This paper presents the results of the Knowledge Graph Construction Workshop 2025 Challenge with RMLMapper, a reference in-memory JAVA implementation of the RML specification (RMLio) before RML got taken up by the W3C Knowledge Graph Construction (KGC) Community Group (RMLkgc). We updated RMLMapper to translate R2RML and RMLio into the latest version of RMLkgc. We participated in the challenge for the modules RML-Core, RML-IO and RML-LV. RMLMapper largely supports RML-Core and RML-IO (90% and 65%). For RML-LV, we added RML-view-to-CSV as a first step to our knowledge graph construction pipeline to materialize RML logical views (RML-LV) as CSV files. This way, RMLMapper largely supports RML-LV (81%). We currently cannot claim full coverage of the RMLkgc specification, however, this challenge enabled us to uncover the current gaps in our implementation, and allowed us to provide basic RMLkgc support in a mature JAVA implementation. Our continued participation in the community group will allow us to remain up to date with the latest developments in RMLkgc and further identify future implementation priorities for the RMLMapper.
W
The third edition of the Knowledge Graph Construction Workshop (KGCW) challenge1 is a next iteration
of the KGCW2024 challenge’s conformance track2. Its continued aim is to assess the compliance with
the RML specification as established by the W3C Community Group on Knowledge Graph Construction
(from hereon dubbed “RMLkgc”) [
RML’s modules encompass (i) RML-Core, an iteration of the original R2RML functionality (exempt database connection and query functionality); (ii) RML-IO, an iteration of the RMLio data source reference functionality; (iii) RML-IO-Registry, a living document of reference formulations; (iv) RMLFNML, an iteration of the RMLio FnO integration for data transformation functionality; (v) RML-LV, a new module that provides an abstraction layer between RML-IO and RML-Core; (vi) RML-CC, a new module that provides support for RDF Collection and RDF Container generation; and (vii) RML-Star, a new module that provides support for RDF 1.2 triple term generation. The challenge contains test cases for each of these modules, to determine their feasibility and correctness by applying them in implementations.
In this paper, we present the results for RMLMapper, a reference in-memory JAVA implementation of
the RMLio specification. Since our previous submission [
CEUR
ceur-ws.org
CHALLENGE INPUT:
RML mapping CHALLENGE INPUT: source data
RML mapping without logical views logical view converted to
CSV files RML-view-to-CSV
RMLMapper
OUTPUT KGC pipeline: RDF knowledge graph preprocessing library for RML-LV that materializes Logical Views as CSV files and produces an adapted mapping document that can be processed by any RMLkgc engines supporting CSV files as source data format (Figure 1). With release v1.1.03, RML-view-to-CSV has been updated to support most features of the latest version of the RML-LV specification.
We participated in the challenge for the modules RML-Core, RML-IO and RML-LV. We employed RMLMapper v7.3.34 and RML-view-to-CSV v1.1.03.
RMLMapper supports 53 out of 59 RML-Core test cases (89.83%). Failing cases are due to (i) using JSONPath expressions for keys that contain special characters (RMLTC010a-JSON, RMLTC0010b-JSON, RMLTC0010c-JSON), whereas the employed RMLMapper library does not follow these latest IETF draft specifications yet; (ii) no support for the shortcut property rml:subject in combination with a blank node (RMLTC0025a-JSON); (iii) being too lenient in wrongfully ignoring the rml:termType in a constant-valued term map (RMLTC0025b-JSON).
Considering the failing of test case RMLTC0023e-JSON, this is deemed an incorrect test case, and discussion is ongoing5.
RMLMapper supports 41 out of 73 RML-IO test cases (56.16%). Failing cases are due to (i) no support for compressed data sources (RMLSTC0002b, RMLSTC0002c, RMLSTC0002d, RMLSTC0002e); (ii) no support for logical source serialization n-triples (RMLSTC0003); (iii) incorrectly handling CSWV null values (RMLSTC0004a, RMLSTC0004c); (iv) testing implementation of RMLMapper misses support for templated values for SQL test cases (RMLSTC0006a); (v) no support for rml:CurrentWorkingDirectory (RMLSTC0006b); (vi) no support for XML reference formulation with namespaces (RMLSTC0007d); (vii) returning warnings instead of errors (RMLSTC0010a, RMLSTC0010b); (viii) no support for targets in rml:DatatypeMap (RMLTTC0001f, RMLTTC0002q); and (ix) no support for logical target serializations N3, RDF/JSON, and RDF/XML (RMLTTC0004b, RMLTTC0004e, RMLTTC0004f), UTF-16 encoding (RMLTTC0005b), and compression (RMLTTC0006b, RMLTTC0006c, RMLTTC0006d, RMLTTC0006e).
Considering the failing of test cases RMLTTC0002f, RMLTTC0002g, RMLTTC0002h, RMLTTC0002i, RMLTTC0002k, RMLTTC0002l, RMLTTC0002m, RMLTTC0002n, RMLTTC0002r: mistakes in the test cases were discovered concerning correct graph assignment6. Considering the failing of test case RMLTTC0004g, a mistake in the test case was discovered concerning incorrect test data7. By fixing these mistakes8, RMLMapper supports 51 out of 73 RML-IO test cases (69.86%).
RMLMapper in combination with RML-view-to-CSV supports 26 out of 32 RML-LV test cases (81.25%). Failing cases are due to (i) no support for constant-valued and tempate-valued rml:ExpressionFields in RML-view-to-CSVW (RMLLVTC0001b, RMLLVTC0001c); and (ii) no support 3https://github.com/RMLio/rml-view-to-csv/releases/tag/v1.1.0 4https://github.com/RMLio/rmlmapper-java/releases/tag/v7.3.3 5https://github.com/kg-construct/rml-core/issues/203 6https://github.com/kg-construct/rml-io/issues/129, https://github.com/kg-construct/rml-io/issues/130, and https://github. com/kg-construct/rml-io/issues/131 7https://github.com/kg-construct/rml-io/issues/132 8https://github.com/kg-construct/rml-io/pull/133 for natural data type mapping in RML-view-to-CSV (RMLLVTC0004a, RMLLVTC0004b, RMLLVTC0004c, RMLLVTC0004d).
RML-view-to-CSV can not support natural data type mapping because it materializes logical views as CSV files. CSV does not provide any native data types, therefore there is no natural RDF mapping of CSV values upon XSD data types9. A mistake in RML-LV test case RMLLVTC0008b was discovered and corrected10.
We currently can not claim full coverage of the RMLkgc specification for modules RML-Core, RML-IO, or RML-LV, however, this challenge enabled us to uncover the current gaps in our implementation, and allowed us to provide basic RMLkgc support in a mature JAVA implementation. Finally, some errors in the test cases have been discovered. Our continued participation in the community group will allow us to remain up to date with the latest developments in RMLkgc and further identify future implementation priorities for the RMLMapper.
The described research activities were supported by SolidLab Vlaanderen (Flemish Government, EWI and RRF project VV023/10), and the European Union’s Horizon Europe research and innovation program under grant agreement no. 101058682 (Onto-DESIDE).
The author(s) have not employed any Generative AI tools.