A Lifecycle-Based Metadata Model for Improving Documentation, Reuse, and Generation of Declarative Mappings

Abstract

Declarative mappings play an essential role in knowledge graph construction and data integration, enabling data transformation, alignment and interlinking of heterogeneous data sources. Despite their importance, mappings as semantic artefacts are rarely accompanied by the contextual documentation needed to understand why and how they were created. As a result, finding, assessing, maintaining and reusing these mappings becomes difficult. Existing standards such as PROV-O, DCAT, and SSSOM address aspects of provenance and alignment documentation but do not capture the full lifecycle of declarative mappings across different mapping types. This gap limits the ability to reuse, store and discover mappings in repositories, and also limits the ability of automated tools, including Large Language Models (LLMs), to generate reliable mappings without sufficient context. This thesis addresses this gap through the design and evaluation of the Lifecycle-Based Metadata Model for Declarative Mappings (LMMD). LMMD organises metadata according to five phases of the mapping lifecycle: analysis, design, development, testing, and maintenance. The model is designed to apply across mapping types including uplift mappings, ontology alignments, and entity interlinking, and can be used and extended accordingly. LMMD is then formalised as the Mapping Metadata Vocabulary (MMV), an OWL ontology aligned with established vocabularies including PROV-O, DCTERMS, and MQV. MMV is operationalised through MetaSEMAP, a web-based annotation tool that enables practitioners to document and export metadata for mapping artefacts. The thesis presents five evaluation studies. The first study is a community survey that validated the relevance of proposed metadata fields with semantic web community members. The second study assessed usability of MetaSEMAP through task-based testing. The third study examined the conceptual clarity, lifecycle coverage, and modelling quality of MMV through expert review and automated validation. The fourth study investigated whether lifecycle-based metadata supports more informed mapping reuse decisions. Finally, a use case explored whether metadata-guided prompting improves semantic correctness and completeness of LLM-generated mappings. The results demonstrate that LMMD provides a relevant, usable, and valuable foundation for mapping documentation and reuse decisions. The findings indicate that a lifecycle-based metadata model addresses a genuine need in the knowledge graph and data integration community, with broader implications for FAIR data practices, mapping governance, and AI-assisted mapping.