PERKS S2 – Supporting Users to Elicit Tacit Procedures and Build an Actionable Knowledge Graph

In many industrial scenarios, crucial operational knowledge is still undocumented: it lives in the experience of workers rather than in formal procedures. When documentation does exist, it is often unstructured, heterogeneous, and difficult for machines to interpret. The PERKS Solution S2 addresses both situations by enabling users to manually capture tacit procedures from scratch and, when documents are available, to refine and complete the procedures automatically extracted by Solution S1 (read more about S1 at https://perks-project.eu/2026/02/02/perks-s1-turning-procedural-documents-into-actionable-knowledge-graphs/).

Developed by Cefriel, S2 builds upon the opensource rapidtriples component to support humanintheloop knowledge acquisition, ensuring that procedural knowledge becomes accurate, structured, and semantically aligned with the Procedural Knowledge Ontology (PKO, available at https://w3id.org/pko# and see also the ontology GitHub repo at https://github.com/perks-project/pk-ontology).

Capturing Tacit Knowledge When No Documentation Exists

A key feature of S2 is its ability to guide users in creating a complete procedure from zero. Through dynamic and adaptive forms, domain experts can articulate their tacit knowledge step by step, without having to deal with RDF, ontologies, or technical representations.
This ensures that even procedures that were never written down can be transformed into machinereadable, actionable knowledge graphs.

When documentation is available, S2 seamlessly connects to the outputs of S1: users can start from the automatically extracted draft and refine, correct, or complete it.
This manual refinement and completion anticipate the more advanced humanintheloop validation workflows that will be introduced in Solution S3.

How S2 Works

S2 implements a hybrid workflow that blends user expertise with automated support:

• Manual knowledge entry via adaptive JSONSchemabased forms.
• Refinement of S1 autoextraction when documents exist.
• Humanintheloop review and enrichment, paving the way for S3’s validation workflows.
• Automatic RDF generation using declarative mapping rules aligned with PKO.
• Creation of highquality training data to continuously improve AI extractors.

This approach makes S2 flexible across use cases while ensuring highquality, ontologycompliant procedural knowledge.

The BEKO Use Case: Faster and HigherQuality Procedures

S2 has been piloted by the PERKS industrial partners in their use cases. BEKO experimented it on LockOut TagOut (LOTO) procedures.
The experimentation covered both scenarios:

1. Building a procedure from scratch (no document available)

Operators used S2’s adaptive forms to manually construct a full LOTO procedure.
The results were significant:

• S2 enabled much faster knowledge capture:
  half a day vs. two full days with traditional methods.
• The resulting procedure was more complete and higher in quality, thanks to guided steps and semantic alignment.

2. Completing a procedure proposed by automatic extraction (when a PDF existed)

The initial draft produced by S1 was preloaded in S2, and users corrected and enriched it.
This hybrid workflow proved intuitive and efficient, helping operators surface tacit details not present in the document.

Feedback from safety managers and operators highlighted:

• improved user experience,
• more detailed and reliable procedures,
• smoother transition between automatic and manual knowledge capture.

Why S2 Is a Key Building Block

S2 strengthens the PERKS Actionable Knowledge Graph by making procedural knowledge:

• complete – even when no documentation exists,
• trustworthy – thanks to structured human input,
• actionable – aligned with PKO and ready for downstream AI modules,
• selfimproving – producing curated training data for extraction systems.

Learn More and Try Out S2 and the Technology Behind it

• Experiment and build a PKO-compliant procedure (also exporting it in RDF) with our S2 demo for procedure collection: https://perks-project.github.io/pko-rapid-triples.
• Learn more about the technology behind S2 by reading the following paper: Mario Scrocca, Alessio Carenini, Valentina Anita Carriero and Irene Celino: “rapid-triples: Customisable and Dynamic Forms for Semi-automatic Knowledge Collection“, in proceedings of the 1st Workshop on Bridging Hybrid Artificial Intelligence and the Semantic Web (HAIBRIDGE), co-located with ISWC 2025, CEUR-WS Vol. 4093, available at https://ceur-ws.org/Vol-4093/Paper4hai.pdf.
• Discover the core component supporting S2, rapidtriples, available as open source at: https://github.com/cefriel/rapid-triples.

Cefriel Research Team

The PERKS Solution S2 has been designed and developed by the Cefriel research team: Mario Scrocca, Valentina Carriero, Antonia Azzini, Ilaria Baroni, and Irene Celino.