A SEMANTIC RESOURCE BASED APPROACH FOR STAR SCHEMAS MATCHING

International Journal of Database Management Systems (IJDMS ) Vol.10, No.6, December 2018
DOI : 10.5121/ijdms.2018.10602 15
A SEMANTIC RESOURCE BASED APPROACH FOR
STAR SCHEMAS MATCHING
Elhaj Elamin1
Amer Alzaidi2
and Jamel Feki2
1
Department of Computer Science, Sudan University of Science and Technology,
Khartoum, Sudan
2
University of Jeddah, FCIT, Jeddah, Saudi Arabia
ABSTRACT
The hybrid approach is widely used in constructing data warehouse (DW) schemas. It relies on a complex
process for matching two sets of multidimensional star schemas: schemas built from business requirements
(BR-Star schemas) and schemas constructed on the organization data source (DS-Star schemas). Using a
semantic resource during this matching helps solving heterogeneity problems. This paper suggests a semi-
automatic approach for the construction of approved star schemas by matching DS-Star schemas with BR-
Stars, and by using WordNet as a semantic resource for solving heterogeneity issues. This approach
consists of three steps: i) Match DS-Stars with BR-Stars; ii) Involve the DW designer for approval of the
matching process; and then iii) Generate approved star schemas. We have defined two Boolean functions
and four semantic metrics for the matching process. We have developed a software prototype for testing
our approach.
KEYWORDS
Star schemas matching, Hybrid approach, Data warehousing, Semantic resource.
1. INTRODUCTION
The data warehouse (DW) has been considered as a vital technology for modern decision support
systems (DSS) for organizations. Indeed, the DW offers efficient capabilities for supporting
decision-makers. Despite the valuable efforts and attempts from researchers devoted to
developing DW approaches, several DW projects failed [1][2]. In fact, researchers agree that any
successful attempt to develop a DW should consider two features namely i) the construction of
the DW multidimensional schema by using a hybrid approach relying on user requirements and
data sources; and ii) the use of semantic resource to overcome the heterogeneity issues
complicating the DW construction process[3][6][9][12].
Furthermore, it is commonly admitted that the intervention of the decision-maker during the DW
construction process greatly improves the quality of the DW schema. Obviously, a full
automation of the design process may lead to inaccurate results; for instance, defining rules for
extracting facts and dimensions automatically may be misleading, as real-world entities (tables,
objects…) and relationships both may have common characteristics and therefore may play the
role of fact or dimension. Therefore, it is helpful to let the decision-maker or the DW designer
intervene for interpreting and evaluating the correctness of designed schemas[17].
Furthermore, it is widely admitted that the star schema represents the keystone structure for DW
modelling. This is due to its simplicity, efficiency and its intuitive shape when compared to other

16
multidimensional schemas (e.g., constellation or snowflake schemas).A multidimensional star
schema has one fact modelling a business activity of interest for decision-makers and n (n>1)
dimensions surrounding the fact [4].Relying the construction of the DW schema on a hybrid
approach requires a matching process between BR-Stars(star schemas designed from business
requirements) and DS-Stars(star schemas designed on the data source). Indeed, despite the
agreement among DW community about the basic methods for constructing the DW, we reveal
three issues in the literature review. First, there is no agreement about which technique is
recommended for performing the matching process, if a technique could be adopted. Secondly,
which kind of semantic resource should we if any? Thirdly, whether the approach considers BR
first as in [7][9][10] or the DS first as in [5] or combines both BR and DS simultaneously [6][8].
These issues motivated us to investigate a new hybrid and semi-automatic approach focusing on
the matching process of multidimensional star schemas; this approach aims to help DW designers
to design star schemas closely related to BR and DS, and that are subject to low effort of
adaptation/validation, therefore enhancing the DW quality and reducing costs. Our proposed
approach differs from the literature approaches since:
It relies on a semantic resource; in fact, we have elected WordNet for the matching, this choice
enables us to avoid building specific domain ontology; consequently, this will shorten the
design/development time of the DW.
The matching process accepts as input two complementary sets of star schemas; this is to consider
both BR and DS simultaneously and, therefore build star schemas closely related to users’
requirements and the organization’s data source.
This paper is organized as follows: Section 2 studies works related to the hybrid approaches for
the DW design, as well as matching methods. In Section 3, we introduce our proposed approach
for generating approved star schemas as result of matching DS-Stars and BR-Stars. Section 4 is
dedicated to results presentation. Finally, Section 5 concludes this paper and enumerates some
ongoing issues and related perspectives.
2. RELATED WORK
The DW construction process is complex, tedious and time-consuming [1]; these difficulties do
not came from nonsense. Indeed, each step in this construction process consists of sub-steps,
which can be carried out using various methods and/or techniques. Additionally, these steps
should coined together in a systematic manner to produce a satisfying multidimensional model
(i.e., schema). To shed light in this complexity, as an example, we cite the matching process
between the DS and users’ requirements; for this purpose, in [8][10] the authors base their work
on a graph technique. In [5] authors use search patterns, whereas [7] uses three multidimensional
normal forms (noted 1MNF , 2MNF and 3MNF) to define a set of Query/View/Transformation
(QVT) relations for accomplishing the agreement between the multidimensional model obtained
from user requirements ,and the DS. In not far away for the matching process, the proposed works
vary in whether they first consider user requirements or DS structure (i.e., schema). As an
example, in [10] [13] the authors first consider requirements then reconciling them against the
DS; other works begin with considering first the DS and then the requirements [5]. Additionally,
the solutions vary in the way the semantic resource had been applied; in this trend, the authors in
[5] develop domain ontology in order to automate the generation of star schemas; authors in [9]
use a global ontology. On behave of automating the DW construction process; existing
contributions vary between generating the star schema manually or through a semi-automatic
approach, while others try to automate the full process beginning from generating user
requirements and DS models as well as the matching process that produces final star schemas. In

17
order to highlight the needs in generating DW star schemas, we give more attention to the
approaches of the related works.
In [6], a hybrid-automated approach has been proposed where the authors use a graph to model
the DS, as well as SQL queries for requirements representation. An important limitation in their
work is that it relies on an expert for writing the queries. Additionally, their approach ignores the
heterogeneity problems raised in the DW design activities. Hence, a basic component for
constructing the DW is missing, namely a semantic resource.
Authors in [7] proposed a hybrid DW approach, they use two conceptual multidimensional
models; on one hand, they first defined a conceptual multidimensional model for capturing user’s
requirements and, on the other hand, they used multidimensional normal forms to define a set of
Query/View/Transformation relations. They reconcile the user requirements model with that of
DS to ensure its correctness. The authors succeed in applying a systematic approach for
developing the multidimensional model. However, in behave of requirements they concentrate on
business goals related to DW users; nevertheless, there is no means for using semantic resource.
In [8], authors automatically generate a set of multidimensional schemas from the DB schema of
the operational information system that satisfy user requirements expressed in terms of SQL
queries. They used a multidimensional graph to store multidimensional information about the
query. In their approach, SQL queries are accepted if they generate anon-empty set of
multidimensional schemas. The drawback of this approach is that there is no template (i.e., style)
for writing the queries. Additionally, the task of expressing user requirements as SQL queries
needs a skilled person in SQL who knows precisely the DS schema.
A hybrid approach is proposed in [5]; its distinctive feature is an automatic analysis of the DS that
leads to the design of the DW schema. Additionally, it belongs to the group of works supporting
the multidimensional design from ontologies. However, the design process of the DW consumes a
lot of effort and time because this approach relies on ontology as driving force for generating the
multidimensional model. Moreover, the authors described their approach as a reengineering
process; hence, the most choices in this situation as ontology language are UML (Unified
Modelling Language) or ERD (Entity-Relationship diagram). The problem here is that generating
ontology from these sources needs a heavy pre-process that delays the DW constructing.
Additionally, the full automation of the DW schema generation in this work decreases the chance
for intervening the DW designer to confirm the generated multidimensional elements.
An automatic hybrid approach is proposed in [9] where authors used algorithms for matching user
requirements with a global ontology. However, in the matching process, their approach ignore the
elements that do not fully match or are under a given threshold. As well, they assume the
existence of a global ontology, whose construction complicates the design of the DW.
A nearest approach to ours is suggested in [13] where the authors use a Goal-Question-Metric
technique for capturing users’ requirements by means of interviews. The capturing goals are then
aggregated and redefined in means of abstraction sheets; from these sheets, the star schema is
generated. As well, they use the graph for constructing star schemas from the DS. Despite their
efforts on using hybrid approach, their approach suffers from the following limitations: first, in
behave of requirements, they use interview technique in capturing user requirements, but they do
not define style for formalizing the goals; this may lead to goals expressed in divergent formats,
which complicates the process of star schema construction. Secondly, in behave of generating star
schemas from DS, authors dedicated an algorithm for exploring the E/R model. Therefore,
mapping this model to a connectivity graph without using the reverse engineering process may
result in poorly defined star schema elements. Additionally, the approach is manual and does not
rely on a semantic resource.

18
Summing up, the DW development process needs additional investigation overcome these
miscellaneous gaps. Table 1 summarizes the techniques used in hybrid approaches for the
purposes of reconciling the DS with users’ requirements, the degree of automation, and the use of
a semantic resource.
Table 1. Comparison between proposed hybrid approaches.
Work reference Matching Technique Automation
Use of semantic
resource
Romero, et al., 2006 [8] Graph Automatic No
Romero, et al., 2010[5] Search patterns Automatic Yes
Mazón, et al., 2007[7] Multidimensional normal forms No No
Romero, et al., 2010[6] Graph/SQL queries Automatic No
Thenmozhi, et al., 2012[9] Algorithms Automatic Yes
Di Tria, et al., 2015 [10] Graph Automatic Yes
Bonifati,et al., 2001[13] Graph and Goal/Question metrics No No
3. PROPOSED APPROACH
Recent research works for constructing the DW greatly concentrate on two features: the first
feature concerns the use of a hybrid approach; that means the design process considers both users’
requirements and DS data model in order to produce a DW multidimensional schema. The second
feature is to solve the heterogeneity problems by using a semantic resource. From our viewpoint,
an important issue is how to apply these two features so that the resulting DW helps the
organizations offering a reliable multidimensional model. We propose a matching approach for
the alignment of star schemas issued from two complementary contributions in the literature: One
dealing with the generation of star schemas from a relational DS model [16], and a second
contribution tackling the construction of star schemas from BRs [14].Our matching approach
relies on a semantic resource to overcome the heterogeneity between DS-Star schemas and BR-
Star schemas. In the literature, ontology solves the problems of heterogeneity; however, domain
ontology is not frequent especially for DWsing; in addition, its construction is neither easy nor
rapid. Indeed, building domain ontology increases the time and the cost for developing the DW.
As with our approach we aim to decrease the DW developing time, we have selected WordNet as
a ready open source semantic resource for solving the heterogeneity between a DS-Star’s
elements and a BR-Star’s elements. More precisely, our proposed approach consists of three
steps: i) Matching DS-Star schemas with BR-Star schemas; ii) Involvement of the DW designer;
and iii) Generation of approved star schemas. Figure 1 depicts the framework for our suggested
approach.
Hereafter, we detail steps for generating approved star schemas.

Figure 1. Proposed framework for generating approved star schemas from BRs and DS models
3.1. Matching DS-Star Schemas with BR
Schema matching is a complex and time
Star schema elements (fact, dimension, etc...) with BR
detect and suggest solutions for unmatched between schemas elements such as diversity in names
of elements. Naturally, we optimize the matching
the two schemas’ elements. So then, we identify ‘similar schemas’ (i.e., schemas analyzing the
same business activity, having identical or synonym fact names, and common dimensions) and
then we match each couple of similar schemas. To do so, we develop an algorithm called
MatchStars, as well we define two Boolean functions for checking whether two names of fact or
dimensions are identical/synonyms or not. Additionally, we define four semantic metrics to
measure the similarity between two star schemas. To assist the DW designer solving the problem
of unmatched elements, our approach allows him/her to intervene and matches them manually. It
is worth mentioning that we have treated the extraction process of DS
schemas in previous a work [17].
Hereafter we introduce the notation we use:
Notation
− FName(S): A function that returns the Fact Name from a star schema S.
− SBF: A set composed of three lists:
o List of fact names from BR
o List of Dimension names in each fact from BR
o List of parameter names in each dimension in each BR
− SDF: A set composed of three lists:
o List of fact names from DS
o List of Dimension names in each fact from DS
Star Schemas with BR-Star Schemas
Schema matching is a complex and time-consuming process [15]; it aims to match correctly DS
Star schema elements (fact, dimension, etc...) with BR-Star schema elements. Furthermore, we
of elements. Naturally, we optimize the matching step not by computing a Cartesian product of
e of similar schemas. To do so, we develop an algorithm called
the similarity between two star schemas. To assist the DW designer solving the problem
is worth mentioning that we have treated the extraction process of DS-Star schemas and BR
schemas in previous a work [17].
Hereafter we introduce the notation we use:
FName(S): A function that returns the Fact Name from a star schema S.
SBF: A set composed of three lists:
List of fact names from BR-Stars
Dimension names in each fact from BR-Stars
List of parameter names in each dimension in each BR-Stars
SDF: A set composed of three lists:
List of fact names from DS-Stars
List of Dimension names in each fact from DS -Stars
19
match correctly DS-
Star schema elements. Furthermore, we
step not by computing a Cartesian product of
e of similar schemas. To do so, we develop an algorithm called
the similarity between two star schemas. To assist the DW designer solving the problem
schemas and BR-Star

o List of parameter names in each
− SCF: A set composed of three lists:
o List of all fact names Common to BR
o List of Dimension names Common to BR
o List of parameter names Common to BR
For the next illustrations, Figure 2 depicts an example of three simple DS
BR-Star schemas.
Figure 2. Example of threeBR
Boolean Functions for Matching DS
Hereafter we highlight for each of the two Boolean function Identical and Synonym, its objective,
syntax, and then we give the result of its application on our running example of figure2. Note that
the results of applying these functions will be append to the SCF set of lists.
1) Function Identical (e1, e2): returns
False otherwise; e1, e2 may be fact names or dimension names. When true, this function
means that the match of these elements makes sense.
Note that, we apply this function in two steps of the matching process; firstly on fact names, and
secondly, on dimension names of identical facts having identical names. We append the common
fact name and the common dimension names into SCF Fact list and Dimension list.
For our running example (Figure 2), the function Identical (Bookings, Bookings) is returns True;
so then we reuse this function on the dimensions of the two Bookings facts in BR
Star. Identical (Date, Date) is also True, so we append the Bookings
well as the common dimension Date in Dimension list to SCF.
2) Function Synonym (e1, e2): returns
False otherwise.
List of parameter names in each dimension in each DS -Stars
SCF: A set composed of three lists:
List of all fact names Common to BR-Stars and DS-Stars.
List of Dimension names Common to BR-Stars and DS-Stars
List of parameter names Common to BR-Stars and DS-Stars.
illustrations, Figure 2 depicts an example of three simple DS-Stars schemas and three
Figure 2. Example of threeBR-Star Schemas and three DS-Star Schemas.
atching DS-Stars and BR-Stars
for each of the two Boolean function Identical and Synonym, its objective,
the results of applying these functions will be append to the SCF set of lists.
returns True if the two-element names e1 and e2 are
may be fact names or dimension names. When true, this function
means that the match of these elements makes sense.
this function in two steps of the matching process; firstly on fact names, and
fact name and the common dimension names into SCF Fact list and Dimension list.
r our running example (Figure 2), the function Identical (Bookings, Bookings) is returns True;
so then we reuse this function on the dimensions of the two Bookings facts in BR-
Star. Identical (Date, Date) is also True, so we append the Bookings fact to the SCF Fact list as
well as the common dimension Date in Dimension list to SCF.
returns True if the two element names e1 and e2 are synonym, and
20
Stars schemas and three
for each of the two Boolean function Identical and Synonym, its objective,
are identical and
may be fact names or dimension names. When true, this function
this function in two steps of the matching process; firstly on fact names, and
r our running example (Figure 2), the function Identical (Bookings, Bookings) is returns True;
-Star and DS-
fact to the SCF Fact list as
synonym, and

21
If Synonym (e1, e2) is True considering facts, we reuse it again on their dimensions. If we find
synonym dimensions, we append the synonym fact name and the synonym dimension names of
this fact to the Fact list and Dimension list in SCF respectively.
For our running example (in Figure 2), Synonym (Buy, Purchase) is True for the two facts;
therefore we continue to look for the synonym dimensions between the facts Buy (in DS-Star) and
Purchase(in BR-Star). Thus, since Synonym (Customer, Client) returns True for the two
dimensions, we append the Purchase fact to the Fact list in SCF as well as the synonym Client
dimension to the Dimension list in SCF.
Table 3 depicts the result of the Identical and Synonym functions on the example in Figure 2.
Note that, so far, only the lists in SCF have elements (i.e. fact, dimension) while SDF and SBF
remain empty.
Note that, the DW designer will be asked to treat the uncommon facts and uncommon dimensions
later manually in the matching phase. It is worth mentioning that the uncommon dimensions have
two cases: in one hand, the dimension name is loadable from the DS, but it is not considered by
the business user in BR. On the other hand, the dimension name is not loadable with data from
the DS, but it is required by the BR business.
Inorder to support the tractability of our approach, we highlight the main steps of the matching
process in the MatchStars algorithm. We used the open library Rita that takes two words, it
returns whether these two words are synonyms or not based on a threshold we have defined (less
than 1). The threshold value 1 causes synonym words to be different (see table 4).
Algorithm MatchStars
Aims: Matches the elements of BR-Star Schemas with the elements of DS-Star Schemas.
Input:
- S-BR: Set of star schemas from business requirements.
- S-DS: Set of star schemas from data source.
- fr: a fact in a star schema belonging to S-BR.
- fd: a fact in a star schema belonging to S-DS.
- DIM1: Set of dimensions of fr.
- DIM2: Set of dimensions of fd.
Output: SCF, SBF and SDF
Begin
Boolean flag= false;
String str="";
Integer i=1, j=1;
Foreach fact fr in S-BR do
Foreach fact fd in S-DS do
If (synonym (fr.name, fd.name) or Rita (fr.name,
fd.name) <1)
Foreach Dimension d1 in fr do
Foreach Dimension d2 in fd do
If (synonym (d1.name, dim2.name) or Rita
(d1.name, dim2.name) <1)
SCF[i].Dimension=d1.name
Increment i;
Endif
Endfor

22
Endfor
Flag = true
Str = fr.name
break
Endif
Else
Flag = false
If(rita(fr.name, fd.name)<1)
SCF[j].Fact=fr.name
Endfor
If flag
SCF[j].Fact=str
Else
SBF[j].Fact=fr.name
SDF[j].Fact=fd.name
Increment j;
Endif
Endfor
End
Table 3 depicts the result of applying the MatchStars algorithm on examples in Figure 2 when
different threshold values are taken.
Note that, in our example in Figure 2 we have equal number of facts for both BR-Stars and DS-
Stars. Inorder to increase the capabilities of our approach, assume now that BR-Stars are limited to
the first two facts (Bookings and Purchase) and the DS-Stars as it is (containing three facts). In this
case we append the additional fact (Payments) in the DS-Stars to the SDF. Table 4 depicts the
result for this case.
In the upcoming paragraphs we want to measure the matching between the DS-Stars and BR-
Stars. Semantic metrics is appropriate tool for this measurement.
Metrics for Measuring the Matching Between DS-Stars and BR-Stars:
In order to measure the matching between DS-Stars and BR-Stars, we define four metrics namely:
Common Fact/s (CF),Ratio of common fact/s (RCF), Common Dimension/s (CD)and Ratio of
common dimension/s (RCD).
The objective of the first metric (CF) is to returns the number of common fact/s between DS-Stars
and the BR-Stars. The syntax of this metric is as follows:
|)()(| DSSBRSCF −∩−=
Note that, S-BR represents the set of stars from business requirements; S-DS is the set of stars
from data source.
We can calculate the ratio of common fact/s between the DS-Stars and BR-Stars by the second
metric (RCF):
|)()(| DSSBRS
CF
RCF
−∪−
=

23
Note that, the denominator in this metric means:
|)()(||)(||)(| DSSBRSDSSBRS −∩−−−+−
The common dimension/s (CD) between (DS-Stars) and (BR-Stars) (CD) can be defined by first
determining the common dimension/s between each couple of the similar facts between (DS-Stars)
and (BR-Stars), then, the union of these common dimensions will give the common dimensions.
Note that, we need the number of common dimensions, so we use the cardinality concept. So the
syntax of our third metric will be as follows:
U
CF
i
ii fdDimfrDimCD
1
21 |))()((|
=
∩=
Here CF represents the number of common facts (metric 1); Dim1(fr) and Dim2(fd) represents the
set of dimensions of a fact inBR-Stars and the set of dimensions of a fact in DS -Stars respectively.
To measure the ratio of common dimensions we suppose that there are n facts in BR-Stars, and m
facts in DS-Stars; we define (RCD) metric as follows:
  = = ==
∩−+
= n
i
m
j
CF
ji
jiji fdDimfrDimfdDimfrDim
CD
RCD
1 1 1,1
2121 |)()(||)(||)(|
In addition to the previous functions and metrics, and in order to enhance the capabilities of our
system we define rules for empty fact and/or empty dimension. An empty fact (factless fact)
means the fact in star schema may contain no measure/s; empty dimension means the dimension
may contain no parameters. The following rules deals with such cases:
Rules for empty facts and/or empty dimensions
R1) For each fact f .in BR-Stars or in DS-Stars ifthe fact contains no measure (m), delete the fact
name (f).
o
0)( =fNofM
R2) For each parameter (pm) in each dimension (dx) in BR-Stars or in DS-Stars: if the number of
parameters in the dimension=0, delete the dimension (dx/dy).
Finally, we have three sets of lists as in table 3; namely, the set of common facts (SCF), the set of
business requirements fact (SBF), and the set of data source fact (SDF). Our consideration will
devote to the set of common fact (SCF) lists, which will be used to be the nuclear of the approved
star schemas. The other two set of lists will be matched manually by the DW designer. The result
of the manual matching will be added to the set of common fact lists (SCF).Note that the final
result of the matching process is a set of approved star schemas.
In the previous functions and metrics we use the semantic resource WordNet to overcome the
heterogeneity in the matching process. In this subsection we detail the benefits of using the
semantic resource.

24
Usage of a Semantic Resource
The matching of star schemas in the DW requires the usage of a semantic resource, the aim is to
identify whether a name of a given concept such as fact or dimension is semantically equivalent to
another or not. In our framework, we use the free and open source WordNet as a general semantic
resource. The reason behind using WordNet is its simplicity; hence, we can decrease the total time
for constructing the multidimensional model. The role of WordNet is obvious in the previous
functions: ENIdentical (e1, e2),ENSynonym (e1, e2).
As well, we use Rita, which is free and open source library designed to be simple but have a
powerful features [11]. For us, RiTa is suitable since it can be integrated with WordNet database,
another reason is that RiTa can implement in java. RiTa works by taking two words under testing
and check their resemblance by referencing the WordNet database and return the distance between
the two words semantically; if the distance equals 1, this means that there is no relation between
those words; if the distance is 0, this is indicate that the two words are synonym (have the same
meaning). There is another variation as fraction for the variable distance, this fraction is in [0..1].
In our framework we use the value (less than 1) as threshold; so, whenever the distance between
the two words is less than 1, this means that those two words are synonyms. The word can be in
form of noun or verb. Table 9 depicts usefulness of WordNet to overcome the heterogeneity that
may arise in the matching process.
Table 2. Fact/Dimension Synonym
The word Fact/dimension The synonym
Bookings Fact Reservation/Engagement
Payment Fact defrayment
Buy Fact purchase
Customer Dimension Client
City Dimension Metropolis
Note that, there will be six facts or more, thanks to WordNet, now the number of facts is only
three, as the three facts are synonyms. This considerable redundancy may hinder the design of the
multidimensional model. Indeed, the full automation of generating the approved star schemas may
be inaccurate. The involvement of DW designer is helpful in this process; the second step in our
approach is to intervene the DW designer.
3.2. Involvement of the DW Designer
Note that, our approach is semi-automatic; Hence, the DW designer can intervene to reflect the
correctness of the generated multidimensional elements. His / Her role in this step is twofold: on
one hand, the DW designer checks the set of common facts to confirm the generated elements in
means of semantic confirmation (see Figure 1). When the number of facts is reasonable this task
will not require much time and effort from the DW designer. On the other hand, the DW designer
manually matches the elements in the set of business Requirements fact list with those elements in
the set of data source fact list .The DW designer can play this role only when there exist one or
more business Requirements facts not matching any of the data source facts, or vice versa.
Generally, this operation could be very simplified when an appropriate semantic resource is used;
i.e., a semantic resource that correctly describes the business domain will produce better results
than a general resource such as WordNet. The result of the manual matching may concern facts,
dimensions and parameters; the DW designer adds these elements to the set of common fact list;
finally, the combination is forming the approved star schemas.

25
3.3. Generating approved star schemas
The final output of our framework is a set of approved star schemas; these schemas should satisfy
what the decision maker’s needs, as well as, loaded correctly with data from the operational data
source. Moreover, to enhance schema validity, we defined multidimensional constraints such as
avoid empty facts, as well as empty dimensions.
The sources of our approved star schemas are three sets namely the set of common facts (SCF)that
contains the common facts and its consequences as dimensions and parameters in both BR-Star
schemas and the DS-Star schemas. The second set is the set of business requirements facts (SBF)
which includes the facts and dimensions found only in BR-Star schemas. The third set is the set of
data source facts (SDF) that contains all facts with their dimensions found only in the DS-Star
schemas. Note that, in the matching process there will be two cases: the first one happens when
all elements of BR-Star schemas and the elements of DS-Star schemas are common.
Consequently, the approved star schemas encompass the elements in the set of common fact list.
In the second case, there will be unmatched element(s) in business requirements facts list or in
data source facts list. Here, the DW designer performs a manual matching process. We have
tested our prototype with various examples of star schemas, to generate resulted approved star
schemas.
4. RESULTS
The result of this paper is approved star schemas which represent the conceptual model for the
DW. To construct this model, we made a matching process between the star schemas generated
from business requirements in our previous work [14] and the star schemas issued from the data
source [16]. We use star schema in the matching process since it has a simple structure and
powerful features, for our knowledge, our work is first one to elaborate semantic metrics and use
star schema for matching purpose. To enhance our results we used WordNet as semantic
resource, the aim behind using WordNet is its simplicity as free an open source dictionary and
hence we can shorten the period of constructing the multidimensional model semantically. In
order to increase the approvability of our model, our approach gives the DW designer the chance
to intervene to confirm the acceptance of the concepts (facts, dimensions, etc) in the matching
process. The generated results (see Figures 3, 4 in the appendix) show the capabilities of our
approach for generating the approved star schemas. Our defined functions and metrics covered
the various cases that may a raised in the matching process.
5. CONCLUSION AND FUTURE WORK
Recently, researchers in the DW design area agree that the hybrid approach and semantic resource
considered as mandatory factors for designing the DW. But still there is no agreement about how
to generate the multidimensional model for the DW. In this paper, we generate approved star
schemas representing a multidimensional model by applying a matching process between the star
schemas issued from the business requirements and the star schemas generated from the data
source. For our knowledge, few works used the star schemas for the matching process. Our
approach encompasses three steps: the first step is Matching DS-Star schemas with BR-Star
schemas; the second step is involvement of the DW designer; the last step is the generation of
approved star schemas. Our contribution consists in defining two Boolean functions for the
matching process and four metrics for measuring its correctness. Moreover, we design
MatchStars algorithm showing the main steps of our approach. In order to produce accurate
approved star schemas, our approach allows the DW designer to intervene in the matching
process, as well as in the confirmation of the approved star schemas. To complete the picture, we
used WordNet as a semantic resource, the generated results shows the usefulness of WordNet to

26
reduce the redundancy of the multidimensional elements. We have used java NetBeans to build
our prototype; additionally, we use RiTa Library to be integrated with WordNet.
As a further work, we will complete the designing of the DW by expanding our conceptual model
to involve the logical and the physical models. As well, we will use domain specific semantic
resource to minimize the amount of human intervention.
REFERENCES
[1] F. Bargui, H. Ben-Abdallah, and J. Feki. "A natural language-based approach for a semi-automatic
data mart design and ETL generation." Journal of Decision Systems (2016): 1-36.
[2] G.Matteo and S. Rizzi. "A methodological framework for data warehouse design." Proceedings of the
1st
ACM international workshop on Data warehousing and OLAP. ACM, 1998.
[3] M. Thenmozhiand K. Vivekanandan. "An Ontological Approach to Handle Multidimensional Schema
Evolution for Data Warehouse." International Journal of Database Management Systems 6.3 (2014):
33.
[4] Ch. Adamson. Star Schema: The Complete Reference. US: McGraw-Hill Osborne Media, 2010
[5] O. Romero and A. Abelló. "A framework for multidimensional design of data warehouses from
ontologies." Data & Knowledge Engineering 69.11 (2010): 1138-1157.
[6] O. Romero and A. Abelló. "Automatic validation of requirements to support multidimensional
design." Data & Knowledge Engineering 69.9 (2010): 917-942.
[7] J. Mazón, J. Trujillo, and J. Lichtenberger. "Reconciling requirement-driven data warehouses with
data sources via multidimensional normal forms." Data &Knowledge Engineering 63.3 (2007): 725-
751
[8] O. Romero and A. Abelló "Multidimensional design by examples." InternationalConference on Data
Warehousing and Knowledge Discovery. Springer Berlin Heidelberg, 2006.
[9] M. Thenmozhi and K. Vivekanandan. "An ontology based hybrid approach to derive
multidimensional schema for data warehouse." International Journal of Computer Applications 54.8
(2012).
[10] F. Di Tria, E. Lefons, and F. Tangorra. "Academic Data Warehouse Design Using a Hybrid
Methodology." Computer Science & Information Systems 12.1
[11] https://rednoise.org/rita/index.php , Visited 20/3/2017.
[12] M. Thenmozhi and K. Vivekanandan. "A tool for data warehouse multidimensional schema design
using ontology." Int. J. Comput. Sci. Issues (IJCSI) 10.2 (2013): 161-168.
[13] A.Bonifati, F.Angela, S. Ceri, A.Fuggetta and S. Paraboschi."Designing data marts for data
warehouses." ACM transactions onsoftware engineering and methodology 10.4 (2001): 452-483.
[14] E.Elamin, S. Alshomrani, and J. Feki. "SSReq: A method for designing Star Schemas from decisional
requirements." Communication, Control, Computing andElectronics Engineering (ICCCCEE), 2017
International Conference on. IEEE, 2017.
[15] HH. Do, S. Melnik, and E. Rahm. "Comparison of schema matching evaluations." Net. ObjectDays:
International Conference on Object-Oriented and Internet-Based Technologies, Concepts, and
Applications for a Networked World. Springer, Berlin, Heidelberg, 2002

27
[16] E.Elamin, A. Altalhi, and J. Feki. “Heuristic Based Approach for Automating Multidimensional
Schemas Construction”. International Journal of Computer and Information Technology (ISSN:
2279–0764) Volume. 2017 November.
[17] E.Elamin. “A hybrid Semi Automatic Approach for the Design of Data Warehouse Conceptual
Model”. Ph.D, Sudan University of Science and Technology, Collage of Graduate Studies.
Sudan.October 2018.
APPENDIX
Figure 3. Our system’s result for matching BR-Star Schemas with DS-Stars Schemas
of Figure2.
Figure 4. Our system’s result for matching BR-Star Schemas with DS-Stars Schemas having synonyms.

28
Table 3. Result of the application of the Boolean functions on example in Figure 2
SCF: Set of Common Facts
Lists
SDF: Set of DS Facts Lists SBF: Set of BR Facts
Lists
Fact Dimension Parameters Fact Dimensions Parameters Fact Dimensions
Parameter
s
Booki
ngs
Date
- - - - - - -
Client
Purcha
se
Customer
- - -Concert
Date
Table 4. Result of the application of MatchStars algorithm on example in Figure2 with different values for
threshold.
SCF: Set of Common
Facts Lists
SDF: Set of DS Facts Lists SBF: Set of BR Facts Lists
Thre
shold
Fact
Dimensi
on
Para
meter
s
Fact Dimension
Para
meters
Fact
Dimensio
n
Parameter
< 1
Bookin
gs
Date
-
Paym
ents
Customer
- Sales Date -Payment
Client Date
Purcha
se
Custom
er
-
- -
Concert
Date
Bookin
gs
Date - Paym
ents
Customer
- Sales Date -
1
Payment
Client -
Date
Purch
ase
Customer
- -Concert
Date
Table 5. Result generated after applying MatchStars algorithm for example 2 having BR-Stars limited the
first two facts (BOOKINGS and PURCHASE)
SCF: Set of Common Facts
Lists
SDF: Set of DS Facts Lists SBF: Set of BR Facts Lists
Fact Dimension
Paramete
rs
Fact Dimension
Parameter
s
Fact Dimension
Paramete
r
Book
ings
Date
- Payments
Customer
- - -
Client
Room PaymentM
Concert Date
Purc
hase
Customer
- - -Concert
Date

A SEMANTIC RESOURCE BASED APPROACH FOR STAR SCHEMAS MATCHING

More Related Content

Similar to A SEMANTIC RESOURCE BASED APPROACH FOR STAR SCHEMAS MATCHING (20)

Recently uploaded (20)

A SEMANTIC RESOURCE BASED APPROACH FOR STAR SCHEMAS MATCHING