We use the Aurora Universities SDG Classifier on works' title/abstract text to classify works as contributing to the UN Sustainable Development Goals.
When we first launched the feature any work with a score above 0.1 for an SDG was given a tag of that SDG. However, we heard from users that this lead to an undesired balance of high recall (high likelihood of getting all publications from that SDG), with vastly lower precision (many publications erroneously tagged as contributing to the SDGs).
Through collaborations with Universities in Canada and Europe, we altered the classification cut-off scores and tested fit of results to their expectations. We found that 0.4 was a much better cut-off threshold-- values higher than that starting to miss SDG matches and values lower than that starting to bring in too many incorrect SDG matches. Therefore, we now use 0.4 as the relevancy score cut-off for determining SDG matches.
The great part about using an open-source classifier is that folks around the world can use the same classifier in their internal systems for documents that don't get indexed in OpenAlex (e.g., course syllabi, non-published research, grants).