The volume of data generated in recent years has increased drastically, necessitating a unified schema to integrate multiple data sources into a single format. The RDF Mapping Language (RML) was developed to define the structure of knowledge graphs (KGs). Over time, various extensions have been introduced to enhance RML's functionality, creating a need for a new specification that consolidates these extensions. Track 1 of the KGCW 2023 Challenge dataset addresses this need by providing a comprehensive set of test cases to ensure that knowledge graph creation engines comply with the updated RML specification. This paper reports on the conformance evaluation of SDM-RDFizer using this dataset, highlighting its capabilities and areas for improvement in achieving full RML compliance.
evaluates the results using the Track 1 dataset.
This paper is organized into three additional sections. Section 2 provides an overview of SDMRDFizer, including its techniques, data structures, and physical operators for optimizing KG creation. Section 3 details the results of the challenge, including the dataset definition and the necessary improvements to meet the test cases. Finally, Section 4 presents the conclusions and future steps for SDM-RDFizer.
SDM-RDFizer [
Track 1 of the KGCW 2024 Challenge4 aims to inspire new methods and techniques for incorporating the new RML specification into existing KG creation engines. This dataset comprises five sets of test cases. RML-Core: This set includes basic test cases originally defined in the RML test cases5 to test the compliance of KG creation engines. These cases have been updated to reflect the new specification and utilize CSV, JSON, XML files, and relational databases (MySQL and Postgres) as data sources. RML-FNML: This set contains test cases that use functions to transform data, employing a series of pre-defined functions to execute these transformations. RML-Star: This set incorporates RDF-Star6 test cases, generated from the RML-Star test cases7 in accordance with the new specification. RML-IO: This set includes a wide variety of remote data sources such as endpoints, compressed files, and JSON and XML files. It also defines outputs for specific properties in various formats like Turtle, RDF/JSON, JSON-LD, and multiple compressed formats like Zip and Tar. The name reflects its focus on Input and Output. RML-CC: This set is comprised of collections and containers.
This work presents the results of executing RML-Core, RML-FNML, RML-Star, and RML-IO with SDM-RDFizer. RML-CC will be incorporated at a later date. Table 1 shows the total number
4https://zenodo.org/records/10973433 5https://rml.io/test-cases/ 6https://www.w3.org/2021/12/rdf-star.html 7https://zenodo.org/records/6518802
Set RML-Core RML-FNML RML-Star RML-IO RML-CC
Total of test cases in the dataset and which cases were passed and failed by SDM-RDFizer. The full results are available on GitHub 8.
RML-Core comprises test cases covering the fundamentals of RML, such as the definition of classes, the use of rml:template and rml:reference, the execution of parent triples maps and joins, and the definition of data types, languages, and graphs. SDM-RDFizer implements three operators to execute diferent types of mappings.
To extract data from various data sources, SDM-RDFizer employs several Python libraries: csv for CSV files, json for JSON files, xml for XML files, mysql-connector for connecting to MySQL databases, and psycopg2 for connecting to Postgres databases.
To parse the new specification, the parser query is updated to replace the rml prefix with its new namespace, replace all mentions of the R2RML namespace with the RML namespace, and update the definition of rml:logicalSource to include the use of the rml:path and rml:root clauses, replacing rml:query with rml:iterator. The rdflib library is used to execute the parser query. SDM-RDFizer successfully performs all 238 test cases from this set.
RML-FNML consists of test cases that use functions to transform values, including tasks like
replacing strings, transforming strings to lower and upper case, concatenating strings, and
more. These test cases demonstrate the use of RML+FnO in the new specification. SDM-RDFizer
incorporates strategies from FunMap [
To parse TMs with functions, a new parser query is explicitly defined for extracting the functions from the TMs. This allows for proper handling of nested functions, as each function is extracted individually, and nested functions are called from the function’s parameters. SDM-RDFizer successfully performs all 14 test cases from this set.
RML-Star comprises test cases that use RDF-Star, an extension of RDF that introduces a new term, the quoted triple, which can be used as either the subject or the object of a triple. Therefore, RML-Star presents rml:quotedTriplesMap, enabling the definition of quoted triples in a KG. SDM-RDFizer implements a new operator to generate quoted triples, allowing for recursive application since quoted triples can contain other quoted triples.
Another challenge in these test cases was using joins in the rml:subjectMap. These cases were handled similarly to joins in the rml:objectMap, using the OJM operator for join execution and PJTT for storing the results. The parser query was expanded to recognize rml:quotedTriplesMap. SDM-RDFizer successfully performs all 18 test cases from this set.
RML-IO consists of test cases that cover a wide range of remote data sources, including compressed files, JSON and XML files, and data extracted from SPARQL endpoints. Some cases introduce the concept of outputting certain triples to specific output files, which may need to be compressed or translated into diferent formats, such as JSON-LD, Turtle, etc. SDM-RDFizer uses the requests library to collect data from remote sources. For SPARQL endpoints, the SPARQLWrapper library connects and executes SPARQL queries, with the results converted to a format similar to CSV. When dealing with compressed files, SDM-RDFizer downloads the file locally and decompresses it using the appropriate library for the format, such as the zip library for Zip files.
These test cases introduce the concept of defining an alternate output within the TM. These can be specified in the rml:subjectMap, rml:predicateMap, rml:objectMap, rml:languageMap, rml:datatypeMap, or rml:graphMap. Based on its location, triples will be outputted to the alternate output file. If defined in the rml:subjectMap, all valid triples generated from this TM will be sent to that particular output. If the alternate output is defined elsewhere, only triples with that specific property will be outputted there. Any remaining triples not destined for an alternate output will be sent to the original output file, which is defined at the start of the SDM-RDFizer execution. When handling these alternative output files, SDM-RDFizer prioritizes generating them over the standard output and can compress them. Finally, SDM-RDFizer converts the output files into various RDF formats, such as RDF/XML, JSON-LD, etc., using the rdflib library. SDM-RDFizer successfully executes 65 of the 67 test cases. The two failures occurred because SDM-RDFizer cannot upload the generated triples into a SPARQL endpoint.
The KGCW 2024 Challenge Track 1 dataset evaluates the compliance of state-of-the-art engines with the new formulation of RML. It comprises 366 test cases across five categories: RML-Core, RML-IO, RML-FNML, RML-Star, and RML-CC. SDM-RDFizer successfully executes 335 of these test cases, fully covering RML-Core, RML-FNML, RML-Star, and all but two from RML-IO. To achieve this, SDM-RDFizer introduced a new parsing query, strategies from FunMap, and an operator for generating inner triples for RML-Star. Moving forward, SDM-RDFizer aims to address the remaining RML-CC cases by incorporating new methods. For this purpose, a new operator will defined, which will behave diferently based on whether it is transforming a list, bag, or sequence. Additionally, a new data structure will be developed for intermediate results in RML-Star to avoid repeated inner triple generation, and an optimized parser will be implemented to manage the increasing complexity.
With these improvements, SDM-RDFizer is set to become a fully compliant RML engine.
This work has been partially supported by the Federal Ministry for Economic Afairs and Energy of Germany (BMWK) in the project CoyPu (project number 01MK21007[A-L]). Leibniz Association partially funds Maria-Esther Vidal in the "Leibniz Best Minds: Programme for Women Professors", project TrustKG-Transforming Data in Trustable Insights with grant P99/2020.