ESCO: Towards a Semantic Web for the European Labor Market Johan De Smedt Martin le Vrang* Agis Papantoniou Havenkant 38 Joseph II straat 27 Havenkant 38 3000 Leuven, Belgium 1049 Brussels, Belgium 3000 Leuven Belgium +32 477 47 59 34 +32 22 96 20 47 +32 479 76 09 62 j.de-smedt@computer.org martin.le-vrang@ec.europa.eu apapanto@central.ntua.gr And indeed, we saw many applications sprouting that are built on ABSTRACT Linked Data: Maps with the live position of underground trains2, The Semantic Web has a huge potential when it is used to Crime heat maps3, websites showing you how your government organize market processes. The labour market is an excellent spends your taxes4 and apps that guide you to the nearest public example where it can add value. By enhancing communication toilet5. While all these applications are useful and qualify as between employers and job seekers in the digital age, market wonderful things, we believe Linked Data will unleash its full processes become more efficient and more people can find the potential when it is used to organize market processes. If we can right jobs for them. The multilingual classification of European use Linked Data to match supply and demand on markets more Skills, Competences, Qualifications and Occupations (ESCO) is a efficiently, the economic impact will be huge. central building block for an ecosystem of semantic assets on the labour market. In this paper we explain how the ESCO data model While the Semantic Web has a great potential on the markets for was designed, building on the Simple Knowledge Organization goods and services, it might benefit the labour market even more. System. We further explain how new versions of the datasets are The labour market has some characteristics that make it a good published and how they can be used by applications. Finally, we case for the Semantic Web to add value. This market is quite summarize the next steps in the on-going work on ESCO. different from going to the supermarket and buying potatoes. If you need potatoes to cook dinner, you do not need to describe Categories and Subject Descriptors each potato individually. And even if you need specific ones, it is pretty straightforward. By naming the variety and the size you H.3.1 [Content Analysis and Indexing], H.3.3 [Information have described what you want. The labour market is different. Search and Retrieval], I.2.4 [Knowledge Representation People on the labour market need to describe all offers Formalisms and Methods], H.3.5 [Online Information individually. No person is like another one and almost no job is systems], J.1 [Administrative Data Processing]. like another one. And it is quite challenging to describe jobs and job seekers, so that in the end you can find the right person for the General Terms right job (or the other way round). Unlike with the potatoes, the Design, Economics, Standardization, Languages most important characteristics you need to describe are intangible. How do you explain the company culture and working Keywords atmosphere? The communication skills of a person and what role ESCO, Job Labour Market, Linked Open Data, Open Standards, he/she can play best in a team? This is where the semantics come Knowledge Organization System, RDF, DCAT in. Communication on the labour market means that the potential Copyright is held by the author/owner(s). employer and the candidates need to develop a shared WWW2015 Workshop: Linked Data on the Web (LDOW2015). understanding of these intangible characteristics. LDOW 2015, May 19, 2015, Florence, Italy. At the same time, it is very important to get it right, for both, the 1. POTENTIAL OF THE SEMANTIC WEB employer and the job seeker. Once the decision for a transaction has been taken and an employment contract has been signed, FOR THE LABOR MARKET correcting a wrong decision is very costly for both sides. When Tim Berners-Lee explained the potential of Linked Data at In the past ten to twenty years the Web changed how these TED, he predicted that “it will be used by other people to do processes are organized on the labour market. Exchange between wonderful things, in ways that they never could have imagined.”1 supply and demand on the labour market is now mainly organized * The views expressed in this article are purely those of the author 2 and may not in any circumstances be regarded as stating an http://traintimes.org.uk/map/tube/ official position of the European Commission. 3 http://crimeview.psi.enakting.org/ 1 Tim Berners-Lee: The year open data went worldwide. 4 http://www.publicspending.gr/ https://www.ted.com/talks/tim_berners_lee_the_year_open_ 5 data_went_worldwide/transcript?language=en http://greatbritishpublictoiletmap.rca.ac.uk/ through electronic means. Recruitment is done via web based In order to describe concepts, ESCO established a specific data tools: Internet portals and company websites have become the two model8, which expresses sets of types, predicates and their most important recruitment channels. Social media is still low in meaning. The model is based on the popular Simple Knowledge market share, but gaining quickly. At the same time, computing Organization System (SKOS) [2] ontology which is used for power increased so that information technology is prepared to representing knowledge organization systems, like thesauri, process large amounts of data. It is no longer a limiting factor in taxonomies and classification schemes9. SKOS offers a way to job matching or job search over a large number of data records. describe and organize concepts (through hierarchical relationships, associative relationships and collections). By The digitization of the labour market also influences the way defining sets of predicates, the concepts can be enriched with people search for information. With the rise of social media a metadata. large amount of data about individuals, i.e. potential employees, becomes accessible to employers. This changes the traditional, The ESCO classification is an extension of the SKOS ontology. prevalent recruitment model, in which employers explain their Table 1 provides an overview of the different meta layers of the requirements and wait for the right candidate to knock on their ESCO data model. door. Employers are increasingly looking for candidates themselves: Active sourcing becomes an important recruitment ESCO concepts are subclasses of SKOS concepts, with some strategy. This requires employers to understand the skills profiles additional metadata properties allowing to structure the ESCO of potential candidates and use this information in their search. pillars. The esco:GroupConcept class gives structure to the top level of an ESCO pillar, and is not used to tag any document. In It is still the personal contact and the job interview that counts for contrast, the esco:MemberConcept is the class of concepts used to the actual recruitment decision, but all the searching and filtering tag CVs, job postings and similar documents. Each of the three process is being done by electronic tools. In this first step, both, ESCO pillars has its own specific concept type: candidates and recruiters narrow down their search from “esco:Occupation”, “esco:Skill” and “esco:Qualification”. These thousands or millions of offers to a few dozen. Modern concept types are subclasses of the ESCO concepts. The actual information technology can help to make a good selection when content of ESCO consists of instances of these specific concept narrowing down the search, but for that we need to move away types. The occupation “flight attendant” is an example for an from describing job vacancies and job seekers’ profiles as instance of the “esco:Occupation” concept. Table 2 provides documents. The offers on the labour market need to be described examples of concepts covered by ESCO. as data. And it needs to be done in a meaningful way that creates a common understanding and not confusion. Finally, each of the more than 10 000 concepts defined by ESCO comes with a group of terms that can be used to refer to the The Semantic Web can help us to overcome this communication concept. These cover all of the 24 languages of ESCO, adding up challenge on the labour market. As Tim Berners Lee put it, the to a total of more than 200 000 terms. For each language ESCO “Web does not just connect machines, it connects people.”6 And defines a preferred term that should preferably be used when on the labour market that is what we need in order to revolutionize referring to the concept (skosxl:prefLabel), but also other terms, market processes, to make them more efficient and finally, to such as synonyms, abbreviations or gender-specific names of the improve peoples’ lives. This is the Semantic Web at its best. occupation (“skosxl:altLabel” and “skosxl:hiddenLabel”). Table 3 shows examples of terms for the ESCO concept “flight attendant”. 2. AN ECOSYSTEM OF VOCABULARIES Table 1 - Meta layers of the ESCO data model Describing offers on the labour market requires complex Meta Example of a concept Examples of sub classes semantics. In order to explain the characteristics of a job offer, layer class (or subclasses) (M4 and M3) and employers need to convey a whole array of information. They and an instance usage instances of this concept need to describe the work environment, the products and services M4 SKOS concept esco:Concept, they work with, the specific tasks of the future employee, the esco:GroupConcept, knowledge, skills and competences that are needed to perform esco:MemberConcept these tasks, salary, characteristics of the employment contract, the M3 esco:GroupConcept esco:Occupation; location of the company, etc. To describe all this in a machine esco:Skill; readable way, we need to use vocabularies covering different esco:Qualification domains. This is what we call the "ecosystem of vocabularies". M2 esco:Occupation “Flight attendant”; The heart of this system is the multilingual classification of “Baker”; European Skills, Competences, Qualifications and Occupations “Veterinary nurse” (ESCO). It covers three different domains – the three “pillars” of M1 “Flight attendant” Use of the occupation ESCO: i) occupations, ii) knowledge, skills and competences, and concept to describe a iii) qualifications. [1] Each of the three pillars contains concepts specific job offer and terms that help to describe offers on the labour market in a M0 Real life case - standardized and machine readable manner using Linked Open (description of a job Data (LOD)7. seeker or job offer) 6 http://webfoundation.org/about/community/knight-2008-tbl- 8 speech/ http://data.europa.eu/esco/model 7 9 http://www.w3.org/TR/ld-bp/#MACHINE http://en.wikipedia.org/wiki/Knowledge_Organization_Systems Table 2 - Examples of concept instances covered in ESCO ESCO concept type esco:Occupation esco:Skill esco:Qualification Number of concept 5 380 5 737 7 examples instances in the ESCO pilot version (ESCO v0) Examples  Kitchen assistant  Image editing  Project Management  Opera singer  Garden landscaping Professional (PMP)  Web designer  Gas welding  European Certificate of Floral  Tannery worker  Vehicle damage inspection Design  Tax auditor  Family law  European/ international  Editor in chief  Hydrology Welding Practitioner  Jeweller  Cope with pressure  Cisco Data Centre  Security guard  Persuade, Collaborate on Application Services Design  Sheep breeder tasks Specialist  Waiter/waitress semantic assets of different domains. When describing job offers Table 3 – Examples for terms in the different language or job seekers on the labour market, concepts of ESCO can variants of the ESCO concept “flight attendant” therefore easily be combined with information from other Label type Language Label vocabularies, taxonomies and classifications. ESCO makes use of existing semantic assets in three ways: skosxl:prefLabel English Flight attendant 1. Mapping to semantic assets of the same domain: Several skosxl:altLabel English Steward national classification systems describe occupations and/or knowledge, skills and competences. Usually they cover the labour skosxl:altLabel English Stewardess market of one specific country10. ESCO provides a reference skosxl:altLabel English Air cabin crew member vocabulary that allows creating machine-readable correspondence tables (“mappings”) between ESCO concepts and the concepts of skosxl:prefLabel German Flugbegleiter/in these classifications. By using mappings of the type skos:closeMatch or skos:exactMatch to ESCO, partners can refer skosxl:altLabel German Steward/ess to ESCO when exchanging information. This way they provide skosxl:prefLabel Greek Αεροσυνοδός meaningful information, even if they use different classification systems on a national level. They achieve semantic … … … interoperability. ESCO occupations of type esco:MemberConcept also map to the International Standard Classification of ESCO is organizing the concepts through hierarchical and Occupations (ISCO)11 versions of 1988 and 2008, a statistical associative relations. These relations are established between classification maintained by the International Labor Organization concepts within one ESCO pillar (e.g. occupation – occupation), (ILO). To this end they use a sub property of skos:broadMatch but also between the different ESCO pillars (e.g. occupation – (esco:memberOfISCOGroup). skill). Within one pillar, ESCO uses hierarchical relations of the type broader (BT) and narrower (NT). By connecting the higher 2. Tagging to semantic assets of a different domain: ESCO tags its level concepts (“esco:GroupConcept”) to member concepts concepts with semantic assets, that are covering different (“esco:MemberConcept”), ESCO builds a hierarchical structure of domains, but that can be used as metadata for ESCO concepts. occupations and of skills and competences. Across the pillars These tagging relations are all specialisations of the Dublin Core ESCO expresses how the different concepts interrelate. Relations Metadata Initiative (DCMI) “dct:subject” property. By tagging between the occupations and the skills pillar show for instance, ESCO concepts with other classifications and frameworks ESCO which knowledge, skills and competences employers typically indicates for example: request when recruiting for a specific occupation. All these relations are specialisations of “skos:related”. Their meaning is specified by using a specializing property or using an instance of esco:Relationship (a detailed relation type specification). 10 For example: “Répertoire opérationnel des métiers et des Next to the concepts of the three ESCO pillars, ESCO makes use emplois (ROME)” in France: http://www.pole- of other “semantic assets”. These semantic assets can be emploi.fr/candidat/le-code-rome-et-les-fiches-metiers- controlled vocabularies, frameworks, classifications, etc. Some of @/suarticle.jspz?id=15734; “Klassifikation der Berufe” in these concepts are created in the context of ESCO, others are Germany: https://statistik.arbeitsagentur.de/Navigation/ reused from existing taxonomies. Statistik/Grundlagen/Klassifikation-der-Berufe/KldB2010/ KldB2010-Nav.html; “Berufssystematik” of the While ESCO describes occupations, knowledge, skills, “Arbeitsmarketservice Österreich (AMS)” in Austria: competences and qualifications, it also links to other semantic http://www.ams.at/bis/OnlineHilfe.php?noteid=38. assets. This allows building on existing work and relating to 11 http://www.ilo.org/public/english/bureau/stat/isco/  The level of an ESCO qualification by tagging it with a actors in the labor market as well as in the education and training level from the European Qualifications Framework sectors. (EQF)12. The EQF makes levels of qualification systems The actual content of the ESCO classification along with its comparable across Europe. supporting taxonomies has been developed by domain experts,  The field of study of an ESCO qualification by tagging which are called the ESCO Reference Groups. These groups bring it with the Fields of Education and Training (FoET)13. together experts from different economic sectors and include  The relevance of an occupation, knowledge, skill or employers, education and training providers, trade union competence for one or more sectors of economic representatives, job recruiters, and sector skills council members. activity by tagging it with the Statistical Classification Each Reference Group is supported by taxonomy creation experts. of Economic Activities in the European Community (NACE) , Rev.2 of 200814. The Reference Group’s work is complemented by that of the ESCO Cross-sector Reference Group. This group develops a 3. Using existing semantic assets to create new ESCO concepts: vocabulary for transversal skills and competences, which are Classifications or frameworks covering other domains can be subsequently applied by the Reference Groups to transferable useful when defining new ESCO concepts, either by importing elements of occupational profiles. them or by using them in facets. An example for this is the implementation of language skills in ESCO. The concepts The remaining editorial work on the ESCO classification will be covering foreign language skills in ESCO make use of two facets developed through an "online consultation process" (cf. section that both build on existing semantic assets. The Common 5.2). European Framework of Reference for Languages: Learning, Teaching, Assessment (CEFR)15 includes different dimensions of 3. PUBLISHING ESCO AS LINKED OPEN learning a language (reading, listening, speaking, etc.). These are DATA added as a facet “language usage” to the “foreign language” concept in ESCO. At the same time, the concept can be combined In order for ESCO to become a de facto standard and an integral with a list of languages, that partly stems from the list of part of an emerging Semantic Web for the labour market, the languages maintained by the Publication Office of the European classification needs to be published in a user friendly way. In Union16 and to the library of congress17 (by using particular, ESCO needs to: “{skos:exactMatch}”) and that is mapped to codes of the ISO 639  Allow for an easy integration into existing IT systems; standard (by using “{dcterms:identifier}”).  Be easy to link to other semantic assets so that it can By building relations between concepts of different domains and fulfil its role as a reference vocabulary for data by linking ESCO to other semantic assets, an “ecosystem of exchange; controlled vocabularies” for the labour market emerges. This is  Ensure that content is well managed and quality assured shown in Figure 1. before publication;  Ensure that continuous updates of the classification do In addition to ESCO, other semantic assets can be used to not lead to high administrative overhead for ESCO annotate job vacancies or CVs. The Nomenclature of Territorial users. This requires a versioning mechanism that Units for Statistics (NUTS)18 can for example be used to specify ensures backwards compatibility, persistent uniform the location of an employer in a job vacancy. They are, however, resource identifiers (URIs) and transparent and not directly related to the ESCO classification. machine-readable information about the different ESCO 2.1 Collaborative content creation versions. The European Commission is developing ESCO together with In order to achieve this, the ESCO platform is built to service a stakeholders like employment services, employer federations, network of Knowledge Organization System (KOS) for trade unions, and professional associations. This ensures that the publishing Open Data according to the Linked Data methodology expertise of these various and important organizations feeds into [3]. Its main components are the thesaurus management system, the project and that the end result will meet the needs of key data ingestion, Application Programming Interfaces (APIs)19 , and indexing services. Figure 2 illustrates these components. The European Commission manages the content of the ESCO 12 http://ec.europa.eu/ploteus/search/ classification in a thesaurus management system. Once the site?f[0]=im_field_entity_type%3A97# European Commission decides to release a new version of ESCO, the publication process starts. The first step is to export the new 13 http://www.uis.unesco.org/Education/Documents/ ESCO version from the taxonomy management system. Through isced-fields-of-education-training-2013.pdf an API service, the system generates and assigns persistent URIs. 14 All entities (i.e. instances of data sets, concept schema, concepts, http://data.europa.eu/esco/ConceptScheme/NACErev2/cs collections, terms, relations …) receive their own URI that serves http://ec.europa.eu/eurostat/web/nace-rev2 as permanent identifier. URIs are maintained throughout all 15 https://europass.cedefop.europa.eu/en/resources/european- versions of the thesaurus, even if concepts become obsolete or are language-levels-cefr asp replaced with other concepts in a later version. 16 http://publications.europa.eu/resource/authority/language 17 http://id.loc.gov/vocabulary/iso639-1 18 19 http://ec.europa.eu/eurostat/web/nuts/overview https://ec.europa.eu/esco/web/guest/escopedia/- http://data.europa.eu/esco/ConceptScheme/NUTS2008/cs /escopedia/ESCO_API When a new ESCO dataset version is exported from the taxonomy Master dataset of the new ESCO version; management system, it is registered in an internal Data Catalog Vocabulary (DCAT)20 register. The taxonomy management  Publication of this dataset in the public DCAT register system provides the dataset with a version record referenced by an (extended with the Dataset Versioning Ontology) which object property (ds-vers:hasVersionRecord) defined by the is available as LOD; Dataset Versioning21 ontology and the DCMI property referring to  Indexed HTML pages for each concept that are being the Version History Set (dct:isVersionOf). published in the ESCO portal23.  ESCO data in an RDF store serving a SPARQL The export is followed by an ingest process. During this process endpoint. the ingest component receives the exported dataset, validates it  JSON-LD page for every ESCO entity. and applies the rules defined in the ESCO data model, such as  An XML report detailing the difference between the inferences, faceting and inheritance. The ingest component then previous published version and the new version. generates a Resource Description Framework (RDF) version of ESCO that can be machine processed at different levels: SKOS, Thanks to this process, the ESCO platform can provide any type SKOS-XL, ISO-THES22 and ESCO. The ESCO pillar concepts of ESCO data in RDF, HTML, TTL and JSON-LD formats. receive associative relationships between them, are further mapped to supporting taxonomies and finally they are tagged, as explained in section 2. As a result of the ingest process the following datasets are produced: 20 http://www.w3.org/TR/vocab-dcat/ 21 http://purl.org/iso25964/DataSet/Versioning/#objectproperties 22 http://purl.org/iso25964/skos-thes 23 http://www.niso.org/schemas/iso25964/ https://ec.europa.eu/esco/home Figure 1 - Examples of semantic assets in the ecosystem of controlled vocabularies for the labor market Figure 2 - ESCO Architecture Snapshot of the Linked Open Data Publication Chain The ingest process also generates a file that includes the indexed several sources flourish, while the major search engines deliver content (SOLR24) of all JSON-LD files of a specific version of poor results in job search. The use of a standardized terminology the ESCO dataset, together with its master RDF. This “war” file to annotate these documents is one step towards a semantic web is available as a DCAT distribution of the ESCO dataset. The for the labour market. Coupled with schemata or syntactic file allows users to build local applications using any specific standards, such as schema.org or HR-XML, information can be version of ESCO. Each user can take a decision based on its easily exchanged between partners. own needs, when to upgrade to the latest version of ESCO. Once information on job seekers and job vacancies is published Except for languages of the Library of Congress, none of the or exchanged as data, it can be used in specialized services and semantic assets used by ESCO (cf. section 2) are currently apps. Competence-based online job matching algorithms can for published as LOD25. Even though for some of them LOD instance help to find the best match between jobseeker and job, publishing efforts are on-going, the publishing strategies for based on the individual professional profile of the jobseeker and these semantic assets vary and range mainly from PDF, Excel the employers’ needs. Imagine that the information on the and CSV files to downloadable RDF. In order to be able to link several million vacant jobs in Europe is semantically annotated to these classifications within the ESCO ecosystem, the ESCO and can automatically be compared with the professional platform transformed these semantic assets into SKOS format, profiles of 26 million unemployed people in order to find the assigned an ESCO URI to each entity and published a LOD best match. copy of the semantic asset. In cases where a taxonomy is already available in SKOS, ESCO uses “skos:exactMatch”. ESCO also With its sophisticated vocabulary of knowledge, skills and provides a mechanism to manage versions of these semantic competences, ESCO will support exactly that. ESCO will enable assets. Their copy in the ESCO ecosystem inherits the applications that allow employers to describe their requirements versioning functionality. These are explicitly detailed using precisely and in a machine-readable format. On the other side of “ds-vers:VersionHistorySet”. There is one instance per concept the labour market, it helps to understand the experience of scheme, holding one “ds-vers:VersionHistoryRecord” for each jobseekers: the knowledge, skills and competences they obtained published dataset version (see earlier in this section). in formal learning, but also “on the job”. By putting the knowledge, skills and competences in the centre, ESCO will 4. ADDED VALUE OF ESCO help to increase opportunities for jobseekers without formal qualifications. It will also make visible, which knowledge, skills As a result of the publishing process ESCO is available in LOD and competences are transferable to a different job context and and free to use for everyone. For ESCO to become a de facto thus promote professional mobility on the labour market. standard it is crucial to maintain the classification over time, so that it keeps pace with new developments on the labour market ESCO will allow building better automated search and matching or in the education and training sector. At the same time, the services for the labour market. Jobseekers can use these as self- versioning mechanism of ESCO ensures backward and forward services to find the best job themselves. At the same time, this compatibility. This makes it a perfect tool for the semantic will free up resources of public employment services, to focus annotation of labour market documents, such as job vacancies, on the jobseekers that really need the assistance of human curricula vitae or training offers. By annotating these documents advisers. with ESCO, the information becomes language independent and However, the possibilities of the semantic web do not stop with machine readable. The annotation transforms the documents into job matching and searching. By crossing domains, we can go structured data. And thanks to the metadata, descriptions and even further than just finding the best match on the labour language variants of ESCO concepts it is easier for market market. We can start to analyse skill gaps, both on the level of participants to create a shared understanding about the intangible the economy and on an individual level. In labour market characteristics of job offers and job seekers. intelligence, the ESCO classification can help to collect and This opens new avenues for cooperation on the labour market. interpret information on skills supply and skills demand, e.g. in Since the use of ESCO creates a shared understanding about job real time big data analyses. This can help to identify emerging offers and job seekers, partners on the labour market can skills needs as well as skills gaps and surpluses across Europe. exchange information across different languages and On an individual level, e-Services can analyse the professional information systems. New partnerships on the labour market profile of job seekers and try to identify what skills they are will form, simply because it is easier to share information. Over lacking to succeed on the labour market or to progress in their time, this will create pressure to tear down isolated data silos career. They can suggest possible career paths to individuals. and open up the information on vacant jobs. Service providers can use this information to provide targeted information on trainings offers that would really pay off for a At the same time, the machine readability of structured data job seeker. This way, electronic tools can provide information allows for new services. While companies and employment and feedback that helps jobseekers to increase their services publish millions of job vacancies, searching this vast employability. information is still difficult. Faceted or semantic search of job vacancies is usually possible within one integrated database, but 5. NEXT STEPS not across all the job vacancies published on the Internet. That is why aggregators, such as Indeed.com, who integrate data from As these examples of applications building on ESCO demonstrate, the ESCO classification is not developed as a standalone tool. It needs to be used in applications that serve jobseekers, employers or labour market professionals. In the 24 http://lucene.apache.org/solr/ framework of EURES, the network of public employment 25 services, the European Commission proposed a revamped http://www.w3.org/TR/ld-bp/#MACHINE interoperability system that allows for a better cross-border (Direct URI resolution is missing in most cases) cooperation of public employment services. 26 For ESCO to best support these applications, the on-going work is focusing on the 7. REFERENCES following steps: [1] Le Vrang, M., Papantoniou, A., Pauwels, E., Fannes, P., 5.1 Assessment of ESCO in "fit-for- Vandensteen, D., and De Smedt, J. 2014. ESCO: Boosting purpose" applications. Job Matching in Europe with Semantic Interoperability, Computer, vol. 47, no. 10, pp. 57–64. The current version of ESCO needs to demonstrate its added value in practical applications. From pilot and test applications [2] Miles, A., and Bechhofer, S., 2009. SKOS Simple we can learn if ESCO is "fit-for-purpose" and how it can be Knowledge Organization System Reference, World Wide improved further. The implementation of the first ESCO APIs as Web Consortium (W3C) recommendation, central and local service supports building such applications. www.w3.org/TR/skos-reference. Examples include an undergoing pilot application within the [3] Bizer, C., Heath, T., and Berners-Lee, T. 2009. Linked scope of the Interoperability Solutions for European Public Data—The Story So Far, Int’l J. Semantic Web and Administrations (ISA)27 initiative and a pilot project with four Information Systems, vol. 5, no. 3, pp. 1–22. European public employment services28. [4] European Commission, 2013. ESCO: European 5.2 Finalising the full content of ESCO Classification of Skills/Competences, Qualifications and While a pilot version of ESCO (ESCO v0) was published in Occupations, http://bookshop.europa.eu/en/esco-european- October 2013 [4], the full content of ESCO is still under classification-of-skills-competences-qualifications-and- development. A large part of the content has been finalized occupations-pbKE0313496 through editorial work supported by field experts. Some demo [5] Brabham, D.C., 2013. Crowdsourcing, MIT Press releases of the finalized parts of the refined classification are planned for 2015. The European Commission aims to complete the remaining content of the classification until end of 2016. Stakeholders will contribute to this process through an “open consultation”, similar to a "crowd sourcing" [5] approach. 5.3 Continuous update of the classification Once the full version of ESCO is published, it needs to be continuously updated to keep pace with new developments on the labour market and in the education and training system. Emerging occupations, changing skill needs and new qualifications need to be reflected in ESCO so that it can fulfil its role as reference terminology for the European labour market. ESCO will need to put in place a system that allows to further develop the content in line with changing realities, but also a continuous development process. Such a system could combine different methodologies for updating content, such as machine learning, crowdsourcing and editorial work carried out by experts. 6. ACKNOWLEDGEMENTS The authors would like to thank Konstantin Kharlov for his support during the preparation of the paper, the TenForce technical team and the ESCO Secretariat team for their ongoing effort in the context of the ESCO project. 26 Procedure 2014/0002/COD; COM (2014) 6: Proposal for a Regulation of the European Parliament and of the Council on a European network of Employment Services, workers' access to mobility services and the further integration of labour markets; http://eur-lex.europa.eu/procedure/EN/1041603. 27 https://joinup.ec.europa.eu/ 28 https://ec.europa.eu/esco/escopedia/- /escopedia/ESCO_mapping_pilot