A Graph-RAG inspired approach to extract Taxonomies applicable to Digital Product Passports

Abstract

A global, cross-continental extension of the European DPP system, suitable for global value chains, is currently under way as organized by the International Organization for Standardization. To operationalize these federated structures in a scalable manner, a pragmatic solution lies in the formal definition of domainspecific taxonomies and ontologies rooted directly in the underlying legal texts. To address this semantic gap, Natural Language Processing (NLP) frameworks based on Large Language Models (LLMs) offer powerful automated synthesis capabilities. Using Grahp-RAG inspired pipelines an ontology and a taxonomy pertinent to various European regulations is constructed and published as turtle files allowing for feeding interoperability pipelines of federated DPP systems.