Matching: Human Resource Ontologies
To improve job portal functionality we utilise Semantic Web technologies to semantically annotate job
postings and applicants’ profiles in order to increase market transparency together with avoiding the
bottleneck of a central database. In a Semantic Web-based recruitment application the data exchange
between employers, job applicants and job portals is based on a set of shared vocabularies describing
domain relevant terms: occupations, industrial sectors and skills. These commonly used vocabularies have
been formally defined by means of a so-called Human Resource ontology (HR-ontology).
The implementation of the HR-ontology was realized by translating several semi-structured
input formalisms and coding text-based classification standards to OWL.
In order to support common industry practice and to maximize the integration of job seeker profiles and
job postings from different organizations the ontology underlying the Semantic Web job portal had to be
aligned to established domain-specific standards and classifications. We reused some of the most relevant
classifications being deployed by national and international agencies:
Our HR ontology is modelled using OWL and descriptions of job postings and applicants’ profiles are
stored in RDF using the vocabulary defined by the HR ontology.
- HR-BA-XML is the official German extension of Human Resource
XML (HR-XML), the most widely used standard for process documents. HR-XML is a library of more than 75
interdependent XML schemes defining particular process transactions, as well as options and constraints
regulating the correct usage of the XML elements.
- Classification of Industry Sector
(WZ2003) is a German classification standard for economic activities.
- Occupation Code (Berufskennziffer – BKZ) is a German version of SOC System. It classifies workers
into 5597 occupational categories according to occupational definitions.
- KOWIEN is a skill ontology which defines concepts representing competencies required to describe
job position requirements and job applicant skills.
Task & Results
In the HR scenario, the domain
specific knowledge is represented in the form of various concept hierarchies (skills, occupation classification,
industry sectors, etc.) and can be used to determine the semantic similarity between concepts. The algorithm
should allow us to compare job descriptions and applicants’ profiles based on their semantics.
We compare one job description including skills, occupation types and industry sectors with all applicants’
profiles which also contain skills, occupation type and industry sector (as well as another way round: compare
one job seeker description with all available openings). Results are ranked according to semantic similarity in
a range 0:1. Hence the provided algorithms must support1:n matching. The ranking will be made according to our weightings.
The current application (cf. black parts in the Fig. 1) uses an algorithm which is based on the similarity between two concepts determined by the
distance between them at the same time reflecting the respective positions of the concepts in the concept hierarchy.
The modified application (cf. red parts in the Fig. 1) uses ontology matcher for automatic generation of a job matcher (Fig. 1 - red matching engine)
depending on the ontologies. The generated matcher is to be used in place of the original one (Fig. 1 - black matchig engine).
Fig. 1 The job-application matching case: existing (in black) + modified (in red) application
The matching alogrithms as well as their parameters should be provided to the organizers (Jerome.Euzenat(at)inrialpes.fr) which will run the algorithms on the ontologies and obtain the alignment.
Furthermore the generated job matcher will be used to compare a set of about 250 job ofersag with about 250 applicants' profiles.
The current approach will be used as the baseline to compare submissions.
Submissions can use any matching approach desired, including the use of additional, external sources. However these sources may only be accessed during
the matchings and must not be manually pre-tuned to the given ontologies.