator that suggest a number of Brinkman topics for a dissertaton based on the contents of the dissertation.
For centuries, the Koninklijke Bibliotheek (KB), the Dutch National Library, has been collecting and storing all publications of Dutch authors as well as any publication on the Netherlands. To ensure findability of its books, maps, and other items, dozens of employees manually attach keywords and other metadata to these items. Recently digital publications are also being collected. These digital sources allow for extraction of text and this provides opportunities for automated or computer-assisted assignment of keywords and metadata.
ICT with Industry Workshop 2019
To explore these opportunities, the KB offered a use case to the ICT with Industry Workshop 2019 focusing on the research question: “To what extent can scientific texts automatically be labeled with relevant keywords?” The goal was to automatically suggest keywords from the Dutch Brinkman thesaurus (‘Brinkeys’) to PhD dissertations from six Dutch universities. The suggested keywords are to be used by KB annotators to quickly select the relevant keywords, which could save a significant amount of time in the day-to-day work of the annotators, as well as increase the inter-annotator-reliability. During the workshop, we developed and tested a number of methods to compare different approaches. Please find the full report in pdf with the results of the workshop on this website.
To demonstrate the possibilities of applying techniques such as fasttext, we built this tool ‘Brinkeys’. It showcases an online system where a limited number of predefined dissertations can be analysed.
This information is based on the poster presentation by Alex Brandsen at the ICT Open 2019 conference, see Brandsen, A. (2019), Brinkey-generator - Computer Assisted Assignment of Thesaurus Topics for Scientific Texts. ICT Open 2019, Hilversum, The Netherlands
Please also see the more elaborate report which described the research process around Brinkeys.
When citing this tool we request you cite it as follows:
Brandsen, A., Kleppe, M., Veldhoen, S., Zijdeman, R., Huurman, H., Vos, H. De, Goes, K., Huang, L., Kim, A., Mesbah, S., Reuver, M., Wang, S., Hendrickx, I. (2019), Brinkeys. KB Lab: The Hague, the Netherlands
After selecting a dissertation, the interface will suggest a number of possible Brinkman topics, as predicted by the fasttext method.
The user can then view this list, as well as parent and child topics of the predicted keywords, and select which they think are relevant by comparing it to the title and abstract of the dissertation. This is to simulate how the system might be used by KB annotators in the near future.
The keywords selected by the user are then compared to the actual annotation in the KB system with a score for overlap, which would give the user an idea of how ‘good’ their selection is.