=Paper=
{{Paper
|id=Vol-2542/MOD-KI5
|storemode=property
|title=Towards a Case-Based Decision Support System for Recruiting Processes using T-Shapes
|pdfUrl=https://ceur-ws.org/Vol-2542/MOD-KI5.pdf
|volume=Vol-2542
|authors=Jakob Michael Schoenborn, Pascal Reuss, Carsten Wenzel, Klaus-Dieter Althoff,Pascal Reuss,Carsten Wenzel,Klaus-Dieter Althoff
|dblpUrl=https://dblp.org/rec/conf/modellierung/SchoenbornRWARW20
}}
==Towards a Case-Based Decision Support System for Recruiting Processes using T-Shapes==
https://ceur-ws.org/Vol-2542/MOD-KI5.pdf
Joint Proceedings of Modellierung 2020 Short, Workshop and Tools & Demo Papers
Workshop on Models in AI 165
Towards a Case-Based Decision Support System for
Recruiting Processes using T-Shapes
Jakob Michael Schoenborn1,3, Pascal Reuss1,3, Carsten Wenzel,2 Klaus-Dieter Althoff1, 3
Abstract: Modeling the individual set of skills a person possesses is a difficult task. However, most
people seek to find a job that suits them best, despite being unaware of which skills she inherits and
what defines herself as a person. Recruiters are knowledgeable in terms of extracting these information
by an informal phone call or when carrying out an assessment center. However, we suggest to set
a step beforehand and provide a case-based decision support system by using T-Shapes as a model
for matching personal skills with job requirements. We provide a first overview on how to structure
and model the four knowledge containers of case-based reasoning and how we use these knowledge
containers to obtain the most similar T-Shape (situation). This T-Shape can then be reused by the
recruiter, providing her with all necessary information at a first glance.
Keywords: Case-based Reasoning, Decision Support, T-Shapes, Recruiting
1 Introduction
The process of finding the most suitable occupation can often be daunting, given the huge
range of different possible career paths. Especially for students who are about to finish their
compulsory education, it is often unclear to them which occupation suits them the best,
given their individual interests and skills. Schools and universities are working together
with career services to address this challenge, for example by carrying out assessment
centers. However, only 17 % of 2010 to 2016 graduates of the Purdue University (USA)
report that these were helpful as the Gallup-Purdue index study shows [Ga16]. On a similar
note, only 15 % of German employees report an emotional commitment to their working
places [Ni18], resulting in 85 % employees to be likely to change their current occupation
given the right opportunity. These different audiences are the targets for the recruiting
domain in general. For the last years, and especially in the STEMM (Science, Technology,
Engineering, Mathematics, and Medicine) domains, the popularity of headhunting increased.
The majority of large German companies rely on headhunting to find the right applicants
for the right position [FH17, Mi00]. Headhunting itself is not a new method in terms of
recruiting. However, due to the rising popularity and especially possible range of platforms
like XING4 and LinkedIn5 (among others), individuals are able to present themselves and
1 Intelligent Information Systems Lab, University of Hildesheim, Hildesheim, Germany
2 Department of Information Systems and Enterprise Modelling, University of Hildesheim, Hildesheim, Germany
3 Competence Center Case Based Reasoning, German Center for Artificial Intelligence, Kaiserslautern, Germany
4 URL: https://www.xing.com
5 URL: https://www.linkedin.com/
Copyright © 2020 for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
166 Schoenborn et al.
their skills to the publicity, providing the headhunters with the required initial information
[FH17].
In the recent literature, there is an uprising interest in automating this process or at least
support the so-called screening phase of the recruitment process and to use the digitization
to decrease the requirement of paper application management. Espenakk et al. presented
a lazy learning approach by considering user profiles as feature vectors using attributes
such as occupation, skill, language among others combined with case-based reasoning
(CBR) [EKK19]. Earlier stages of recruitment using CBR has been contributed by Siraj
et al. [Si11] and Rafter et al. [RBS00]. Another feature vector approach by using machine
learning techniques can be seen by Faliagka et al., solving the candidate ranking problem
using LinkedIn profiles and linguistic analysis [Fa12].
The mapping between the skills of a possible applicant to the current open job position
is still challenging. To tackle this challenge, we suggest the usage of T-Shapes. These
T-Shapes are considering two different dimensions: vertical for skills with rather deepened
knowledge; and horizontal for broader skills, for example, where the applicant has gathered
experiences but would not be considered as an expert (unlike vertical skills). An example of
T-Shapes can be seen in Fig. 2. T-Shapes have been introduced by David Guest in 1991
[Gu91], but received more attention after Tim Brown, CEO of IDEO (a global design
company) mentioned using these models. The contents (attributes) of the T-Shape can either
be manually inserted by the person herself (the recruiter) or can be gathered in form of the
result of an assessment center. Some larger companies require to pass an assessment center
before the application process itself begins. The results of these assessment centers aim to
show strengths and weaknesses of considered applicants, which in turn could be transferred
into T-Shapes. We propose to use these models as case structures for the methodology of
CBR.
Case-based reasoning uses experiences of past situations, stores them in a case base, and
reuses the stored experiences for future, similar problems. The so-called CBR-cycle (see
Fig. 1) consists of four steps: Retrieve, Reuse, Revise, Retain and is accompanied by four
knowledge containers: similarity measurements, vocabulary, case base and adaptation
knowledge [Ba12, Ri03].
The first step, retrieve, uses the input situation of a user - the current situation. Based on
the attributes and their values which describe the situation, the best matching stored case
from the case base will be retrieved, using similarity measurements. During the second step,
the retrieved case will be proposed to be reused, thus, provided to the user. The user might
accept this solution or reject it due to various reasons, e.g., the similarity between the input
and the retrieved situation is not high enough. Either way will trigger the next step, revise,
where the proposed case or the knowledge containers can be adjusted to provide a tested or
repaired case to the user. During the last step, retain, the knowledge engineer may decide to
store the case for further usage, or to store adjustments of the similarity measurements to
Case-based Decision Support for Recruiting Processes using T-Shapes 167
Fig. 1: Case-based reasoning (CBR)-cycle by A. Aamodt and E. Plaza [AP94].
prevent the case base from growing too large, running into a risk of redundancy or lowered
maintainability.
2 Using T-Shapes for domain modeling in CBR
As mentioned above, we suggest the combination of T-Shapes and CBR. Fig. 2 depicts an
exemplary T-Shape of a person considering herself as very knowledgeable in case-based
reasoning. Nevertheless, other areas such as penetration testing and cryptography are also
a feature of the skill set of this person, for example by possessing certain certificates.
However, this model would suggest that the person considers her own main interest and main
contribution in the area of case-based reasoning. Regarding the attribute-value pairs and the
suggested granularity as well as the scope of the used attributes, we consider T-Shapes as a
simplistic depiction of a certain real world application which has been developed to serve a
concrete purpose. It reduces the complexity of the original domain to generalise information
in a way which is simple to understand and, thus, enables different users to communicate
about complex problems. T-Shapes are an example of a model due to its simplified depiction
of a persons set of skills. Using this intentionally broad definition of a T-Shape, the emphasis
is on simplifying the very complex domain of competence management. Dealing with every
facet of competence management is a complex task whereas the possibility to apply a first
168 Schoenborn et al.
Fig. 2: Examplary T-Shape model showing the skill set of a person.
filter and then moving further into detail is expected to be beneficial regarding time until a
final decision (not) to recruit has been made. We use this simplification to store T-Shapes as
cases in the case base to ensure an efficient retrieval process in terms of computational time
while still considering complex problems by using the knowledge containers.
For each attribute in the horizontal domain, the width of each skill represents the knowledge
level of the corresponding skill: the larger the width, the more deepened the skill. This
opens up the question on the model itself: Would not only one dimension be enough
whereas the largest container represents in which skill the user can be considered as an
expert? Indeed, there are multiple different variations of shapes, such as the I-Shape (which
would represent the mentioned idea), the π-Shape, as well as the M-Shape (which suggests
two/three depth-skills, which can be seen analogous to the top 3 skills a premium member
can name on XING). However, the depth skill does not necessarily needs to be limited
to one skill as suggested in Fig. 2 but rather can split up into multiple sub-domains, for
example, CBR regarding maintenance, knowledge management or gaming. Nevertheless, it
is important to list it as one attribute to emphasize the main interest area of an applicant and
to categorize appropriately during the first fielding step.
Skills can be divided into different categories, for example: Technical skills and social skills.
The former are usually easier to quantify in terms of certificates, number of completed
projects, and years of experience. However, as also often listed in job positions, social
skills such as being able to work in a team but also to be self-reliant are also required.
Nevertheless, experience made by hobbies or side-jobs are important criteria of a persons
personality and thus need to be contained in the T-Shape model. An exemplary approach
can be seen by the work of Wilkin et al. who mapped activities like “has supervised the
work of others” to scores for the five attributes “Intelligence, Interpersonal, Leadership,
Motivation, Usefulness” whereas Leadership scored as the highest [WC12]. By comparing
the complexity of soft skills in contrast to hard skills, there is a higher complexity caused by
dependencies between multiple soft skills recognizable. As an example we consider the skill
of problem solving thinking: on the one hand, this can be regarded as an independent skill
but also requires a lot of other skills, such as the ability to grasp complex interrelationships,
the ability to plan, and the ability to think and act in a complex way. By using completion
Case-based Decision Support for Recruiting Processes using T-Shapes 169
rules, however, we are able to complete requests in cases where more complex soft skills
require lower soft skills.
The IT skill pyramid, for example, is used to classify the dependency on skills [PT17].
It divides skills into three tiers: The lowest tier describes the fundamental knowledge
needed to start a career in the IT field, including for example the above mentioned problem
solving thinking. Additionally, general topics like the knowledge about the usage of certain
established technologies are covered in the first tier. Tier two contains all the technical skills,
knowledge and abilities required to work in the IT field, such as an understanding of the
software and hardware and the underlying architecture. The third level describes the skills
that require a comprehensive overview and knowledge of the IT sector [PT17]. Hard skills
are mostly unique, such as graduation, university degrees and certificates.
Within T-Shapes, the knowledge is available in an unstructured form. In order to use this
knowledge in CBR, we have to transfer it into the structured form of CBR in hindsight
of the four knowledge containers as introduced. We propose the following structure of a
case C containing a vertical, most weighted, skill V, a set of horizontal skills H containing
technical- and soft-skills, and a job position P as solution to the case:
C = (V, {H}, P) (1)
As soon as each attribute has been structured or mapped into a measurable way by
using similarity measurements, for example, by numeric functions, symbolic matrix
representations, Boolean comparisons or taxonomies of social skills (as described above
regarding the matching of soft-skills), we can fill the case base with cases of individual
persons and another case base with cases of job advertisements to compare against. Within
the vocabulary, the structure of the used model is defined as well as all values that may
belong to these attributes. Additionally, it is important to store synonyms to provide a proper
matching between skills and solutions of the case, for example, by using a thesaurus with
synonymous of job positions6.
The knowledge container of adaptation knowledge contains information on how a T-Shape
can be used to solve a query case. As a concrete example, a person with a master degree
in meteorology has a high similarity to a data analyst due to her personality profile. The
skills of recognizing patterns and dealing with large amounts of data are among the top
qualifications for both persons and positions. From this point, it could be deduced that a
meteorologist might also be suitable for the position of a data analyst. The reverse conclusion,
however, is not always granted. Due to the lack of domain knowledge about meteorology, a
trained data analyst will not be able to make reliable predictions about the weather forecast.
In the case base, known T-Shapes with their corresponding job descriptions are stored. All
existing attributes from the job description as well as created T-Shapes are converted into
attribute-value pairs. This differentiation and way of modelling allows us the comparison
6 e.g., as provided by John Carty: https://github.com/johnpcarty/Thesaurus-of-Job-Titles
170 Schoenborn et al.
between two T-Shapes, thus, enabling to highlight the most similarity between those two.
Additionally, we can also provide the most dissimilar case. By enabling the user to provide
feedback, we may be able to adjust the similarity measurements (as one of the knowledge
containers) to increase the accuracy in general and, in the best case, create completion rules
to also enable the system to provide a good solution, despite a barely filled T-Shape.
3 Conclusion and Outlook
We provided a first glance on our approach to support recruiters and individual persons
by finding an accurate job position. Due to the early stage of progress, it should be clearly
noted that an improvement of our domain model is currently in progress. A far-reaching
vocabulary needs to be created automatically with the help of text mining methods. After
the refinement of the vocabulary, the newly acquired attribute values have to be entered
into our similarity measures. This can be done with a framework like FEATURE-TAK
(Framework for Extraction, Analysis, and Transformation of UnstructuREd Textual Aircraft
Knowledge) that extends knowledge containers of a CBR system in a semi-automated way.
The framework was originally developed to analyze, extract, and transform unstructured or
semi-structured textual knowledge from failure descriptions in the aviation domain. The
extracted knowledge items then were used to extend the vocabulary, the similarity measures,
and the case bases of a decision support system for diagnosis and maintenance of Airbus
aircraft [Re18]. While it was developed for textual aircraft knowledge, the framework can
be extended and configured to be used for other textual knowledge, too. The creation of a
further prototype and its evaluation is also planned to gain feedback of the application and
to measure the users acceptance in terms of allowing algorithmic decision making taking
place in crucial decisions of an individual person regarding her individual development and
career path.
Case-based Decision Support for Recruiting Processes using T-Shapes 171
Bibliography
[AP94] Aamodt, Agnar; Plaza, Enric: Case-based Reasoning: Foundational Issues, Methodological
Variations, and System Approaches. AI Commun., 7(1):39–59, March 1994.
[Ba12] Bach, Kerstin: Knowledge Acquisition for Case-Based Reasoning Systems. PhD thesis,
University of Hildesheim, 2012.
[EKK19] Espenakk, Erik; Knalstad, Magnus Johan; Kofod-Petersen, Anders: Lazy Learned Screening
for Efficient Recruitment. In: Case-Based Reasoning Research and Development - 27th
International Conference, ICCBR 2019, Otzenhausen, Germany, September 8-12, 2019,
Proceedings. pp. 64–78, 2019.
[Fa12] Faliagka, Evanthia; Ramantas, Kostas; Tsakalidis, Athanasios; Tzimas, Giannis: Application
of Machine Learning Algorithms to an online Recruitment System. 01 2012.
[FH17] Flöthmann, Christoph; Hoberg, Kai: Career Patterns of Supply Chain Executives: An
Optimal Matching Analysis. Journal of Business Logistics, 38, 2017.
[Ga16] Gallup: Great Jobs. Great Lives. The Value of Career Services, Inclusive Experiences and
Mentorship for College Graduates. 2016. Last validation: 11/28/2019.
[Gu91] Guest, David E.: Personnel Management: The End of Orthodoxy? British Journal of
Industrial Relations, 29(2):149–175, 1991.
[Mi00] Mileham, Patrick: The ‘science’ of headhunting. Drug Discovery Today, 5(4):161 – 163,
2000.
[Ni18] Nink, Marco: Engagement Index Germany 2018. 2018.
[PT17] Patacsil, Frederick; Tablatin, Christine: Exploring the importance of soft and hard skills as
perceived by IT internship students and industry: A gap analysis. Journal of Technology
and Science Education, 7:347, 09 2017.
[RBS00] Rafter, Rachael; Bradley, Keith; Smyth, Barry: Personalised Retrieval for Online Recruit-
ment Services. 04 2000.
[Re18] Reuss, Pascal; Stram, Rotem; Althoff, Klaus-Dieter; Henkel, Wolfram; Henning, Frieder:
Knowledge Engineering for Decision Support on Diagnosis andMaintenance in the Aircraft
Domain. In (Nalepa, Grzegorz J.; Baumeister, Joachim, eds): Synergies Between Knowledge
Engineering and Software Engineering, volume 626. Springer, pp. 173–196, advances in
intelligent systems and computing edition, 2018.
[Ri03] Richter, Michael M.: Fallbasiertes Schließen. Görz, Günther; Rollinger, Claus-Rainer;
Schneeberger, Josef (Hrsg.): Handbuch der Künstlichen Intelligenz, 4:407–430, 2003.
[Si11] Siraj, Fadzilah; Mustafa, Norhashimah; Haris, Megat; Yusof, Shahrin; Muhammad Ashraq,
Salahuddin; Hasan, Md Rajib: Pre-selection of Recruitment Candidates Using Case
Based Reasoning. Computational Intelligence, Modelling and Simulation, International
Conference on, 0:84–90, 09 2011.
[WC12] Wilkin, Christa L.; Connelly, Catherine E.: Do I Look Like Someone Who Cares? Recruiters’
Ratings of Applicants’ Paid and Volunteer Experience. International Journal of Selection
and Assessment, 20(3):308–318, 2012.