Source-linked AI summary
ArCo: the Italian Cultural Heritage Knowledge Graph
Valentina Anita Carriero, Aldo Gangemi, Maria Letizia Mancinelli, Ludovica Marinucci, Andrea Giovanni Nuzzolese, Valentina Presutti, Chiara Veninata
TL;DR
Italian cultural heritage has potential for linked open data, and the official General Catalogue offers a curated source for a knowledge graph. This paper presents ArCo, a knowledge graph built from that catalogue with supporting conversion and documentation resources. ArCo is described as a robust but evolving Semantic Web resource.
Problem
The paper identifies the curated Italian General Catalogue as a potential hub for a reliable, rich knowledge graph of Italian cultural heritage.
Source: §1 · p. 2Method
ArCo combines a detailed ontology network and linked-data dataset with catalogue-to-RDF software, documentation, examples, and testing resources.
Source: §1 · p. 2Results
ArCo is presented as a robust Semantic Web resource for Italian cultural heritage, while remaining an evolving project that can be improved and enriched.
Source: §8 · p. 11Takeaways & Limitations
ArCo’s design and evaluation address provider and consumer requirements, while reusing or aligning with relevant cultural-heritage ontologies.
Source: §1 · p. 2Some naturalistic-heritage characteristics remain unmodeled, pictures are not yet in the dataset, and most data remain in Italian despite 17,906,639 entities having an English label.
Source: §8 · p. 11
Abstract
from arXiv · showhide
ArCo is the Italian Cultural Heritage knowledge graph, consisting of a network of seven vocabularies and 169 million triples about 820 thousand cultural entities. It is distributed jointly with a SPARQL endpoint, a software for converting catalogue records to RDF, and a rich suite of documentation material (testing, evaluation, how-to, examples, etc.). ArCo is based on the official General Catalogue of the Italian Ministry of Cultural Heritage and Activities (MiBAC) - and its associated encoding regulations - which collects and validates the catalogue records of (ideally) all Italian Cultural Heritage properties (excluding libraries and archives), contributed by CH administrators from all over Italy. We present its structure, design methods and tools, its growing community, and delineate its importance, quality, and impact.
1 Bringing the Italian Cultural Heritage to LOD
Italy’s curated General Catalogue provides a potential foundation for linked cultural-heritage data, and this paper presents ArCo as a knowledge graph and supporting reuse package. ArCo combines detailed cultural-heritage models and catalogue data with conversion software, documentation, examples, tests, and online access.
1 Bringing the Italian Cultural Heritage to LOD
- The General Catalogue aims to cover Italian cultural heritage except libraries and archives; ICCD had assigned over 15 million catalogue numbers and collected about 2.5 million records.About 0.8 million records were available for consultation on the official website.Source: §1 · p. 2
- ArCo’s ontology network and linked-data dataset describe about 0.8 million cultural properties in approximately 169 million triples.The network models cultural heritage at a fine-grained level.Source: §1 · p. 2
- ArCo provides software that converts ICCD-standard-compliant catalogue records into linked data, enabling automatic, frequent updates and facilitating reuse.The release also includes documentation of requirements and ontology models, running examples with testable SPARQL queries, a test suite, and a SPARQL endpoint.Source: §1 · p. 2
- ArCo is presented as having reliable provenance and high-quality ontologies developed through eXtreme Design, evaluation, elicited requirements, and reuse or alignment with relevant heritage ontologies.The data also link to about 18,700 entities in other linked-data datasets.Source: §1 · p. 2
- The authors describe ArCo as an evolving project with aspects that can be improved.They state that improvement areas are discussed later in the paper.Source: §1 · p. 3
2 The official catalogue of Italian Cultural Heritage
ArCo’s source is the ICCD-maintained General Catalogue, whose records are submitted by authorized contributors and undergo automatic and scientific validation. The records use XML, where information is represented as strings, so RDF modeling requires identifying entities and defining classes and URI rules.
2 The official catalogue of Italian Cultural Heritage
- The General Catalogue contains 2,735,343 records, of which 781,902 are publicly consultable; other records may concern private, at-risk, or not-yet-assessed properties.Its 487 authorized contributors submit records through SIGECweb, where submissions undergo automatic standards checks and scientific validation by appointed experts.Source: §2 · p. 3
- ArCo also draws on consumer use cases and ICCD cataloguing standards covering multiple cultural-property types.These include archaeological, architectural and landscape, photographic, musical, natural, numismatic, scientific and technological, historical and artistic properties, among others.Source: §2 · p. 3
- Because XML records express information as strings, RDF modeling creates individuals for entities such as an artwork’s author and technical status.The modeling process defines ArCo classes from sets of fields and rules for generating individual URIs.Source: §2 · p. 3
3 Using eXtreme Design for developing ArCo Knowledge Graph
ArCo was developed through an iterative eXtreme Design process that joins customer requirements, ontology design, and independent testing. The process also covers RDF production, refactoring, and incremental releases.
3 Using eXtreme Design for developing ArCo Knowledge Graph
- XD assigns ODP selection and implementation to a design team, testing to a separate team, and requirements elicitation to a customer team.The teams translate requirements into competency questions and other constraints that guide development.Source: §3 · p. 4
- Requirements arrive as user stories and competency questions, with stakeholders adding perspectives beyond ICCD and GitHub issues feeding continuous feedback.Incremental releases let teams address emerging requirements and errors in subsequent versions.Source: §3 · p. 4
- ArCo’s seven modules use direct and indirect ODP reuse, annotate reused patterns with OPLa, and include a new pattern for recurrent events.Modules cover coherent domain areas and subsets of requirements.Source: §3 · p. 4
- The XML2RDF Converter maps ICCD-standard XML to RDF using XSLT stylesheets, while URI generation derives IDs from selected record elements or uses an existing unique identifier.Generated individuals undergo LIMES entity linking to minimize duplication.Source: §3 · p. 5
- Testing combines competency-question verification, expected-inference checks, and error provocation, alongside regular HermiT reasoning for incoherence.Injected inconsistencies and missing inferences expose absent ontology axioms; testing proceeds in parallel with modeling.Source: §3 · p. 5
- Testing issues return to the design team for refactoring and consistency checks, and the testing team validates changes before release.Modules are versioned alpha, beta, or stable according to internal and external testing progress.Source: §3 · p. 5
4 ArCo Knowledge Graph
ArCo packages a modular cultural-heritage ontology network and linked-data dataset with tools and materials for access, use, and reuse. Its modules model distinct cultural-property information, and the dataset includes external links.
4 ArCo Knowledge Graph
- ArCo’s release package combines the knowledge graph with documentation, software, and online services intended to support consumption, understanding, and reuse.The paper describes the KG as the union of its ontology network and linked-open-data data.Source: §4 · p. 5
4.1 How to use ArCo
- Each release provides user guides, ontology source and documentation, a SPARQL endpoint, query examples, and an RDFizer for converting ICCD-standard XML to ArCo RDF.Releases are distributed in Docker containers on GitHub and online.Source: §4.1 · pp. 5–6
4.2 ArCo Ontology Network
- ArCo’s seven ontology modules connect through owl:imports; arco and core provide top-level and cross-module concepts, while catalogue preserves record provenance and dynamics.The remaining modules address cultural events, denotative descriptions, locations, and context descriptions.Source: §4.2 · p. 6
- The network directly reuses Cultural-ON and OntoPiA and indirectly reuses CIDOC-CRM, EDM, BIBFRAME, FRBR, FaBiO, FEntry, and OAEntry.Figure 3 distinguishes ArCo modules from directly and indirectly reused ontologies.Source: §4.2 · p. 6, Figure 3 · p. 6
- ArCo represents some concepts both as n-ary relations for temporal or other high-level modeling and as shortcut binary relations for lightweight navigation.For material, hasMaterial is a property-chain shortcut over technical-status and characteristic relations.Source: §4.2 · p. 6
- The arco module partitions cultural properties into tangible and intangible types, while catalogue represents records, their associated properties, and record versions.The core module supplies general-purpose concepts used across the network.Source: §4.2 · p. 7
- Location models multiple time-indexed perspectives, denotative description models measured or detected characteristics, context description models associated attributes, and cultural-event includes recurrent events.The recurrent-event pattern is ArCo-defined; a thorough description is deferred to a dedicated publication and the ODP portal.Source: §4.2 · pp. 7–8
4.3 ArCo dataset
- 169,151,644 triples make up the ArCo dataset; Table 1 summarizes extensions for prominent ArCo concepts.The release also includes links to 18,746 distinct external entities, limited to authors and places.Source: §4.3 · p. 9, Table 1 · p. 9
5 Evaluation and Impact of ArCo
ArCo is evaluated for usability, logical consistency, and requirements coverage, with testing issues addressed by the design team. The paper also reports a growing user community and independent organizational uses.
5.1 Evaluation
- ArCo’s usability indicators include bilingual labels and comments, detailed documentation, 395 restrictions, 130 disjointness axioms, and 37 alignments with seven external ontologies.The ontology also contains 59 classes and properties directly reused from other ontologies.Source: §5.1 · p. 9
- 53 competency questions produced SPARQL queries with expected results; all identified issues were fixed, and the design team resolved 35 GitHub issues.Testing also found three issues in 18 inference tests and 14 missing axioms in 29 error-provocation tests.Source: §5.1 · p. 9
5.2 Potential impact of ArCo
- At least five organizations use ArCo in independent projects, including Google Arts & Culture, Regesta.exe, Synapta, OnData, and InnovaPuglia.Reported activity also includes 496 release-site visits from 170 distinct users between January and March 2019, and 1,262 endpoint queries from March 25 to April 4, 2019.Source: §5.2 · p. 10
6 ArCo: availability, sustainability, and licensing
ArCo is available through MiBAC, its SPARQL endpoint, GitHub, LOV, and Zenodo, with a DOI and a CC BY-SA 4.0 license. MiBAC, CNR, ICCD experts, and an active community contribute to its maintenance and evolution.
6 ArCo: availability, sustainability, and licensing
- ArCo’s public GitHub repository is version-controlled, and the knowledge graph is licensed under CC BY-SA 4.0.Zenodo provides DOI 10.5281/zenodo.2630447; MiBAC commits to maintaining and evolving ArCo, with CNR support and ICCD training.Source: §6 · p. 10
7 Cultural Heritage and Knowledge Graphs: related work
ArCo expands cultural-heritage linked open data with Italian data and ontology models for issues that existing vocabularies do not sufficiently cover. Its terminological coverage exceeds that of CIDOC-CRM and EDM against ArCo’s competency questions.
7 Cultural Heritage and Knowledge Graphs: related work
- ArCo enriches the cultural-heritage linked-data landscape with Italian data and ontology models for overlooked issues.The paper situates it alongside resources including CIDOC-CRM, EDM, the Rijksmuseum collection, and the Getty Vocabularies.Source: §7 · p. 10
- CIDOC-CRM and EDM did not meet ArCo’s requirements, including modeling paleopathology diagnosis, coin issuance, and musical-instrument classification.Source: §7 · p. 10
- Coverage was computed by matching a 66-term vocabulary extracted from ArCo’s competency questions against the tested ontologies.The study used Sketch Engine for keyword extraction and Silk’s substring metric with a 0.5 threshold for matching.Source: §7 · p. 10
- ArCo’s terminological coverage is 0.68, versus 0.29 for CIDOC-CRM and 0.12 for EDM.Coverage was measured against ArCo’s competency questions.Source: §7 · pp. 10–11
- ArCo aligns with CIDOC-CRM and EDM, and directly reuses Cultural-ON and OntoPiA, among other vocabularies.It also aligns with BIBFRAME, FRBR, FaBiO, FEntry, and OAEntry.Source: §7 · p. 11
8 Conclusion and ongoing work
The paper presents ArCo as a robust linked-data knowledge graph for Italian Cultural Heritage that remains under development. Ongoing work addresses identifier ambiguity, unmodeled heritage details, translations, and supporting tools.
8 Conclusion and ongoing work
- ArCo is presented as a robust Italian Cultural Heritage knowledge graph published as linked open data, but still open to improvement and enrichment.Source: §8 · p. 11
- The URI strategy may assign the same identifier to distinct entities, although the authors expect few ambiguous cases relative to dataset size.They are applying heuristics and expert validation to identify cases, while exploring external-linking methods.Source: §8 · p. 11
- Some naturalistic-heritage characteristics remain unmodeled, and available pictures of cultural entities are not yet included in the dataset.Source: §8 · p. 11
- ArCo’s data are in Italian; 17,906,639 entities have English labels, and completing English translation is a first step.The authors are considering automatic translation to support the expert translation work.Source: §8 · p. 11
- Planned work includes additional tools for cultural-heritage data owners and addressing requirements from library and archive domains.Source: §8 · p. 11