Let’s invert it! From RDF to relational data with R2RML

Arcangelo Massari, Anastasia Dimou

Keywords: R2RML, knowledge graph construction, mapping invertibility, relational databases

Abstract: Knowledge graph construction from relational databases relies on mapping languages, such as R2RML, to transform source data into RDF. While the forward transformation is well investigated, the backwards transformation of reconstructing the original data from an RDF graph using the same mapping remains largely unexplored. This paper presents an algorithm for R2RML mapping inversion that generates SPARQL queries from the mapping document to reconstruct the source data, along with a proof-of-concept implementation. We validate the algorithm against the W3C R2RML test suite applied in reverse: all test cases with well-formed and supported mappings are correctly inverted, while the remaining cases, which reveal unsupported features and information loss that prevent reconstruction, are analyzed. Based on these results, we discuss the conditions under which R2RML mappings can be inverted and the structural factors that prevent inversion.

Submission Number: 14