Knowledge graphs (KGs) are a powerful way of representing information for digital humanities. However, non-technical users often struggle at the outset of exploration, a challenge defined as the Initial Exploration Problem. The Tús Maith framework addresses this issue through curated natural language questions and answers (CuQAs) created from Competency Questions (CQs) that aim to convey the scope of a KG and provide meaningful entry points into it. While prior work has explored using large language models (LLMs) for CQ template generation, the template-filling step, where questions and answers are instantiated with entity information, remains a key challenge. In this paper, we evaluate whether LLMs have the capacity to support domain experts in this stage, focusing on the Virtual Record Treasury of Ireland (VRTI) KG, where accuracy, provenance, and robustness are crucial for practical use. Using structured JSON inputs derived from popular search terms and expert-authored templates, we generated and assessed 24,900 question-answer pairs across four LLMs (GPT-5, DeepSeek-V3.1, Gemini 2.0 Flash, Qwen-2.5-72B) under two provenance conditions (basic vs. full). Our evaluation considers slot fidelity, semantic similarity, completeness, hallucination rates, and runtime efficiency, with statistical tests conducted per run per LLM, and additional batch-level analysis (n = 68) to isolate provenance requirement effects. We further show that a lightweight JSON validation check is an effective proxy for ground truth semantic evaluation of factual question-answer pairs. These LLM-generated, validated questions form an intermediate step in the lifecycle from abstract CQ templates to filled-in questions and answers intended to be reviewed and refined by the VRTI KG’s domain experts (historians) to produce the final user-facing questions (CuQAs). To demonstrate the practical impact, we present a prototype (TMv1) of the Tús Maith framework and highlight the design implications for curator-facing interfaces: provenance-transparent interaction, validation-integrated workflows, and performance-transparent model selection.

Declarative mappings support semantic interoperability across heterogeneous data sources, enabling tasks such as ontology alignment, data uplift, and interlinking. However, mapping projects are often documented only through their final artefacts, with limited structured metadata describing their context, development process, and evolution. This paper presents MMV (Mapping Metadata Vocabulary), a lifecycle-based ontology designed to represent metadata for declarative mapping projects across different mapping types. MMV captures contextual and procedural metadata spanning analysis, design, development, testing, and maintenance, while reusing established Semantic Web vocabularies. The paper describes the design of MMV, illustrates its instantiation, and outlines a planned expert evaluation.

Summary This article provides a brief introduction to the “digital windfall” of Irish historical data and discusses the innovative technologies and methods that the VOICES project will harness, including a knowledge graph and AI-informed approaches. Next, two detailed case studies, which reflect “work in progress” on records from the Irish Court of Chancery and testamentary records, are explored. The article reflects on the possibilities that machine learning affords historians and the processes that are being developed to avail of these opportunities. It is argued that the transparent and ethical use of innovative technologies, grounded in the Semantic Web, Linked Open Data, and artificial intelligence, can enable a new paradigm of historical research, which allows for the recovery in a holistic and integrated way of the marginalized voices and lived experiences of women in early modern Ireland.

The Virtual Record Treasury of Ireland (VRTI) was established in 2016 as an initiative to reconstruct the Public Record Office of Ireland that was lost in the civil war fire in 1922. The initiative has become a nationwide effort to re-discover, digitize, reconstruct and catalogue the lost archives of the subsequent years. Leveraging information technologies, historians have constructed the one of the largest historical archives in Ireland, contributing to preserving the heritage of a nation. The VRTI has implemented a data-centric approach to exploiting the Knowledge Graphs (KG) capabilities for organizing, cataloguing and sharing knowledge across different resources. However, interacting with a KG requires the creation of complex and time-consuming queries which limits the uptake of the information. Therefore, it was decided to design an intuitive user interface to facilitate exploration of the Knowledge Graph for Irish History (KGIH). The KGIH is a five star linked open data resource, the first of this kind to exist for Irish historical research. The VRTI-KG explorer provides a straightforward method for casual users and researchers to interact with the KG and apply it in real-world use cases. This paper presents the uptake and impact of the explorer and outlines recent enhancements designed to provide historians with more control over the information presented on the interface.

This article examines how benchmarking digital methods can advance historical research by recovering women’s lives from fragmented and underused archives. It focuses on the Funeral Entries held in the Genealogical Office of the National Library of Ireland, a rich but understudied manuscript collection compiled by the Ulster King of Arms between the late sixteenth and early eighteenth centuries. These records, which document death dates, kinship networks, and social affiliations, contain an unusually high proportion of women for the early modern period (about 38 percent), offering rare insight into gendered experiences of death, family, and memory. The article emerges from the ERC-funded VOICES project, which develops AI-powered approaches to recover women’s voices and experiences from early modern sources (VOICES Project, 2023; https://voicesproject.ie). As part of this work, we present a benchmarking experiment using the Funeral Entries to assess the ability of Handwritten Text Recognition (HTR) and Named Entity Recognition (NER) models to cope with early modern orthography, multilingual naming practices, and manuscript variability. These experiments not only illuminate the constraints of existing tools but also demonstrate how benchmarking can generate reusable workflows, inform the creation of annotated gold standards, and support the production of FAIR-aligned humanities data. Rather than offering a completed dataset, the article argues for the historiographical value of iterative benchmarking. We show how evaluating and refining computational methods reframes questions of archival visibility, evidentiary status, and the interpretive potential of genealogical records, positioning the Funeral Entries not merely as instruments of male lineage but as essential sources for recovering early modern women’s social worlds.

Knowledge Graphs (KGs) and their machine learning counterpart, Knowledge Graph Embedding Models (KGEMs), have seen ever-increasing use in a wide variety of academic and applied settings. In particular, KGEMs are typically applied to KGs to solve the link prediction task; i.e. to predict new facts in the domain of a KG based on existing, observed facts. While this approach has been shown substantial power in many end-use cases, it remains incompletely characterised in terms of how KGEMs react differently to KG structure. This is of particular concern in light of recent studies showing that KG structure can be a significant source of bias as well as partially determinant of overall KGEM performance. This paper seeks to address this gap in the state-of-the-art. This paper provides, to the authors' knowledge, the first comprehensive survey exploring established relationships of Knowledge Graph Embedding Models and Graph structure in the literature. It is the hope of the authors that this work will inspire further studies in this area, and contribute to a more holistic understanding of KGs, KGEMs, and the link prediction task.

Telecommunication networks pose challenges for root cause analysis due to complexity, heterogeneous data, and high event volumes. Existing models like RDF and LPG provide only partial solutions when used alone. This paper explores the Semantic-enhanced Programmable Graph (SPG) approach via the OpenSPG engine, which combines property graph flexibility with RDF-inspired semantics. Applied to a telecom use case, OpenSPG shows advantages for event-centric modeling but remains limited in maturity, tooling, and semantic interoperability, highlighting directions for future semantic-enabled network management.

Large Language Models (LLMs) are increasingly used to automate knowledge engineering tasks such as generating RDF mappings. While promising, their outputs often lack semantic precision, syntactic correctness, and contextual metadata. This paper investigates whether structured metadata aligned with the mapping lifecycle can improve the quality and reusability of LLM-generated mappings. We present a metadata model that covers key phases of the mapping process and integrate it into the MetaSEMAP tool to support context-aware prompting. Using real-world uplift scenarios, we compare RML outputs generated from unguided prompts with those informed by lifecycle metadata. Our initial findings show that guided prompts consistently produce syntactically valid, semantically rich, and FAIR-aligned mappings. These results highlight the potential of structured metadata to guide LLMs toward generating higher-quality and reusable semantic artifacts in knowledge graph construction.

The growing threat of disinformation and misinformation across digital platforms has intensified the demand for systems capable of producing verifiable and trustworthy outputs. With the widespread adoption of Large Language Models (LLMs) for a variety of tasks, the requirement to provide accurate and fact-verifiable answers is increasing daily. GraphRAGs have become a powerful approach for solving complex tasks that require factual context to deliver accurate and explainable answers. However, Knowledge Bases (KB) used to provide factual and contextual knowledge are composed of thousands or millions of statements, which limits the size of inputs that an LLM can handle, typically by the number of input tokens supported by the model. This work addresses the problem of Fact-Checking by injecting Knowledge Graph Embedding (KGE) vector representations into LLMs using a Retrieval Augmented Generation (RAG) approach to obtain more accurate results. The results show a notable difference in the quality of the results with two different vector representations and two KB construction methods.

Non-technical users often face a significant barrier when first attempting to explore complex knowledge graphs (KGs). We define this challenge as the Initial Exploration Problem, characterised by three interrelated barriers: ontology opacity, query incapacity, and scope uncertainty. This paper investigates how large language models (LLMs) can support domain experts in addressing this problem by automatically generating template-style competency questions (CQs) for the Virtual Record Treasury of Ireland (VRTI) KG. These templates are not user-facing themselves, but serve as scaffolding for creating curated questions (CuQs), expert-validated, natural language questions that help new users begin meaningfully exploring the graph. We evaluate two LLMs (GPT-4o and Gemini 2.0 Flash) across twelve prompt configurations varying in scope and framing, and assess question quality using both semantic similarity to expert-authored CQs and detailed expert review. Our findings highlight how prompt design influences LLM output, and underscore the value of combining automated generation with expert curation. Ultimately, we propose a practical pipeline to support the creation of exploratory entry points tailored to user needs, helping domain experts craft better questions, and helping users take their first steps into meaningful KG exploration.

This version explores the intersection of Knowledge Graphs (KGs) and Large Language Models (LLMs) with a focus on enabling scalable, efficient, and trustworthy AI applications in industrial contexts. As generative AI rapidly evolves, integrating symbolic and neural methods becomes essential to address challenges such as explainability, data alignment, and system robustness by gathering academic researchers and industry practitioners to discuss practical solutions and future of Semantic Web technologies in the era of foundation models.

Comprehensive and well-structured metadata annotation and documentation are essential for supporting the reuse and maintenance of declarative mappings throughout their lifecycle. This paper presents a usability evaluation of MetaSEMAP, a tool designed to facilitate the annotation of declarative mappings, including uplift mappings, ontology alignment, and interlinking. While MetaSEMAP and its underlying metadata model are still under active development, the focus of this work is on evaluating how users interact with the tool and interpret the metadata concepts it introduces. The evaluation investigates MetaSEMAP's ability to support metadata annotation using real-world scenarios such as the Virtual Record Treasury of Ireland. Participants provided feedback on the tool's usability and their preferences for metadata representation, including RDF-star and Named Graph. The results reveal both strengths and areas for improvement, offering valuable insights for the development of more effective mapping annotation tools. This work supports efforts to improve interoperability and sustainability in mapping practices, with alignment to FAIR principles as a longer-term goal.

SPARQL is a powerful but complex language for querying knowledge graphs, motivating research into natural language-to-SPARQL generation using large language models (LLMs). While large, proprietary LLMs excel at this task, their resource requirements can limit practical deployment. This paper evaluates smaller, open-source LLMs (0.5B–9B parameters) with quantization methods (8-bit and 4-bit compression) to balance computational efficiency and query generation performance. Our findings demonstrate that 8-bit quantization can maintain or enhance performance in smaller models, whereas 4-bit quantization leads to notable degradation, especially for larger models. This highlights the potential of quantized, smaller LLMs for SPARQL generation in resource-constrained scenarios and provides insights for optimizing specialized NLP tasks.

Objectives This study aims to describe the data structure and harmonisation process, explore data quality and define characteristics, treatment, and outcomes of patients across six federated antineutrophil cytoplasmic antibody-associated vasculitis (AAV) registries. Methods Through creation of the vasculitis-specific Findable, Accessible, Interoperable, Reusable, VASCulitis ontology, we harmonised the registries and enabled semantic interoperability. We assessed data quality across the domains of uniqueness, consistency, completeness and correctness. Aggregated data were retrieved using the semantic query language SPARQL Protocol and Resource Description Framework Query Language (SPARQL) and outcome rates were assessed through random effects meta-analysis. Results A total of 5282 cases of AAV were identified. Uniqueness and data-type consistency were 100% across all assessed variables. Completeness and correctness varied from 49%–100% to 60%–100%, respectively. There were 2754 (52.1%) cases classified as granulomatosis with polyangiitis (GPA), 1580 (29.9%) as microscopic polyangiitis and 937 (17.7%) as eosinophilic GPA. The pattern of organ involvement included: lung in 3281 (65.1%), ear-nose-throat in 2860 (56.7%) and kidney in 2534 (50.2%). Intravenous cyclophosphamide was used as remission induction therapy in 982 (50.7%), rituximab in 505 (17.7%) and pulsed intravenous glucocorticoid use was highly variable (11%–91%). Overall mortality and incidence rates of end-stage kidney disease were 28.8 (95% CI 19.7 to 42.2) and 24.8 (95% CI 19.7 to 31.1) per 1000 patient-years, respectively. Conclusions In the largest reported AAV cohort-study, we federated patient registries using semantic web technologies and highlighted concerns about data quality. The comparison of patient characteristics, treatment and outcomes was hampered by heterogeneous recruitment settings.

With the upcoming enforcement of the EU AI Act, documentation of high-risk AI systems and their risk management information will become a legal requirement playing a pivotal role in demonstration of compliance. Despite its importance, there is a lack of standards and guidelines to assist with drawing up AI and risk documentation aligned with the AI Act. This paper aims to address this gap by providing an in-depth analysis of the AI Act's provisions regarding technical documentation, wherein we particularly focus on AI risk management. On the basis of this analysis, we propose AI Cards as a novel holistic framework for representing a given intended use of an AI system by encompassing information regarding technical specifications, context of use, and risk management, both in human- and machine-readable formats. While the human-readable representation of AI Cards provides AI stakeholders with a transparent and comprehensible overview of the AI use case, its machine-readable specification leverages on state of the art Semantic Web technologies to embody the interoperability needed for exchanging documentation within the AI value chain. This brings the flexibility required for reflecting changes applied to the AI system and its context, provides the scalability needed to accommodate potential amendments to legal requirements, and enables development of automated tools to assist with legal compliance and conformity assessment tasks. To solidify the benefits, we provide an exemplar AI Card for an AI-based student proctoring system and further discuss its potential applications within and beyond the context of the AI Act.

Knowledge Graphs (KGs) have become ever-more important for modelling biomedical information, as their intrinsic graph structure matches the structure of many biological interaction networks. Together with KGs, Knowledge Graph Embeddings (KGEs) have shown immense potential to learn biological data and predict new, in-band facts about the data the KG describes. However, recent literature has suggested several major deficits to KGEs: that they have an extremely short 'receptive field' of data they use to make predictions and that their learning is guided by memorising graph structure, not learning latent semantics. Moreover, while several studies have suggested that graph structure and KGE model choice affect optimal hyperparameters, the exact relationship of hyperparameters to learning remains unknown and is instead solved using a computationally intensive hyperparameter search. In this paper we introduce TWIG (Topologically-Weighted Intelligence Generation), a novel, embedding-free paradigm for simulating the output of KGEs that uses a tiny fraction of the parameters. TWIG learns weights from inputs that consist of topological features of the graph data, with no coding for latent representations of entities or edges. Our experiments on the UMLS dataset show that a single TWIG neural network can predict the results of state-of-the-art ComplEx-N3 KGE model nearly exactly on across all hyperparameter configurations. To do this it uses a total of 2590 learnable parameters, but accurately predicts the results of 1215 different hyperparameter combinations with a combined cost of 29,322,000 parameters. Based on these results, we make two claims: 1) that KGEs do not learn latent semantics, but only latent representations of structural patterns; 2) that hyperparameter choice in KGEs is a deterministic function of the KGE model and graph structure. We further hypothesise that, as TWIG can simulate KGEs without embeddings, that node and edge embeddings are not needed to learn to accurately predict new facts in KGs. Finally, we formulate all of our findings under the umbrella of the "Structural Generalisation Hypothesis", which suggests that "twiggy" embedding-free / data-structure-based learning methods can allow a single neural network to simulate KGE performance, and perhaps solve the Link Prediction task, across many KGs from diverse domains and with different semantics.

BACKGROUND:Antineutrophil cytoplasmic antibody (ANCA)-associated vasculitis is a heterogenous autoimmune disease. While traditionally stratified into two conditions, granulomatosis with polyangiitis (GPA) and microscopic polyangiitis (MPA), the subclassification of ANCA-associated vasculitis is subject to continued debate. Here we aim to identify phenotypically distinct subgroups and develop a data-driven subclassification of ANCA-associated vasculitis, using a large real-world dataset. METHODS:In the collaborative data reuse project FAIRVASC (Findable, Accessible, Interoperable, Reusable, Vasculitis), registry records of patients with ANCA-associated vasculitis were retrieved from six European vasculitis registries: the Czech Registry of ANCA-associated vasculitis (Czech Republic), the French Vasculitis Study Group Registry (FVSG; France), the Joint Vasculitis Registry in German-speaking Countries (GeVas; Germany), the Polish Vasculitis Registry (POLVAS; Poland), the Irish Rare Kidney Disease Registry (RKD; Ireland), and the Skåne Vasculitis Cohort (Sweden). We performed model-based clustering of 17 mixed-type clinical variables using a parsimonious mixture of two latent Gaussian variable models. Clinical validation of the optimal cluster solution was made through summary statistics of the clusters' demography, phenotypic and serological characteristics, and outcome. The predictive value of models featuring the cluster affiliations were compared with classifications based on clinical diagnosis and ANCA specificity. People with lived experience were involved throughout the FAIRVASVC project. FINDINGS:A total of 3868 patients diagnosed with ANCA-associated vasculitis between Nov 1, 1966, and March 1, 2023, were included in the study across the six registries (Czech Registry n=371, FVSG n=1780, GeVas n=135, POLVAS n=792, RKD n=439, and Skåne Vasculitis Cohort n=351). There were 2434 (62·9%) patients with GPA and 1434 (37·1%) with MPA. Mean age at diagnosis was 57·2 years (SD 16·4); 2006 (51·9%) of 3867 patients were men and 1861 (48·1%) were women. We identified five clusters, with distinct phenotype, biochemical presentation, and disease outcome. Three clusters were characterised by kidney involvement: one severe kidney cluster (555 [14·3%] of 3868 patients) with high C-reactive protein (CRP) and serum creatinine concentrations, and variable ANCA specificity (SK cluster); one myeloperoxidase (MPO)-ANCA-positive kidney involvement cluster (782 [20·2%]) with limited extrarenal disease (MPO-K cluster); and one proteinase 3 (PR3)-ANCA-positive kidney involvement cluster (683 [17·7%]) with widespread extrarenal disease (PR3-K cluster). Two clusters were characterised by relative absence of kidney involvement: one was a predominantly PR3-ANCA-positive cluster (1202 [31·1%]) with inflammatory multisystem disease (IMS cluster), and one was a cluster (646 [16·7%]) with predominantly ear-nose-throat involvement and low CRP, with mainly younger patients (YR cluster). Compared with models fitted with clinical diagnosis or ANCA status, cluster-assigned models demonstrated improved predictive power with respect to both patient and kidney survival. INTERPRETATION:Our study reinforces the view that ANCA-associated vasculitis is not merely a binary construct. Data-driven subclassification of ANCA-associated vasculitis exhibits higher predictive value than current approaches for key outcomes. FUNDING:European Union's Horizon 2020 research and innovation programme under the European Joint Programme on Rare Diseases.