The ZBW collects and indexes publications in economics and business studies. In order to make them accessible and to ensure that they can be found in EconBiz, they are enhanced with metadata, e.g. on authors and publication types, but also on their content, via keywords (subject indexing).
The AI service AutoSE has been in use at ZBW since July 2021. It automatically assigns suitable keywords from the STW Thesaurus for Economics to new publications in EconBiz within a turnaround time of two hours. The machine learning models are regularly improved and retrained – the most recent training cycle was executed in September 2024.
The service also contributes to a better cooperation between humans and machines ("human in the loop"): the concepts generated by AutoSE are also made available to our subject indexing experts via a platform called "Digitaler Assistent" as suggestions that they can then adopt or reject. The experts also give assessments for the suggested concepts, helping to train the AI models and continuously improve the quality of their output.
AutoSE is already proving to be very successful:
- As in previous years, AutoSE has also annotated a large number of publications that were already part of ZBW holdings with subjects generated by the newly trained model, either overwriting subject indexing by older models or (in the case of 200.000 publications) annotating them with STW concepts for the very first time.
- Since AutoSE was introduced, keywords have been assigned to more than 1.9 million titles.
- This represents approximately 30% of ZBW's holdings and more than 50% of all English-language holdings.
Sounds interesting? Find out more about AutoSE here.