Cultural Heritage (CH) data have become increasingly prominent within the Semantic Web, yet dataset discoverability remains limited due to fragmentation across platforms and a lack of a standard platform to give visibility to published data. This demo paper presents the main features of CHeCLOUD (Cultural Heritage Linked Open Data Cloud): the first domain-specific subcloud of the Linked Open Data Cloud specifically devoted to aggregate and enhance access to Knowledge Graphs (KGs) and ontologies related to CH. CHeCLOUD provides a centralized catalog of 192 curated CH KGs and ontologies and computes a FAIR score for each of them, relying on an automatic and periodic assessment of KGHeartBeat. CHeCLOUD currently ofers RESTful APIs, metadata browsing, and interactive graph visualizations to support FAIR evaluation. Additionally, CHeCLOUD features a semi-automated submission pipeline that engages users and maintainers through GitHub-based workflows. CHeCLOUD aims to foster reuse, interoperability, and findability within the CH community, while ofering a reusable infrastructure to support similar thematic hubs across other domains.
tion of che), it builds on the successful experience of other domain-specific sub-clouds, such as the
Linguistic LOD Cloud [
CHeCLOUD is a domain-specific subcloud within the LOD ecosystem that indexes linked datasets,
KGs, and ontologies relevant to the CH domain, covering tangible, intangible, and natural heritage in
accordance with UNESCO’s definition [
Out of the 1, 658 datasets originally listed in the LOD Cloud snapshot, 147 were initially identified as potentially relevant to CH. This collection was subsequently refined and expanded to 192 datasets through expert validation, manual resolution of broken links, and targeted searches in additional repositories (e.g., DataHub, Zenodo, and institutional portals). With this scale, CHeCLOUD currently ranks as the third-largest subcloud in the LOD Cloud ecosystem, following the Life Sciences and Linguistic subclouds.
While Lieto et al. [
CHeCLOUD ofers a coarse-grained exploration interface that provides a high-level overview of the
KGs and ontologies included in the catalog. Implemented as a visual dashboard, it aggregates metadata
and FAIRness assessment results, supporting maintainers in monitoring the overall health and structure
of the cloud. The dashboard displays a variety of statistical insights, such as the number of datasets per
CH subcategory, the availability of SPARQL endpoints and RDF dumps, the most commonly adopted
ontologies, license types, media formats for data dumps, and the availability of access mechanisms (e.g.,
SPARQL endpoints, RDF dumps), enabling analyses similar to those performed on the broader LOD
Cloud [
Figure 2 shows a snapshot of the dashboard, illustrating that CHeCLOUD is composed predominantly
of datasets, with only 10% being ontologies. Approximately 70% of the datasets relate to tangible
heritage, while natural heritage is the least represented category—an imbalance aligned with trends
observed in related studies [
Resembling the well-known visualization provided by the LOD Cloud, the CHeCLOUD landing page displays catalogued datasets as an interactive graph. Figure 1 illustrates only the connected component, while isolated datasets, i.e., those not linked to others, are accessible in the online version. Nodes are color-coded according to CH subcategories (generic, tangible, intangible, natural), following the UNESCO classification, and their size reflects the number of incoming links (in-degree). In the online interface, hovering over a node highlights its connections and displays a tooltip with dataset details, e.g., its full name and in-degree. Moreover, users can locate specific datasets through a keyword-based search, as in Figure 4 (1), which matches terms found in the dataset’s name, ID, description, or keywords.
Once a dataset is selected, users are directed to a metadata visualization page that includes both
descriptive metadata and FAIRness metrics, automatically updated on a weekly basis by
KGHeartBeat [
Contributors can support CHeCLOUD (i) by proposing the inclusion of new datasets, or (ii) by suggesting updates to the metadata of datasets already indexed in the catalog.
To streamline the integration process, CHeCLOUD employs a semi-automated submission and validation workflow. New datasets can be submitted through a dedicated metadata form. Upon submission, a LLM (currently Gemini 2.5 Pro, though interchangeable) automatically evaluates the dataset’s relevance to the CH domain and assigns it to the appropriate subcategory. This classification is based on an analysis of the dataset’s title, description, and keywords. A human reviewer subsequently validates both the metadata and the assigned classification to ensure accuracy. Once approved, the system creates a new branch and pull request (PR) in the CHeCLOUD GitHub repository, ensuring transparency and traceability. Contributors can track progress and engage with maintainers via the PR interface. The only manual step required of maintainers is the approval and merging of the request, after which the dataset is automatically synchronized and integrated into the catalog. Newly accepted datasets are also connected via REST APIs to KGHeartBeat, which calculates weekly quality metrics. FAIRness scores for all datasets, including new ones, are updated accordingly. Note that due to the weekly evaluation cycle, newly added datasets may appear in CHeCLOUD with their FAIRness scores after up to one week.
The same workflow applies to metadata updates for indexed datasets where the form is pre-filled with current metadata, allowing contributors to selectively edit the information, as in Figure 4 (3).
This section illustrates a sample workflow that will be demonstrated during the live session, showcasing how a user interacts with the CHeCLOUD platform. A recording of the demonstration is also available at: https://shorturl.at/6o0EZ.
Imagine the data steward of the Amsterdam Museum preparing a grant proposal to publish or reuse Linked Data describing their institution and its collection. To strengthen the proposal, the institution seeks to align with established best practices and reference successful examples in the domain. CHeCLOUD provides a valuable resource throughout this process: • Identifying relevant datasets. The steward begins by accessing the CHeCLOUD dashboard (Figure 2) to explore datasets related to cultural heritage. By consulting the FAIR score table, they can perform a coarse-grained comparative analysis of datasets based on their overall FAIR score and individual principle scores. This allows them to discover high-quality datasets in similar domains, providing concrete references for reuse, alignment, or inspiration. • Evaluating their own dataset. Returning to the graph-based visualization (Figure 1), the user searches for Rijksmuseum or a broader keyword such as Netherlands via the search bar (Figure 4 (1)). Upon selecting the dataset, the detail page (Figure 4 (2)) reveals that it currently has a low FAIR score, particularly due to a missing score in the Accessibility dimension. • Identifying gaps and justifying improvements. By using the on-demand textual explanation to understand the root causes of the dataset’s FAIRness (Figure 4 (3)), and by inspecting the metadata directly within CHeCLOUD, the steward identifies a key issue: a missing or incorrect SPARQL endpoint. They propose a correction via the metadata update workflow (Figure 4 (4)). After approval by a maintainer and subsequent re-evaluation by KGHeartBeat, the updated FAIR score becomes visible (Figure 4 (5)). This not only improves the dataset’s quality but also provides evidence of active curation, supported by collaborative contributions. Such transparency can strengthen the proposal’s impact and help users make informed decisions.
This scenario demonstrates how CHeCLOUD supports stakeholders in benchmarking, improving, and promoting Linked Data datasets, ultimately contributing to better data quality and more informed decision-making within the CH sector.
This demo overviews CHeCLOUD, a catalog of CH KGs and ontologies, designed as a thematic subcloud within the LOD Cloud. It ofers tailored interfaces for producers, consumers, and contributors, enabling FAIRness monitoring, metadata updates, and cloud exploration via dashboards and graph views. The infrastructure, available on GitHub, is reusable for creating subclouds in other domains. FAIRness assessments are automated through integration with KGHeartBeat.
Limitations and Future Work. At present, no comparative studies have been conducted to demonstrate superior performance and enhanced usability of this cloud in relation to other similar resource aggregators, such as the LOD Cloud. Furthermore, the entire process of FAIRness assessment currently relies exclusively on KGHeartBeat, and no comparison has been carried out with alternative tools that enable the evaluation of resource FAIRness.
Dataset discovery is currently semi-automated, relying on LLMs and metadata, currently limiting the platform’s scalability and its long-term sustainability. Future plans include content-based classification, advanced search with adjustable FAIR weights, and improved explainability with corrective suggestions.
The newly indexed datasets and their updated metadata are not yet reflected in the LOD Cloud. However, this integration can be readily achieved, as CHeCLOUD supports the export of the entire catalog as a JSON file that adheres to the same structure as the original LOD Cloud, thereby facilitating seamless future incorporation.
This work was partially supported by the European Alliance NEOLAiA (Project 101124794: “NEOLAiA – Transforming Regions for an Inclusive Europe").
During the preparation of this work, the author used ChatGPT in order to: Grammar and spelling check.