A project using the University of Oxford's "Their Finest Hour Online Archive" evaluated three Natural Language Processing approaches—Named Entity Recognition, Keyword Extraction, and Topic Modelling—to automate keyword extraction at scale. The study tested these methods across techniques ranging from traditional statistical models to modern generative AI neural networks.
- Quantitative and qualitative findings indicate that NLP approaches offer real potential for keyword extraction in crowdsourced collections, though no single method provides a complete solution.
- Model choice was found to significantly shape the results of the extraction process.
- Open-weight, extractive models emerged as best suited for responsible deployment compared to generative AI.
The authors argue that automated keyword extraction raises distinct stewardship responsibilities because metadata is the direct product of engagement with living contributors. They caution that while generative AI has abstractive potential, it introduces accountability risks that managers must weigh carefully.