TABular Semantic Enhancement Blueprint (TAB-SEB) v1

Project website link: https://ariannamorettj.github.io/tab_seb/ Overview Purpose. The workflow blueprint supports semantic enhancement of Cultural Heritage and GLAM metadata by converting tabular, semi-structured, or externally sourced data into RDF to increase interoperability. Secondary aim. It also targets usability: the benefits of the Semantic Webare exposed to stakeholders who do not use Semantic Web tooling directly through mediated access mechanisms (e.g., precompiled queries and web interfaces). Intended users (of the workflow blueprint). Multidisciplinary digitisation teams (domain experts + digital humanists/technical staff), with a separation of responsibilities that preserves feasibility in real projects. Intended users (of the workflow products). (1) Specialised users (capable of independently interacting and enhancing LOD), (2) domain experts (of the relevant domain for the data, without technical expertise), and (3) general audence (interacting with the data without domain expertise or technical competency) Research output framing. The contribution is formalised as a transferable workflow blueprint independent of any specific execution environment; executable implementations are treated as validation artefacts and instantiations. Detailed description: This section builds on the workflow structure introduced in the methodology chapter, in the subsection Definition of the Workflow’s Main Purpose and Steps Identification. That section established the final high-level architecture: a central core step that converts tabular data into RDF, complemented by an upstream phase for data retrieval and harmonisation via metadata crosswalks, and a downstream phase for publication and dissemination. Here, the complete resulting workflow is presented as the main research output. Each step is examined procedurally, specifying expected inputs and outputs, providing non-blocking tool suggestions, and defining testing and validation practices. The workflow’s primary purpose is to support the semantic enhancement of cultural heritage and GLAM metadata by enabling the conversion of tabular, semi-structured, and externally sourced data into RDF, thereby increasing interoperability. A secondary aim concerns usability: the workflow is designed so that the benefits of Semantic Web technologies can be leveraged by stakeholders who do not operate Semantic Web tooling directly, through mediated access mechanisms such as precompiled queries and web-based interfaces built on the resulting graph. The intended users are multidisciplinary digitisation teams in which domain experts from the humanities and GLAM institutions collaborate with digital humanists and technical staff. The workflow is structured to separate responsibilities in a way that preserves feasibility in real projects: domain experts contribute modelling-relevant knowledge through collaborative data collection and structured inputs, while technical contributors implement conversion and publication steps without requiring full domain mastery. The expected outputs, therefore, operate at two complementary levels: at the data level, workflow execution produces RDF datasets that are openly reusable by technical audiences and suitable for integration into broader Linked Open Data ecosystems. and at the access level, the workflow supports the generation of user-facing dissemination layers that expose the same data through interfaces intended for mixed audiences, enabling exploration without requiring SPARQL fluency. The research contribution is formalised as a workflow blueprint, defined as a transferable operating model independent of any specific execution environment. Executable implementations are treated as validation artefacts and context-specific instantiations. This distinction follows established workflow-publication practices in which the blueprint constitutes the stable methodological contribution, while implementations provide evidence of feasibility and boundary conditions. The blueprint is expressed as a chain of phases and checkpoints, with explicit decision points and traceable deviations, and is designed to support human-in-the-loop reproducibility and iterative refinement. Monitoring, evaluation, and testing are embedded as workflow-native quality-control activities, and intermediate outputs and diagnostics are treated as accountable artefacts rather than incidental by-products. Operationally, the blueprint is organised into three macro-phases followed by a supplementary evaluation step that enables re-iteration when input data are updated or when output quality requires improvement: preliminary data preparation; conversion to RDF as the core materialisation step; and data dissemination. Each macro-phase is defined by explicit input and output expectations, associated decision points, and integrated quality-control checkpoints to sustain traceability and methodological reliability across heterogeneous domains. The full workflow blueprint is released on Protocols under a CC0 license, where it is kept updated. The workflow’s general structure, which outlines the main blueprint steps, is illustrated below. A more in-depth analysis of the procedural tasks for each step is available in the workflow document on Protocols. This analysis includes links and examples of specific implementations, the expected inputs and outputs, optional suggestions for tools, and methods for testing and validation. Title: TABular Semantic Enhancement Blueprint (TAB-SEB). Authors/creators: Arianna Moretti. Resource (private link for reviewers): https://www.protocols.io/private/FB400EA3134211F1BB740A58A9FEAC02 (to be removed before publication). Resource (official release): https://doi.org/10.17504/protocols.io.eq2ly5qrrvx9/v1. Partially documented in: Moretti, Arianna. “Defining a Workflow for Semantic Enhancement of Cultural Heritage Metadata.” Proceedings of IRCDL 2026 (CEUR-WS Workshop Proceedings), 2026. The workflow blueprint is presented here as a versioned and citable, transferable, technology-agnostic but execution-oriented methodological artefact released on Protocols.io, with explicit phases, declared inputs and outputs, decision points, and integrated quality-control checkpoints. Its technical function is to provide a stable coordination layer across heterogeneous implementations by connecting tabular templates, extraction and harmonisation procedures, identifier-enrichment routines, a CSV-to-RDF materialisation core based on mappings, configurations, and functions, and downstream dissemination components such as RDF releases, SPARQL endpoints, semantic access layers, and provenance packages. Within this framework, the blueprint constitutes the persistent methodological research object, whereas the case-study implementations represent step-specific executable instantiations that document feasibility, reuse conditions, and extension points. Below, a summary of the general logic and internal organisation of the workflow is introduced. However, for detailed consultation of the resource in its full workflow dimension, including step-by-step articulation, examples, linked artefacts, supporting documentation, warnings, and implementation-oriented guidance, reference should be made to the Protocols.io release itself. The first section of the blueprint covers (1) Preliminary tabular data preparation. (1.1) Input data identification establishes the operational entry point of the workflow by determining whether the project begins from newly collected tabular data or from pre-existing external sources, and by fixing the semantic and procedural assumptions under which later RDF materialisation can remain deterministic. This step includes both the reuse of tabular templates and conventions when data are collected ex novo, and the assessment of alternative access channels and extraction strategies when the workflow starts from external sources. In both cases, and especially where third-party data are reused, it also requires identification of the licence constraints governing reuse, including institutional policies, source licensing terms, rights-holder requirements, and dissemination restrictions. This step also encompasses data extraction and the generation of an exploratory sample dataset, accompanied by monitoring and automated tests, as well as subsequent human inspection and feedback collection aimed at identifying harmonisation needs. (1.2) Harmonisation and cleaning, then curate the tabular layer through header translation, schema crosswalking, unification of heterogeneous exports, value normalisation, separator management for multivalued cells, deduplication, conflict detection, cardinality checks, and identifier validation, producing curated tables, transformation logs, and diagnostic reports. (1.3) Dataset enrichment with persistent identifiers and links to external sources increases linkability, reduces ambiguity, and prepares the dataset for semantic reconciliation. (1.4) Definition of research-oriented requirements and querying desiderata shifts the dataset from a descriptive inventory to a research-oriented data product by formalising the entities, relations, filters, aggregations, and dissemination requirements that the later semantic layer must support. (1.5) Production of purpose-oriented optimised subsets together with traceability artefacts applies where the full dataset is too broad, heterogeneous, or computationally inconvenient for a given analytical purpose. (1.6) Publication of the resulting tabular datasets and complementary materials with persistent identifiers under an open licence turns the curated tabular layer into a stable, citable, and reusable intermediate research output prior to RDF generation. The second section encompasses (2) RDF graph materialisation. (2.1) RDF graph generation from curated data covers the conversion of the curated tabular layer into RDF under an explicit target model and mapping strategy. Its inputs include the cleaned tables, the target application profile or data model, mapping files, runtime configurations, and any function libraries or orchestration logic required to handle heterogeneous values, multivalued fields, partial input availability, and graph merging. This step produces RDF serialisations together with execution diagnostics and traceable mapping artefacts. (2.2) Validation assesses the structural and semantic consistency of the generated graph, including checks for duplicate IRIs, type conflicts, temporal inconsistencies, and violations of expected property patterns. (2.3) Testing complements validation through qualitative and quantitative procedures, including code-level checks, runtime and performance monitoring, regression-sensitive diagnostics, human inspection of sample outputs, and feedback-based refinement cycles. Taken together, these steps define the materialisation core as a controlled transformation environment rather than as a single conversion run. The third section is dedicated to (3) RDF data dissemination. (3.1) Publication of the RDF serialisation through citable and FAIR-compliant releases stabilises the graph as a persistent research output by depositing the RDF data with versioning, identifiers, and explicit reuse conditions. (3.2) Publication of a SPARQL endpoint provides stable programmatic access to the graph through a queryable semantic service. (3.3) Configuration and publication of a semantic access layer for dissemination addresses the need for mediated access beyond specialist users by enabling website- or portal-based interaction, precompiled queries, navigable views, and public-facing documentation without presupposing SPARQL literacy. (3.4) Integration of a provenance management and change-tracking approach preserves auditability, citability, and interpretability across versions by documenting execution conditions, tracking transformations, and supporting the long-term intelligibility of released data products. The fourth and final section concerns (4) Overall quality check. (4.1) User feedback gathering captures adequacy and usability issues emerging across templates, outputs, and dissemination layers. (4.2) Qualitative and quantitative evaluation of the workflow and its outputs consolidates the evidence base generated throughout the process, including validation artefacts, performance reports, output statistics, sample-based inspections, and stakeholder observations, in order to assess both output quality and workflow adequacy. (4.3) Reiteration planning in case of refinement needs or data updates formalises the conditions under which the workflow must be partially or fully re-executed, specifying which phases can be reused unchanged, which artefacts must be regenerated, and which deviations from previous runs must be documented. In this way, the blueprint is not limited to describing a one-off conversion procedure, but functions as a maintainable lifecycle model through which reusable implementations can be attached to specific procedural positions and mobilised as execution-oriented reference artefacts within a single, versioned methodological framework.