Decide how to deal with keywords for FLoRA

[LukasWallrich]

It would be good to be able to search the database - but manually extracting keywords is a pain.

We could:

• take the OpenAlex keywords, but they truly are a mess. Consider https://api.openalex.org/w3126001372 and https://api.openalex.org/works/doi:10.1038/s41562-023-01655-0 as two fairly random examples.
• LLM code keywords based on journal, title & abstract - will certainly be less nonsensical, but not reproducible - but author keywords are not systematic either. Not sure if I would do that for articles where we don't have access to the abstract
• extract keywords from articles - but that would likely have to be the original rather than the replication, and only a subset of them has open versions.

Let's keep the OpenAlex keywords in the pipeline but note that they might get removed in the future?

[LukasWallrich]

I just find this one too funny not to share directly - definitely an article where ice cream is the most important keyword

this is what cheap (small) LLMs would give us

Gemini 3 Flash: Less-is-better effect; Evaluability hypothesis; Joint vs. separate evaluation; Preference reversal; Consumer judgment

Claude Haiku: less-is-better effect; evaluability hypothesis; joint vs. separate evaluation; preference reversal; attribute evaluation

ChatGPT 5 Nano: less-is-better effect; evaluability hypothesis; consumer judgment; separate vs juxtaposed evaluation; decision making

and Kimi K2 (probably the "smartest" cheap model): Less-is-better effect; Evaluability hypothesis; Separate evaluation; Judgment and decision making; Context effects

Overall, they stay quite close to the text (with the simple prompt "Give me 3-5 keywords that describe this research. They should be useful for searches/filters in a database" and the abstract only) - so it somewhat depends what we want the keywords to be useful for. If we want them for specific clustering/searches (e.g. lighthouse-style) this seems good. If we want filters, maybe we also need to ask for broader categories