Decision Tree Transformation for Knowledge Warehousing
Rim Ayadi
1,2
, Yasser Hachaichi
1,2
, Saleh Alshomrani
3
and Jamel Feki
1,3
1
Multimedia, InfoRmation Systems and Advanced, Computing Laboratory, Sakiet Ezzit, Tunisia
2
University of Sfax, Sfax, Tunisia
3
Faculty of Computing and IT, University of Jeddah, Jeddah, Saudi Arabia
Keywords:
Heterogeneous Knowledge, Knowledge Harmonization, Knowledge Warehouse, Data Mining, MOT, Meta-
models, Transformation Rules, Source Model, Target Model.
Abstract:
Explicit knowledge extracted from data, formalized tacit knowledge from experts or even knowledge existing
in business sources may be in several heterogeneous formal representations and structures: as rules, models,
functions, etc. However, a knowledge warehouse should solve this structural heterogeneity before storing
knowledge. This requires specific tasks of harmonizing. This paper first presents our proposed definition and
architecture of a knowledge warehouse, and then presents some languages for knowledge representations as
particular the MOT (Modeling with Object Types) language. In addition, we suggest a metamodel for the
MOT, and a metamodel for the explicit knowledge obtained using decision trees technique. As we aim to
represent knowledge having different modeling formalisms into MOT, as a unified model, then we suggest
a set of transformation rules that assure the move from the decision tree source model into the MOT target
model. This work is still in progress, it is currently completed with tranformations for additional.
1 INTRODUCTION
In recent years, there has been a growing interest in
knowledge both for personal and professional life.
Thus, many companies have devoted time and efforts
to discover relevant information either on themselves
or touching their customers and competitors.
Hence, knowledge is a power means which helps
companies to improve their decision making process
and to better manage competitive situations. Further-
more, the evolution of the company in its environ-
ment, in addition to its external challenges, pushes it
to capitalize knowledge in order to be more efficient
and then to increase profitability. From this situation
was born the need to gather knowledge into a repos-
itory commonly called knowledge warehouse (KW).
A KW has a specific architecture providing the neces-
sary infrastructure to extract, store and handle explicit
knowledge (Nemati et al., 2002).
In fact, explicit knowledge existing in the com-
pany and used in decision-making process has ini-
tially heterogeneous formats. In order to put different
types of knowledge together in the KW, we need to
harmonize and homogenize. A means to do this we
have elected the usage of the semi-formal graphical
language for knowledge modeling called MOT (Mod-
eling with Object Types (Paquette, 2002)). The MOT
language relies on a set of typed knowledge units and
links allowing homogeneous modeling of heteroge-
neous explicit knowledge. Our objective is to repre-
sent knowledge having different models from source
models (i.e., decision tree, association rules, cluster-
ing) into the unified target model which is the MOT.
In this paper we focus on the transformation of deci-
sion tree knowledge into the MOT model.
The rest of this paper is organized as follows: Sec-
tion 2 introduces the context of this work and our def-
inition and architecture for a KW. Section 3 presents
some languages for knowledge representations and
then presents the MOT language and its metamodel.
Section 4 proposes the metamodel for models of ex-
plicit knowledge obtained by decision tree technique.
In addition, it defines transformation rules from this
metamodel into the MOT metamodel. Finally, sec-
tion 5 concludes the paper and gives some relevant
perspectives.
2 CONTEXT
For business intelligence purposes, we suggest to
gather knowledge (i.e., tacit and explicit knowledge)
616
Ayadi R., Hachaichi Y., Alshomrani S. and Feki J..
Decision Tree Transformation for Knowledge Warehousing.
DOI: 10.5220/0005380506160623
In Proceedings of the 17th International Conference on Enterprise Information Systems (ICEIS-2015), pages 616-623
ISBN: 978-989-758-096-3
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
into a knowledge warehouse (KW) for which we give
our definition and architecture. Our suggestions rely
on several works of the literature (Kerschberg, 2001),
(Dymond, 2002), (Nemati et al., 2002), (Qing-lan and
Zhi-jun, 2009) and (Irfan and uddin Shaikh, 2010)
that have tried to introduce the KW concept and pro-
pose a basic architecture for a KW. However, these
authors did not provide a precise definition of a KW
and did not detailed its architecture. Indeed, in their
works, a KW is considered as a data warehouse (Dy-
mond, 2002) (Nemati et al., 2002) described with
three layers: capture, storage and access to content.
For example, (Dymond, 2002) considers that the stor-
age structure is referred to as a knowledge base and
is constructed as a tree with objects at the nodes. Ob-
jects are packages containing data in ”attributes” and
blocks of program code in ”methods”.
In addition, most of proposed architectures are
presented in an abstract way and are composed of
three layers (Data source layer, Knowledge manage-
ment layer and Knowledge presentation layer) (Ker-
schberg, 2001) (Dymond, 2002) (Nemati et al., 2002)
(Irfan and uddin Shaikh, 2010). These architectures i)
do not present the interaction between decision mak-
ers and the KW (i.e., How to exploit and update stored
knowledge?) and ii) do not specify the sources of
knowledge: are they exist in business sources of com-
panies, are they extracted from existing data or are
they acquired through the formulation of tacit knowl-
edge from individuals working in these companies.
Furthermore, in the absence of a complete defini-
tion of KW, we rely on the following extracts so that
we can give our definition of a KW:
The knowledge can be found in multiple repos-
itories under multiple heterogeneous representa-
tion (Kerschberg, 2001) such as databases, docu-
ments, computer programs and even in people’s
heads (Dymond, 2002);
The knowledge extracted and stored in the KW
should be explicit (Kerschberg, 2001) (Nemati
et al., 2002), i.e., formal and systematic in order to
be easily communicated and shared (Nonaka and
Takeuchi, 1995);
The KW can be used as a clearinghouse of knowl-
edge to be used throughout the organization by the
employees to support their knowledge intensive
decision-making activities (Nemati et al., 2002).
Based on these three extracts that we consider
incomplete and often reflecting a particular view-
point, we provide our definition. In our viewpoint,
a Knowledge Warehouse gathers explicit knowledge
that may come from multiple sources having heteroge-
neous formats and relating to several business activ-
ities within a domain. This knowledge is unified and
integrated in order to support an intelligent decision-
making process (Ayadi et al., 2013).
Based on this definition, we draw multi-layer core
architecture for a KW (cf., Figure 1). In this archi-
tecture, we find the following tasks located at three
layers (Ayadi et al., 2013).
Data Acquisition: This step is interested in col-
lecting the initial data from companies belonging to
one or more sectors.
Knowledge Extraction: Extracts the hidden
knowledge from the initial data by using multiple
knowledge extraction techniques such as data mining
techniques (Zaki and Meira, 2014).
Tacit Knowledge Explicitness: This step provides
experts with knowledge models to express their tacit
knowledge into explicit knowledge in order to be used
by a computerized decision process.
Knowledge Harmonization: The harmonization
aims to standardize knowledge expressed in hetero-
geneous formats before being loaded into the KW. It
is based on a transformation process that transforms
knowledge from a source model into a common tar-
get model (i.e., MOT).
Knowledge Storage: Storing knowledge accord-
ing to a KW model is crucial for the computerized
decision process. Naturally, stored knowledge could
be later updated by the KW administrator in order to
manage knowledge evolution over time.
Knowledge Exploitation: Once the KW is loaded,
the usage of its content is the ultimate step for intelli-
gently solving decision problems.
As a further step in the KW life cycle is the KW
maintenance. After a decision was taken and evalu-
ated, if the decision maker is satisfied then he vali-
dates it, otherwise he can inform the KW administra-
tor to update knowledge that led to invalid decisions.
In the remaining of this paper, we focus on
the knowledge harmonization as a keystone task for
knowledge warehousing. Knowledge has often sev-
eral representation forms: tacit knowledge of experts
or knowledge extracted through data mining tech-
niques. For this reason, we elected the MOT language
as a unified language for knowledge representation.
3 KNOWLEDGE
REPRESENTATIONS
In the literature, there are several languages for repre-
senting knowledge (Fensel et al., 1994): i) the infor-
mal language allows knowledge explicitness in form
of sentences, their specifications contain ambiguity
and contradictions and lack precision, ii) the formal
DecisionTreeTransformationforKnowledgeWarehousing
617
Figure 1: Basic architecture for a knowledge warehouse (Ayadi et al., 2013).
language is a set of strings of symbols that may be
constrained by rules that are specific to it. It is used,
among other things, for the precise definition of data
formats and the syntax of programming languages.
These languages are hard to understand and it is very
difficult to extract an intuition about the functional-
ity of a system given the huge amount of details of
a formal specification only and iii) the semi-formal
language is a knowledge expression formalism whose
grammatical and semantic flexibility makes it user-
friendly (H
´
eon, 2012). It has advantages compared
to informal language because it provides a represen-
tational guidance which structures the modeling pro-
cess. On the other hand, compared to the formal lan-
guage, the semi-formal language offers greater ease of
use and offers the possibility of expanding the num-
ber of people qualified to represent their knowledge,
without the help of knowledge engineers.
Since our work aim to model knowledge for deci-
sion makers and due to the advantages of semi-formal
languages we choose using a semi-formal language
that allows a graphical representation of knowledge
and that facilitate the knowledge modeling. In the re-
mainder, we will study the best known semi-formal
languages to choose the most appropriate to our work
(Paquette, 2002).
Semantic Trees are used to categorize concepts
and provide instances for these concepts. The most
general concepts are on top of the tree and the most
specific concepts are arranged hierarchically below.
The upper links indicate a relationship between a
class and its subclasses while the lower links are links
of belonging of an instance to a class. This type of
graphs leaves several possible interpretations since it
does not specify the nature of the link.
Concept Maps were developed by Joseph D. No-
vak in the 70s and they are also used to represent rela-
tionships between concepts. They also have the form
of a graph, but not necessarily of hierarchical nature.
This time, the nature of each link is specified by a la-
bel. These maps use non-oriented links that leads to
complicate interpretation. Then, the imprecise choice
of labels can make it difficult the knowledge transfer.
Semantic Networks were invented by Allan
Collins and Ross Quillian in 1969. These networks
are graphical representations analogous to concept
maps but they use oriented links. Semantic networks
use two types of nodes i) nodes representing taxo-
nomic knowledge (concepts and their instances) and
ii) others representing the properties of these con-
cepts. Furthermore, they also employ two types of
links i) structural links (e.g., is-a, has, part-of) and ii)
links specific to the domain semantic. However, these
networks complain about the lack of standard termi-
nology and standard graphical representation.
MOT (Modeling with Object Types) Language,
designed by Paquette ((Paquette, 2002), (Paquette,
2010)) in collaboration with researchers at the re-
search center LICEF
1
, provides a graphical formal-
ism in order to facilitate the representation of explicit
knowledge of various formats as well as the articula-
tion of experts’ tacit knowledge. The MOT language
allows for a quality gain in knowledge representation,
since it has greater expressiveness to represent pro-
cedural and strategic knowledge other than declara-
tive knowledge (H
´
eon et al., 2010) (H
´
eon, 2012). It
enables to represent abstract and factual knowledge
as well as the relationship between them using ori-
ented typed links. Thus, MOT is used to represent
several types of models: facts systems, taxonomies,
procedural systems, methods, processes, multi-agent
1
Research Center: Laboratory of Cognitive Informatics
and training environments (LICEF) of the Quebec Teleuni-
versity.
ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems
618
systems, etc (Paquette, 2002). Being a language that
overcomes problems of the others semi-formal lan-
guages, we elected the MOT to represent knowledge
in the KW. The next sub-sections describe the alpha-
bet, semantics, grammar and meta-model of MOT.
3.1 The MOT Alphabet, Semantics and
Grammar
The MOT alphabet includes two types of graphical
symbols for knowledge and relationships (i.e., links).
It allows representing knowledge by geometrical fig-
ures that differentiate their types and their abstraction
levels (cf., figure 2):
Figure 2: MOT shapes used for knowledge representation.
Abstract Knowledge: knowledge that represent
classes of objects (Paquette, 2002). There are
three types of abstract knowledge: Concept, Pro-
cedure and Principle;
Factual Knowledge (Facts): refers to a tangi-
ble entity to describe concrete objects (Paquette,
2002) (H
´
eon, 2011). There are three types of
facts: Example, Trace and Statement.
From the semantic viewpoint, different types of
abstract and factual knowledge are expressed as fol-
lows (Paquette, 2002) (H
´
eon, 2012):
Concept (declarative/conceptual knowledge) rep-
resents a class of objects. Indeed, it is the abstrac-
tion of a concrete object called Example;
Procedure (procedural knowledge) describes the
sets of operations for acting on the objects. The
instantiation of a procedure gives a Trace;
Principle (strategic knowledge) makes causal
links between objects, determining conditions that
may apply to the implementation of actions and
representing agents that act on something. Instan-
tiating a principle gives a Statement.
Relationships are represented by directional links
between knowledge. MOT covers a set of seven typed
links (Paquette, 2002) (H
´
eon et al., 2010):
Link I: Instantiation link connects an abstract
knowledge to a fact;
Link C: Composition link connects knowledge to
one of its components or its constituent parts;
Link S: Specialization link connects two abstract
knowledge of the same type, one of which is a-
kind-of the other;
Link P: Precedence link associates two procedu-
ral knowledge or strategic knowledge such as the
first must be completed and evaluated before the
second starts;
Link I/P: Input/Product link used to associate a
procedural knowledge and a conceptual knowl-
edge in order to represent the input or the product
of procedural knowledge;
Link R: Regulation link from strategic knowledge
to another knowledge specifies a constraint, re-
striction or rule that governs knowledge;
Link A: Application link is an association between
two domains: the first domain plays the role of
meta-knowledge for the second. Thus, the link A
allows the combination of a fact to a knowledge.
In addition to rules of good practices, the MOT
language offers integrity rules for each type of link
(cf., Table 1) (Paquette, 2002) (H
´
eon et al., 2010)
(H
´
eon, 2011) . These rules define the valid relation-
ships between different types of knowledge.
3.2 The MOT Metamodel
Recall that our goal is the representation of explicit
knowledge expressed in heterogeneous formats into
the MOT language; we propose the metamodel of this
language (cf., Figure 3) based on its descriptions in
(Paquette, 2002). The MOT model is composed of n
(n0) Links and m (m0) Knowledge unit(s) (KU).
Each KU has a name name KU and can be either
an Abstract knowledge unit (AKU) either a Factual
knowledge unit (FKU). In turn, an AKU can be a Con-
cept, Procedure or Principle. Each AKU can be in-
stantiated to many FKU which can be an Example,
Trace or Statement. In MOT, a Link can be of type {I,
C, P, S, I/P, R, A}. It links a KU source to a KU desti-
nation. we note that the Link C may have a cardinality
represented as a value written in the arrow head and
which is 1 by default.
To meet our needs and as we aspire that the MOT
meta-model covers all types of explicit knowledge,
we extend its meta-model by adding two new classes
called Position and Principleprocedure depicted in
red color in Figure 3. The class Position has the
attribute number which represents the position of a
2
The link R between a concept and a principle is an Add
in (H
´
eon et al., 2010) (H
´
eon, 2011) (H
´
eon, 2012). It is not
part of the original grammar MOT proposed by Paquette
(Paquette, 2002).
DecisionTreeTransformationforKnowledgeWarehousing
619
Table 1: Integrity rules for MOT links.
Destination Abstract knowledge Factual knowledge
Origin Concept Procedure Principle Example Trace Statement
Concept C, S I/P R
2
I, C
Procedure I/P C, S, P C, P I, C
Principle R C, R, P C, S, P, R I, C
Example A A A A,C A,I/P A
Trace A A A A, I/P A, C, P A, C, P
Statement A A A A, R A, C, R, P A, C, R, P
Figure 3: The extended MOT metamodel.
KU. This class is useful when defining transformation
rules from the decision tree metamodel where each
decision node of the tree has a position (cf., section
4.1). The class Principleprocedure represents a KU
that can be at the same time a Principle and Procedure.
This new AKU will be useful to transform knowl-
edge from the association rules model into the MOT
model. In this MOT metamodel, we have taken into
account transformations for three models of the three
most used techniques (ie, decision trees, association
rules, clustering). Note that we are working on how
to integrate knowledge coming from other data min-
ing techniques (e.g., neural networks, support vector
machines), therefore, the presented meta-model can
be extended by incorporating more concepts (i.e., at-
tributes, classes).
Our objective is to define transformation rules that
assure the move from a decision tree source model
into the MOT target model. Instead of defining the
rules at the model level we position ourselves at a
higher abstraction level so then we define rules at
the metamodel level (MM); thus we gain genericity
of rules. Indeed, our transformation rules associate
a concept of the source MM with its corresponding
concept in the target MM (Van Sinderen et al., 2012).
In the next section we limit ourselves to present trans-
formation rules from the decision tree MM.
4 DECISION TREE TECHNIQUE
In this section we are interested in decision tree as a
knowledge source and its transformation into MOT.
ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems
620
4.1 Decision Tree Metamodel
Figure 4 presents the decision tree metamodel (DT-
MM). In this MM, a Decision tree is composed of De-
cision node(s), Branch(es) and Leaves. Each Decision
node has a name name DN and a position positionN.
Each Branch is a path for a possible decision or oc-
currence and is labeled with a value. Each Leaf has
a position positionF and represents the class label of
the object to be classified. This class can be of type
Conclusion or Action. Each Decision node receives
a single Branch except the root node, and may have
several outgoing Branches. Each Branch associates
with its source one Decision node and with its desti-
nation either a Decision node or a Leaf. A Leaf is the
destination of a single Branch.
Figure 4: Decision tree metamodel (DT-MM).
Figure 5 shows a sample decision tree built on
a set of 14 observations to forecast the behavior of
sports described with 5 attributes (Quinlan, 1993), by
using the decision tree construction technique in data
mining (Althuwaynee et al., 2014). Indeed, the be-
havior of sports is predicted by determining the Play
attribute (represents the leaf of the DT having the yes
or no class). This prediction is based on the weather
data (Outlook of the sky, Temperature, Humidity and
Wind) representing decision nodes of the DT.
Figure 5: A sample decision tree.
4.2 Transformation Rules from DT to
MOT
The transformation rules link between the DT-MM el-
ements and those of the MOT metamodel. Let’s note
that some elements of the DT-MM transforms to AKU
(e.g., principle, procedure) whereas others transforms
to links (e.g., link P, link C) between knowledge units.
Thus, we distinguish two types of rules.
4.2.1 Transformation Rules to Generate
Abstract Knowledge Units (AKU)
In the following transformation rules the term root of
a decision tree (DT) refers the Decision node located
at position zero (highest level).
RD1. The root R of a DT becomes an AKU of type
Principle Pr1. The name of Pr1 is R; its position
is that of R;
RD2. The value of a branch Br1 (n1, n2); i.e., a
branch from node n1 to node n2, of a DT becomes
an AKU of type Principle Pr2. The name of Pr2 is
the concatenation of n1 (name of the source node)
with the value of Br1. The position of Pr2 is that
of the destination node n2 (decision node or leaf)
of Br1;
RD3. A leaf F1 in a DT transforms into i) an AKU
of type Principle Pr3 when the class of F1 is a
Conclusion, and ii) to an AKU of type Procedure
Proc when the class of F1 is an Action. The name
of Pr3 or Proc is the class of the leaf F1. The
position of Pr3 or Proc is positionF of the leaf F1
concatenated with ”.1”.
Note that in this rule, the type of the resulted AKU
(i.e., Principle or Procedure) depends on the seman-
tics of the leaf class F1 (i.e., Conclusion or Action).
4.2.2 Rules for Linking AKU
In a decision tree modeled in MOT each AKU-node
(i.e., Principle or Procedure) is spotted by its posi-
tion. The root has position 0 (zero), its immediate
descendants are numbered sequentially from left to
right including their ancestor number (zero) starting
from 0.1. Further descendants are numbered relative
to their ancestors as 0.1.2.3 for the third descendant
of the second child of the first descendant of the root
(cf., Figure 6).
Figure 6: Position of AKU (Abstract Knowledge Unit).
In order to link all AKU-nodes obtained by rules
RD1 to RD3 we rely on their positions as stated in
rule RD4.
RD4. Two AKU n1 and n2 having their positions x
and x.i respectively will be related by a link from
n1 to n2; where i is a single digit. As an exam-
ple, if x is 0.1.2 (position of n1) and n2 position is
DecisionTreeTransformationforKnowledgeWarehousing
621
0.1.2.4 then a link is constructed starting from n1
towards n2. The type of the link being constructed
is determined by the following two rules:
RD4.1. If n1 is a principle Pr1 issued from rule
RD1 and n2 is a principle Pr2 (obtained with
rule RD2) then their link is of type link-C and
has the default cardinality (one);
RD4.2. If n1 is a principle Pr2 issued from RD2
and n2 is an AKU issued from rule RD2 or RD3
then the type of their link is link-P.
In the next section we develop the DT2MOT algo-
rithm based on the four above rules and we present its
application on the example of figure 5.
4.3 The DT2MOT Algorithm
In this section we develop the DT2MOT algorithm
that transforms a DT to MOT. This algorithm is not
optimized; it shows step-by-step the transformation
of a given DT to MOT and indicates in comments of
each method the used transformation rule (cf., Algo-
rithm 1). Figure 7 shows AKU obtained from the DT
of figure 5 by applying the DT2MOT algorithm from
line 3 to line 13. In this figure, the position and the
rule of each extracted AKU are depicted.
Figure 7: AKU and their positions (obtained from tree of
figure 5).
Figure 8 shows the MOT model obtained from the
DT of figure 5. In this model, AKU of figure 7 are
linked by applying the DT2MOT algorithm from line
14 to line 19.
Algorithm 1: DT2MOT algorithm.
Input: Decision Tree DT
Output: MOT model
1 Declaration: j N*, i is a position of a KU
conform to notation used in RD4
2 begin
3 foreach decision node dn DT do
4 if (positionN of dn = 0) then
/* node located at position
zero is the Root */
5 Transform root to principle (dn,
Pr1) /* this method
corresponds to rule RD1
where dn and Pr1 are
respectively the input and
the output */
6 Break;
7 foreach branch br DT do
8 Transform branch to principle (br, Pr2)
/* rule RD2; br: input and
Pr2: output */
9 foreach leaf f DT do
10 if (class of f is a Conclusion) then
11 Transform leaf to principle (f, Pr3)
/* rule RD3.i; f: input
and Pr3: output */
12 else
/* class of f is Action */
13 Transform leaf to procedure (f,
Proc) /* rule RD3.ii; f:
input and Proc: output
*/
/* Note that the calls for the four
previous methods could be
executed in parallel, they
generate AKU with their
positions */
/* Linking the AKU task */
14 foreach AKU n1 do
15 foreach AKU n2 6= n1 do
16 if (position of n1 = 0 and position of
n2 = 0.j) then
17 Link by linkC (n1, n2) /* rule
RD4.1; n1: source of
linkC and n2:
destination of linkC */
18 else if (position of n1 = i and
position of n2 = i.j) then
19 Link by linkP (n1, n2) /* rule
RD4.2; n1: source
and n2: destination
*/
ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems
622
Figure 8: Decision tree of figure 5 transformed into MOT.
5 CONCLUSION AND
PERSPECTIVES
In order to load explicit knowledge in the knowledge
Warehouse (KW) we need to unify their modeling and
solve structural heterogeneity. To represent knowl-
edge having different models into a unified one, we
have elected the semi-formal language for graphical
knowledge representation called Modeling with Ob-
ject Types (MOT) for which we have designed its
metamodel. In addition, we have defined in this pa-
per the Decision Tree (DT) metamodel and we have
suggested a set of appropriate rules that allows trans-
forming DT model into MOT model. Furthermore,
we have developed an algorithm describing the prin-
ciple of these transformations that we have illustrated
with an example.
We are working on several extensions: how to
transform into MOT language, other kinds of ex-
tracted knowledge by using other techniques such as
associations rules, ECA rules, non-monotonic knowl-
edge, clustering, neural networks, support vector ma-
chines, etc. Then, how to integrate these many MOT
models produced from different source models (e.g.,
Decision tree, association rules, clustering) and what
are the operators that should be used to query and
maintain an MOT knowledge Warehouse. Using the
KW during an intelligent decision-making process is
the ultimate objective of this research.
REFERENCES
Althuwaynee, O., Pradhan, B., Park, H.-J., and Lee, J.
(2014). A novel ensemble decision tree-based chi-
squared automatic interaction detection (chaid) and
multivariate logistic regression models in landslide
susceptibility mapping. Landslides, 11(6):1063–1078.
Ayadi, R., Hachaichi, Y., and Feki, J. (2013). Vers des
entrep
ˆ
ots de connaissances : D
´
efinition et architec-
ture. In Conf
´
erence sur les Avanc
´
ees des Syst
`
emes
D
´
ecisionnels ASD’2013.
Dymond, A. (2002). The knowledge warehouse: The next
step beyond the data warehouse. In Data Warehousing
and Enterprise Solutions. SAS Users Group Interna-
tional 27.
Fensel, D., Landes, D., Neubert, S., and Studer, R.
(1994). Integrating semiformal and formal methods in
knowledge-based systems development. In Proceed-
ings of the 3rd Japanese Knowledge Acquisition Work-
shop JKAW’94 (Hatoyama, Japan, Nov. 7-9), pages
73–87.
H
´
eon, M. (2011). Guide du langage de mod
´
elisation par
objets typ
´
es mot. Cotechnoe inc, Qu
´
ebec, Canada.
H
´
eon, M. (2012). Ontocase, une approche d’
´
elicitation
semi-formelle graphique et son outil logiciel pour la
construction d’une ontologie de domaine. In 5i
`
eme
Gestion des connaissances dans les soci
´
et
´
es et les or-
ganisations, Montr
´
eal (Qu
´
ebec), Canada.
H
´
eon, M., Basque, J., and Paquette, G. (2010). Valida-
tion de la s
´
emantique d’un mod
`
ele semi-formel de
connaissances avec ontocase. In Acte des 21
`
emes
Journ
´
ees Francophones d’Ing
´
enierie des Connais-
sances, page 55
`
a 66, N
ˆ
ımes, France.
Irfan, R. and uddin Shaikh, M. (2010). Enhance knowl-
edge management process for group decision making.
In Proceedings of the 2010 Second International Con-
ference on Computer Engineering and Applications -
Volume 01, ICCEA ’10, pages 66–70. IEEE Computer
Society.
Kerschberg, L. (2001). Knowledge management in hetero-
geneous data warehouse environments. In Interna-
tional Conference on Data Warehousing and Knowl-
edge Discovery, pages 1–10. Springer.
Nemati, H. R., Steiger, D. M., Iyer, L. S., and Herschel,
R. T. (2002). Knowledge warehouse: an architectural
integration of knowledge management, decision sup-
port, artificial intelligence and data warehousing. De-
cision Support Systems, 33(2):143 – 161.
Nonaka, I. and Takeuchi, H. (1995). The Knowledge-
Creating Company: How Japanese Companies Cre-
ate the Dynamics of Innovation. Oxford University
Press, New York.
Paquette, G. (2002). Mod
´
elisation des Connaissances
et des Comp
´
etences: Un Langage Graphique Pour
Concevoir et Apprendre. Presses de l’Universit
´
e du
Qu
´
ebec, Sainte-Foy.
Paquette, G. (2010). Visual Knowledge Modeling for Se-
mantic Web Technologies: Models and Ontologies.
IGI Global, Hershey, PA.
Qing-lan, H. and Zhi-jun, H. (2009). Research on cost
control dss based on knowledge warehouse. In Pro-
ceedings of the 2009 Sixth International Conference
on Fuzzy Systems and Knowledge Discovery - Volume
07, pages 357–361. IEEE Computer Society.
Quinlan, J. R. (1993). C4.5: Programs for Machine Learn-
ing. Morgan Kaufmann Publishers Inc., San Fran-
cisco, CA, USA.
Van Sinderen, M., Johnson, P., Xu, X., and Doumeingts, G.
(2012). Enterprise Interoperability: 4th International
IFIP Working Conference, IWEI 2012, Harbin, China,
September 6-7, 2012. Proceedings. Springer.
Zaki, M. J. and Meira, W. (2014). Data Mining and Anal-
ysis: Fundamental Concepts and Algorithms. Cam-
bridge University Press.
DecisionTreeTransformationforKnowledgeWarehousing
623