A Knowledgebased Solution for Core Competence Evaluation inHumanCapital Intensive Companies
Simona Colucci(Politecnico di Bari, Bari, Italy andD.O.O.M. s.r.l., Matera, Italys.colucci@poliba.it)Eugenio Di Sciascio(Politecnico di Bari, Bari, Italy,disciascio@poliba.it)Francesco M. Donini(Universit`a della Tuscia, Viterbo, Italy,donini@unitus.it)
Abstract:
Determining ﬁelds of excellence in the knowhow of knowledge intensive companiesis often a crucial decisional process, aimed e.g., at identifying the competence to be strengthened or to invest on in a long term strategy. In this paper we propose a semanticbased approachfor automatic extraction of such a specializing knowledge, usually called Core Competence inknowledge management literature. The proposed approach exploits Description Logics as formalism for the representation of knowledge sources and implements novel reasoning services, inparticular
informative common subsumers
speciﬁcally devised for Core Competence evaluation.
Key Words:
Common Subsumers, Core Competence, Description Logics.
Category:
M.4, M.7, H.3.3
1 Introduction
Since ﬁrst investigations on the role of knowledge in humancapital intensive companies, the capability to focus on a signiﬁcant portion of the organizational knowhowhas been identiﬁed as a crucial asset for business success. Such a belief is at the basis of the socalled
Resourcebased Theory
of the ﬁrm [Wernerfelt, 1995], according towhich unique company capabilities should be exploited to achieve competitive advantage [Barney, 1991, L. Halawi and McCarthy, 2005, Meso and Smith, 2000]. In particular, the term Core Competence was introduced [Hamel and Prahalad, 1990] to denotesuch a specializing portion of organizational knowhow. It is intuitive that the hardnessof identifying such an intellectual capital increases with the size of the company andwith the ambiguity of company knowhow descriptions.In recent years we have been investigating knowledgebased approaches and solutions for a speciﬁc ﬁeld of knowledge management, namely skills and competencemanagement, in the framework of Description Logics (DLs)[Baader et al., 2002] and
Proceedings of IKNOW ’08 and IMEDIA '08Graz, Austria, September 35, 2008
semantic technologies, both exploiting classical inference services and introducing newones [Colucci et al., 2007b]. As it is nowadays wellknown, semanticbased technologiesaskforcompanyintellectualcapitaltobeunambiguouslydescribedinformalrepresentations, according to a shared vocabulary provided by ontologies modeling skills domain. In particular, our solutions employ DLs for knowledge representation and exploitDL reasoning services to infer new knowledge on the elicited descriptions. Obviously,once company knowhow has been formally represented in a common knowledge basein terms of individual proﬁle descriptions and knowhow, such a repository could beexploited to extract the most characterizing portion of company intellectual capital,
i.e.
company Core Competence. Nevertheless this is easier said than done, as wellknownreasoning services fail to provide such information. As we show later on, the apparently best suited inference service for the above task, the
Least Common Subsumer
[Cohen et al., 1992], shows clear limits.In this paper we therefore propose new nonstandard inference services on collections of individual proﬁle descriptions formalized in DL, for the automatic extractionof company Core Competence. Such speciﬁcally developed reasoning services are introduced in Section 2. Two different Core Competence evaluation approaches are thendetailed in Section 3, before closing the paper with conclusions.
2 New Services Deﬁnition
In the automated Core Competence extraction we propose, we refer to
ALN
(Attributive Language with Number Restrictions) for formally describing knowledge sources of a company.
ALN
provides a limited set of constructs, which allow for describing theknowledge domain by combining the basic elements of a DL, namely
concept names
,representing objects of the domain —
i.e.
ProductionManagement,AssetAllocation,Creativeness
and
AssetManager
— and
role names
, representing possible binary relationships among concepts,
i.e.
knows,isAbleTo
. Every DL includes two concepts,
and
⊥
, representing a concept interpreted by the whole domainand by an empty set, respectively.
ALN
allows also for
qualiﬁed universal restrictions
—
i.e.
∀
knows
.
AssetAllocation
denotes an advanced knowledge in Asset Allocation — and
number restrictions
—
i.e.
(
≥
3
knows
)
,
(
≤
2
isAbleTo
)
denote thepossession of at least three skills and at most two abilities — over roles. By using suchconstructs it is possible to detail concept
inclusions
and
deﬁnitions
, which constitutethe intensional knowledge of a DL system, what is called a
TBox
in DL and
ontology
in knowledge representation. For example the inclusion
AssetManager
Manager
asserts that the set of asset managers in the domain is included in the one of managers;theconceptdeﬁnition
AssetManager
≡
Manager
∀
knows
.
AssetAllocation
givesinstead to managers endowed with Asset Allocation knowledge the name of Asset Manager, like the deﬁnition
Manager
≡ ∀
knows
.
Management
gives the name manager to subsets of domain possessing Management knowledge.
260
S. Colucci, E. Di Sciascio, F. M. Donini: A ...
Every DL allows for basic reasoning services inferring new knowledge from thedescriptions elicited in the TBox; in particular
satisﬁability
and
subsumption
are deﬁned for every DL. In a nutshell, satisﬁability checks for internal coherency of conceptdescriptions, evaluating the consistency of elicited information; subsumption checksinstead whether a concept description is more generic than another one. Formally, subsumption is deﬁned as follows, with respect to a domain interpretation function
I
:
Deﬁnition 1 (Subsumption)
Given two concept descriptions
C
and
D
and a TBox
T
in a DL
L
, we say that
D
subsumes
C
w.r.t.
T
if for every model of
T
,
C
I
⊂
D
I
. Wewrite
C
T
D
, or simply
C
D
if we assume an empty TBox.Having a collection of concept descriptions in a DL
L
, the problem of determining the Least Common Subsumer(LCS) of the collection has been proposed by Cohen,Borgida and Hirsh [Cohen et al., 1992] as a nonstandard reasoning service. By definition, the LCS of a collection of concept descriptions represents the most speciﬁcconcept description subsuming all of the elements of the collection. Formally, we recallthe following deﬁnition:
Deﬁnition 2 (LCS,[Cohen and Hirsh, 1994])
Let
C
1
,...,C
n
be
n
concepts in a DL
L
. An
LCS
(
C
1
,...,C
n
)
, is a concept
E
in
L
such that the following conditions hold:
(i)
C
i
E
for
i
= 1
,...,n
(ii)
E
is the least
L
concept satisfying (i),
i.e.,
, if
E
is an
L
concept satisfying
C
i
E
for all
i
= 1
,...,n
, then
E
E
If the collection contains employee proﬁle descriptions, as in our reference scenario,the LCS represents the competence shared by all the employees in the collection. Sucha concept description is a good candidate for determining the Core Competence of thecompany at a ﬁrst sight. Nevertheless the need for the LCS to subsume each conceptin the collection causes its corresponding description to be too generic in most cases:if a competence has to be shared by the whole company personnel it needs to be quitegeneric. As a toy example, consider a small company in which only the following threeemployees work:–
Nick
:
AssetManager
∀
isAbleTo
.
Creativeness
–
Frank
:
∀
knows
.
AssetAllocation
∀
isAbleTo
.
Creativeness
–
Robert
:
Engineer
∀
isAbleTo
.
Creativeness
The only LCS of such a collection is Creativeness ability, which might result a notmuch signiﬁcant knowledge. If we instead give up such a full skill coverage and acceptthe assumption that Core Competence needs to be possessed by a signiﬁcant portion of company personnel, more interesting results can be achieved. Obviously the requireddegree of coverage may be set by company management. To this aim, we propose andintroduce new reasoning services.
Deﬁnition 3 (kCS)
Let
C
1
,...,C
n
be
n
concepts in a DL
L
, and let be
k < n
. A
k
Common Subsumer
(
k
CS) of
C
1
,...,C
n
is a concept
D
such that
D
is an LCS of
k
concepts among
C
1
,...,C
n
.If the example company management decides that
2
/
3
of the employees have to
S. Colucci, E. Di Sciascio, F. M. Donini: A ...
261
possess some knowledge to consider it part of the Core Competence, Asset Allocationknowledge represents a commonality between two employees (according to the deﬁnitions at the beginning of the section) and then a Core Competence. Of course alsoCreativeness ability is a
k
−
CS
of the collection, but it does not add any informativecontent to the LCS: for this reason we distinguish in the following kCommon Subsumers adding informative content to LCS.
Deﬁnition 4 (IkCS)
Let
C
1
,...,C
n
be
n
concepts in a DL
L
, and let
k < n
. An
Informative
k
Common Subsumer
(IkCS) of
C
1
,...,C
n
is a
k
CS
E
such that
E
is strictlysubsumed by
LCS
(
C
1
,...,C
n
)
.Among possible
IkCSs
, some are characterized by maximal cardinality of the setof subsumed concepts: in our example scenario, if we set
k
= 3
Asset Allocationknowledge stops being a
k
−
CS
and the only common subsumer is Creativeness ability,which is not informative by deﬁnition. We deﬁne in the following concepts like AssetAllocation as best informative common subsumers (with
k
= 2
).
Deﬁnition 5 (BICS)
Let
C
1
,...,C
n
be
n
concepts in a DL
L
. A
Best InformativeCommon Subsumer
(BICS) of
C
1
,...,C
n
is a concept
B
such that
B
is an Informative
k
CS for
C
1
,...,C
n
, and for every
k < j
≤
n
every
j
CS is not informative.For collections whose LCS is equivalent to the universal concept
, the followingdeﬁnition makes also sense:
Deﬁnition6(BCS)
Let
C
1
,...,C
n
be
n
conceptsinaDL
L
.A
BestCommonSubsumer
(BCS) of
C
1
,...,C
n
is a concept
S
such that
S
is a
k
CS for
C
1
,...,C
n
, and for every
k < j
≤
n
every
j
CS
≡
.Consider for example a new employee :–
Fred
=
Manager
∀
knows
.
ProductionManagement
.The only LCS of the collection including the four employees is the universal concept. On the contrary, for
k
= 3
we have Creativeness ability as kcommon subsumer,which is informative w.r.t. the LCS (it is equivalent to the universal concept) and best:if we add one unit to
k
the kCS reverts to the universal concept.
3 Solutions to Core Competence Evaluation Problem
In this paper we provide two processes for Core Competence evaluation: the ﬁrst oneexploits the services introduced in Section 2 to discover unknown ﬁelds of excellenceof the company; the second one checks for the possession of a list of known target competencies by a signiﬁcant portion of company personnel and explains how to reach thetarget in case the check fails.Both of the approaches ask for the concepts to be written in
Concept Components
according to the following rules. If
C
is a concept description in a DL
L
, with
C
written ina conjunction
C
1
···
C
m
, the
Concept Components
of
C
are deﬁned as follows: if
C
j
, with
j
= 1
...,m
is either a concept name or a negated concept name or a numberrestriction, then
C
j
is a
Concept Component
of
C
; if
C
j
=
∀
R
.
E
, with
j
= 1
...,m
,
262
S. Colucci, E. Di Sciascio, F. M. Donini: A ...
then
∀
R
.
E
k
is a
Concept Component
of of
C
, for each
E
k
concept component of
E
.The deﬁnition of
Subsumers Matrix
in the following is preliminary to both processes of Core Competence evaluation.
Deﬁnition 7 (Subsumers Matrix)
Let
C
1
,...,C
n
be a collection of concept descriptions
C
i
in a Description Logic
L
and let
D
j
∈ {
D
1
,...,D
m
}
be the
Concept Com ponents
deriving from a set of concepts. We deﬁne the
Subsumers Matrix
S
= (
s
ij
)
,with
i
= 1
...n
and
j
= 1
...m
, such that
s
ij
= 1
if the component
D
j
subsumes
C
i
,and
s
ij
= 0
if the component
D
j
does not subsume
C
i
.Referring to Subsumers Matrix, we deﬁne:
Concept Component Signature
(
sig
D
j
): set of indeces of concepts
C
1
,...,C
n
subsumed by
D
j
; observe that
sig
D
j
⊆{
1
,...,n
}
;
Concept Component Cardinality
(
T
D
j
): cardinality of
sig
D
j
, that is, how many concepts among
C
1
,...,C
n
are subsumed by
D
j
. Such a number is
ni
=1
s
ij
;
Maximum Concept Component Cardinality
(
M
S
): maximum among all conceptcomponent cardinalities, that is,
M
S
=
max
{
T
D
1
,...T
D
m
}
;
Second Maximum Concept Component Cardinality
(
PM
S
): maximum among thecardinalities of concept components not subsuming all
n
concepts in the collection(
PM
S
=
max
{
T
D
j

T
D
j
< n
}
); by deﬁnition
PM
S
< n
;
Common Signature Class
(
sig
Dj
): concept formed by the conjunction of all conceptcomponents whose signature contains
D
j
:
{
D
h

sig
D
j
⊆
sig
D
h
}
3.1 Core Competence ExtractionDeﬁnition 8 (Collection Subsumers Matrix)
Let
C
1
,...,C
n
be a collection of concept descriptions
C
i
in a Description Logic
L
. We deﬁne the
Collection SubsumersMatrix
as a Subsumers Matrix in which
D
j
∈{
D
1
,...,D
m
}
are the concept components deriving from all concepts in the collection.In the following we deﬁne, with respect to a collection of concept descriptions,
BCS
the set of BCSs,
BICS
the set of BICSs,
ICS
k
the set of IkCSs, given
k < n
and
CS
k
the set of kCSs, given
k < n
. In [Colucci et al., 2008] we proposed Algorithm 1 determining the sets
BICS
,
CS
k
,
ICS
k
,
BCS
of a collection
{
C
1
,...,C
n
}
of conceptsin
ALN
, whose Subsumers Matrix is given as input. In order to understand the rationale of the proposed algorithm, consider the company with the four employees (Nick,Frank, Robert and Fred) in the tiny example in Section 2. The concept componentscoming from the collection of employees are:
D
1
=
∀
knows
.
Management
,
D
2
=
∀
knows
.
AssetAllocation
,
D
3
=
∀
isAbleTo
.
Creativeness
,
D
4
=
Engineer
,
D
5
=
∀
knows
.
ProductionManagement
. The collection subsumers matrix is shown in thelefthand side of Figure 1. If
k
= 2
, the only components with cardinality at least equalto
k
are
D
1
,
D
2
and
D
3
and then their common class signature is added to the set
CS
k
(line 3), which contains the
k
−
CSs D
1
,
D
2
D
3
and
D
3
. The check in line 4 resultstrue for all of the three components, given that the only concept subsuming the fouremployee proﬁles is the universal concept; the three concepts in
CS
k
are then added to
S. Colucci, E. Di Sciascio, F. M. Donini: A ...
263