Abstract:
Effective integration of heterogeneous statistical datasets remains a key challenge in semantic data publishing. Traditional approaches, ranging from ETL pipelines to OLAP and ontology-based solutions, are effective in many settings, but they can become difficult to reuse when multidimensional datasets evolve, use different structures, or rely on distinct controlled value sets. This paper presents an RDF Data Cube-specific modeling and tooling approach for semantic integration of multidimensional statistical data. The approach shifts part of the integration effort to the modeling stage, where datasets are described through modular Data Structure Definitions (DSDs), reusable cube components, SKOS codelists, metadata and hierarchical component relations. To operationalize this approach, we present CubeModeler, a semantic modeling environment that supports DSD construction, component and codelist management, dataset description, DSD-driven RDF transformation and integrated SPARQL access. Two use cases, basketball statistics and environmental/public-administration data, demonstrate how the approach supports different integration settings: a relatively stable domain with recurring entities and a decentralized domain with heterogeneous providers, codelists, and DSD structures. Representative SPARQL query patterns show how generated RDF cubes can be integrated through shared components, contextual resources, codelists and component hierarchies. The evaluation shows that the main benefit of CubeModeler is the organization of the modeling and querying effort into reusable artifacts and repeatable workflows. Once components, codelists, DSDs, and query patterns are defined, later datasets and integration tasks can reuse them through targeted extensions, RDFization workflows, or parameterized query patterns.
