Harmonizing Research Artifacts from Open Repositories to Enhance Interdisciplinary Research and RDM Infrastructure

The Nationale Forschungsdateninfrastruktur (NFDI)\footnote{https://www.nfdi.de/} initiative supports scientific disciplines by promoting sustainable and FAIR research data management. NFDI4DataScience (NFDI4DS)\footnote{https://www.nfdi4datascience.de/} plays a central role within the consortium by enabling the integration of AI and data science driven applications into research workflows. It emphasizes the handling of digital objects such as machine learning models, datasets, publications, and source code, aiming to democratize access to these artifacts by open science and FAIR principles. This integration fosters interdisciplinary collaborations and reshapes the research landscape across the domains\cite{schimmler2023nfdi4ds}. To support these, open repositories like Hugging Face\footnote{https://huggingface.co/}, GitHub\footnote{https://github.com/}, arXiv\footnote{https://arxiv.org/}, and Zenodo\footnote{https://zenodo.org/} make digital research artifacts widely accessible. These platforms host large volumes of content, often linked to scientific publications. This underscores the increasing importance of research artifacts within the scholarly ecosystem and highlights their contribution to broadening the scope of the research data management life cycle, as emphasized by NFDI4DS. However, these repositories also exhibit rich, heterogeneous metadata, essential for the researchers across disciplines to understand, evaluate, and reuse models and source code. Among these, Hugging Face stands out as a leading platform for hosting machine learning models. As of March 2025, it hosts over 1.5 million models up from approximately 500,000 in February 2024 illustrating the exponential growth and adoption of across domains \cite{suryani2025hugging,suryani2024exploration}. Hugging Face models typically include structured metadata like task, pipelines, licensing, multilingualism, datasets, and publication links. When combined with traditional scholarly metadata, these heterogeneous features offer a unique opportunity to foster interdisciplinary research and promote the reuse of research artifacts \cite{suryani2024exploration}. This highlights the potential for research repositories liking and harmonization, an approach that can contextually exploit metadata across repositories, considering Hugging Face, it could facilitate research communities covering: \begin{itemize} \item Models with publications. \item Models linked to datasets and source code. \item Linking research artifacts across repositories i.e. Author with Models. \item Research Portfolios and Model Provenance. \item AI-centered Research Artifact Community Detection \end{itemize} To address these exciting aspects, the underlying metadata must be represented in a diversified structure that captures inherent heterogeneity. Knowledge Graphs (KG) to provide an exciting semantic modeling mechanism covering diverse entities and their relationships. The ever-growing landscape of digital objects within the AI and DS domains imposes challenges concerning their usability and interoperability within open source repositories. We propose the Hugging Face Knowledge Graph (HFKG), a semantic infrastructure specifically designed to integrate structured metadata from Hugging Face and respective outgoing links to external scholarly resources. By linking models with associated publications, HFKG enables richer contextualization and enhanced interdisciplinary discovery of AI and data science artifacts. In addition, HFKG aims to cover advanced querying utilizing semantic relationships, entity disambiguation to resolve ambiguities across heterogeneous metadata, provenance tracking, and relationship extraction, significantly enhancing interdisciplinary information retrieval and knowledge discovery. Moreover, aligning strategically with the NFDI4DS consortium, HFKG could serve as an extensible framework, interoperating with established research data infrastructures such as the GESIS Knowledge Graph by schema alignments, thus aiming to bring AI and data science artifacts under a harmonized and FAIR research data infrastructure.