autorml 0.1.2

AutoRML: A framework for automatic RML mapping generation using semantic table annotations

AutoRML

AutoRML is a system that automatically generates declarative mappings (YARRRML and RML) and constructs RDF knowledge graphs for tabular data. It does so by utilizing third-party semantic table annotation systems, mapping produced semantic annotations to declarative mapping rules.

AutoRML can be used with different semantic table annotation systems. It has been tested with MTab and TorchicTab. UPDATE: MTab has dropped support. Only TorchicTab is currently available.

Installation

Python versions 3.9, 3.10, and 3.11 are tested and recommended. In case of conflicts, create a new virtual environment. For example, if you use conda, run:

conda create -n autorml_env python=3.11

conda activate autorml_env

AutoRML installation:

pip install autorml

Usage

Usage: 
python -m autorml <options>

Options:
-h,--help                           Show options information
-i, --input_table                   Input table in CSV format
-sta, --sta_system                  Used semantic table annotation system (default: torchictab)
-m, --materialize                   Enable RDF materialization 
-sa, --save_annotations             Save semantic table annotation system results
-af, --annotations_folder           Output annotations folder (default: annotations)
-oa, --annotations_output           Output annotations file (default: annotations.json)
-mf, --mappings_folder              Output mappings folder (default: mappings)
-oy, --yarrml_output                Output YARRRML mappings file (default: mappings.yml)
-or, --rml_output                   Output RML mappings (default: mappings.rml.ttl)
-rf, --rdf_folder                   Output RDF folder (default: rdf)
-okg, --rdf_output                  Output RDF file (default: kg.nt)
-ds, --delete_sem                   Delete supporting semantically enhanced table after termination

Example usage

You can find some tables to test AutoRML in the examples folder. To run AutoRML for one of the examples, go to the repository root and run:

python -m autorml -i "examples/tables/cities.csv" -m -sa -af "examples/annotations" -mf "examples/mappings" -rf "examples/rdf" -ds

After AutoRML terminates, explore the examples folder to see the results of AutoRML's automated declarative knowledge graph construction. The examples/annotations folder will include the semantic annotations generated by the used semantic table annotation system. The examples/mappings folder will include the human-readable YARRRML and the RML declarative mappings produced by AutoRML. The examples/rdf folder will include the RDF knowledge graph automatically generated from the input data source, utilizing AutoRML's RML mappings. You can explore and query the generated graph, as well as search for its relationships and entities in Wikidata, TorchicTab's target knowledge base to semantically annotate the tables.

The open-source version of AutoRML currently only supports TorchicTab as a semantic table annotation system, since MTab is no longer available. AutoRML is designed to support different semantic annotation systems, so you can also try hosting your own or contact AutoRML authors to test your tables.

Evaluations

To see AutoRML results for different dataset collections and semantic table annotation systems, unzip and explore evaluations/evaluations_data.zip or visit our dedicated AutoRML Evaluations Zenodo repository: (TODO add Zenodo link when we submit the paper. Currently the Zenodo repository exists but is private). The evaluation files contain semantic annotations, mappings and RDF knowledge graphs automatically generated by AutoRML, using two different semantic table annotations systems: MTab and TorchicTab, and five differnet collections of CSV tables.

Annotation accuracy:

Annotation automation:

Contact

Ioannis Dasoulas: [e-mail]

Anastasia Dimou: [e-mail]