Improving Expressive Power in Modeling Data Warehouse and OLAP Applications - Evolving Application Domains of Data Warehousing and Mining - page 35

Figure 14. Role-playing dimensions in SSAS a) dimension panel and b) solution explorer panel

grouping according to dimension members, i.e.,
according to the order number shown in Figure 3
for the level called SalesOrderNumber .
SSAS allows the designer to add a new dimen-
sion using a fact table as a source table. After
choosing the attribute of interest (in our example,
SalesOrderNumber ), the Dimension Usage (see
Figure 15a) automatically defines a relationship
between the newly-created dimension and fact
table as Fact type (Figure 15 b).

as the number of account holders. This induces
the so-called double-counting problem.
The double-counting problem arises since the
schema does not respect multidimensional normal
forms 8 (MNFs) (Lehner et al., 1998; Lechtenböger
and Vossen, 2003). These forms determine condi-
tions that ensure correct measure aggregation in the
presence of generalized hierarchies. In particular,
one of the conditions of the first multidimensional
normal form (1MNF) requires that each measure be
uniquely identified by the set of leaf levels provid-
ing the basis for correct schema design; 1MNF is
then used to define the remaining MNFs.
As can be seen in our example in Figure 16,
only the time and account determine the amount
and cashback points of a transaction. Therefore,
the schema in Figure 16 is not in the 1MNF and
this schema should be transformed. Two differ-
ent options exist as shown in Figure 17. The first
option (Figure 17a) includes an additional fact
relationship, while the second option (Figure
17b) creates a non-strict hierarchy (Malinowski
& Zimányi, 2008).
Mapping both conceptual schemas from Figure
17 to the relational model will give the following
tables: three tables for representing each dimension
( Time , Account , and Customer ), one Transaction

Multi-valued or Many-to-
Many dimensions

Some practitioners (e.g., Kimball & Ross, 2002)
and scientists (Luján-Mora et al., 2006; Pedersen
et al., 2001; Song et al., 2001) refer to the so-called
“multivalued dimensions” or “many-to-many
relationships between facts and dimensions” in
order to represent the situation where several
members of a dimension participate in the same
instance of a fact relationship. An example is
shown in Figure 16; it is used for the analysis of
a transaction performed over clients' accounts.
Since an account can be shared among different
clients, aggregation of the measures Amount and
Cashback points will count them as many times