An RDF Dataset Description Model
for Expressing Vocabulary Usage Patterns

Nandana Mihindukulasooriya, María Poveda-Villalón, Raúl García-Castro, Asunción Gómez-Pérez

Ontology Engineering Group, Universidad Politécnica de Madrid

Acknowledgments:
4V: Volumen, Velocidad, Variedad y Validez en la Gestión Innovadora de Datos

Current State

Datasets
- LOD Cloud crawled data(900K+ documents)
- LOD Laundromat + LDF endpoints (658K+ datasets)
- datahub /w rdf (1792 datasets)
- OpenLink Software LOD Cache

Dataset descriptions
- Generic - DCMI, DCAT
- RDF - VoID, Aether VoID extension, HCLS Distribution Level Description, LODStats, RDFStats, ProLOD++,ExpLOD, ABSTAT

Dataset searche engines
- datahub, lodsearch.org, Sindice, Semplore, VisiNav, Semantic Sitemaps, DING!, ...

Motivation

Use Case I - Dataset discovery

As an RDF data consumer
I want to discover datasets that satisfy the requirements of a given task
- Input: keywords, SPARQL query, Shapes (SHACL, ShEx)
- Output: relevant datasets
so that I can use data from the LOD cloud without spending days of manual inspection.

Use Case I - Example

I want to find training data for a ML classifier that categorizes restaurants according their cuisine based on the restaurant's description.

	    PREFIX schema: <http://schema.org/>

select ?restaurant ?description ?cusine  where {
 ?restaurant a schema:Restaurant;
     schema:description ?description;
     schema:servesCuisine ?cusine .
}

SPARQL Query

	    @prefix schema: <http://schema.org/> .
@prefix ex: <http://example.org/vocab> .
@prefix sh: <http://www.w3.org/ns/shacl#> .

ex:RestaurantShape a sh:Shape ;
	sh:targetClass schema:Restaurant;
	sh:property [
		sh:predicate schema:description;
		sh:minCount 1 ;
	];
	sh:property [
		sh:predicate schema:servesCuisine;
		sh:minCount 1 ;
	] .

Shape (SHACL)

Motivation

Use Case II - Vocabulary Usage Reports

As an ontology engineer / vocabulary developer
I want to know how my vocabulary is used in datasets in practice
so that I can understand mismatches or problems and improve it.

Use Case II - Example

Given an ontology (e.g., SSN Ontology)
Where is it used
- How many datasets / instances use each class?
- How many datasets / triples use each property?
How is it used
- What are the other classes used with instances of each class?
- What are the common subject types of each property?
- What are the common object types of each property?
- Do people use my vocabulary terms in a more generic/specific way?
- Do people replace part of ontology with another one?
- Do people frequently violate some restrictions?

Motivation - Contd.

As an RDF dataset consumer, I want to understand the content and the structure of a dataset so that I can start using it or perform queries without spending hours doing exploratory queries.
As an RDF dataset consumer / producer, I want to compare datasets (different datasets or multiple versions of the same dataset) so that I can understand the differences and changes.

Can we extend dataset descriptions by expressing vocabulary usage patterns?

Vocabulary Usage Patterns

An implict schema (shapes) for a given dataset
Class analysis
- # of instances, associated properties with their cardinalities, equivalent/super classes
Property analysis
- # of triples, estimated domains / ranges, value ranges/patterns for data type properties
Abstract triple pattern analysis
Language analysis for strings