Hierarchies and Relative Operators

in the OLAP Environment

Elaheh Pourabbas, Maurizio Rafanelli
Istituto di Analisi dei Sistemi ed Informatica - CNR, Viale Manzoni 30, 00185 Roma, Italy
e-mail: {pourabbas, rafanelli} @ iasi.rm.cnr.it
(a) one or more measured data, each representing the result of
Abstract the application of an aggregation function on raw data. Their
In the last few years, numerous proposals for modelling and
numerical values are called measures;
querying Multidimensional Databases (MDDB) are proposed. A (b) a set of dimensions, which provide a qualitative description
rigorous classification of the different types of hierarchies is still of the measured data and are also called metadata, i.e. data
an open problem. In this paper we propose and discuss some
about data.
different types of hierarchies within a single dimension of a cube.
Since most proposed models have such constraints as
These hierarchies divide in different levels of aggregation a single
"dimensions are linguistic categories corresponding to different
dimension. Depending on them, we discuss the characterization ways of looking at the information", then each dimension is a
of some OLAP operators that refer to hierarchies in order to simple concept hierarchy. A different treatment is proposed in [ 1]
maintain the data cube consistency. Moreover, we propose a set
where the authors proposed a data model that provides support
of operators for changing the hierarchy structure. The issues for multiple hierarchies along each dimension and for ad hoc
discussed provide modelling flexibility during the scheme design
aggregates, as well as a few algebraic operators. In this paper, we
phase and correct data analysis. deal more about multiple hierarchies, and we introduce the
multiplicity of a hierarchy as a semantic variant of the simple
I. Introduction one. Sometimes, dimensions are organized in hierarchies in
which there are different aggregate levels [11]. Some design and
Recently, the business strategies have given a strong relevance to computation problems can arise when the mapping between
new research areas, such as Data Warehousing and On-Line- different aggregate levels of a hierarchy is not complete. We
Analytical-Processing, OLAP [2]. The OLAP concept was distinguish this type of hierarchy from that of where the mapping
proposed for rendering very large, historical (statistical) between dimension levels is complete, and then accordingly, we
databases in multidimensional perspectives, and it is oriented to introduce the concepts of the partial and total classification-
decision making for business users. The connection between hierarchies. An OLAP system concerns mostly simple data cube,
analyzing business data and socio-economic data (generally i. e., it is a simple structure to collect in a single "scheme" all the
known as statistical data) is not obvious, but both of them deal multidimensional aggregate and non-aggregate data relative to a
with multidimensional data sets, and both are concerned with defined event (e.g., Sales, and so on). The values in each cell of
statistical summarizations over the dimensions of the data sets. this data cube are some "measures" of interest.
Similarities and differences between OLAP and Statistical In this context, through some examples, we will discuss different
databases are presented in [13]. The concept of types of operations using the well known OLAP operators, and
multidimensionality (or n-dimensionality) of these datasets, and we propose their specialization to solve some problems which
in particular, of aggregate data [12], as well as the concepts of arise in particular situations.
dimension (often called category attribute, descriptive variable, The paper is structured as following: Section 2 gives an overview
character, etc.) and of measure (often called summary attribute, on basic concepts. Section 3 discusses the different types of
quantitative data, variable, etc.) have been already discussed [9, hierarchies of a cube. Section 4 introduces the characterization of
12]. Recently, in literature, many authors proposed some OLAP operators on hierarchies. Section 5 gives a set of
multidimensional data models and query languages. Gray et al. operators that refer to changing the hierarchy structure. Finally,
in [3] proposed the data cube operator as extension to SQL Section 6 concludes.
which generalized the histogram, cross-tabulation, roll-up, drill-
down, and sub-total constructs found in most report writers. 2. An overview on basic concepts
In [7] the authors formalized a multidimensional data model for
OLAP, and developed an algebra query language called In literature, different sets of basic concepts and operators were
Grouping Algebra. The relative multidimensional cube algebra is proposed. In this paper, we will refer to the multidimensional
proposed in order to facilitate the data derivation. Gyssens et al. data structure and to a set of minimal basic operators described
in [4] presented a tabular database model and discussed a tabular in the following.
algebra as a language for querying and restructuring tabular data. A Cube is "a group of data cells arranged by the dimensions of
Lehner in [5] discussed the design problem that arose when the the data" [8]. It represents a logical view of multidimensional
OLAP scenarios became very large and they proposed a nested data.
multidimensional data model useful during schema designing and A dimension is "a structural attribute of a cube that is a list of
multidimensional data analysis phases. members, all of which are of a similar type in the user's
In literature, multidimensional data are characterized by having perception of the data" [8]. The set of the cube dimensions
two different types of attributes: represents the relative data multidimensionality.
A hierarchy is a set of variables which represent different levels
of aggregation of the same dimension and which are linked
32 SIGMOD Record, Vol. 29, No. 1, March 2000
between them by a mapping. A typical example of hierarchy is Example 2.3 Consider the cube of Figure 1. Suppose that the
City ~ State ~ Region ~ Country. phenomenon considered in it is "Cars sales" and that the
A measure is a particular dimension of the cube [1], which measure values represent the cars sold in USA by City, Vendor
represents the extensional fashion of the phenomenon described (dimension 1), and Year (dimension 3). By this operation we can
by the cube, and which is, in general, a numeric value. Assigning push, for example, Vendor instances into the cells of the cube,
a value to each dimension of a cube, the measure is obtained by so that in them we will find a couple of values (in our case, for
a mapping from this assignment. In Figure 1 these concepts are example, we will find <Smith, 2,738> . . . . . etc.). It is different
graphically represented. from the operator defined in [ 1]. In fact, we delete the dimension
1 (Vendor) of the cube, while in [1] it is maintained also as
Dimension I
dimension, that is, it is duplicated into the measure of the
Title [ l)imep.s~n .~> Loc~tio¢l~City cube. ~1
The pull operator is the converse of the previous one. It creates a
new dimension converting the element, specified from it, which
is in the measure.
Example 2.4 Let us consider the database described in the
previous example. By this operator we can extract the element of
Rcgioll~ ~ Nol~h West the measure specified in the operation (for example, "Cars
sales") transforming it in a dimension of the cube. The result is a
new cube where the dimension 1 becomes "Cars sales" and the
measure becomes "Vendor". O
City ~ P" Lewimon ...Bangor Cleveland ... Akron San Frar~ig¢ Los Angeles Phoe~x Tucson
The slice (or Destroy Dimension [1]) operator deletes one
Imctr~otaal
spat©
Extensioaal
....
dimension of the cube, so that the sub-cube derived from all the
remaining dimensions is the slice result that is specified. It is
Figure 1. Example of a data cube with hierarchy. equivalent to the summarization operator of the statistical
In a multidimensional database different cubes are stored, each of database operators [ 10].
which is defined with different dimensions. The domains of these Example 2.5 Let us consider the cube of Figure 1 with an
dimensions consist of a set of values (instances). We define additional dimension "Model". This operation allows to cut one
primitive domain of a variable the set of all the possible values specified dimension, recomputing all the values of the new
that this variable can assume in the database. This means that measure in each cell of the cube. For example, slice Model
every variable of a hierarchy has its own domain whose values deletes this dimension from the cube and recomputes the values
are a subset of, or coincide with the values of its primitive of the measure in the single cell of the resulted cube that
domain. The OLAP operators defined in literature [1, 2, 3, 8] and becomes, in this case, a bidimensional table. O
considered in this paper are roll-up, drill-down, push, pull, slice,
dice, and select. We briefly describe them in the following. The dice (or Restriction [1]) operator restricts the dimension
The roll-up operator decreases the detail of the measure, value domain of the cube removing from this domain those
aggregating it along the dimension hierarchy [ 1]. It is equivalent values of the dimension that are specified in the condition
to the classification operator of the statistical database operators (predicate) expressed in the operation. It is equivalent to the
[10]. "Roll-up involves computing all of the formula-based restriction operator of the statistical database operators [ 10].
relationships of data for one or more dimension". Note that Example 2.6 Let us consider the cube of Figure 1 and suppose
because in this paper, we consider only data obtained from count that the Year domain is <1990, 1991, 1992, 1993, 1994, 1995,
and sum function application, then this computation is a sum. 1996, 1997, 1998>. This operation allows to cut the part of the
Example 2.1 Consider the hierarchy of Figure 1. The roll-up domain instances of one dimension of this cube which are
operation allows to change level from City to State, recomputing specified in the operation. For example, dice Year = <1990,
the values of the measure. 13 1991, 1992 > carries out the removing of the above mentioned
instances from the domain of the dimension Year, restricting it to
We would point out that we consider a city as a municipality the remaining values (1993 . . . . . 1998). I3
area. It means that each state consists of a set of municipality
(form territorial, population, etc., point of view)- The select operator is the dual of the dice operator. It carries out
The drill-down operator is a binary operator [ 1] which considers the restriction operation removing from this domain those values
the aggregate cube joined with the cube that has more detailed of dimension that do not satisfy the condition (predicate)
information and increases the detail of the measure going to the expressed in the operation.
lower level of the dimension hierarchy. Example 2.7 Let us consider the Example 2.6. This operation
Example 2,2 Consider the hierarchy of Figure 1. The drill-down restricts the dimension value domain of the cube maintaining in
operation allows to pass from State to City, retrieving the values this domain those values of the dimension that are specified in
of the measure which were previously stored in the same cube. 13 the condition expressed in the operation. For example, select
Year =<1990, 1991, 1992> carries out the selection of these
The push operator is used to convert a dimension into the values into the domain, so that the new domain values consists of
relative measure in order to manipulate it or to consider it as new exactly these values. I3
measure. Combining with the Pull operator, it can exchange
measure and dimension and, then, allows to treat uniformly both
of them.

SIGMOD Record, Vol. 29, No. 1, March 2000 33

 Sales
3. Characterization of hierarchies Class Cil)" Vendor Year 1997 1998
Alcoholic : Los Angles Smith I(X)00 12000
Hierarchy is fundamental to data warehouse and OLAP N e w York Wong 20000 16000
environment. A hierarchy is an effective form of knowledge Washington MC Donald 23000 17000
Atlanta Laarent 50000 60O00
representation for encoding prior domain knowledge relevant to
data cube. In a simple form, a hierarchy shows the relationships D~ s~: .... ~i000
Detroit Clifford t 9(3000 18000
between domains of values. Each operation on hierarchy can be Non Los Angles Snlth 20900 14500
regarded as a mapping from one domain to a smaller domain. alcoholic New York Wong 12300 32OO9
Washington M c Donald 87000 23890
In OLAP environment, hierarchies are used to conceptualize the Atlanta Laurent 23100 49000
process of generalizing data as a transformation of values from
Dallas Backer 56000 34500
one domain to values of another domain by means of drill- Detroit Clifford 21000 30OOO
down/roll-up operators. Figure 2. Example of a data cube
In this section, we discuss the hierarchies from two different
perspectives: mapping between domain values, i. e., leading to Ikvarag~
consider total and partial classification hierarchies and hierarchy All Drink
structure. The later case, treats multiple and multiplicity of
hierarchies. ~s AlcOholic - - Non Alcoholic

Bee~ Spirits Wine Liquor " Bott Juice Tea Soft Dfink Coffe
3.1. Classification hierarchies
Dimensions have often been associated with different
hierarchically organized levels. These levels correspond to
different granularities of viewing data. The name of each level is
expressed by the corresponding variable name. Generally, the
shift from a lower (more detailed) level to a higher (more
State Maitre ..... Massacn " ' "
aggregate) level is carried out by a mapping. A mapping between
two variables can be complete or incomplete. In the first case the II_r/
( y ~ ~ ..LosAn~ee . . . . . .
i
hierarchy is called total classification hierarchy, and in the
second case it is called partial classification hierarchy. We give Figure 3. The hierarchies along the dimensions Beverages, and
the following definitions: Location (at left) and the relative domain value (at right)

Definition 1 A mapping between two variables of a hierarchy Definition 3 A partial classification hierarchy on a given
defines a containment function if each variable instance of a dimension is a hierarchy in which between at least one adjacent
lower level corresponds to only one variable instance of a higher couple of variables there is no full mapping.
level and each variable instance of a higher level corresponds to
Example 3.2 Consider the chain store example we gave in
at least one variable instance of a lower level. In such case, this
Example 3.1. Suppose that the chain stores of the above
mapping is called full mapping.
mentioned company in the state of California are located only in
Definition 2 A total classification hierarchy on a given some of cities of this state (see Figure 4).

[
dimension is a hierarchy in which between each adjacent couple Location

of variables there is a full mapping. C~ntry U.S.A.

The containment function respects the summarizability

conditions (disjointness and completeness) of multidimensional
databases described in [6] and in [ 11 ]. As known in literature, a
hierarchy is intensionally represented by a partial ordered set.
City ... Boston ... ~ s c o S a n l o s e ~ .............
Then, a total classification hierarchy is any subset that defines a
total order. Figure 4. Domain values of the City level
Example 3.1 Let us consider a nation-wide drink company that Then, the domain value of City level along the Location
owns chain stores located in all cities. Assume that all stores in dimension is restricted with respect to that shown in Figure 3.
the chain sell the same beverages. Sales data are collected yearly, Accordingly, these cities are not listed in the table of Drink
i. e., at the end of each year, each member store reports the total sales. ~1
sales amount of each drink to the regional headquarters. Figure 2
These two types of hierarchies will influence the result of queries
shows part of the data reported in 1997 and 1998. Q
for which the summarization operations will be needed. Details
of this fact are discussed in a further section.
The hierarchies along the dimensions Location, and Beverages
Note that, in this paper, we consider only hierarchies in which no
are represented below both in intensional level and extensional
overlapping exists among domain instances of each variable.
level.
As shown in Figure 3, for a domain value of a level on location
dimension all domain values of the lower level are defined, i. e., 3.2. Multiplicity of a hierarchy and multiple
it is a total classification hierarchy. This is completely in hierarchies
accordance with the hypothesis made in Example 1, where in all
cities of the given country such a drink store is located. One of the more important problems regarding the hierarchies
refers to their definition. In this section we propose a set of
definitions in order to fix a reference point in their study.
34 S I G M O D Record, Vol. 29, No. 1, March 2000
Firs~ of all, we distinguish between multiplicity o f a hierarchy kic~rchies variable imzaaces
and multiple hierarchy.
Definition 4 Let H and H~ be two hierarchies. HI is a {
multiplicity of H if its level domains are the same as the H
g3
level domains and the variable name associated to each level
of H 1 is a specialization of the variable name associated to (a}

the corresponding levels o f H .

Example 3.3 Let us consider a location hierarchy defined as:
City ~ Province ~ Region. A possible multiplicity o f this
hierarchy is City o f residence ~ Province o f residence O
Region of residence. I:l (b)

Definition 5 Let H ~, H z . . . . . H , be a set o f hierarchies.

Figure 6. Example o f path generation between levels
This set forms a multiple hierarchy if each of them has at
least one variable in common with another hierarchy of the
same set. 4.1. Case of the Roll-up operator
Example 3.4 Let us suppose that we have four hierarchies, As mentioned above, this operator decreases the detail o f the
labelled (a), (b), (c), and (d), as illustrated in Figure 5. The measure, aggregating it along the dimension hierarchy. A
hierarchy labeled (d) is a multiple hierarchy, where the level problem arises when a variable relative to a level of the hierarchy
Province is the same for (a) and (b) and the level Region is the is not complete (i.e., case of partial classification hierarchy).
same for (a) and (c). El In the following we consider what happens when this operator is
applied to a total classification hierarchy and, then, to a partial
?~ntr~
Countr¢
C~mrv classification hierarchy.

\
rI \ Example 4.1 Let us consider the data cube represented in Figure
"% / / 2. In Figure 7, an its "multidimensional" view is illustrated. I:l
# of sales
USA
TCity
I
City
I
City i•
" in 1997 in

dors
(a) (b) (c) (d~

Figure 5. Example of a multiple hierarchy

Definition 6 Let H be a hierarchy. The hierarchy obtained
from deleting one or more non terminal variable or level o f
instance of the measure
H is a derived hierarchy. Class stored into a cube cell
Example 3.5 From the multiple hierarchy (d) shown in Figure 4,
we obtain the following derived hierarchies: City ~ Province
Country, City ~ Region -~ Country, City ~ Country, City Vendors (Smith, Wong, Mc Donald, Laurent, Cliffords, Chen) I
Zone, City ~ Region -~ Country, City ~ Province
Figure 7. Multidimensional view of Drink Sales data cube
Country, and City ~ State ~ Country.
Let us suppose to formulate a query defined as below:
Specifically, in the case o f partial classification hierarchies, the "Select Vendors for which the total Sales is >10000 units in
variable instances o f derived hierarchies are the instances o f each State o f the West"
variables that are adjacent to the instances o f deleted levels and
between which a connected path can be defined. For example, in This query is solved in the following way:
Figure 6-(b) are reported the variable instances o f derived Roll-up from City to Region, Select Region = West, Drill-
hierarchies obtained from variable instances o f the partial down from Region to State, Push Vendors, Pull # of sales,
classification hierarchy illustrated in Figure 6-(a) that satisfy the Dice # of sales "<10000". I:l
above mentioned condition. Note that in some cells of the resulted cube (see Figure 8) null
4. Characterization of OLAP operators on values can appear. This demonstrates that some instances of
hierarchies "Vendors" are not defined.
Let us suppose, now, that the classification hierarchy relative to
Recently different authors proposed a set of OLAP operators,
"City ~ State ~ Region" has, as domains the cities o f
which are defined on data cube and which produce as output a
new cube [1, 2, 9]. In this section, we discuss the operators California, with only the instances "San Francisco, San Jose, San
involved in manipulating dimensions with hierarchies in order to Diego". This one is a subset o f the primitive domain of City, in
introduce some important modifications and specializations. which all the cities o f California (San Francisco, San Jose, San
Diego, Los Angeles, etc.) are stored.

SIGMOD Record, Vol. 29, No. 1, March 2000 35

 Shates of Wesl means that we lose the exact information on the real period to
Vmtdors with total which the result should refer. To overcome this mistake we
sales :,19,~0 units
in each w~l stats ot introduce a specialization of the Slice operator, called Partial
the USA in 1997 ~ s a l e s Slice (or P-Slice) and defined below. Q
Definition 8 The P-Slice operator removes the dimension on
drink which it is applied transforming it in an implicit dimension.
The only value of the implicit dimension domain is the set
valued of all the values that formed the domain of the
Mc Donald
removed dimension.
All
Similarly to the solution proposed for the Roll-up operator, the
same Note is added to the title of the cube.
hates of West California, Arizona, Utah ....... Nevad~
According to this definition, the above query is now solved in the
t ofsales 101410 114820 118460 121270 152950 192760 22461 ] following way:
P-Slice Year.
Figure 8. The result of the query
The result is now a cube with the same dimensions of the primary
In particular, when the operator Roll-up from City to Region is
one, where the title becomes "Drink sales by Class, Vendor, and
applied, no information is stored about the non completeness of
Year where Year is a subset of the primitive domain".
the domain of City relative to California. This means that for
For symmetric reasoning of terminology we use the term Total
California the number of vendors for which the total Drinks sold
Slice (or T-Slice) for the well known Slice operator
in 1990 is >10,000 units refers only to the cities of San
If, instead, the Year domain in the considered cube coincided
Francisco, San Jose, and San Diego and not to all the cities of
with its primitive domain, then the previous query would be
California. Then, since this information is not specified
solved in the following way:
anywhere, the answer for this state is wrong.
T-Slice Year.
A solution to that is to save the information about the domain
values that cause the non-completenesses of the hierarchy. This The cardinality of the resulting cube is now decreased of one
can be obtained in two different ways. The former consists of dimension, since, by removing Year no information is lost.
adding a Note (the clause where <variable name> is-a subset of
the primitive domain) to the title of the cube. In the case of 5. An enlargement of the operator set
Figure 8, the title becomes "Vendors with total Drinks sales referring to hierarchies
>10,000 units in each state of the West in 1997 in USA, where
city of California is-a subset of the primitive domain". The later In this section we propose a set of operators able to change the
consists of adding the same Note to each variable of the primary configuration of a hierarchy extending or reducing its
hierarchy whose level is higher with respect to the level of the level number, adding a new multiplicity, and creating a multiple
variable with the incomplete domain. In the same Figure 8, we hierarchy. For formalizing them, let us consider l I and 12 be
have to add the clause where city of California is-a subset of the two adjacent levels of a given hierarchy defined as Ii --* 12 • Let
primitive domain to the variables State, Region, and Country. us discuss them.

4.2. Case of the Slice operator 5.1 Insert level

As mentioned above, the slice operator reduces the The Insert level operator allows to add a new level to a hierarchy,
dimensionality (or cardinality) of a cube eliminating one (giving the variable name, the domain instances, and the
dimension through its multidimensional space. This fact is not relationships between this level and, respectively, the higher and
always true because, if we delete a dimension whose domain is a the lower adjacent levels in the hierarchy). The insertion of a
subset of the primitive domain, we lose information and the new level denoted by I i between the above mentioned levels is
resulting cube of this operation contains incorrect data.
represented through the symbol Insertlevel ll211i ([i,I ..... ]i.n ;R i ) '
Before discussing this situation, we need to introduce the implicit
dimension definition. where li. 1..... I~., represents the inserted level domain instances

Definition 7 We call implicit dimension any dimension of a and

data cube which has only one instance in its definition Ri = (I/2,1 (Ilia (ll 1,1 ..... I11 ,ql ) ..... lli,Pl (Ill .ql+l ..... Ill ,q2 ))'
domain. This instance can be one value or multi-valued.
.... l li,qk=h (Ill .qph_l+l ..... It I,q, =n )))
Example 4.2 Let us consider the cube of Figure 2, where the
primitive domain of the dimension Year assumes the values
where the instances of levels 12, li, Ii are divided in
<1990, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998>. This
respectively, k, h , and n subsets and p j , q v represent a
means that they are all the possible values that this dimension can
assume in the database. Instead, the value domain of Year in the generic set of li, and I I levels instances where j = 1..... h and
considered cube is <97, 98>. v = 1,...,n.
Let us suppose that the following query is carried out:
Example 5,1 Let us consider the Location dimension shown in
"Give me the drink sales in all Cities by Class and Vendor"
Figure 2. Let us suppose to insert the variable County between
It is solved in the following way:
the variables City and State. We have to define its domain
Slice Year.
values, as well as the mapping relationship between City and
In this case if the slice operator deletes the dimension Year, the County, and between County and State (see Figure 9). This is
result seems to refer to the whole primitive domain of Year. This obtained by the following formula:
36 SIGMOD Record, Vol. 29, No. 1, March 2000
 State satisfied. Then, we defined the concepts of multiplicity of a
lnsertlevel City Count" ( Green . Orange .... ;R Couno' ) where
R County = (...,( California (Green (San Francisco ..... Richmond ), given hierarchy and of multiple hierarchies.
=1 Based on the definitions and concepts proposed in this paper, we
.... Orange (Los Angels ..... Oxnard ) ),... ) discussed the characterization of the OLAP operators involved in
Location the hierarchy manipulation. In particular, depending on the type
of hierarchy, we studied the different behaviour of the Roll-up
and the drill-down operators in order to keep the consistency of
data that is the result of the queries. We also characterize the
slice operator, defining the implicit dimension concept and
specializing the operator in two different types: P-slice and T-
" LL,_ .... " slice. Finally, we proposed an enlargement of the operator set,
specific for the hierarchy manipulation, which are the Insert
lnsfr~level level, the Delete level, the Add multiplicity, and the Add level
Figure 9. Example of Insert level operator operators. For each situation discussed, clarifying examples are
5.2 Delete level given.
T h e Delete level operator redefines a hierarchy as a subset of the
existing one, deleting a variable with its relative domain. This References
operation re-creates the mapping relationships between the two
levels (higher and the lower) adjacent to the deleted variable.
[1] Agrawal R., Gupta A., Sarawagi S., Modelling
Multidimensional Databases, in 13th International
The deletion of a level Ii is denoted by
Conference on Data Engineering, (ICDE97, Birmingham,
Deletelevel l~li(l~. I ..... ti. . ) where 11 and t2 are, respectively, U. K., April 7-11, 1997), 232-243.
the relative lower and higher level in the given hierarchy. This is [2] Codd E.F., Codd S.B., Salley C.T., Providing OLAP (on-
the converse operation of the insert level operation. Note that, in line analytical processing) to user-analysts: An IT
this case the relationships between level are defined mandate. Technical report, 1993.
automatically by the system. [3] Gray J., Bosworth A., Layman A., Pirahesh H., Data cube:
a relational aggregation operator generalizing group-by,
5.3. Add multiplicity cross-tabs and subtotals, in 12th IEEE International
The Add multiplicity operator duplicates a given hierarchy, Conference on Data Engineering (ICDE96, New Orleans,
changing the name of the variables in order to specialize them, Louisiana, Feb. 26-Mar 1, 1995), 152-159.
as already discussed in section 3.2. Let H be a hierarchy. This [4] Gyssens M., Lakshmanan L.V.S., Subramanian I.N.,
operator is represented through Tables as a paradigm for querying and restructuring,
Addmultiplicity H ( additionalname ) where additionalname is ACM-PODS 1996, Montreal, Canada, 93-103.
a string to be added to each variable names of the hierarchy in [5] Lehner W., Modeling large scale OLAP scenarios,
order to specialzing the last one. Advances in Database Technology-EDBT98, Valencia,
Spain, Lecture Notes in Computer Science, Springer
Example 5.2 Let us consider the example 3.3. The hierarchy Verlag Pub., Vol. 1377, 53-167.
City of residence -~ Province of residence ~ Region of [6] Lenz H-J., Shoshani A., Summarizzability in OLAP and
residence is obtained by statistical data bases, in 9th International Conference on
Addmultiplicity Location ( o f residence ) [=l Scientific and Statistical Data Management (SSDBM97),
1997.
5.3. Add level [7] Li C., Wang X.S., A data model for supporting on-line
T h e A d d level operator creates a new hierarchy starting from a
analytical processing, in Proceedings of Conference on
given hierarchy. It creates a new variable and its relative domain, Information and Knowledge Management, November,
and defines the mapping relationship between this and the
1996, 81-88.
variable of the starting hierarchy. These two hierarchies define, [8] OLAP Council, The OLAP glossary.
in this way, a multiple hierarchy. Let I n be a level to be added
http://www.olapcouncil.org. The OLAP Council 1997.
to a hierarchy defined as I l -~ 12 -~... -. In_ I . T h e A d d level [9] Rafanelli M., Ricci F. L., Proposal of a logical model for
operator is represented through statistical database, Proceedings 2 ~ International.
A d d l e v e l t I--,t 2 ~...~t,_~ In (1 a .... Itn, R i ) Workshop on SDBM'83, Los Altos, CA, 264-272.
[101 Rafanelli M., Ricci F.L., Mefisto: a functional model for
gives as result a hierarchy l i -, 12 - * - . . ~ In- 1 ~ l n •
statistical entities, IEEE Transactions on Knowledge and
Example 5.3 An example of multiple hierarchy has been already Data Engineering, August 1993, Vol.5, No.4.
shown in Figure 5-(d). El [ 11 ] Rafanelli M., Shoshani A., STORM: A STatistical Object
Representation Model, Proceedings of 5th Intern. Confer.
on SSDBM, Charlotte, NC, April 1990, Lecture Notes in
6. Conclusions Computer Science, Springer Verlag Pub., Vol. 420.
In this paper, after a brief overview of basic concepts relative to [ 12] Shoshani A., Wong H.K.T., Statistical and Scientific
the multidimensional data structures and to a set of OLAP Database Issues" IEEE Transactions on Software
operators, which are object of the discussions that followed, a Engineering, October 1985, VoI.SD-I 1, No.10.
characteriazation of hierarchy is proposed. We distinguished [13] Shoshani A., OLAP and Statistical Databases: Similarities
between hierarchies in which the mapping between different and Differences, in Proceedings of ACM PODS 97, 1997,
levels is full and hierarchies in which this condition is not 185-196.

$ I G M O D Record, Vol. 29, No. 1, March 2000 37