Mastodawn

Can we use qualitative data in machine learning pipelines?
For #EHBEA2026, I prepared an abstract summarizing a few ideas about technical and epistemic prerequists. We claim that a) yes, it can be useful to integrate expert knowledge (example: reducing search space, and yes, this is no new sauce)
b) automatic extraction of said knowledge necessitates a closer view on what qualitative data is and how it is usually created --->

Show thread

Daniela Gawehns 1d ago

--> Qual data is created by #human curation, interpretation and repeated investigation. Any (semi) automated extraction mechanism will have to at least mimick the guardrails that qual researchers have to ensure rigor of their work. On top of it, some of those tools (e.g. #positionality statements) will also be useful for other (quant) parts of the pipeline.
!! This would assume a context where we allow researchers to bring their biases and beliefs into the research process.!!!

Show thread

Daniela Gawehns

Sidenote: if the #automated knowledge (e.g., a #parameter to reduce the search space) is very similar to the human input, and if the researcher holds a #positivist view on things -- they are epistemically welcome to use this additional parameter for their research/ in their pipeline.
I would argue that they shouldnt boast with having used/integrated #qualitativeData though, it is a parameter learned from (in most cases) textual data.

Show thread

Daniela Gawehns 1d ago

You can read the full text of the short and long abstract on zenodo:

https://zenodo.org/records/19237858

I ll be presenting these ideas in a 10 min short talk. Feedback is welcome, especially from #qualitativeresearch : did we portray the process of data creation more or less acurately? Did we miss an important tool to ensure rigor?

Be kind, I am dipping my feet for the first time into these waters 😅
I am also hoping to develop this into a full paper!

Abstracts of the Short Talk "What would qualitative data do to machine learning pipelines?"

Submission to EHBEA 2026, Leiden. The abstract was accepted for a short talk in the AI track of the conference. Abstracts as submitted to the conference (short form) and as extended for the book of abstracts (long form).

Zenodo