The Spanish DBpedia is a data source used initially to support the Spanish community. However, our logs show that the Spanish language goes beyond Spanish speakers and many non-Spanish speakers use the Spanish DBpedia on a daily basis. In the last months we have made two important enhancements to the Spanish DBpedia: (1) we publish a nonstandard dataset containing the type of resources that in the standard distribution have no type, and (2) we update automatically our data every week by using the DBpedia databus. In this way, we satisfy a frequent request made by companies and we foster the usage of the Spanish language, the second mother language by the number of speakers (after Chinese), and the second in scienti c papers (after English).
1.1
The data published by the Cervantes Institute in its 2020 report [
The DBpedia project has long generated semantic information from English
Wikipedia. Since June 2011, the information generation process has extracted
information from Wikipedia in 111 of its languages, but only 18 languages have a
DBpedia chapter with a website. One of them is Spanish. The DBpedia
Internationalization Committee has assigned a website and a SPARQL [
As we have mentioned previously, DBpedia extracts data from 111 di erent language editions of Wikipedia. Then, for each language we have a knowledge base (a \Knowledge Graph" in modern terminology, abbreviated as KG). The largest DBpedia KG is extracted from the English edition of Wikipedia, with around 400 million facts (triples) that describe 3.7 million resources (Wikipedia entries). The DBpedia knowledge graphs that are extracted from the other 110 Wikipedia editions together consist of 1.46 billion facts and describe 10 million additional resources. Therefore, two-thirds of the information in DBpedia comes from non-English Wikipedias.
From a technical perspective, the DBpedia project maps Wikipedia infoboxes [
In addition to the regular releases, the project maintains a live knowledge base which is updated whenever a page in Wikipedia changes. DBpedia sets 27 million links pointing to many external data sources (e.g. Wikidata, Yago, Freebase) and thus enables data from these sources to be used together with DBpedia data. Several hundred data sets on the Web publish RDF links pointing to DBpedia and thus make DBpedia the central interlinking hub in the Linked Open Data (LOD) cloud2. 1.3
Around 40% of the resources (entries) in the Spanish Wikipedia are not pointed
(do not have links) by the English Wikipedia [
The contributions made to the Spanish DBpedia during the last year are described in the next sections. 2.1
Each DBpedia resource usually has more than one type. For example, the resource Cervantes has types Agent, Person and Artist, this is because the DBpedia ontology de nes a well-known hierarchy of classes.
A previous study [
In the study mentioned previously, made by members of our research group, it was presented a new technique that improved the overall quality of the English DBpedia dataset by (1) providing type(s) to those resources lacking a type, and (2) adding more specialized types to already typed resources. Following the previous example about Cervantes, this means to infer that it also has the type Writer, a more speci c type in the DBpedia hierarchy, although this fact was not in the original dataset. These inferred types are stored in a new dataset so that now it is made publicly available in Spanish and English.
This method surpasses the so-called SDTypes dataset as shown in gure 1.
The small circle represents the number of types predicted by the SDType
approach (1.38M). The big circle represents our approach, which produces 56.7%
more types (2.15M). Our approach predicts most of the types predicted by the
SDTypes approach[
App2 C5.0 - 156.7% 2,151,900 only SDtype - 3.5% → 48,152 common - 96.5% → 1,325,252 only App2 C5.0 - 60.2% → 826,760 SDtype - 100% 1,373,404 Fig. 1. Overlapping between the types predicted by our approach (light gray) and the ones predicted by the SDTypes approach (dark gray).
The contribution of the present work is to notify that we have overcome a
technical issue that limited the number of resources that could be analyzed. This
limitation restricted our study to old versions of DBpedia, in which we had less
than 2.4 million resources. The master thesis [
In the past, the update process was based on a manual process in which we had to download the Spanish datasets provided periodically by DBpedia. This periodicity was the refresh rate of the DBpedia extraction process, and was in the range of months.
However, now we have an automated process and we get fresh data
every week. This is possible by using the DBpedia Databus [
The idea behind using Databus to get our data is the fact that it is able
to generate metadata about datasets and then upload this metadata, so that
anybody can query, download, derive, and build applications with this data
via the Databus. Since the integration of data is easy with the Databus, many
additional datasets have been integrated and loaded alongside DBpedia for the
world to query. The process to load datasets on the bus comprises these phases:
{ Acquisition: data is downloaded from the source and logged in.
{ Conversion: data is converted to N-Triples and cleaned (Syntax parsing,
datatype validation and SHACL).
{ Mapping: the vocabulary is mapped on the DBpedia Ontology and converted
(this was being done for Wikipedia's Infoboxes and Wikidata [
{ The possibility to compare all data via a web service. 3
We share with the community the enhancements made to the Spanish DBpedia
in the last year. This has been the biggest change since its creation in 2011.
The changes include a new user interface, automatic weekly updates, and the
creation of a high-quality dataset on resource's types. All of this is publicly
available (code, datasets and instructions) on the Internet [