We present the ArCo, GVP and DBpedia Linking Initiative (AGDLI), a research activity within the project SMARTOUR: intelligent platforms for tourism, funded by the Italian Ministry of University and Research. Our initiative is aimed at linking ArCo's cultural entities to the well known Getty Vocabulary Program and DBpedia ontologies, with the main goal of providing a semantically rich representation of the Italian cultural heritage for tourism-related knowledge-based applications. In this paper we provide a detailed description of the initiative and describe the current research developments and outcomes.
Nowadays, we are observing an increasing number of novel semantically-enabled
and knowledge-based applications. Hence, Linked Open Data are more and more
gaining the attention from public administrations and industries all over the
world. In this paper3, we describe the ArCo, GVP and DBpedia Linking
Initiative (AGDLI ). Our initiative is a research activity part of the SMARTOUR:
intelligent platform for tourism project (see Section Acknowledgements). The
main goal of the initiative is to study semi-supervised methodologies to
generate semantically rich de nitions of Italian cultural heritage entities, to be used
in di erent knowledge-based tourism related applications, such as recommender
systems [
Commons License Attribution 4.0 International (CC BY 4.0).
4 http://wit.istc.cnr.it/arco/?lang=en.
5 https://www.getty.edu/research/tools/vocabularies/.
DBpedia [
In Figure 1, we depict an excerpt of the ArCo schema. In this diagram we highlight some of the properties of ArCo CulturalProperty entities that we link to the GVP ontology. Speci cally, we are investigating semi-supervised methodologies to automatically: i) mine and link the dc:type and rdfs:label properties of CulturalProperty instances to the AAT ; ii) link the cities of the addresses of CulturalProperty instances to the TGN ; iii) link the agents of authorship attributions of CulturalProperty instances to the ULAN ; iv) normalize the date intervals of CulturalProperty instances into a machine readable format7, such as the Open Date Range Format 8. We note that ArCo is in Italian, while the GVP is mainly in English, which represents an additional challenge of our linking initiative. In Figure 2, we depict an example of mining concepts from ArCo entities' textual descriptions and linking them to corresponding concepts in the AAT. In this task: 1. we automatically translated from English to Italian the AAT terms. To this end, we used the Google Translate API9. Note that, we preserved the original Italian terminology when already provided by the AAT; 6 https://www.dbpedia.org/. 7 This initiative's aim is intended to provide a ready to use resource for time-based tourism applications. 8 Dublin Core Collection Description: Open Date Range Format http://www.ukoln.ac.uk/metadata/dcmi/date-dccd-odrf/. 9 https://cloud.google.com/translate/. skos:subject
AAT gvp:Concept * type
label dc:type rdfs:label
dc:date arco-arco:CulturalProperty date dc:date
DCCD ODRF clvapit:hasGeometry arco-location:hasCulturalPropertyAddress
arco-cd:hasAuthorshipAttribution clvapit:Geometry clvapit:Address arco-cd:AuthorshipAttribution arco-location:hasCoordinates clvapit:hasCity
arco-cd:hasAttributedAuthor arco-location:Coordinates clvapit:City l0:Agent arco-location:lat arco-location:long rdfs:label
rdfs:label lat
long Prefixes arco-arco: <https://w3id.org/arco/ontology/arco/> arco-location: <https://w3id.org/arco/ontology/location/> arco-cd: <https://w3id.org/arco/ontology/context-description/> clvapit: <https://w3id.org/italia/onto/CLV/> dbpedia: <http://dbpedia.org/resource/> dc: <http://purl.org/dc/elements/1.1/> gvp: <http://vocab.getty.edu/ontology#> l0: <https://w3id.org/italia/onto/l0/> owl: <http://www.w3.org/2002/07/owl#> rdfs: <http://www.w3.org/2000/01/rdf-schema#> skos: <http://www.w3.org/2004/02/skos/core#> city
author owl:sameAs owl:sameAs
owl:sameAs owl:Class
gvp:PhysPlaceConcept gvp:PersonConcept TGN
2. we applied standard text pre-processing techniques (e.g., tokenization, lowercasing) to ArCo entities' textual descriptions. To this end, we adopted the Stanford NLP API10; 3. we automatically collected all the occurrences of Italian AAT terms in ArCo entities' rdfs:label and dc:type resulting preprocessed textual properties. In this step, for each ArCo's entity, we obtained a collection of links with AAT concepts. As a result, we obtained a collection of ambiguous links to all the AAT concepts having the same skosxl:literalForm 11 (see the example of as described in Figure 2,). 4. since these tasks are error-prone, we performed a manual re nement of the translated AAT 's terms, xing translation errors and adding synonyms, singular, plural and hypernymous forms for terms occurring in the the textual properties of missing linked ArCo's entities; 5. we repeated steps 3 and 4 until an adequate coverage was reached. 10 https://nlp.stanford.edu/software/. 11 https://www.w3.org/TR/skos-reference/skos-xl.html#literalForm. To link ArCo's cities to TGN and DBpedia we performed string matching12 with the corresponding terms and entities. At the time of writing, we are investigating on e ective linking methodologies of ArCo's I0:Agent s with ULAN entities, and on dc:date normalization. 3
Current Outcomes and Conclusions In this paper, we introduced the AGDLI initiative. As a result, we obtained13: { the automatic translation in Italian of the 55K AAT terms; { a total of 5.6 M triples (skos:relatedMatch and skos:related ) linking the 98.2% (by dc:type) and the 99.9% (by rdfs:label ) of arco-arco:CulturalProperty entities to candidate AAT concepts; { a total of 6.6 K triples (skos:relatedMatch) linking the 86.3% of clvapit:City instances to candidate TGN entities; iv) 4.7 K novel owl:sameAs relations, now linking the 100% clvapit:City to DBpedia. 12 We applied di erent similarity measures e.g., string edit distance-based similarity. 13 Resources are available under Creative Commons Attribution 4.0 International (CC BY 4.0) at https://sites.google.com/unitelmasapienza.it/agdli/. As already introduced in Section 2, the next planned activities are aimed at both linking ArCo's authorship attributions to ULAN entities and normalizing the CulturalProperty 's dc:date.
Moreover, we are planning to apply semi-supervised methodologies for the disambiguation of the generated candidate links. For instance, generated links to AAT concepts can be re ned with semi-supervised word sense disambiguation approaches, while the generated matches with TGN candidates can be disambiguated based on the distance between the geographical coordinates of ArCo and TGN entities.
Further plans of the AGDLI initiative include, among others, the
application and investigation of knowledge graph completion methodologies [
This work was carried out within the research project "SMARTOUR: intelligent platform for tourism" funded by the Italian Ministry of University and Research with the Regional Development Fund of European Union (PON Research and Competitiveness 2007-2013).