Employment Services (ESs) are becoming more and more important for Public Administrations where their social implications on sustainability, workforce mobility and equal opportunities play a fundamental strategic importance for any central or local Government. The EU SEEMP project aims at improving facilitate workers mobility in Europe. Ontologies are used to model descriptions of job offers and curricula; and for facilitating the process of exchanging job offer data and CV data between ES. In this paper we present the methodological approach we followed for reusing existing human resources management standards in the SEEMP project, in order to build a common “language” called Reference Ontology.
Nowadays there is an important amount of investment in human capital for economic
development. Human resources management refers to the effective use of human
resources in order to enhance organisational performance [
Employment Services (ESs) are becoming more and more important for Public Administrations where their social implications on sustainability, workforce mobility and equal opportunities play a fundamental, strategic importance for any central or local Government. The goal of the SEEMP1 (Single European Employment MarketPlace) project is to design and implement an interoperability architecture for public eEmployment services which encompasses cross-governmental business and decisional
processes, interoperability and reconciliation of local professional profiles and
taxonomies, semantically enabled web services for distributed knowledge access and
sharing. The SEEMP project relies on WSMO [
A major bottleneck towards e-Employment applications of Semantic Web technology and machine reasoning is the lack of industry-strength ontologies that go beyond academic prototypes. The design of such ontologies from scratch in a textbook-style ontology engineering process is in many cases unattractive for two reasons. First, it would require significant effort. Second, because the resulting ontologies could not build on top of existing community commitment. Since there are several human resources management standards, our goal is not to design human resources ontologies from scratch, but to reuse the most appropriate ones for public eEmployment services developed on the framework of the SEEMP project. In this paper we present the methodological approach we followed for reusing existing human resources management standards like NACE2 (Statistical Classification of Economic Activities in the European Community), ISCO-88 (COM)2 (International Standard Classification of Occupations, for European Union purposes) and FOET2 (Classification of fields of education and training), among others.
This paper is organized as follows: Section 2 presents some related work. Next section 3 explains the adopted methodological approach to build the SEEMP Reference Ontology from standards/classifications and already existing ontologies, and then in section 4 an overall perspective of the resultant SEEMP Reference Ontology is shown. Then section 5 describes some considerations with respect to the building process of the local ontologies taking as starting point the Reference Ontology and the ES data sources. Finally, section 6 offers some final conclusions, and poses the future work that, among other things, will serve to validate the ideas proposed in this paper. 2
Currently the Human Resource Semantic Web applications are still in an experimental phase, but their potential impact over social, economical and political issues is extremely significant.
COKE is described in [
Regarding main standardization initiatives in the HR sector, the HR-XML
consortium has built up a library of more than 75 interdependent XML schemas
which define the data elements for particular HR transactions, as well as options and
constraints governing the use of those elements [
Finally there is an effort described in [
In this section we describe the adopted approach to build the SEEMP Reference
Ontology; a preliminary version is described in [
4. Ontology evaluation; in this activity we evaluate the ontology content.
This activity states why the ontology is being built, what its intended uses are, and
who the end-users are. For specifying the ontology requirements we used the
competency questions techniques proposed in [
Competency Questions. These questions and their answers are both used to
extract the main concepts and their properties, relations and formal axioms of the
ontology. We have identified sixty competency questions; they are described in
detail in [
In order to choose the most suitable human resources management standards for modeling CVs and job offers, the following aspects have been considered: The degree of coverage of the objects identified in the previous task, this aspect has been evaluated taking into account the scope and size of the standard. However, a too wide coverage may move us further away from the European context; therefore we have tried to find a tradeoff between this aspect and the following one: the current european needs, it is important that standard focuses on the current European reality, because the user partners involved in SEEMP are European, and the out coming prototype will be validated in European scenarios; and the user partners recommendations, in order to asses the quality of the standards, the opinion of the user partners is crucial since they have a deep knowledge of the employment market.
Besides, when choosing the standards, we also took into account that the user partners of SEEMP selected the ICT domain for the prototype to be developed in SEEMP. Hence, the chosen standards should cover the ICT domain with an acceptable degree. In the case of the occupation taxonomy, as it will be shown, we have chosen one standard, but then we have taken some concepts coming from other classifications, in order to obtain a richer classification for the ICT domain.
When specifying job offers and CVs, it is also necessary to refer to general purpose international codes such as country codes, currency codes, etc. For this aim, the chosen codes have been the ISO codes, enriched in some cases with user partners’ classification.
Finally, the representation of job offers and CVs also require temporal concepts such as interval or instant. So, in order to represent these concepts in the final Reference Ontology, the DAML time ontology3 was chosen.
This activity states how we enrich the human resources management standards. In
order to make possible the enrichment of the standards, it was necessary to import
them into the ontology engineering tool WebODE [
Once we transformed the standards into ontologies, the next step is to enrich them
introducing concept attributes and ad hoc relationships between ontology concepts of
the same or different taxonomies. We perform this task, doing the following. We
created from scratch the Job Seeker Ontology (models the job seeker and his/her CV
information), and the Job Offer Ontology (models the job vacancy, job offer and
employer information); following some HR-XML[
The evaluation activity makes a technical judgment of the ontology, of its associated software environments, and of the documentation. We will evaluate the Reference Ontology using the competency questions identified in the first task. 4
The Reference Ontology4, described in this section, will act as a common “language”
in the form of a set of controlled vocabularies to describe the details of a job posting
and the CV of a job seeker. The Reference Ontology was developed following the
process described in detail in section 3 and with the ontology engineering tool
WebODE [
Ten of them were obtained after wrapping the original format of the 4 The Reference Ontology is available at: http://droz.dia.fi.upm.es/seemp/ (Username: seemp and password: employer) • standard/classification, using ad hoc translator or standard/classification.
The connections between the ontologies by means of ad hoc relationships.
EURES ICSOEF6392 IEOSUNCREOET-8S8 COM wrapper for each Skill Ontology
Language
Ontology ishaassscoocnitartaecdt ttyope /
subClass-Of ReLgaubloautorryhas work condition / is associated to Ontology has contract type / is associated with
ishaassswoocriaktceodnwdiitthion / Competence Ontology Compensation
Ontology
EURES requires competence / is associated with
pensation/ haissacsosmociatedwith
LE FOREM ISO 4217
Economic Activity
Ontology NACE Rev. 1.1
Fig. 1. Main ad-hoc relationships between the modular ontologies.
Next we provide the conceptualization of the mentioned ontologies. • Job Seeker Ontology. This ontology models knowledge of job seeker and CV information used in the SEEMP Project. The Job Seeker Ontology imports concepts from the education ontology, language ontology, economic activity ontology, compensation ontology, geography ontology, driving license ontology, labour regulatory ontology and skill ontology, and these imported concepts are used to connect the Job Seeker Ontology with the other ontologies. Error! Reference source not found. shows all the ad hoc relationships whose domain is a concept belonging to the Job Seeker Ontology (concepts of the Job Seeker Ontology are drawn in yellow). Examples of the relationships can be: ‘Job Seeker has driving license Driving License’ (with the Driving License concept from the Driving License Ontology), ‘Job Seeker has education Education’ (with the Education concept from the Education Ontology), ‘Job Seeker has mother tongue Language’ (with the Language concept from the Language Ontology), etc. • Job Offer Ontology. This ontology models knowledge of job vacancy, employer and job offer information used in the SEEMP Project. The Job Offer Ontology imports concepts from the education ontology, language ontology, economic activity ontology, compensation ontology, geography ontology, driving license ontology, labour regulatory ontology and skill ontology, and these imported concepts are used to connect the Job Offer Ontology with the other ontologies.
Language
Competence speaks/isspokenby h a i s s m motheorthtoenrglaunegoufage/ Examples of the relationships can be: ‘Job Vacancy requires driving license Driving License’ (with the Driving License concept from the Driving License Ontology), etc.
Competence Work Condition Compensation has work condition / is associated with has compensation / is associated with req isausisreoscicaotemdpwetenc
ith e/ in/ with islocsastoecdiated isa is located in / is location of belongs to / has job vacancy
Vacancy
Education / ion ith t a w c d du ate e i c s o ire ss u a req is
/ hasjaosbscoactieagteodrywith is ICT Vacancy f lass-O subC has job category / is associated with reqiusiraessswocoirakteedxpweitrhience /
Computing Professionals subClass-Of has job category / is associated with Occupation
Requested
Work Experience hbaeslovJnaogcbsaVtnoaccy/ancyisahsassocciaotnetdrawctitthypiseh/aasssoacciativteitdywseitchtor/ has job category / is associated with Location
Organization
Contract Type
Sector • Compensation Ontology. This ontology models knowledge of wages and salaries used in the SEEMP Project. It is based on the ISO 42175. The ISO 4217 is expressed in HTML format. It is a list of 254 currency names and codes. The resultant Compensation Ontology has 2 concepts: Currency and Salary. For every currency element specified in the ISO 4217 a different instance of the Currency concept is defined. So, the Currency concept has 254 instances. An example instance of the Currency concept is UNITED STATES - US Dollar. The ad hoc relations defined in this ontology are has currency between the Salary and Currency concepts, and its inverse one. • Driving License Ontology. This ontology models knowledge of Driving License domain. It is based on the levels recognized by the European Legislation6. This classification is expressed in HTML format and it is a list of 12 kinds of driving licenses. The resultant Driving License Ontology just has the Driving License concept; and for every kind of driving license specified in the European Legislation a different instance of the Driving License concept is defined. An example instance of the Driving License concept is A1 Light weight motorcycle. The Driving License concept has two relations: ‘Driving License is associated with Job Vacancy’ (with the Job Vacancy concept from the Job Offer Ontology) and ‘Driving License is driving license of Job Seeker’ (with the Job Seeker concept from the Job Seeker Ontology). • Economic Activity Ontology. This ontology models knowledge of economic activities and sectors. It is based on the NACE Rev. 1.17. This standard is expressed in MS Access database format and it is a classification of 849 economic activities. The resultant Economic Activity Ontology has 849 concepts. In this case we have defined a concept for every element of the NACE taxonomy in order to preserve the hierarchy. In the Economic Activity Ontology, the most general concept is the Activity concept. This concept is organized in the taxonomy based on the NACE. The Activity concept has four relations: ‘Activity is associated with Objective’ (with the Objective concept from Job Seeker Ontology), ‘Activity is associated with Offered Work Experience’ (with the Offered Work Experience concept from Job Seeker Ontology), ‘Activity is associated with Job Vacancy’ (with the Job Vacancy concept from Job Offer Ontology) and ‘Activity is associated with Requested Work Experience’ (with the Requested Work Experience concept from Job Offer Ontology). • Occupation Ontology. This ontology models knowledge of occupations and job categories. It is based on the ISCO-88 (COM)8, ONET8 and European Dynamics classification of occupations. ISCO-88 (COM) and ONET are expressed in MS Access database format; European Dynamics classification of occupations is stored in an ORACLE database table. ISCO-88 (COM) is a classification of 520 occupations; ONET is a classification of 1167 occupations and the European Dynamics classification has 84 occupations. The resultant Occupation Ontology has 609 concepts. For this ontology we have extended manually the ISCO-88 (COM) classification with European Dynamics and ONET classifications for ICT
occupations. In this case we have defined a concept for every element of the resulting extended taxonomy in order to preserve the hierarchy. This ontology defines a concept taxonomy based on the aforementioned standards, in which the most general concept is the Occupation concept. The Occupation concept has four binary relations: ‘Occupation is associated with Objective’ (with the Objective concept from the Job Seeker Ontology), ‘Occupation is associated with Offered Work Experience’ (with the Offered Work Experience concept from the Job Seeker Ontology), ‘Occupation is associated with Job Vacancy’ (with the Job Vacancy concept from the Job Offer Ontology) and ‘Occupation is associated with Requested Work Experience’ (with the Requested Work Experience concept from the Job Offer Ontology). • Education Ontology. This ontology models knowledge of education level and education fields. The education fields are based on the FOET9 and the education levels are based on the ISCED9710 (International Standard Classification of Education); both of them are expressed in MS Access database format. FOET has 127 education fields and ISCED97 has 7 education levels. The resultant Education Ontology has 130 concepts. For the education levels we have defined the Education Level concept; and for every education level specified in ISCED97 a different instance of the Education Level concept is defined. For the education fields we have defined a concept for every element of the FOET taxonomy in order to preserve the hierarchy. The concept Education has four binary relations: ‘Education has education level Education Level’, ‘Education has education field Education Field’, ‘Education is associated with Job Vacancy’ (with the Job Vacancy concept from the Job Offer Ontology) and ‘Education is education of Job Seeker’ (with the Job Seeker concept from the Job Seeker Ontology). • Geography Ontology. This ontology is based on the ISO 316610 country codes and the European Dynamics classifications: Continent and Region. The ISO 3166 is expressed in XML format; Continent and Region classifications are stored in ORACLE database. The ISO 3166 has 244 country codes and names; Region classification has 367 regions and Continent classification has 9 continents. The resultant Geography Ontology has four concepts, a Location as main concept, which is split into three subclasses: Continent, Region and Country. For every country element specified in the ISO 3166 a different instance of the Country concept is defined, so the Country concept has 244 instances. For every region element specified in the Region classification a different instance of the Region concept is defined, so the Region concept has 367 regions. Finally for every continent element specified in the Continent classification a different instance of the Continent concept is defined. An example instance of the Continent concept is EU – Europe. An example instance of the Country concept is SPAIN – ES. An example instance of the Region concept is 9 Available through RAMON Eurostat's Classifications Server at http://ec.europa.eu/comm/eurostat/ramon/ 10 http://www.iso.org/iso/en/prods-services/iso3166ma/index.html
Galicia. The Country concept has four binary relations: ‘Country is nation of Job Seeker’ (with the Job Seeker concept from the Job Seeker Ontology), ‘Country is residence of Job Seeker’ (with the Job Seeker concept from the Job Seeker Ontology), ‘Country is located in Continent’ and ‘Country has region Region’. • Labour Regulatory Ontology. This ontology is based on the LE FOREM11 classifications ContracTypes and WorkRuleTypes, both of them expressed in XML format. ContractTypes classification has ten contract types and WorkRuleTypes has 9 work rule types. The resultant Labour Regulatory Ontology has 2 concepts. For every type of work condition or contract type considered by LE FOREM, a different instance of one of these two concepts (Contract Type or Work Condition) is included in the ontology. An example instance of the Contract Type concept is Autonomous. An example instance of the Work Condition concept is Partial time. The Work Condition concept has 2 binary relations: ‘Work Condition is associated to Objective’ (with the Objective concept from the Job Seeker Ontology) and ‘Work Condition is associated with Job Vacancy’ (with the Job Vacancy concept from the Job Offer Ontology). The Contract Type concept has two binary relations: ‘Contract Type is associated to Objective’ (with the Objective concept from the Job Seeker Ontology) and ‘Contract Type is associated with Job Vacancy’ (with the Job Vacancy concept from the Job Offer Ontology). • Language Ontology. This ontology is based on the ISO 639212 and the Common European Framework of Reference (CEF)13. The ISO 6392 is expressed in HTML format and CEF is a description in PDF format. The ISO 6392 has 490 language codes and CEF has 6 language levels. The resultant Language Ontology has 3 concepts: Language, Language Level and Language Proficiency. For every language element specified in the ISO 6392 a different instance of the Language concept is defined, so the Language concept has 490 instances. For every language level element specified in the CEF a different instance of the Language Level concept is defined, so the Language Level concept has 6 instances. An example instance of the Language concept is eng – English. An example instance of the Language Level concept is A2 – Basic User. The Language concept has three relations: ‘Language is mother tongue of Job Seeker’ (with the Job Seeker concept from the Job Seeker Ontology), ‘Language is spoken by Job Seeker’ (with the Job Seeker concept from the Job Seeker Ontology) and ‘Language is evaluated by Language Proficiency’ (with the Language Proficiency concept from the Language Ontology). • Skill Ontology. This ontology models knowledge of Skills and abilities. It is based on European Dynamics Skill classification. This classification has 291 skills and 11 LE FOREM is an user partner of the SEEMP project, http://www.leforem.be/ 12 http://www.iso.org/iso/en/prods-services/popstds/languagecodes.html 13 http://www.cambridgeesol.org/exams/cef.htm it is stored in an ORACLE database table. The resultant Skill Ontology has 2 concepts: Skill concept with its subclass ICT Skill. For every skill element specified in the European Dynamic classification a different instance of the ICT Skill concept is defined. An example instance of the ICT Skill concept is Hardware programming. The Skill concept has two relations: ‘Skill is associated with Job Vacancy’ (with the Job Vacancy concept from the Job Offer Ontology) and ‘Skill is skill of Job Seeker’ (with the Job Seeker concept from the Job Seeker Ontology). • Competence Ontology. This ontology defines a concept called Competence as a super class of the imported concepts Skill, Language Proficiency and Driving License. The Competence concept has three binary relations: ‘Competence is associated with Vacancy’ (with the Vacancy concept from the Job Offer Ontology); ‘Competence is competence of Candidacy’ (with the Candidacy concept from the Job Seeker Ontology) and ‘Competence requires Education’ (with the Education concept from the Education Ontology). • Time Ontology. This ontology is based on DAML ontology14 and it is expressed in OWL format. The main concepts of this ontology are Instant and Interval, which are subclasses of Temporal Entity. Instant is linked to Interval through the properties of begins, ends, inside and begins or in. Instant is also linked to an instant temporal description, which is a concept with the properties of second, minute, hour, day, month, year and time zone. Interval has the subclass proper interval, which is related with itself through the relations ‘interval equals’, ‘interval before’, ‘interval starts or finishes’, etc. Proper intervals can be concatenated through the relation ‘concatenation’.
Finally we present the Reference Ontology statistics. The Reference Ontology is composed of thirteen modular ontologies. The Reference Ontology has 1609 concepts, 6727 class attributes, 60 instance attributes, 94 ad hoc relationships, 1674 instances and 20 axioms. 5
In this section we provide some guidelines for the building process of the local ontologies, each ES uses its own Local Ontology, which describes the employment market in its own terms. Based on the proposed SEEMP architecture, the possible options for building the local ontologies are: building local ontologies from the RO, and building local ontologies as a reverse engineering process from ES schema sources. Next, these options will be explained. 5.1 Building local ontologies from the Reference Ontology The building process is structured/guided by the architecture of the Reference Ontology and scoped with applications needs. In this sense, we will need to extend some already defined elements, to remove unnecessary elements, or to add new application dependent elements that appear in each ES schema source. The result of this should be a RO friendly "local" ontology. Thanks to this similarity, mappings between local ontologies and RO will not be complex. But on the other hand, mappings between local ontologies and ES schema sources will be complex. Regarding the evolution and change propagation dimension we have: • • •
Changes in the RO imply a change in the mappings between local and RO as well as probably changes in the mappings between the local ontologies and the ES schema sources.
Changes in the RO imply a change in the local ontology; in this case, the mappings RO – local ontology would remain as they were. The mappings between the local ontologies and the ES schema sources should be updated. Changes in the ES schema sources imply changes in its local ontology (probably the part that is not a mirror of the RO) and the mappings between local ontologies and ES schema sources, and probably minor changes in the mappings between local ontology and the RO. This last consequence is especially interesting because the changes in the ES schema sources will be the most frequent in the scenario posed by SEEMP. 5.2 Building local ontologies as a reverse engineering process from the ES schema sources In this case, mappings between local ontologies and schema resources should not be complex. On the other hand, complex mappings will appear between the Local and RO. The building process requires more sophistication of knowledge engineering and good acquaintance of all the data and their structures of the application: not easily found skill set in ES or any other operational/research organizations. Regarding the evolution and change propagation dimension we have: • Changes in the ES schema sources imply a change in the local ontologies and, consequently, in mappings between sources and local ontologies, but not necessarily in mappings between local and the RO. •
Changes in the RO imply changes in the mappings between local ontologies and the RO, but it is not necessary to modify anything at the ES level.
In SEEMP project we adopt a different option depending on the part of the local ontology to be built. On one hand, we select option 1 (building local ontologies from the RO) for Job Seeker and Job Offer ontologies and other general purpose ontologies like, for example, the Time Ontology. On the other hand, we select option 2 (building local ontologies as a reverse engineering process from ES schema resources) for Occupation, Education, Economic Activity, Language, Compensation, Labour Regulatory, Skill and Driving License ontologies.
The reason of selecting option 1 for Job Seeker and Job Offer ontologies is because
there are not significant differences between these ontologies and the way how each
ES structures job seeker and job offer information. Consequently mappings between
local ontologies and RO will be simple, but mappings between local ontologies and
ES schema sources will be complex. Furthermore, for the job seeker and job offer
information local ontologies will share the same vocabulary (see [
The reason of selecting option 2 for the ontologies mentioned above is because each ES may have its own classification systems for the related information. Nevertheless, it may happen that the local ontology shares some classification with the RO (as there will happen in the European scope with the driving license classification). In that case, the reverse engineering process for that classification will be skipped, and that part of the RO will be reused. By using option 2, mappings between local ontologies and RO will be complex, but mappings between local ontologies and ES schema sources will be simple. 6
In this paper we have presented the methodological approach we followed for reusing existing human resources management standards/classifications in the SEEMP Project. We also described the resultant RO which acts as a common “language” in the form of a set of controlled vocabularies to describe the details of a job posting and the CV of a job seeker. The RO was developed with the proposed methodology and with the ontology engineering tool WebODE. Finally we have provided some guidelines for the building process of the local ontologies, and we conclude that the best option for building the local ontologies is building them following a hybrid approach that employs the best option for each part of the RO.
An important conclusion of the work that we have carried out is that we can reuse human resource management standards in new applications following a systematic approach. Moreover, it is clear such a reuse can save time during the development of the whole system. However, it is not always possible to reuse a standard in a straightforward way, because sometimes the ideal standard does not exist for different reasons (different scope, outdated, etc.), and it is necessary to extend some “imperfect” standard with additional terminology coming from other standards or ad hoc classifications. As future work, we will complete the development of the local ontologies for the SEEMP project. This work allows us to confirm that proposed systematic approach not only supports the creation of a RO from standards/classifications properly, but it also facilitates the creation of the local ontologies related to this RO, since the building process of these local ontologies can take advantage of the already existing RO.
Acknowledgments. This work has been partially supported by the FP6 EU SEEMP Project (FP6-027347).