Quantifying and qualifying the human-in-the-loop task in the Digital Product Passport production process

ACL ARR 2025 February Submission1297 Authors

Abstract: The Digital Product Passport 4.0 (DPP 4.0) initiative drives the transition from linear to circular economies. From 2027, it will enable seamless data exchange be-tween manufacturers, suppliers and end-of-life actors using Asset Administration Shell-compliant digital twins (DTs). However, creating and validating these DTs is tedious, error-prone, and potential-ly overwhelming even for experts due to the complexity of formal descriptions and data. We turn this disadvantage into an advantage by leveraging the hierar-chical DT structures as an interlingua to communicate with a Large Language Model (LLM). Through automatic tree navigation, prompts are generated for substructures, which an LLM is asked to correct/revise. Clear cases run automati-cally, while a human-in-the-loop (HiTL) is involved for necessary decisions through specific dialogues. We quantify and qualify the HiTL-tasks and, most im-portantly, show how to automatically ob-tain the appropriate context for the user-centered decision-making dialogues. A prototype implementing key aspects is currently under evaluation.

Paper Type: Short

Research Area: NLP Applications

Research Area Keywords: user-centered ai-based dialog, reviewing and correction of Digital Twins, Industry 4.0, Asset Administration Shell

Contribution Types: NLP engineering experiment, Position papers, Theory

Languages Studied: English, German

Submission Number: 1297