[Tooling] New work item for SWG4.4: "AI-assisted Metadata Generation Tool"

[peterlubrich]

Idea:
We propose a tool for generating metadata formatted in mobilityDCAT-AP in a (semi-) automatic way using AI and semantic technologies.
This developement is embedded in the NAPCORE IT Tools Activity.
At this activity, the proposed tool was listed under the category NAPCORE Tools for Metadata ("Automatic metadata extraction").

We have done a brainstorming on such tool with Cefriel, see PPT attached:

• Background & motivation
• Expected functionalities
• Next steps

260303_NAPCORE AI Metadata Tool_Brainstorming.pptx

Proposed procedure:

• Identification of interested NAPCORE partners
• Discussion & conceptualization
• Development and piloting

[EdNDW]

Thanks for opening this and sharing the Cefriel brainstorming, good to see this moving forward.

One thought on scope: the current proposal focuses on extraction from content data (e.g. DATEX II payloads). That works well for
structured domain formats, but many providers publish CSV, GeoJSON, GTFS, or SIRI feeds where the payload carries limited metadata signals.
A documentation or landing page often contains exactly what the content file doesn't: licence, publisher, geographic scope, thematic classification.

I've been exploring a complementary direction in my own time: combining the dataset endpoint with a documentation URL as dual input, using AI to pre-populate a mobilityDCAT-AP form that the provider then reviews and completes. I built a proof-of-concept using AI-assisted development. I'm not a developer myself, so the code quality is not production-ready, but the workflow and concept are functional enough to demonstrate the idea, including RDF output and a SHACL validation step.

I'd like to contribute this concept to the conceptualization phase and explore whether it could be developed further, potentially in co-creation with Cefriel and other interested partners. From my position at the Dutch NAP (NTM/NDW) I'd also be interested in piloting this in a real NAP context.

Is there a next step or working session where we could align these ideas?