MULTI-DIMENSIONAL MODELING
Formal Specification and Verification of the Hierarchy Concept
Ali Salem
Faculty of Sciences of Sfax,University of Sfax, Sfax, Tunisia
Faiza Ghozzi
Institute Superior of Informatics and Multimedia of Gabes,University of Gabes, Gabes, Tunisia
Hanene Ben-Abdallah
Faculty of Economics and Management of Sfax, University of Sfax, Sfax, Tunisia
Keywords: Multi-dimensional model, OLAP, Constraints,
Meta-Model, formal Specification, Hierarchy.
Abstract: The quality of a data mart (DM) tightly depends on the quality of its multidimensional model. This quality
dependen
ce motivated several research efforts to define a set of constraints on the DM model/schema.
Currently proposed constraints are either incomplete, or informally presented, which may lead to ambiguous
interpretations. The work presented in this paper is a first step towards the definition of a formal framework
for the specification and the verification of the quality of DM schemas. In this framework, the quality is
expressed in terms of both the syntactic well-formedness of the DM schema as well as its semantic
soundness with respect to the DM instances. More precisely, this paper first formalizes in Z the constraints
pertinent to the hierarchy concept; the formalization is treated at the meta-model level. Secondly, the paper
illustrates how the formalization can be instantiated and the constraints are verified for a particular sample
model through the theorem prover Z\eves.
1 INTRODUCTION
Face to the international, unrestrained economic
competition, an increasing interest in decision
support systems (DSS) has emerged over the last
decade. These latter assist a decision maker in
extracting data pertinent to their analysis interests
from their transactional systems (called On-Line
Transactional Processing-OLTP systems). Unlike
transactional systems, the majority of DSS relies on
OLAP (On-Line Analytical Processing) systems
where data are often stored in multidimensional
databases. In such databases, data is organized by
center of interests (Facts) and examined according to
various axes of analysis (Dimensions) represented
through analysis prospects (Hierarchies) (Kimball,
2002). These multidimensional databases are often
organized in terms of two type’s storage areas: a
data warehouse (DW) that regroups all data required
for any potential analysis requirements, and/or a set
of data marts (DM) each of which regroups data
extracts required to one particular analysis
requirement.
In order to assist in the development of DM/DW,
several
multidimensional models have been
proposed to specify DM/DW schemas, e.g., the basic
star model, its generalization the constellation
model, the snow flake model, etc. (Hurtado, 2002]
(Lechtenbörger, 2003). On the other hand, evidently,
the quality of a DM/DW depends tightly on the
quality of its schema. In this context, we consider
that the quality of a schema those properties that can
be expressed in terms of the schema’s syntactic (or
structural) specification as well as its semantic
correctness (or soundness). The syntactic quality of
a schema ensures that elements of the
multidimensional model are correctly used together;
for example, the acyclicity constraint disallows the
existence of a level cycle in the hierarchy. The
soundness constraint defines a hierarchical link of
dependence between the dimension instances...
317
Salem A., Ghozzi F. and Ben-Abdallah H. (2008).
MULTI-DIMENSIONAL MODELING - Formal Specification and Verification of the Hierarchy Concept.
In Proceedings of the Tenth International Conference on Enterprise Information Systems - DISI, pages 317-322
Copyright
c
SciTePress
The quality of a schema has been addressed by
the definition of a set of constraints at the model
level. That is, several researchers (cf., (Hurtado,
2002),(Lechtenbörger, 2003), (Ghozzi, 2003) have
defined a set of rules that a schema must respect in
order to produce either a syntactically correct
schema, or a sound schema with respect to data
instances. The rules are defined on the structures and
structural elements of a schema. Nevertheless, these
works have not proposed a mechanism to validate
and verify these rules.
In this paper, we present the first steps towards
the development of a formal framework for DM/DW
modeling and verification. On one hand, this
framework relies on the precise definition of the
constraints ensuring the syntactic and semantic
correctness of a DM/DW schema. On the other
hand, its exploits the formal definition in order to
provide for a means to verify both types of
correctness. More specifically, in this paper, we first
present the formal definition of the Hierarchy
concept at the meta-model level in the Z language
(Spivey, 1992); secondly, we illustrate how the
constraints can be instantiated for a particular model
and verified using the Z/eves theorem prover
(Saaltink, 1999).
The remainder of this paper is organized as
follows. In Section 2, we first overview current
proposals of constraints for DM/DW schemas;
secondly, we present our approach of constraint
definition and verification. In Section 3, we present
the set of constraints pertinent to the hierarchy
concept and their formalization in Z. In Section 4,
we show how to instantiate the constraint for a
particular model and how to verify the correctness of
the constrained model through Z/eves. Finally,
Section 5 summarizes our contributions and outlines
ongoing work.
2 RELATED WORKS
During our survey of the previous works in this
domain, we field studied the hierarchy concept and
the constraints related to this concept. Defining the
hierarchies classification of certain dimension
attributes is crucial because these classification
hierarchies provide the basis for the subsequent data
analysis. Since a dimension attribute can also be
rolling up to more than one other attribute, multiple
classification hierarchies and alternative path
hierarchies are also relevant (Trujillo, 2001).
According to (Lehner, 1998), in the context of
statistical databases and on-line analytical
processing as well, classification hierarchies provide
a basis for defining aggregate data. (Part, 2006)
confirms that hierarchies are crucial to
multidimensional modeling since they are used in
conjunction with aggregation functions to aggregate
(“rollup”) or detail (“drill-down”) measures. These
quotations prove the importance related to the
hierarchy concept. In (Malinowski, 2004), the
authors present a conceptual classification of
hierarchies and propose graphical notations for them
based on the ER model. With respect to dimensions,
every hierarchy classification level is specified by a
class. An association of classes specifies the
relationships between two levels of a hierarchy
classification. The only prerequisite is that these
classes must define a Directed Acyclic Graph
(DAG) rooted in the dimension class (constraint
{dag} placed next to every dimension class). The
DAG structure can represent both alternative path
and multiple hierarchies classification (Lujàn, 2002).
In the GMD model (Franconi, 2004), the authors
describe the hierarchy by an order function between
the different dimension attributes. (Abello, 2006)
presents a multi-dimensional model object oriented,
and defines a hierarchy as aggregation relation
between the different dimension attributes.
Otherwise, few works formally define
hierarchies but they mainly discuss the
summarizability conditions and offer some solutions
to correct measure aggregations in presence of the
so-called heterogeneous hierarchies (Hurtado, 2001).
In (Hurtado, 2002), the authors propose a set of
constraints to solve the aggregation problem. These
constraints are related to the hierarchical structuring
of the dimension attributes and the dimension
instances. We note an explicit and complete
definition of the hierarchy concept in the works of
(Ghozzi, 2003). These works will make the basis of
our formal specification.
The constraints expressed in these works differ
from a model to another. This difference resides, on
the one hand, in the level of expression of the
constraint (Meta-model, model) and on the other
hand, in the level of checking or safeguarding of the
constraint. Moreover, there is no consensus on the
whole constraints to take into account. This
dissension on the level of the constraints expression
in these various works poses a true problem
touching with the coherence of the data to
incorporate. In other words, it can lead to incoherent
results of analyses.
The goal of our work is to lead to a consistent
formal specification of a multidimensional Meta-
model in constellation. Thus, we offer the designers
a means to check their models.
ICEIS 2008 - International Conference on Enterprise Information Systems
318
In this paper, we present a formal specification
of the hierarchy concept and a domain check
validation of this specification with Z/Eves proover.
We prove, as well, the consistence of this
specification with an initial state theorem (Spivey,
1992).
3 CONSTRAINTS RELATED TO
THE HIERARCHY CONCEPT
The constraints expressed at the Meta-model level
include the essential constraints to maintain the
coherence model. They are related to the basic
concepts and independent of any application. This
type of constraints can be classified according to
basic concepts' of the multidimensional model; fact,
dimension, Hierarchy and constellation.
Among these constraints we differentiate
between the constraints related to the hierarchy
concept. These constraints can be classified in two
categories according to their checking level:
Structure constraints and instances constraints.
3.1 Structure Constraints
These constraints describe the rules to scheduling
hierarchy attributes:
Unicity of identifier and “All” attributes
(Ghozzi, 2003). For example, in supplier
dimension (Fig1) we can find several Supplier
whish have the same identifier. This can
generate ambiguities when we query
multidimensional data because facts will be
related to more than one Supplier. In addition,
All attribute is defined to enclose a hierarchy
(Fig 1)
The identifier is the attribute of the finest
granularity and “All” is the attribute of highest
granularity (Fig 1) (Ghozzi, 2003). In a
hierarchy, attributes are classified according to a
partial order (roll up). For example, in supplier
hierarchy (Fig 1), ID determines City
determines Country etc. Only ID can determine
all the information related to a Supplier. The
attributes are classified from the finest
granularity to the highest granularity. “All”
attribute is used to enclose a hierarchy.
Figure 1: Attributes ID and ALL.
Nonempty hierarchy (Ghozzi, 2003): Each
hierarchy has at least two levels of parameters;
ID and the “All” attribute.
Acyclicity (Franconi, 2004) (Abello, 2006)
(Hurtado, 2002) (Carpani, 2001) (Ghozzi,
2003): this constraint disallows the existence of
a level cycle in the hierarchy. For example, for
the supplier hierarchy (Fig 2), we notice the
existence of the cycle (continent, city). This
cycle generates a problem of redundancy during
the data analysis.
Figure 2: Acyclicity.
Connection to the top (Hurtado, 2002) (Ghozzi,
2003) this constraint expresses that all
parameters, except All, have at least a father (a
parameter of less fine granularity) (Fig 3). This
constraint guarantees the order structure of
parameters in dimension.
Figure 3: Connection to the top.
3.2 Instances Constraints
The hierarchical structure between the parameters is
also applied on their members. Thus, several works
speak about the dimension diagram of instances.
The constraints on this level describe the relations
between the various values of the attributes.
Partition (Hurtado, 2002) (Ghozzi, 2003). To
each parameter member corresponds one and
only one member among those of each
parameter successor in the hierarchy. In a
hierarchy, each member must determine the
successor member. If a parameter member
corresponds to more than one successor member
we can not determine this successor member.
For example (fig 4), the city of Paris (member
of city parameter) should not belong at the same
time to France and Italy (member of country
parameter).
Figure 4: Partition.
Id City
All
Supplier
Child
Father
Supplier
Id
City
Country
All
Continent
Paris
Italie
France
Country
City
Supplier
Id
City
Country
Continent
All
MULTI-DIMENSIONAL MODELING - Formal Specification and Verification of the Hierarchy Concept
319
Soundness (Ghozzi, 2003): for each parameters
couple in the Hierarchy, there are at least two
members belonging to this couple in such a way
that these two members are dependent (fig 5). In
our schema, this dependence is materialized by
the existence of a relation between these
members.
Figure 5: Soundness.
4 FORMAL SPECIFICATION IN
LANGUAGE Z
The formal specification of our model allows the
expression of the constraints in an exact and specific
way, thus offering the means of validating and
checking them.
The selected language of specification is Z
language. It is based on the set theory and
mathematical logic (Spivey, 1992). The set theory
used includes the standard operators of the sets, the
Cartesian products and the sets of power.
Mathematical logic is a first order predicate
calculus. A schema Z is composed of two parts: a
part for the declaration and a part for the predicates
representing the constraints on the declared
variables.
4.1 Specification
We start by defining the two types NomH and Dom:
The first includes Hierarchy names and the second
includes the various attribute values in dimensions:
[NomH, Dom]
Then, the Standard free type, which is used for
classification of dimension attributes as weak
attributes, parameters, identifier or “All”:
Type ::weak Parameter ID All
The Relval relation things to see the various
relations between the attributes values:
Relval: Dom Dom
The Weight function assigns to each attribute a
weight:
Weights: AttDim Type
Each dimension attribute includes a finished set of
values (Dom). We define it as a compound type.
AttDim
val:Dom

The formal definition of a hierarchy in language Z
results in the schema named Hierarchy where: N is
the name of the hierarchy; Att is a finished whole of
dimension attributes AttDim. ParamH is a sequence
describing the attributes hierarchy. A sequence, in
language Z, can be considered as a function whose
field is an adjoining subset of the natural numbers.
The predicate part of the Hierarchy schema gathers
constraints connected to:
The unicity of : identifier [1],
Country
City
France Paris
The unicity of the parameter All [2],
Nonempty hierarchy [3]: Each hierarchy has at
least two levels of parameters in the definition
of the hierarchy.
The identifier is the attribute of the finest
granularity [4],
The All attribute is of the largest granularity [5],
Acyclicity [6]: The existence of a cycle in the
hierarchies is forbidden.
Soundness [7]: ParamH defines a hierarchical
link of dependence between the parameter
members of a dimension.
Partition constraint [8] [9]: to each parameter
member corresponds one and only one member
among those of each successor parameter in the
hierarchy.
Connection to the top: All the parameters,
except “All”, have at least a father (a parameter
of less fine granularity). This constraint makes
it possible to ensure the passage from a level to
another. It is checked by definition of ParamH
as a sequence.
Hierarchy
N: NomH
Att: AttDim
ParamH: seq AttDim

[1]
1
x: Att Weights x = ID
[2]
1
x: Att Weights x = All
[3] # ParamH 2
[4] x: Att x ran ParamH ParamH 1 = x
  Weights x = ID
[5] x: Att x ran ParamH ParamH # ParamH = x
Weights x = All
[6] i, j: 1 .. # ParamH i Ð j ParamH i Ð ParamH j
[7] i, j: 1 .. # ParamH j = i + 1
 v, w: Dom v ParamH i . val w ParamH j . val
 v w Relval
[8] i, j: 1 .. # ParamH; v: Dom j = i + 1 v ParamH i . val
 w1, w2: Dom
w1 ParamH j . val
w2 ParamH j . val
v w1 Relval
v w2 Relval
w1 = w2
[9] v: Dom; i: 1 .. # ParamH - 1 v ParamH i . val
ICEIS 2008 - International Conference on Enterprise Information Systems
320
 w: Dom w ParamH i + 1 . val  v w Relval

4.2 Proof Reading
The proof involves a demonstration that the various
requirements upon the data type are consistent and
not contradictory. To show that the requirements are
consistent, we have only to show that the constraint
part of the state schema is satisfiable. This is usually
achieved by proving an initialisation theorem: we
show that an initial state, at least, exists.
In this phase, we will test the consistency of our
specification. The step to be followed in this phase
is to define a correct hierarchy and to prove that the
latter checks the constraints expressed on the level
of the Hierarchy schema.
For that, we start with the instantiation of the
Clas_Cons Hierarchy of Car dimension (Fig 6):
Figure 6: The Clas_Cons Hierarchy of dimension Car.
Among the attributes of Car dimension, we have
Immat, Model, Mark, Power and All. The Immat
attribute is the identifier. The attributes Model and
Mark are classified as parameters. Power is a weak
attribute. These attributes are classified according to
the hierarchy Clas_Cons (Fig 6).Each attribute
contains a set of values related to each other
according to the relation Relval (Fig 6).
The description of this example in Z is written to
use the Axiom Box Schema. The form of definition
includes a constraint upon the object being
introduced. Such definitions are said to be
axiomatic, as the constraint is assumed to hold
whenever the symbol is used: it is an axiom for the
object. In the Z notation, our example description is:
Immat, Model, Mark, all, Power: AttDim
clas_Cons: NomH
T102, T103, T104, Fiesta, Clio, Golf, Ford,
Renault, Volkswagen, vall: Dom

Weights Immat = ID
Weights Mark = Parametre
Weights Model = Parametre
Weights all = All
Weights Power = Faible
Immat . val = T102 T103 T104
Model . val = Fiesta Clio Golf
all . val = vall
Mark . val = Ford Renault Volswagen
Relval
= T102 Fiesta Fiesta Ford Ford vall T103
Clio
Clio Renault Renault vall T104 Golf Golf
Volkswagen
Volkswagen vall
Then, we define the HierarchyInstance schema
which will play the role of a hierarchy instances. In
addition, we must assign the various values
necessary to the various sets already declared to the
level of the Hierarchy schema.
HierarchyInstance
Hierarchy

N = clas_Cons
Att = Immat Model Mark all Power
ParamH = Immat Model Mark all

At last, we proved the theorem of following
initialization state:
theorem ConsistanceHierarchie
Hierarchy HierarchyInstance
To prove this theorem we use the Z/Eves Prover.
This prover is semi-automatic. His is bases on set
theory and first order logic.
In the environment of demonstration, one can use
several commands to indicate to Z-Eves tracks of
possible demonstrations. For some, it is a question
of clicking, for others; we have to introduce other
sub-theorems and axioms to assert the prover.
In our case to prove the initial state theorem we
proceeded as follows (fig 8):
invoke: this command replaces all the schema
used in predicate by their definition.
prove: this command allows various options of
demonstration and simplification.
use: is used to explicitly call another theorem or
an axiom.
prove by reduce: it has the same role as the
prove command, but it simplifies even better.
case and next: in our proof, there are many
predicates to be shown. The best solution is to
use these two commands which make it possible
to show them separately.
instantiate: this command is useful to instantiate
one or more quantified variables.
Car
Immat
Model
Mark
All
Power
T102
T103
T104
Fiesta
Clio
Golf
Ford
Renault
Volkswagen
vall
MULTI-DIMENSIONAL MODELING - Formal Specification and Verification of the Hierarchy Concept
321
Figure 7: Proof script for theorem ConsistanceHierarchy.
Finally, we succeeded in proving the theorem of the
initialization state. We have, so, to prove that there
is not a contradiction to the constraints specified in
the predicate part of the Hierarchy schema.
Consequently, we proved the consistence of our
formal specification of the hierarchy concept.
Figure 8: Proved theorem ConsistanceHierarchy.
5 CONCLUSIONS
In this paper, we gathered all the constraints related
to the hierarchy concept judged to be essential to
maintain the coherence of the data to incorporate
and ensure the integrity of the structures and the
multidimensional data compared to these
constraints. We defined these constraints at the
meta-model level. In addition, we proposed to
classify these constraints according to two
categories: the constraints related to the structure
and the constraints related to the instances. Further
more, we precisely formalized in Z the Hierarchy
concept. This formalization provides the designers
with the means of validating the hierarchical
structuring of the dimension attributes of their
models.
REFERENCES
Abelló A., Samos J., Saltor F. “YAM2: a
multidimensional conceptual model extending UML”.
Information Systems. Vol 31, p 541-567, 2006
Carpani F., Ruggia R., “An Integrity Constraints
Language for a Conceptual Multidimensional Data
Model”. In 13
th
International Conference on Software
Engineering & Knowledge Engineering (SEKE’01),
Argentina, 2001.
Codd E. F., Codd S.B., Salley C.T., "Providing OLAP (On
Line Analytical Processing) to Users-Analysts: An IT
Mondate", Rapport technique, E.F. Codd and
Associates, 1993.
Franconi E. and Kamble A., “The GMD Data Model and
Algebra for Multidimensional Information” Advanced
Information Systems Engineering, 16
th
International
Conference, CAiSE 2004, Riga, Latvia, June 7-11,
2004, Proceedings.
Ghozzi F., Ravat F., Teste O., Zurfluh G., “Modèle
Dimensionnel à Contraintes”. In Revue des Sciences et
Technologies de l'Information, Série RIA- ECA,
Hermes –Lavoisier, Vol. 17, N. 1-2-3, p.43-56, 2003.
Hurtado C., “Mendelzon A., Reasoning about
summarizability in heterogeneous multidimensional
schemas”, in: Proc. of the 21
st
ACM Int. Conf. on
Management of Data and Symposium on Principle of
Databases Systems, p. 169–179, 2001.
Hurtado C., Mendelzon A., "OLAP Dimension
Constraints". In 21
st
ACM SIGACT-SIGMOD-
SIGART Symposium on Principles of Database
Systems (PODS’02), Madison, USA, p. 169-179,
2002.
Kimball R., Ross M., "The Data Warehouse Toolkit",
Wiley, New York, 2
th
edition, 2002.
Lechtenbörger J., Vossen G., "Multidimensional normal
forms for data warehouse design". In Revue
Information Systems, Vol. 28, N. 5, p. 415-434, 2003.
Lujàn S, Trujillo J., Song, "Extending the UML for
Multidimensional Modeling" The Unified Modeling
Language: 5
th
International Conference, Dresden,
Germany, September 30 - October 4, 2002,
Proceedings.
Lehner W., Albrecht J., Wedekind H., “Normal forms for
multidimensional databases”, in: Proc. of the 10
th
Int.
Conf. on Scientific and Statistical Database
Management, p. 63–72, 1998.
Malinowski E., Zima´nyi E., “OLAP hierarchies: A
conceptual perspective”, in: Proc. of the 16
th
Int. Conf.
on Advanced Information Systems Engineering, p.
477–491, 2004.
Prat N., Akoka J., Comyn-Wattiau I., “A UML-based data
warehouse design method” Decision Support Systems,
Volume 42, Issue 3, p. 1449-1473, 2006
Trujillo J. C., Palomar M., Gómez J., Song:I "Designing
Data Warehouses with OO Conceptual Models". In
IEEE Computer, Vol. 34, N.12, p. 66- 75, 2001.
Spivey J.M., “The Z Notation : a Reference Manual”.
Prentice-Hall, 1992.
Saaltink M.. The Z/EVES 2.0 User’s Guide. ORA Canada,
OneNicholas Street, Suite 1208, Ottawa (Ontario),
K1N 7B7, 1999.
ICEIS 2008 - International Conference on Enterprise Information Systems
322