Semantic Web Technologies in the Recruitment Domain
Ralf Heese, Humboldt-Universitọt zu Berlin, Germany
Malgorzata Mochol, Freie Universitọt Berlin, Germany
Radoslaw Oldakowski, Freie Universitọt Berlin, Germany
Abstract
Due to the large number of job offers published online, it is almost impossible for job seek- ers and job portals to gain an overview of the entire employment market. Since job offers lack semantically meaningful annotations, their location and integration into databases is extremely dificult. In this chapter, we demonstrate how the application of Semantic Web technologies can enable unambiguous identiication of concepts and relationships between concepts and how the e-recruitment process provides advantages for all participants in the market. When comparing job and applicant proiles, this above mentioned identiication through the use of a dedicated matching function is a key element for increasing the preci- sion of search results provided by search engines. Furthermore, it allows for automating and supporting recruitment processes. In this chapter, we present an application scenario and our prototypical implementation discussing the construction of a human resource ontology for annotating job offers and job applications and our matching function.
300 Heese, Mochol, & Oldakowski
Introduction
The cornerstone of an organization’s success is the alignment of all corporate resources with business objectives and strategies (Wright & McMahan, 1992). The key asset of every organization is its employees. The crucial factor for businesses to compete effectively is inding the best person for a given task and developing and leveraging their skills and capa- bilities. Furthermore, the employer has to maximize the impact of training and educational efforts and to align the activities of the employees with corporate objectives as well as to retain top performers by an effective incentive management (Ferris, Hochwarter, Buckley, Harrell-Cook & Frink, 1999).
Whereas most of the chapters in this book deal with employee development through or- ganizational e-learning, this chapter focuses on human resource recruitment, in particular Web-based recruitment. We present an application scenario in which online recruitment processes are streamlined using Semantic Web technologies. Additionally, we describe our prototypical implementation of the technological infrastructure.
Semantic Web technologies can support the unambiguous identiication of concepts and formally describe relationships between concepts thereby allowing representation of data in a more machine-understandable way. In our scenario, job position postings and job applicant proiles are annotated using controlled vocabularies combined with domain ontologies. This opens up new possibilities for better job posting discovery by search engines as well as more intelligent matching of open positions with candidate proiles, which no longer relies on the containment of keywords but exploits domain-speciic knowledge, leading to increased automation of the recruitment process.
This chapter is structured as follows: In the second section we provide some background information concerning the current situation on the electronic job market and point to the shortcomings of today’s online recruitment process. The third section contains a brief in- troduction into the fundamentals of the Semantic Web. Subsequently, in the fourth section, we describe a typical recruitment process seen from the perspective of a company as well as an applicant. We outline what impact the application of the Semantic Web technologies would have on each phase of the process. The ifth section deals with practical aspects of ontology engineering. It begins with a comparison of two ways of building ontologies: from scratch and through ontology reuse. Next, we give some insight into the structure of the human resources (HR) ontology we have developed as the building block for annotating job postings and applicant proiles. The sixth section is devoted to the prototypical realization of the scenario. Here, we describe the general scenario architecture as well as illustrate how the implemented semantic matching functionality enables search and ranking possibilities far beyond simple keyword-based algorithms. Finally, the seventh section summarizes the impact of the application of Semantic Web technologies in the electronic recruitment domain.
E-Recruitment
Over the last few years, the Internet has evolved into the primary recruitment medium. As reported in Keim et al. (2005), 90% of human resource managers in Germany rated the In-
ternet as an important recruitment channel. One reason for this high rate is that the Internet, in particular, reaches applicants who are young and highly qualiied. Despite the fact that companies use more than one channel to publish their job openings, over half of all personnel recruitment is the result of online job postings (Monster, 2003). Although a large number of online job portals have sprung up, dividing the online labor market into information islands, employers still publish their job postings on a rather small number of portals in order to keep costs and administrative effort down. The hiring organizations simply assume that a job seeker will visit multiple portals while searching for open positions.
Alternatively, companies can publish job postings on their own Web site (Mülder, 2000).
This way of publishing, however, makes it dificult for job portals to gather and integrate job postings into their database. Thus search engines such as Google or Yahoo! have become vital in the job search. Furthermore, dedicated search engines, such as worldwidejobs.de (http://www.worldwidejobs.de/), are entering into the market, allowing detailed queries as opposed to the keyword-based search of current search engines. The quality of search results depends not only on the search and index methods applied. Further inluential factors include the processability of the used Web technologies and the quality of the automatic interpreta- tion of the company-speciic terms occurring in the job descriptions. The deiciencies of a website’s machine processability result from the inability of current web technologies, such as HTML, to semantically annotate the content of a given Web site. Therefore, computers can easily display the content of an HTML site, but they lack the ability to interpret the content properly.
Our conclusion is that the information low in the online labor market is far from optimal.
The publishing behavior of companies not only makes it almost impossible for a job seeker to get an overview of all the appropriate openings but also complicates or even prevents job offers from reaching a greater range of potential employees. In this chapter, we show how the shortcomings of the current situation in the online recruitment market can be overcome by utilizing a new Internet vision called “Semantic Web,” leading to higher market transpar- ency and enhanced optimization of the recruitment process.
Semantic Web
The Semantic Web provides a common framework that allows data to be shared and reused across application, enterprise, and community boundaries (World Wide Web Consortium, 2001). The term Semantic Web comprises a family of technologies that enable applications to generate and automatically process metadata describing distributed Internet resources.
The objective is to use the Web as a global distributed database that can be deployed by ap- plications to perform certain tasks automatically (Berners-Lee, Hendler, & Lassila, 2001).
The development of the Semantic Web is a joint effort of scientiic institutions (Massachu- setts Institute of Technology, Stanford, The Institute for Learning and Research Technology (ILRT), etc.) together with top businesses (Hewlett-Packard, IBM, Nokia, etc.) (Quan &
Karger, 2004) and is led by the World Wide Web Consortium (W3C). To realize the vision of the semantic Web networked resources, for example, Web sites, are annotated by structured and machine-understandable metadata, which are assigned a well-deined meaning and are interpreted by means of logic rules.
302 Heese, Mochol, & Oldakowski
The fundamental technologies of the Semantic Web are presented in this section. For naming Web resources and terms uniform resource identiiers (URIs) are used as a global identiication mechanism (layer 1). At the syntactical layer, the exchangeability of semantic information is guaranteed by the use of the XML mark-up language (layer 2). The resource description framework (RDF) (Beckett & McBride, 2004) speciies a data model for publishing metadata on the web (layer 3) and utilizes XML as serialization syntax for data transmission. With RDF schema (RDF/S), the user can develop his own metadata vocabularies. The Web ontology language (OWL) (Dean & Schreiber, 2004) at layer 4 of the technology stack extends RDFS with terms and concepts for an even more expressive knowledge representation in the form of ontologies. Ontologies formally name and describe the central concepts of an application domain and relationships between them. Since ontologies conceptualize a part of the real world, they can serve as means for communication between the users of an application domain.
Generic ontologies (upper ontologies) deine general concepts and relationships between them and are currently being standardized by international standardization committees, for example, SUMO by the IEEE standard upper ontology working group (“Suggested Upper Merged Ontology (Sumo),” 2005), while domain-speciic ontologies are being developed by communities of domain experts, for example, the gene ontology by the Gene Ontology Consortium (“The Gene Ontology,” 2005)
Layer 5 provides the means for specifying sets of deductions that can be made from a col- lection of data, together with formalisms for describing steps taken to reach conclusions from given facts. Finally, the trust layer deines mechanisms upon which applications decide whether or not to trust the given information. These two topmost technology layers are cur- rently being researched and simple application demonstrations are being constructed, e.g.
(Bizer, Cyganiak, Gauss & Maresch, 2005).
These new Internet technologies (Semantic Web technologies) are maturing and moving out from academic applications into the wider industry. This is demonstrated on one hand by the strong and growing interest in these topics shown by various commercial sectors Figure 1. Technology stack of the Semantic Web
(1) Identification of Resources (2) Data Encoding
(3) Semantics (4) Ontology (5) Logic Framework (6) Trust
Unicode, URI XML, XML Schema RDF, RDF Schema DAML + OIL, OWL Inference algorithms Digital signatures Semantic Web applications
and on the other hand by public bodies like the European Commission, which supports the distribution and transfer of these technologies to the business world. One such activity is the Knowledge Web EU Network of Excellence, which has formed an “industry board”
(http://knowledgeweb.semanticweb.org/o2i) to promote greater awareness and faster uptake of Semantic Web technology within European industry. The Knowledge Web together with this board aims to transfer technology from research to industry, to promote ontological technologies, propose technological recommendations, and to meet industrial application needs. Within this context, they have identiied some key sectors for the early uptake of Semantic Web technologies: human resources, health and life sciences (Nixon & Mochol, 2005). We have picked up on the issue of human resource management as a potential ear- lier adopter and developed a Semantic Web-based scenario for e-recruitment for which we implemented a prototype.
Supporting Recruitment Processes with Semantic Web Technologies
Apart from intermediaries, the actors in a recruitment scenario are employers and job ap- plicants. Figure 2 displays the recruitment processes from the viewpoint of an organization as well as an applicant. The actions performed by a company personnel manager or a job applicant are very similar. For instance, both processes contain the generation of a job/ap- plicant proile and the matching of job descriptions with candidate proiles. Therefore, the next sections take mainly the viewpoint of the company into consideration. Each following subsection describes a step of the recruitment process and begins with a description of the current situation. Afterward, we illustrate the utilization of Semantic Web technologies in each step.
Requirements Analysis
The requirements for an open position are usually conceived by the operating department in cooperation with the human resources department. Nowadays, the requirements for a job posting are commonly written in free text as illustrated on the left hand of Figure 3. The usage of free text limits the machine processability of postings in the later phases of the recruitment process. In contrast to a free text description, the utilization of a widespread shared “language” in form of a set of controlled vocabularies as shown on the right hand of Figure 3 would facilitate communication between all parties involved. Personnel managers would use the controlled vocabularies to semantically annotate the details of a job posting, in the same way job applicants would annotate their application. This annotation step opens up the potential for the automation of a variety of tasks along the recruitment process such as inding job offers in a speciic sector of the job market.
304 Heese, Mochol, & Oldakowski
Figure 2. Processes in the recruitment domain
Analyse therequirements Publish the job posting Receive andpreselectapplications Final recruitmentdecision
Generate ajob profile Search manuallyorpublish the profile Write and mailapplications Interview fora job
Employer
Job Applicant
Offering a job position
Searching a job
Publication of the Job Posting
There are a number of channels to publish a job posting on the Internet: (1) commercial job portals, such as Monster (http://www.monster.com/), JobPilot (http://www.jobpilot.com/), or StepStone (http://www.stepstone.com/) compete to publish job postings for a publication fee, (2) the employer’s own Web site, and (3) state labor agencies such as the German Fed-
Figure 3. Example of a job description and an extract of the skill ontology
306 Heese, Mochol, & Oldakowski
eral Employment Ofice (http://www.arbeitsagentur.de/). Various job portals target different groups and are mostly speciic to certain geographic regions. Since there are a lot of portals, a job seeker is unable to get an overview over the entire job market. Additionally, job portals can hardly gather job offers from employers’ Web sites. To improve market transparency several public bodies like the German Federal Employment Ofice (BA) and the Swedish National Labour Market Administration (AMS) (http://www.ams.se/) have set up projects to integrate open positions into a central database. Furthermore, they intend to establish a complete worklow for e-recruitment. In both projects, participating employers use terms from a controlled vocabulary to categorize their postings and submit them to a central da- tabase using variations of the HR-XML (http://www.hr-xml.org/) data format, for example, HR-BA-XML and HR-XML-SE, respectively. The collected postings are published through a central portal and are additionally forwarded to commercial job portals. The problem with these projects is that the entire market depends on a single central database — an approach to which many market participants object and which can be quite error-prone, as experienced by the German project (Crosswater Systems Ltd., 2003).
Using Semantic Web technologies to semantically annotate job postings would increase market transparency together with avoiding the bottleneck of a central database. Organiza- tions would publish their annotated job postings directly on their corporate Web sites using controlled vocabularies and the RDF data format. The postings could then be crawled directly by job portals and thus the centralized approach would be replaced with a distributed one.
Consequently, all job portals would operate on the same information and postings would reach more applicants leading to higher market transparency.
Another beneit from having postings annotated with terms from a controlled vocabulary is that the terms can be combined with background knowledge about an application domain.
Job portals could offer semantic matching services that allow the comparison of job posi- tion postings with applicant proiles based on domain-speciic knowledge instead of merely relying on keywords, like traditional search engines do. For example, if Java programming skills are required for a certain job and an applicant is experienced in Delphi, the matching algorithm would consider this person’s proile a better match than someone else’s who has skills in SQL, because Delphi and Java are closer related than SQL and Java.
Receiving and Preselecting Job Applications
According to Keim et al. (2004), a general tendency has been emerging to move away from the classical job applications toward an electronic version. The latter has the advantage that it allows both sides to check basic requirements automatically, thereby reducing the number of applications to be reviewed manually. About 59% of large-scale enterprises request that job applicants store and maintain their personal proile on the company server for a while (Keim et al., 2005). Enterprises like IBM (https://forms.bpfj.intronet.com/arm/ibm/emea/) and Volkswagen (https://www.vw-personal.de/www/de/bewerbung/onlinebewerbung.html) commonly require online forms to be submitted by job searchers. From the applicants’
viewpoint, this procedure is time-consuming and quite laborious, since they have to re-enter their CV data into each new application form.
Both applicants and employers beneit from the semantic annotation of job applications in addition to the classic free text application. Applicants could reuse their RDF (see third sec- tion) proiles and send them to different employers instead of having to ill in numerous Web forms. Employers use the annotations for automating the preselection process by matching their minimal requirements with the applications. After the employment of a candidate, the annotated job application could serve as the basis data for the company human resource management system.
Another issue of this phase is prooing testimonials of experience and education. Testimonials could be attached to applications in the form of RDF statements that are digitally signed by the issuing university or organization. The process of validating digitally signed testimonials can be automated, further reducing costs.
Interviews and Recruitment Decision
Face-to-face contact with candidates is indispensable for the inal recruitment decision. In the interviews, the personnel manager rates the soft as well as technical skills of a candi- date. Although social competencies can be expressed using ontologies, we believe that an automatic evaluation of these competencies is of little importance. Thus, this phase of the recruitment process will still be done manually in the future.
Applying the Semantic Web to Human Resource Management
In a Semantic Web-based recruitment scenario, the data exchange between employers, appli- cants, and job portals is based on a set of vocabularies that provide shared terms to describe occupations, industrial sectors, and job skills. This common vocabulary is deined in our human resources ontology (HR ontology). Ontologies represent domain-speciic knowledge and might be used to determine semantic similarity between resources such as job applications and job offers. Hence, ontologies are not only used to annotate job postings and applications but for semantic matching as well – a technique that combines annotations using controlled vocabularies with background knowledge about a certain application domain. Ontologies form a key component of our job portal allowing the semantic comparison of job postings with candidate proiles. However, the development of ontologies is far from trivial and the decision to construct ontologies from scratch has to be well considered.
In the following, we describe a methodology for ontology reuse and a model for cost esti- mation of the ontology development process. Thus, ontology engineers can turn to recom- mendations of how to proceed during ontology development and different kinds of support can be engaged.