From Flat Lists to Taxonomies: Bottom-up Concept Scheme Generation in Linked Statistical Data

RDF Data Cube allows the modeling and publishing of Linked Statistical Data (LSD) in the Semantic Web. Often, variable values of such statistical data come in a non-standardized way and represented by too narrow, concrete or wrongly typed literals. Generally, adequate and standard concept schemes for such variables (especially in very specific domains like historical religious denominations, or building types in the pre-industrial era) do not exist and need to be created. This is a manual task that requires lots of expert knowledge and time investment. We present a workflow that combines hierarchical clustering and semantic tagging to automatically build concept schemes in a data-driven and bottom-up way, leveraging lexical and semantic properties of the non-standard dimension values. We apply our workflow in two different use-cases and discuss its usefulness, limitations and possible improvements.