Valuing Prior Learning: Designing an ICT Artifact to Assess Professional Competences Through Text Mining

Purpose – This paper aims to introduce an information and communication technology (ICT) artifact that uses text mining to support the innovative and standardized assessment of professional competences within the validation of prior learning (VPL). Assessment means comparing identi ﬁ ed and documented professional competences against a standard or reference point. The designed artifact is evaluated by matching a set of curriculumvitae(CV) scraped from LinkedIn against a comprehensivemodel of professional competence. Design/methodology/approach – A design science approach informed the development and evaluation of theICTartifactpresented in this paper. Findings – A proof of concept shows that the ICT artifact can support assessors within the validation of prior learning procedure. Rather the output of such an ICT artifact can be used to structure documentation in thevalidationprocess. Research limitations/implications – Evaluating the artifact shows that ICT support to assess documented learning outcomes is a promising endeavor but remains a challenge. Further research should work onstandardized ways todocument professionalcompetences, ICTartifacts capture the semantic content of documents, andre ﬁ neontologies oftheoretical models of professional competences. Practical implications – Text mining methods to assessprofessional competences rely on large bodies of textualdata, andthus a thoroughly builtandlarge portfolio is necessary asinputforthis ICTartifact. Originality/value – Following the recent call of European policymakers to develop standardized and ICT-based approaches for the assessment of professional competences, an ICT artifact that supports the automatized assessmentofprofessionalcompetenceswithinthevalidationofpriorlearningisdesignedandevaluated.


Introduction
The validation of prior learning (VPL) is the process of "assessing and recognizing a wide range of skills and competences which people develop through their lives and in different contexts, for example through education, work and leisure activities" (Bjørnåvold, 2000, p. 216).The European Union supports the validation of prior learning by introducing the Lifelong Learning Strategy (EU, 2006), the European Qualification Framework (EQF) (EU, 2017) and the recommendations on the validation of prior learning (EU, 2012).Having viable and efficient approaches for the assessment of professional competences within the validation of prior learning could help to lower the number of unemployed, increase labor market mobility and facilitate social cohesion within the European Union.
While policy frameworks for the assessment of professional competences within VPL are in place in most of the European countries, providing specific methods and approaches for the assessment proves to be a challenge for policy-making (EU, 2012(EU, , 2017) ) and scientific research (Bohne et al., 2017;Brockmann et al., 2009).In VPL, assessment is the phase in which a person's learning outcomes (i.e.professional competences) are "compared against specific reference points and/or standards" (Cedefop, 2015, p. 18).A standard or reference point is a document that describes which learning outcomes people have to obtain to be qualified on a certain level, e.g. a document that shows what a student should be able to do after finishing training or education.
As we lack innovative approaches to support the assessment of professional competences (Cedefop,p. 20), the European Union calls to develop standardized and information and communication technology (ICT)-based approaches for the assessment of professional competences within the VPL (Cedefop, 2017).Currently, the VPL procedures remain a labor-intensive manual task.The assessment of competences within the VPL has to be done by qualified assessors, who need to be trained to guide individuals through the validation process (Diedrich, 2013).Consequently, it takes weeks or even months to conduct a validation procedure before individuals can show their qualifications to employers.Our research question is: How to automatize the assessment of professional competences within the VPL?Our research objective is to introduce an ICT artifact that supports the assessment of professional competences within the validation of prior learning by matching a documentation of professional competences with a given standard, a predefined theoretical model of professional competences.
In this paper, we draw on a design science research (Hevner et al., 2018;Hevner et al., 2004;Gregor and Hevner, 2013) approach to develop an artifact (i.e. an algorithm) that uses text mining to match a repository of curriculum vitae (CV) with a given theoretical model of professional competences.The designed artifact allows us to compare each CV individually with the predefined theoretical model.We refer to the activities of this artifact as "competence mining."This proof of concept shows that such an artifact may support the assessment of professional competences within the VPL by assigning documented competences to a standard or reference point.Practically, we introduce an artifact that can be applied to automatically match textual data (e.g.portfolios or CV) to a standard or reference point (e.g. a theoretical model or qualification standard according to EQF).Based on previously identified (i.e. made explicit or spoken out) and documented (i.e.written down) evidence, the artifact is able to assess professional competences (i.e.compare them against a standard or reference point).This artifact may help assessors within the VPL procedure as it can give a hint about the candidate's competence profile, thus making the VPL procedure less time-consuming and tedious (Han and Lee, 2016).Theoretically, we add to the debate revolving the standardization of VPL procedures.We find that standardizing VPL to a EJTD 44,2/3 certain extent may diminish the negative effects that the VPL procedures can bring about (Diedrich, 2013).
The remainder of the paper is structured as follows.Section 2 consists of a literature review that introduces related theoretical and practical approaches for the ICT-supported assessment of professional competences.Section 3 more closely describes the design science approach (Hevner et al., 2018;Hevner et al., 2004;Gregor and Hevner, 2013).In section 4, we describe the designed artifact.In section 5, we present a proof of concept of the artifact by matching a repository of CV to a theoretical model of professional competence.In section 6, we discuss the findings and show how they relate back to the research question and objectives.In section 7, we outline potential limitations of the artifact, point out further research endeavors and conclude.
2. Literature review 2.1 The assessment of professional competences within the validation of prior learning A person acquires professional competences mainly through experiences and learning that can be formal, non-formal or informal.Formal learning, occurring in an "organized and structured context (formal education, in-company training, etc.) is designated as learning" (Bjørnåvold, 2000, p. 204) and comparably easy to assess because licenses or degrees are awarded that explicitly specify the learning outcomes.Differently, non-formal and informal learning outcomes are partly tacit (Polanyi, 1966).Non-formal learning, "planned activities that are not explicitly designated as learning, but which contain an important learning element" (Bjørnåvold, 2000, p. 204), is considerably harder to assess as documentation may have different grades of trustworthiness.Informal learning or experiential learning that can "be understood as accidental learning" (Bjørnåvold, 2000, p. 204) is even more situated in the environment (Lave and Wenger, 2011) and occurs in day-to-day activities related to work, family or leisure, including language learning or parenting and more challenging to assess.For example, what a person is able to do is comparably easy to assess based on a university degree, comparably harder to assess based on certifications of massive open online courses or courses from tertiary education and even harder from learning that the person is not aware of.
To validate formal, non-formal and informal learning, tacit knowledge and competences must be made explicit and documented in a social process (Nonaka, 1994;Nonaka et al., 2000) the VPL.Consequently, the VPL usually consists of four phases: identification, documentation, assessment and recognition of prior learning (Cedefop, 2015).First, a qualified assessor supports individuals in identifying previously acquired knowledge, skills and competences from different contexts using reflection (Schön, 1983) and dialogue (Bohm, 2012) with the aim that individuals become increasingly aware of prior achievements.The "discovery and increased awareness of own capabilities is a valuable outcome of the process" (Cedefop, 2015, p. 18).Second, documenting learning outcomes or stocktaking requires people to provide evidence through "building" of a portfolio that tends to include a CV and a career history of the individual, with documents and/or work samples that attest to their learning achievements (Cedefop, 2015, p. 18).Individuals have to approach authorities, peers or former supervisors who are willing to provide evidence of the identified learning (e.g.certificates, licenses, proof of voluntary work).Third, assessment is the phase in which "an individual's learning outcomes are compared against specific reference points and/or standards" (Cedefop, 2015, p. 18).Standards or reference points (Bohlinger, 2017) are set by companies or professional associations and assessment methods range from written, oral or practical tests/examinations to portfolios.Fourth, recognition is the certification of previously assessed learning through the award of a qualification by an authority Valuing prior learning (Cedefop, 2015, p. 18).The identification and documentation of professional competences is crucial for their subsequent assessment (Annen, 2013).Starting from the premise that professional competences have been previously identified and documented, this paper only deals with the assessment phase in the VPL.We identify twowell documentedmain challenges in the assessment of professional competences.First, a validity challenge (Stenlund, 2010): does the artifact assess what it promises to do?A person documenting competence in business and management subjects should show these competences in the relevant dimensions of the assessment.We propose a comprehensive model of professional competences as a standard or point of reference in Section 4. This model isbased on the Occupational Information Network (O*Net; Peterson et al., 2001) able to assess professional competences in all relevant domains (and is not limited to a certain profession).The second challenge is to determine the level of acquired competence by assigning a numerical value to the content dimension (Anderson and Krathwohl, 2001;Dreyfus and Dreyfus, 1987).In other words, how can we determine the level of competence, based on a thorough documentation of prior learning?This challenge refers to determining whether a person is a beginner, intermediate, advanced or expert in a certain field.We refer to established taxonomies of competence development and descriptions of the complexity of learning outcomes (Anderson and Krathwohl, 2001;Bloom et al., 1956;Dreyfus and Dreyfus, 1987;Krathwohl, 2002) to determine the documented level of competence.

The assessment of competences using text mining
Extracting professional competences via content analysis from documents such as job advertisements or CV has a long tradition.We can observe this within the academic literature but also more practical fields [1].We can distinguish between approaches that depart from the occupational side and use job advertisements to examine competence requirements for a specific occupation (Müller et al., 2014;Gallivan et al., 2004;Aken et al., 2010;Todd et al., 1995) and approaches that depart from the analysis of individual CV (Darabi et al., 2018;Gorbacheva et al., 2015;Han and Lee, 2016;Lichtnow et al., 2008;Patel et al., 2017;Valdez-Almada et al., 2017).While extracting competence requirements must depart from the occupational side, the assessment of professional competences must begin with the individual CV.
In recent years, text mining is often used to assess large amount of textual data.Text mining is a form of data mining (Romero and Ventura, 2010;Sachin and Vijay, 2012) and is often used in educational settings.In this context, it is referred to as educational data mining (Romero and Ventura, 2010).It comprises a set of methods to analyze unstructured data such as texts or narrations.Text mining techniques "[. ..] allow to automatically extract implicit, previously unknown, and potentially useful knowledge from large amounts of unstructured textual data in a scalable and repeatable way" (Debortoli et al., 2016, p. 556).In this regard, text mining helps to foster knowledge discovery because very large amounts of data can be analyzed simultaneously (Kobayashi et al., 2018).Text mining usually follows the common steps of other data mining techniques, namely, pre-processing, data mining and post-processing (Debortoli et al., 2016;Kobayashi et al., 2018;Romero and Ventura, 2010).
Concerning the assessment of competence requirements, text mining was used in several studies.Darabi et al. (2018) use text mining to identify skills and qualifications which employers search for in engineering fields by comparing job postings to the O*Net.Debortoli et al. (2014) use latent semantic analysis (LDA) to develop a competency taxonomy of business intelligence and big data jobs based on job advertisements.Karakatsanis et al. (2017) use latent semantic indexing to match job postings on the Web with descriptors from EJTD 44,2/3 the O*Net.They aim at identifying the most in-demand occupations on the job market.Kobayashi et al. (2018) aim at introducing organizational researchers with the fundamental logic underpinning text mining and use topic modeling in a job analysis case study.
We consider a work as related if the approach uses text mining methods to assess individual CV.Table I summarizes these works.While there is a considerable amount of work in the field, the application of text mining procedures on large amount of CV remains, with notable exceptions (Darabi et al., 2018;Gorbacheva et al., 2015;Han and Lee, 2016;Lichtnow et al., 2008;Patel et al., 2017;Valdez-Almada et al., 2017) scarce.These works aim at extracting competences in specific directions, such as engineering education (Darabi et al., 2018), business process management (Gorbacheva et al., 2015), construction work (Han and Lee, 2016), computer science (Lichtnow et al., 2008), computer science and engineering majors (Patel et al., 2017) and software engineering (Valdez-Almada et al., 2017).Thus, extraction of competences is limited to a certain field.We address this limitation by designing an artifact which is, because of the comprehensiveness of its underlying model, able to assess individual competences of several professions.

Method
Our approach draws on a design science paradigm (Gregor and Hevner, 2013;Hevner et al., 2004;Peffers et al., 2007) to guide the development of the artifact.While the natural and social sciences aim to understand reality, "design science attempts to create things that serve human purposes" (Simon, 1996, p. 55).Design science comprises the creation (Section 4) and evaluation (Section 5) of an "innovative, purposeful artifact for a specified, currently unresolved problem domain" (Hevner et al., 2004, p. 82).With the artifacts utility as an ultimate goal in mind, it addresses research challenges through the "building and evaluation of artifacts designed to meet the identified [. ..] need" (Hevner et al., 2004, pp. 79-80).An artifact is "a thing that has, or can be transformed into, a material existence as an artificially made object (e.g.model, instantiation) or process (e.g.method, software)" (Gregor and Hevner, 2013).The design science research process includes six steps: problem identification and motivation; definition of the Valuing prior learning objectives for a solution; design and development; demonstration; evaluation; and communication (Peffers et al., 2007, p. 46).Methodological rigor is achieved by "appropriately applying existing foundations and methodologies" (Hevner et al., 2004, p. 80) in design science.Subsequently, we describe the designed artifact and evaluate the artifact on a set of CV.

Artifact description
The foundation of the artifact to be applied is a comprehensive model of professional competences (see Table AI in Appendix).It merges the normative European competence perspective (Cheetham and Chivers, 1996;Le Deist and Winterton, 2005;Mulder et al., 2007) which focuses on what a person is able to do with the descriptive content model of the O*Net that provides a comprehensive and detailed taxonomy of occupational descriptors (Peterson et al., 2001).The underlying model contains 4 general competence dimensions and 32 subcompetences (Fahrenbach et al., 2019).To create the dictionary (see .In total, the dictionary contains 1,255 descriptors for the 32 sub-competences (average: 39.2 descriptors per sub-competence; minimum: 8; maximum: 213).In case new competences or skills arise (e.g.programming languages), they can be updated in the dictionary.The designed artifact for the assessment of professional competences receives a documentation of learning such as a repository of CV (in principle, it could receive any textual documentation of learning outcomes) and a dictionary of competences (which serves as a standard or reference point for the assessment) and returns a match between them.It has two main activities.In the first activity, the set of CV is processed to generate a bag of word representation for each of the CV.Natural Language Processing (Bird et al., 2009) is used for tokenization (i.e.splitting up sentences into words), removal of stop words (i.e.removal of words that do not meaningfully contribute such as "and" or "or"), lemmatization (i.e.removal on inflectional word endings and return of the dictionary form) and the subsequent collection of relevant words.The second activity receives the bag of relevant word representations for each CV and the dictionary of competences and performs a match of these sets.In this regard, we base our analysis on the classical vector space model (Salton et al., 1975) in which documents (such as CV or standards) are represented as vectors of terms.By matching, we mean that the artifact creates a term-document matrix.A collection of documents is then represented in such a term-document matrix which contains the number of occurrences each term appears in each document (Manning et al., 2008).In other words, the artifact counts the number of coincidence words for each CV, each competence and its sub-competences.The counts are returned as the output of the designed artifact.With the number of matches, it is possible to conduct further analysis and we propose an optional third step.In this step, a thorough statistical analysis can be conducted.It is, for example, possible to rank CV based on a particular competence or to provide an overall description of a CV among all competences.An overview of the designed artifact is given in Figure 1.In sum, we provided an overview on the designed artifact in this section.

Evaluation
In this section, we evaluate the designed artifact on a set of CV gathered from LinkedIn.Section 5.1 describes the data collection and the data set used.Section 5.2 outlines the processing of CV, including data preparation and pre-processing.Section 5.3 outlines the application of the designed artifact.In Section 5.4, we analyze the results of applying the designed artifact on an aggregated level.EJTD 44,2/3

Data collection
We used an openly available data set[3] from a blog post with 1,445 URLs to CV from LinkedIn as primary data source.LinkedIn is a social media platform on which users can create an online portfolio and headhunters or companies can search through these for recruiting purposes (Bastian et al., 2014).LinkedIn is increasingly used for employee selection and hiring (Roulin and Levashina, 2018) but also for research purposes.Although its use is hotly debated, first studies indicate good psychometric properties and validity of information reported on LinkedIn (Roulin and Levashina, 2018).We decided to use CV from this data set, as it is openly available on the Web and we could avoid possible biases, such as selection bias in data collection (Heckman, 1979).To scrape LinkedIn CV, we used an openly available webscraper [4].The original data set provides reference to 1,488 individuals; however, there are only 1,445 URL to LinkedIn CV reported, of which two entries were duplicates.Furthermore, we could not access 8 URLs from the 1,443 links because of deleted accounts or updated privacy settings.In total, we scraped 1,435 CV from the original data set.All CV were stored in JSON arrays and saved on local hard drives.The scraping of CV took place between August 10, 2018 and August 27, 2018.
Each scraped CV is organized in a similar way, consisting of six general categories."General information" includes the name, company, school and a short description or statement of purpose, "jobs" include the names of companies, job titles and job descriptions, "schools" include name of schools and degrees, "details" include personal websites and social media accounts, "skills" include self-assessed skills and endorsements from externals and "allskills" include a list of all reported skills separated by a comma (self-assessed and endorsed).A demographic overview of the scraped profiles is given in Table AIII in the Appendix.The demographic characteristics point at a skewed distribution with regard to gender, ethnicity and place of education.All of the 1,435 individuals work in 192 venture capital firms, either as an associate, principal or partner.Venture capital organizations "raise money from individuals and institutions for investment in early-stage businesses that offer high potential but high risk" (Sahlman, 1990, p. 473).According to literature, successful CEOs in venture capital firms rely on a set of characteristics, which can be summarized by two factors.Factor 1 is described by "general management ability" and factor 2 by "communication and interpersonal skills with a focus on execution and resoluteness" (Kaplan et al., 2012(Kaplan et al., , p. 1005)).Valuing prior learning

Processing curriculum vitae
The 1,435 scraped CV from LinkedIn, stored in the JSON file, served as input for the designed artifact, which has as first activity the processing of CV.We used Jupyter to convert the CV in JSON format to python objects for further processing and text mining.For preparing and pre-processing of the CV, we followed common text mining procedures (Debortoli et al., 2016;Kobayashi et al., 2018).For the text mining itself, we relied on the python library nltk (Bird et al., 2009).We identified the relevant stop words in English from this library (Rajaraman and Ullman, 2011) as well.For natural language pre-processing, we lemmatized the words (Debortoli et al., 2016).To do so, we imported the Word Net Lemmatizer library to extract the non-inflected (canonical or lemma) form of each word (Miller, 1995).We also applied tokenization, which allows to split up documents into sentences and sentences into words (Debortoli et al., 2016).In sum, we followed common text mining procedures to remove words that create noise in the data set.

Define assessment
As outlined above, the assessment of competences entails to compare previously identified and documented competences against a standard or point of reference (Cedefop, 2015;Bjørnåvold, 2000).To do so, the designed artifact counts occurrences of matching words between the repository of CV and the dictionary.The artifact evaluates each word in each of the CV against each word in the dictionary and saves the result in a vector.If there is a match between the CV and the dictionary, the result is stored as "1" in the vector, otherwise as "0."Based on these vectors, we summed up all matches per CV and sub-competence.This activity resulted in a data set with 1,435 rows, indicating the CV and 32 columns indicating the matches for each sub-competence.The subsequent analysis of the artifact is based on the already aggregated data on the level of 32 sub-competences.Using the LinkedIn URL and an ID, we can track each individual in the original and resulting data set.

Analyze assessment
Applying the artifact resulted in a data set with 67,522 matches between the 1,435 CV and the dictionary in total.The average number of matches per sub-competence is 2,110 (min: 18; max: 12,485; median: 926; SD: 2925).Table AIV in the Appendix shows the number of matches per sub-competence.
To get a better overview regarding which sub-competence matched frequently with the CV, the upper part of Figure 2 shows the ordered frequencies (y-axis) of matches per subcompetence dimension (x-axis).The upper part of Figure 2 also shows that CV matched to mostly four different sub-competences [MC9 (business management) accounted for 18.5 per cent, PC3 (suitability based on interests) for 13.1 per cent, DC1 (domain knowledge) for 12.1 per cent and MC5 (performing complex technical activities) for 9.3 per cent of all matches].As a result, 4 out of 32 sub-competence dimensions account for 60 per cent of the matches.
There are several explanations for these results, which are outlined below.First, individuals working as venture capitalists seem to rely on a comparable set of competences, mainly related to business management (as indicated by the prevalence of MC9).In the dictionary, MC9 (business management) was described with terms such as "business; business and management; business administration; accounting; human resource management; HRM; material resource management; organizations; organization; sales; marketing; sales and marketing; economics; office information; enterprise resource planning; organizing systems; economics; administration and management; strategic planning; resource allocation; human resource modelling; resource allocation; [. . .]."We interpret the frequency of matches with MC9 as closely related to the factor "general management EJTD 44,2/3 ability."In this regard, our findings are in line with previous research (Kaplan et al., 2012).Second, the large number of matches in PC3 (suitability based on interests) can be explained theoretically.The underlying theory of occupational interests by Holland (1997) defines, among others, "enterprising interests" which are described by entrepreneurial activities and interest in management.In this regard, the dictionary described PC3 with terms such as "entrepreneur; realistic; pragmatic; social; artistic; enterprise; convention; conventional; hands-on problems; investigation; investigate; problem-solving; thinking; design patterns; teaching; service; entrepreneurship; project management; leadership; business; risk taking; routines; procedures; [. ..]." Third, the large amount of matches in DC1 (domain knowledge) can be explained by the breadth and depth of the entry in the dictionary (213 descriptors).DC1 describes domain-specific knowledge and includes a wide variety of cross-occupational knowledge and school subjects such as "computers and electronics; engineering and technology; biology; psychology; arts and humanities; [. ..]," which explains the number of matches in this domain.Fourth, MC5 (performing complex technical activities) is strongly related to perform skilled activities in technical fields.MC5 is described in the dictionary with "technical activities; skilled activities; coordinated movements; movements; coordination; computers; computer; PC; software; hardware; tools; computer systems; programming; computer programming; data entry; process information; Coding; Code; functions; electronics; [. ..]." Individuals working as venture capitalists seem to have considerable technical experience (given the high number of matches in MC5).This finding can be explained by 27 per cent of individuals with an engineering degree in the original data set.The upper part of Figure 2 also indicates that other competence dimensions match considerably less.For SC8 (conflict management), the artifact returned only 18 matches (0.03 per cent).
The lower part of Figure 2 indicates the number of matches on the x-axis and the number of CV on the y-axis (also in Table AV of the Appendix).Figure 2 shows that a large amount of CV only match to a modest number with the dictionary (117 CV do not match at all, 262 CV show one to nine matches with the dictionary, and only a small number of CV show considerable matches with the dictionary).The average number of matches per CV is 47 (median: 37; SD: 43).This finding can be explained with the fact that many individuals provide only very few information about themselves on their LinkedIn CV (Gorbacheva et al., 2015;Roulin and Levashina, 2018).However, the automatized assessment of professional competences relies on a large repository of documents and textual data (Han and Lee, 2016).These can be written and narrative statements of purpose, a detailed Valuing prior learning description of previous work activities or any other textual document.Thus, the (very few) CV with the most matches have an extensive statement of purpose uploaded to their LinkedIn CV.

Interpretation and application of the artifact
As we set out to answer the research question How to automatize the assessment of professional competences within the VPL?, this section interprets the findings and outlines possible areas of application with two examples from the data set.We argue that a viable answer to the research question and objective can be the designed artifact.Starting from the premise that a person identifies his/her competences and documents them thoroughly in a (guided) self-assessment, the designed artifact is able to match the documents to a predefined standard (the comprehensive competence model).
While we analyzed results on the level of the whole data set in the last section, we take a look at two individual competence profiles in this section.On an individual level, the designed artifact results in a distinct competence profile, such as the green field in Figure 3(a)-(c).In Figure 3(b)-(d), the red line indicates a standard, against which the individual competence profile is assessed (in this case, the standard is of illustrative nature).
To demonstrate the application of the assessment for individual CV, and to assess the level of competence, we refer to common taxonomies which suggest six levels of competence (Anderson and Krathwohl, 2001;Dreyfus and Dreyfus, 1987) in which 1 represents a beginner and 6 represents an expert.To align the results to these taxonomies, we normalized the data set to a scale from 0 to 6 by using the following formula: As the data in Figure 3 is normalized to a scale from 0 to 6, we have to know the overall number of matches to interpret the distribution of assessed professional competences.Figure 3 shows the analysis of two individuals with the most matches in the data set.Figure 3 We introduce three different areas of application for the designed artifact.First, as outlined above, the artifact can be used for the assessment of competences when professional associations set standards for occupational fields on a certain level, such as within the EQF (EU, 2017).Second, the artifact can be used in organizations within human resources allocation or hiring decisions, when searching for a single best individual.For example, an organization defines competence requirements (Campion et al., 2011) (see Figure 3, red lines) and an individual applies with a certain competence profile (see Figure 3, green field).Using the designed artifact, it is possible to select a single best individual for a given standard of competence to point at learning fields in which the individual has to acquire additional competences to fit to the organization's competence requirement.Third, within human resource development, organizations can use the artifact to assess the competences of their employees and tailor specific learning interventions accordingly, based on the gap between the competence profile and a previously set standard (Swanson, 2001).

Limitations and conclusion
In this paper, we designed and proposed an artifact to assess professional competences of individuals to value their prior learning.The artifact applies a text mining algorithm to EJTD 44,2/3 make the assessment of professional competences more efficient and less tedious.The designed artifact can be a part in the VPL procedure.Subsequently, we present limitations and further research.
First, limitations concern the data set we used.All individuals in our data set work for venture capitalist firms within the USA.In this regard, the demographic and professional variety within the data set is limited as can be seen in Table AIII of the Appendix.Further research should apply the designed artifact to textual data from different professions and countries.Further research should, nevertheless, test the designed artifact with several documents of one person outlining the competences in different areas of professional and personal life.Also, many of the CV did not produce a match or produce only a very small number of matches between the repository and the competence model.In this regard, we Valuing prior learning support the call to use long and descriptive or narrative resumes as repository for text mining methods (Han and Lee, 2016).Long, descriptive and narrative CV may also support the assessment of competences in VPL.Second, limitations concern the designed artifact.The artifact should be only used for already identified and documented competences.It has to be pointed out that the artifact does not validate competences automatically, rather it may help in organizing documented professional competences for a reviewer or external assessor.In this regard, the resulting competence profiles may serve as a heuristic for further dialogue between an assessor and a candidate and can be a basis for a thorough psychological assessment.In this regard, the designed artifact is also not a behavioral assessment.To assess behavioral competences, i.e. whether a person is really able to perform a certain occupation, further behavioral simulations have to be conducted (Epstein, 2002).In other words, if we are to find out whether a person is really able to bake, i.e. possesses the necessary experience and tacit knowledge to do so, automatized assessments will only be of little help (Ribeiro and Collins, 2007).In this regard, the occurrences of matches between a body of documents and a standard may serve as an approximation toward competence and should be interpreted as first impression.Furthermore, the normalization of results to a scale from 0 to 6 may distort the results to some extent as the highest number of matches automatically gets assigned the value 6 and the lowest number a value near to 0. Further research should test the designed artifact with different procedures of normalization.Even though we build a dictionary based on a comprehensive model of professional competences (Peterson et al., 2001), further research should aim to use an even more detailed model of professional competence as a standard.

Figure 1 .
Figure 1.Activities of the designed artifact Figure 2. Upper part shows the number of matches per sub-competence dimension.Lower part shows the number of matches per CV (a)-(b) shows an individual with 258 matches in total (min: 0; max: 36 matches).Figure 3(c)-(d) shows a different individual but the same standard as in Figure 3(b).The CV of this individual showed 261 matches in total (min: 0; max: 36 matches).
Figure 3. Two individuals with their specific distribution of assessed competences normalized to a scale from 0 to 6 the "willingness and ability, as an individual personality, to understand, analyse and judge the development chances, requirements and limitations in the family, job and public life, to develop one's own skills as well as to decide on and develop life plans.It includes personal characteristics like independence, critical abilities, self-confidence, reliability, responsibility and awareness of duty, as well as professional and ethical values" (Le Deist and Winterton, 2005, p. 38) PC1

Table I .
Related work in the field Table AII in Appendix), we characterized each of the 32 sub-competences of the underlying model with the descriptors in version 22.2 of the O*Net content model[2]

Table AIV .
"willingness and ability, on the basis of subject-specific knowledge and skills, to carry out tasks and solve problems and to judge the results in a way that is goaloriented, appropriate, methodological and independent.General cognitive competence . . . the ability to think and act in an insightful and problem-solving way" (Le Deist and Winterton, Number of matches per sub-competence Matches per sub-competence dimension PC (N and %) SC (N and %) MC (N and %) DC (N and %)