Home

Awesome

Knowledge Graph Construction with R2RML and RML

We conduct an evaluation of KGC engines considering several R2RML and RML processors to identify their strengths and weaknesses. We (i) perform a qualitative analysis of the distinctive features of each engine, (ii) examine their conformance with the mapping language specification they support, and (iii) assess their performance and scalability using the GTFS-Madrid-Bench benchmark.

Citing:

@inproceedings{arenas2021knowledge,
  title = {{Knowledge Graph Construction with R2RML and RML: An ETL System-based Overview}},
  author = {Arenas-Guerrero, Julián and Scrocca, Mario and Iglesias-Molina, Ana and Toledo, Jhon and Pozo-Gilo, Luis and Doña, Daniel and Corcho, Oscar and Chaves-Fraga, David},
  booktitle = {Proceedings of the 2nd International Workshop on Knowledge Graph Construction},
  year = {2021},
  series = {CEUR Workshop Proceedings},
  publisher = {CEUR-WS.org},
  volume = {2873},
  url = {http://ceur-ws.org/Vol-2873/paper11.pdf},
}

Engines

We test the performance and scalability of a set of KG construction engines:

R2RML-based:

RML-based:

Evaluation resources

GTFS-Madrid-Bench

Using the GTFS-Madrid-Bench and based on the input dataset we create the following distributions to test the engines:

Data can be directly download executing bash scripts/download-data.sh

R2RML and RML test-cases

We use the resources provided by the W3C community on KG-Construction (https://www.w3.org/community/kg-construct/) to run the R2RML and RML test-cases over the selected engines.

Results

We created a comparative framework to gather and compare the information about the engines' features, availabe here; and tested the engines with the mentioned benchmark in terms of time and memory used. The raw data resulting from the evaluation is stored here, and the resulting figures can be seen here.

Authors