Classification of Knowledge Organization Systems with Wikidata

Abstract

This paper presents a crowd-sourced classification of knowledge organization systems based on open knowledge base Wikidata. The focus is less on the current result in its rather preliminary form but on the environment and process of categorization in Wikidata and the extraction of KOS from the collaborative database. Benefits and disadvantages are summarized and discussed for application to knowledge organization of other subject areas with Wikidata.

Keywords

• Knowledge Organization Systems
• Wikidata

-

Wikidata2 is the most recent project of the Wikimedia Foundation. In short, it is a collaboratively edited, free knowledge database that can be read and edited by both humans and machines. A good overview is given by two of Wikidata’s main creators Vrandečić and Krötzsch [ 16 ]. The primary goals of the project are: 1. Centralize links between Wikipedia language editions and other Wikimedia project sites. For instance all Wikipedia articles about “encyclopedia” (in any language) are linked to one Wikidata item with identifier Q5292. These so called sitelinks and other data about the concept known as “encyclopedia” can be looked up at https://www.wikidata.org/wiki/Q5292.3 2. Centralize Infoboxes. More and more manually edited infoboxes (tables with basic, factual information about a topic) are being extended to use Wikidata as their database backend, so the displayed information will be the same in all Wikipedia editions. 3. Provide an interface for rich queries. The content of Wikidata can be queried via a public SPARQL interface at https://query.wikidata.org/ (see figure 1 for an example). Query results are planned to be integrated into Wikipedia and other projects as lists, tables, maps and other forms.

The data model of Wikidata is neither relational nor based on RDF (although mappings to RDF exists) but it reflects the strategy of Wikidata to store statements instead of facts. Each statement should be sourced by references and contradicting statements are not forbidden on purpose. Statements can further be controlled by qualifiers , such as domain and date of validity, eventually supporting n-ary relations between Wikidata items. The Wikidata ontology consists of Wikidata properties, which are defined by community consensus. For instance P571 identifies the property “inception” to state the date when something was created or founded. Labels and scope notes can be edited independently from statements with support of synonyms and homonyms. 3

Knowledge Organization Systems in Wikidata To a large degree, Wikidata contains mappings to other KOSs: links between Wikidata items to VIAF and Geonames are among the most used Wikidata properties with around 100.000 Wikidata each4 and at least 1051 of 2490 Wikidata properties5 refer to external identifier systems of other databases. 6 With 2 https://www.wikidata.org/ 3 In RDF this URL corresponds to the URI http://www.wikidata.org/entity/Q5292. 4 https://www.wikidata.org/wiki/Wikidata:Database_reports/List_of_properties 5 Measured at https://www.wikidata.org/wiki/Special:ListProperties, 2016-05-24 6 See also Voß et al. [ 15 ] for a German introduction to authority data in Wikidata. uniquely identified records about virtually anything, Wikidata can also be seen as a KOS in its own right, so it is included in BARTOC.7 For the scope of this study, Wikidata items about KOS types and KOS instances are of special interest. KOS instances are Wikidata items linked to KOS type items with property “instance of” (P31)“. For instance the Dewey Decimal Classification ( Q48460) is an instance of both a”library classification” ( Q48473) and a “universal classiifcation scheme” ( Q24243801). Both are connected to the item “classification scheme” (Q5962346) with property “subclass of” (P279).

Despite their obvious use for knowledge representation and classification, the subclass and instance properties have no special role in Wikidata. Instead they can freely be used to connect any Wikidata items, only constrained by human intervention of other editors and, hopefully, community consensus in common sense. The lack of stricter rules on use of subclass and instance properties in Wikidata has led to criticism among researchers that try to use it as a formal ontology [ 1, 12 ]. Nevertheless, Wikidata can successfully be queried for a hierarchical list of transitive subclasses of “knowledge organization system” (Q6423319) and additional numbers for each class (figure 1).

SELECT ?item ?itemLabel ?broader ?parents ?size ?sites { { # number of additional superclasses SELECT ?item (count(distinct ?parent)-1 as ?parents) { ?item wdt:P279* wd:Q6423319

OPTIONAL { ?item wdt:P279 ?parent } } GROUP BY ?item } { # number of instances SELECT ?item (count(distinct ?element) as ?size) { ?item wdt:P279* wd:Q6423319 OPTIONAL { ?element wdt:P31 ?item } } GROUP BY ?item } { # number of sitelinks SELECT ?item (count(distinct ?site) as ?sites) { ?item wdt:P279* wd:Q6423319

OPTIONAL { ?site schema:about ?item } } GROUP BY ?item } } OPTIONAL { ?item wdt:P279 ?broader } SERVICE wikibase:label { bd:serviceParam wikibase:language "en" } The following classification of KOS types was extracted from SPARQL (figure Wikidata via 1). The command line tool ‘wdtaxonomy’, created especially to extract and display taxonomies, proved to be hugely useful for analysis [ 14 ]. The outcome, at the time of its creation at August 23rd 2016, contains 189 Wikidata items of KOS types, grouped in a multi-hierarchy. Classes within one level are sorted by their Wikidata identifier, reflecting the relative time when they were added to the database. Items with multiple superclasses within the KOS type hierarchy are shown in italics at the their second occurrence, for instance a ‘plant taxonomy’ is both a ‘biological classification’ and a ‘taxonomy’. Superclasses from other parts of Wikidata are indicated with upwards pointing arrows (↑), for instance ‘datamodel’ is also subclass of ‘model’ and ‘data set’. The numbers right to each class label indicate the current number of instances in Wikidata (if there are any), and after a small plus sign the number of Wikipedia articles or other Wikimedia projects sites linked with the entry (sitelinks). For instance three Wikidata items are marked as instance of ‘semantic network‘ and twenty Wikipedia editions include an article about semantic networks (3+20). knowledge organization system ↑↑↑ · encyclopedia ↑ · · internet encyclopedia ↑ · · encyclopedic dictionary · · · biographical encyclopedia ↑ · · universal encyclopedia · · hypertext encyclopedia · dictionary ↑ · · lexicon ↑ · · lexicographic thesaurus · · orthographic dictionary · · etymological dictionary · · glossary ↑ · · visual dictionary · · explanatory dictionary · · · explanatory combinatorial dictionary · · · monolingual learner’s dictionary · · encyclopedic dictionary · · reverse dictionary · · machine-readable dictionary · · · online dictionary · · · · Wiktionary language edition ↑↑ · · rime dictionary · · rhyming dictionary · · conceptual dictionary · · bilingual dictionary · · slang dictionary · · idiom dictionary · · Anagram dictionary · · author dictionary · · language for specific purposes dictionary · · medical dictionary · · phonetic dictionary · · picture dictionary · · single-field dictionary · · specialized dictionary · · · sub-field dictionary

2+3 299+183 29+27 42+7 59+8 1+2 6+1 182+126 4+38 7+48

+7 1+13 13+35 1+6 4+6 +2 +2 · · · · zoological nomenclature 2+2 · · library classification 26+22 · · decimal classification 6+1 · · faceted classification 1+11 · · universal classification scheme 12 · · specialized classification scheme 10 · · · biological classification 10+112 · · · · plant taxonomy · · · · syntaxonomy +9 · · · · taxonomy of Chordata +1 · · · · · taxonomic classification of Passeriformes +1 · · · · taxonomy of birds 1+2 · · · · · taxonomic classification of Passeriformes · · · linguistic typology 3+42 · · · climate classification 1+14 · · · · genetic climate classification 1+1 · · · · efective climate classification 2+1 · · · classification of wine 1+7 · · · chemical classification 3+8 · · · video game content rating system ↑ 8+17 · · · classification in sports 88+3 · · · · age category in soccer +1 · · · · olympic sailing class ↑ 4+6 · · · · equestrian sports classification +1 · · · · weight class 3+7 · · · · · wrestling weight classes +2 · · · · · lightweight class · · · · age category in athletics 5 · · · · competition class 84 · · · · · glider competition class 9+3 · · · · · group 3+4 · · · · · · rally group 1 · · · musical instrument classification 1 +11 · · · medical classification 8+3 · · · corporate taxonomy +1 · · · economic taxonomy 2+2 · · · · job classification system 7+1 · · · · patent classification 4+2 · · · · industry classification 18+3 · · · · product classification +1 · · · military taxonomy +2 · · · · military casualty classification 6 · · · safety taxonomy +1 · · · astronomy classification 1 · · · · stellar classification 25+59 · · · · asteroid classification 3 · · · · · asteroid family 82+25 · · · · · asteroid spectral type 19+22 · · · · · minor-planet group 1+7 · terminology 3+1 · · nomenclature 7+25 · · · biological nomenclature 12+12 · · · · phylogenetic nomenclature +10 · · · · botanical nomenclature · · · · zoological nomenclature · · · nosography 1+3 · · · chemical nomenclature 3+27 · · · · IUPAC chemical nomenclature 3+29

Summary number of classes: with instance: with sitelink: 189 (100%) 116 (61%) 175 (92%)

Discussion The classification of KOS types extracted from Wikidata is detailed but obviously sketchy in its current form. The system was even more incomplete before large parts of it had been edited by the author, mainly to adjust or add missing English labels and items without any instance or subclass statement. A couple of items were also moved or merged after notification and discussion with the Wikidata community.8 Instead of criticizing usage limitations of Wikidata class hierarchies such as Spitz et al. [ 12 ], or suggesting methods to better spot classification inconsistencies such as Brasileiro et al. [ 1 ], peculiarities of knowledge organization systems based on Wikidata shall be highlighted in the following. First of all, classes and instances are more or less given by existence or nonexistence of Wikidata items with sitelinks: only 14 of 189 classes in the classification above don’t have at least one corresponding Wikipedia article. To some degree new Wikidata items can also be added without sitelink, but this is controversial at least for abstract concepts which have no obvious unique identification. 9 KOS types therefore usually require some Wikipedia article before inclusion. The strong connection to Wikipedia also forbids removal, reinterpretation or merging of concepts that don’t easily fit into a classification, unless one engages in editing Wikipedias. Anyway, KOS derived from Wikidata are build by a bottom-up approach from general encyclopedic concepts. As scope and definition of a Wikidata item vary between Wikipedia language editions, the concepts are fuzzy to some degree. For instance at the moment there are two items for classification as process and classification as result but only one item for both metamodelling and metamodel.10 One should not try to solve all of these cases in structured data, as Graham [ 5 ] warned about application of Wikidata: “We just need to ensure that we aren’t seduced into codifying, categorizing, and structuring in cases when we should be describing the inherent messiness of a situation.” It has already been shown that the category system of Wikipedia is more a thesaurus than a classification [ 13 ]. KOS extraction from Wikidata may also result in less strict hierarchies without strong formal logic. Participation in Wikidata difers from collaborative ontology engineering [ 9 ]. Although the database is often referred to as knowledge base, its practical purpose in many ways is more knowledge organization than knowledge representation. The number of sitelinks can be used as indicator how established or widely known a concept is. The number of instances more depends on which KOS instances are relevant for inclusion in Wikimedia projects and have been classified in Wikidata at all. Despite this bias, instances are very helpful to judge the application of a concept for classification. Both new Wikidata instances and new sitelinks are added to KOS types virtually every day. The benefit of this dynamic growth is better coverage of multiple views and domains, so most KOS extracted 8 The question about ‘taxonomy’ vs. ‘biological classification’ could not be resolved. 9 See https://www.wikidata.org/wiki/Wikidata:Notability 10 “classifying” (Q13582682), “classification” ( Q5962346), and “metamodel” (Q1925081) from Wikidata include aspects of a universal classification and less sufer from opinionated knowledge organization. The downside of this crowd-sourcing is lack of a final consensus. Therefore extraction of KOS from Wikidata is an iterative process that requires continuing review and contribution to Wikidata (figure 2).

extraction Wikipedias discussion This paper introduced the extraction of knowledge organization systems from Wikidata exemplified by a classification of KOS types. With 181 classes the result is probably the most detailled classification of KOS types so far. Before further evaluation and cleanup, however, it can only serve as starting point. The strong grounding in Wikipedia is both benefit and challenge. The system is more build bottom-up from instances instead of top-down from theoretical properties like existing classifications and typologies. Further improvement of the particular classification requires involvement of both the Wikidata community and domain experts in knowledge organization. For instance one could express diferent characteristic of division as Wikidata qualifiers. 11 It is also not sure yet whether the hierarchical structures extracted from Wikidata can better be expressed by other types of KOS such as multi-level models [ 1 ] or thesauri [ 13 ]. Further work on the KOS classification includes alignment with the NKOS KOS Types Vocabulary [ 3 ]. KOS instances have been classified in both Wikidata and BARTOC [ 8 ], connected via Wikidata property “BARTOC ID” (P2689). More elaborated KOS extraction should also incorporate Wikidata qualifiers. Existing tools such as SQID Wikidata Browser12 could further be extended to better support management of KOSs extracted from Wikidata. 11 This has already been started for domain-specific subtypes of classification schemes. 12 http://tools.wmflabs.org/sqid/

References

[1] Freddy Brasileiro et al. “ Applying a Multi Level Modeling Theory to Assess Taxonomic Hierarchies in Wikidata” . In: Proceedings of WWW 2016 Companion . 2016 . doi: 10 .1145/2872518.2891117.

[2] Eva Bratková and Helena Kučerová . “Knowledge Organization Systems and Their Typology” . In: Revue of Librarianship 25.2 ( 2014 ), pp. 1 - 25 . url: http://full.nkp.cz/nkkr/knihovna142_suppl/1402sup01.htm.

[3] Dublin Core Metadata Initiative. NKOS Task Group . KOS Types Vocabulary . Oct. 2015 . url: http://wiki.dublincore.org/index.php/NKOS%5C_ Vocabularies.

[4] Alan Gilchrist . “Thesauri, taxonomies and ontologies - an etymological note” . In: Journal of documentation 1 ( 2003 ), pp. 7 - 18 . doi: 10 . 1108 / 00220410310457984.

[5] Mark Graham . “ The Problem With Wikidata” . In: The Atlantic ( 2012 ). url: http : / / www . theatlantic . com / technology / archive / 2012 / 04 / the - problem - with-wikidata/255564/.

[6] Gail Hodge . Systems of knowledge organization for digital libraries: beyond traditional authority files . Digital Library Federation , 2000 .

[7] ISO 25964 Information and documentation - Thesauri and interoperability with other vocabularies . ISO standard . 2013 .

[8] Andreas Ledl and Jakob Voß . “ Describing Knowledge Organization Systems in BARTOC and JSKOS” . In: Proceedings of International Conference on Terminology and Knowledge Engineering . 2016 , pp. 168 - 178 . url: http://hdl.handle.net/10760/29366.

[9] Claudia Müller-Birn et al. “ Peer-production system or collaborative ontology engineering efort: what is Wikidata? ” In: OpenSym. Ed. by Dirk Riehle . ACM, 2015 , 20 : 1 - 20 : 10 . doi: 10 .1145/2788993.2789836.

[10] Vreda Pieterse and Derrick G. Kourie. “Lists, Taxonomies, Lattices, Thesauri and Ontologies: Paving a Pathway Through a Terminological Jungle” . In: Knowledge Organization 41.3 ( 2014 ), pp. 217 - 229 .

[11] Renato Rocha Souza , Douglas Tudhope, and Maurício Barcellos Almeida. The KOS spectra: a tentative typology of Knowledge Organization Systems . 2010 . url: http://mba.eci. ufmg.br/downloads/ISKO%5C%20Rome% 5C%202010%5C%20submitted.pdf.

[12] Andreas Spitz et al. “ State of the Union: A Data Consumer's Perspective on Wikidata and Its Properties for the Classification and Resolution of Entities” . In: Proceedings of ICWSM 2016 . 2016 .

[13] Jakob Voß . Collaborative thesaurus tagging the Wikipedia way . Tech. rep . 2006 . url: http://arxiv.org/abs/cs/0604036.

[14] Jakob Voß . wikidata-taxonomy 0.2.7 . 2016 . doi: 10 . 5281 / zenodo . 60708. url: https://github.com/nichtich/wikidata-taxonomy.

[15] Jakob Voß et al. Normdaten in Wikidata. 2014 . isbn: 978 -1- 291 -85658-3. url: https://hshdb.github.io/normdaten-in-wikidata/.

[16] Denny Vrandečić and Markus Krötzsch . “Wikidata: A Free Collaborative Knowledgebase” . In: Communications of the ACM 57.10 ( 2014 ), pp. 78 - 85 . doi: 10 .1145/2629489.

[17] Marcia Lei Zeng and Gail Hodge . Taxonomy of Knowledge Organization Sources/Systems . 2000 . url: http://nkos.slis.kent.edu/KOS_taxonomy. htm.