Database Reference
In-Depth Information
Table 2. Facts and measures for the integrated schema issued from hotel room booking and e-Ticket of
Figure 8
Fact
Relevance level
Measure
F-Room
Second
Price
F-Payments
Second
PaymentAmount
TotalPayementDueAmount
TotalPayementDueDate
F-Bookings
First
F-Buy
First
TotalPayementB
really represent keys issued from empty entities.
Table 2 shows all measures extracted from each
extracted fact-table.
those in F-T1 ( i.e. , F-T2 has n-m> 0 dimensions
more than F-T1 ) , then to move them to F-T1 , they
must be aggregated on the set of their uncommon
dimensions. Note that if the dimension set of F-T1
is equal to the dimension set of F-T2 , then the
set of uncommon dimensions between F-T1 and
F-T2 is empty; therefore, T1 is parallel to T2 and
reciprocally. Consequently, the measures of both
facts F-T1 and F-T2 have the same granularity
level and could be seen as two halves of the same
fact; hence, we recommend merging them into a
single fact conventionally called F-R1-R2 .
In our design method, aggregated measures as
well as dimensions used in their calculation are
automatically identified. However, the designer
must intervene to define the necessary aggregation
functions which are semantics dependent.
Measure Identified from Parallel Tables
As mentioned above, a second origin of measures
is parallel tables. We adapted the definition of
parallel tables from the concept of parallel re-
lationships which is specific to the E/R model
(Seba, 2003). In an E/R diagram, a relationship
R1 connected to m entities is said to be parallel
to a relationship R2 connected to n entities ( m≤n )
if the m entities linked to R1 are also linked to
R2 . By analogy, we define the concept of parallel
tables as follows:
Let T1 and T2 be two relationship tables such
as T1 and T2 are connected to m and n ( m≤ n )
tables respectively; T1 is said to be parallel to T2
(noted T1//T2 ) if and only if the primary key of
T1 is included in or equal to the primary key of
T2 (Feki & Hachaichi, 2007-c). Note that, in this
definition, T1 and T2 are assumed to be relation-
ships because entities could not be parallel; this
optimizes the search of parallel tables (Feki, &
Hachaichi, 2007-c). In our running example, there
are no parallel tables.
Let T1 and T2 be already identified as two fact-
tables and let T1 be parallel to T2 . The fact F-T1
(built on T1 ) can receive other measures coming
from the fact F-T2 (built on T2 ) by aggregating
measures of F-T2 before moving them to F-T1 .
Since the measures in F-T2 are more detailed than
Dimension Identification
A dimension is generally made up of a finite set
of attributes that define various levels of details
(hierarchies), whereas others are less significant
but used, for instance, to label results or to restrict
data processed in queries. These latter are called
weak (or non dimensional) attributes. The set of
candidate dimensions for a given fact can be built
either on tables modeling entities or attributes.
Given a fact F-T ( i.e. , fact build on table T ), we
consider every table T1 that represents an entity
and that is directly referred by the table T as a
candidate dimension for F-T . Conventionally, the
name of this dimension is D-T1 and its identifier
Search WWH ::




Custom Search