Using i* modeling for the multidimensional design of data warehouses

1
Using i modeling for the multidimensional design
of data warehouses

• Jose-Norberto Mazón, jnmazon_at_dlsi.ua.es
• Juan Trujillo, jtrujillo_at_dlsi.ua.es
• Toronto, 17th July 2008

2
Contents

• Introduction
• Current research
• Requirements for DWs
• Reconciling with data sources
• Deriving logical representations
• Conclusions and short term research

3
Contents

• Introduction
• Current research
• Requirements for DWs
• Reconciling with data sources
• Deriving logical representations
• Conclusions and short term research

4
IntroductionResearch problem

• Data warehouse
• Integrated collection of historical data in
  support of decision making process
• Multidimensional (MD) modeling
• Fact
• Contains interesting measures of a business
  process
• Dimension
• Represents context of analysis
• Resembles traditional method for database design
• Model at conceptual level
• Abstracting details related to specific
  technologies

5
IntroductionResearch problem
- Integrated collection of historical data in
support of decision makers
OLAP
INTERNAL
DATA MINING
DATAWAREHOUSE
ETL
CUBES
REPORTS
DATA SOURCES
WHAT-IF ANALYSIS
EXTERNAL
6
IntroductionResearch problem
- Integrated collection of historical data in
support of decision makers
OLAP
INTERNAL
DATA MINING
DATAWAREHOUSE
ETL
CUBES
REPORTS
DATA SOURCES
DATA SOURCES
WHAT-IF ANALYSIS
EXTERNAS
7
IntroductionResearch problem
- Integrated collection of historical data in
support of decision makers
OLAP
INTERNAL
DATA MINING
DATAWAREHOUSE
ETL
CUBES
REPORTS
DATA SOURCES
DATA SOURCES
WHAT-IF ANALYSIS
EXTERNAS
- Information needs cannot be understood by only
analyzing data sources
8
IntroductionResearch problem
- Integrated collection of historical data in
support of decision makers
OLAP
INTERNAL
DATA MINING
DATAWAREHOUSE
ETL
CUBES
REPORTS
DATA SOURCES
DATA SOURCES
DECISION MAKERS
EXTERNAS
- Information needs cannot be understood by only
analyzing data sources
9
IntroductionResearch problem
- Integrated collection of historical data in
support of decision makers
OLAP
INTERNAL
DATA MINING
DATAWAREHOUSE
ETL
CUBES
REPORTS
DATA SOURCES
DATA SOURCES
DECISION MAKERS
WHAT-IF ANALYSIS
WHAT-IF ANALYSIS
- Decision making processes must be understood
by designers
EXTERNAS
- Information needs cannot be understood by only
analyzing data sources
10
IntroductionDrawbacks of the state-of-the-art

• Only data sources are analyzed to define the
  conceptual MD model
• Incorrect information needs may be modeled
• Requirements are specified once the conceptual MD
  model is defined (even after the deployment of
  the DW)
• Incorrect MD elements may be modeled
• Requirements and data sources are not reconciled
• Complex ETL processes to populate the DW
• Thus, the DW is not viewed as a valuable resource

11
IntroductionNovelty of our proposal

• 1. Explicit requirement analysis stage
• Focus on decision making processes
• Information requirements
• 2. Transformation to a conceptual MD model
• Model Driven approach
• MD model agrees with decision makers
  expectations
• 3. Reconcile requirement model with data sources
• MD model agrees with data sources
• Completeness
• Faithfulness

12
IntroductionNovelty of our proposal

• 1. Explicit requirement analysis stage
• Focus on decision making processes
• Information requirements
• 2. Transformation to a conceptual MD model
• Model Driven approach
• MD model agrees with decision makers
  expectations
• 3. Reconcile requirement model with data sources
• MD model agrees with data sources
• Completeness
• Faithfulness

13
IntroductionNovelty of our proposal

• 1. Explicit requirement analysis stage
• Focus on decision making processes
• Information requirements
• 2. Transformation to a conceptual MD model
• Model Driven approach
• MD model agrees with decision makers
  expectations
• 3. Reconcile requirement model with data sources
• MD model satisfies decision makers needs
• MD model agrees with data sources
• Completeness
• Faithfulness

14
IntroductionObjectives of our proposal

• Defining a goal-oriented approach for DWs
• Based on i
• Model decision processes
• Decision makers are concerned about GOALS not
  directly DATA
• Traceability to a conceptual MD model
• Align with MDA
• Integrate requirements and data sources

15
MDA

• Model Driven Architecture (MDA)
• Object Management group (OMG) standard
• Using models in software development
• Computation Independent Model (CIM)
• Platform Independent Model (PIM)
• Platform Specific Model (PSM)
• Transformations between models
• Query/View/Transformation language (QVT)
• The code is obtained from PSMs

16
MDA

• Model Driven Architecture (MDA)

Describes user requirements
Contains information about functionality
and structure of the system without taking into
account the technology used to implement it
Includes information about the specific technology
that is used in the implementation of the system
on a specific platform
Every PSM is transformed into code to
be executed, obtaining the final software product.
17
MDA

• Query/View/Transformation language (QVT)
• Declarative part of QVT
• Transformation ?? set of relations
• Relations between metamodels formally defined and
  automatically performed
• Relations applied to models

18
MDA
MODEL 1
Declarative approach of QVT specifies
relationships that must hold between candidate
models
CANDIDATE MODEL
DOMAIN
R
MODEL2
METAMODEL NAME
KIND OF RELATION
WHEN WHERE CLAUSES
19
IntroductionOur proposal
DOLAP 2005 DaWaK 2006 DSS 2008
REBNITA 2005 RIGIM 2007
ER 2006 ER 2007 DKE 2007
20
Contents

• Introduction
• Current research
• Requirements for DWs
• Reconciling with data sources
• Deriving logical representations
• Conclusions and short term research

21
Requirements for DWs

• Goal Oriented Requirement Engineering
• DW supports the decision making process to
  fulfill goals of an organization
• Decision makers are concerned about goals
• Information requirements are obtained by refining
  decision makers goals
• MDA approach
• Information requirements must be derived into a
  conceptual MD model

22
Requirements for DWs

• CIM
• Goals and information requirements
• PIM
• Conceptual MD model
• QVT
• Transformation between models

23
Requirements for DWsDefining a CIM

• Classification of DW goals
• Strategic goals
• Change to a better situation
• Decision goals
• Take appropiate actions
• Information goals
• Related to required information
• Information requirements
• Interesting measures of business process
• Context of analysis

24
Requirements for DWsDefining a CIM

• i framework
• Modeling goals of decision makers and the
  required tasks and resources to fulfil them
• Several decision makers with different goals
• Two extensions of UML
• Profile for i
• Profile for adapting i to the DW domain

25
Requirements for DWsDefining a CIM
26
Requirements for DWsSample CIM
27
Requirements for DWsSample CIM
28
Requirements for DWsSample CIM
29
Requirements for DWsSample CIM
30
Requirements for DWsSample CIM
31
Requirements for DWsSample CIM
32
Conceptual MD model

• UML Profile for MD modeling
• Luján, Trujillo, Song. A UML profile for
  Multidimensional Modeling in Data Warehouses.
  Data and Knowledge Engineering. 2006.
• Class diagram

Stereotype Icon
Fact
Dimension
Base
FactAttribute
DimensionAttribute
Rolls-UpTo ltltRolls-UpTogtgt
33
Conceptual MD model
34
Conceptual MD modelObtaining an initial PIM
35
Conceptual MD modelObtaining an initial PIM
36
Conceptual MD modelSample initial PIM
37
Reconciling with data sources
RECONCILIATION
INITIAL PIM
USER REQUIREMENTS
PIM
DATA SOURCES
PSM
38
Reconciling with data sources

• The MD conceptual model is reconciled with the
  available data sources
• The DW will be properly populated from data
  sources
• The analysis potential provided by the data
  sources is captured by the MD conceptual model
• Redundancies are avoided
• Optional dimension levels are controlled to
  enable summarizability and to avoid inconsistent
  queries
• Reconciliating process is automatically performed
• QVT relations based on Multidimensional Normal
  Forms
• Lechtenbörger and Vossen. Multidimensional normal
  forms for data warehouse design. Information
  Systems 28(2003)

39
Reconciling with data sources
40
Reconciling with data sources
d
1..n
ltltRolls-upTogtgt
r
1
n_t1district, n_t2state
41
Deriving logical representations

• PIM
• UML profile for MD modeling Luján et al. DKE
  2006
• PSM
• Common Warehouse Metamodel (CWM)
• From PIM to each PSM
• QVT transformation

42
Deriving logical representations

• Common Warehouse Metamodel (CWM)
• Resource layer
• Standard to represent the structure of data
  according to certain technologies
• Relational metamodel
• Tables, columns, primary keys, and so on
• Multidimensional metamodel
• Generic data structures
• Vendor specific extension
• Oracle Express extension

43
Contents

• Introduction
• Current research
• Requirements for DWs
• Reconciling with data sources
• Deriving logical representations
• Conclusions and short term research

44
ConclusionsObjectives

• DW projects fail in support decision making
  process
• Requirement analysis stage is overlooked for
  defining a conceptual MD model
• Using i framework together with MDA

45
ConclusionsScientific contributions

• MDA framework
• UML profile for i
• Extension for using i in the DW domain
• Transformations to obtain a conceptual MD model
• Several kind of logical representations
• Multidimensional normal forms
• Reconciling data sources and requirements in a
  hybrid approach
• Eclipse-based prototype

46
Eclipse-based prototype
47
ConclusionsRelated work at LUCENTIA research
group
MDA DKE 2007 DSS 2008
Requirements for DWs RIGiM 2007
CIM
UML profile for Data mining DKE 2007
UML Profile for MD Modeling at DKE 2006
Data sources analysis ER 2007
PIM
Common Warehouse Metamodel
Security DSS 2006 IS 2007
UML for Physical Modeling at JCIS 2006
PSM
48
Short term research

• Studying unstructured decision processes in-deth
  to model them in i diagrams
• Taking advantage of every i feature
• Considering complex mechanisms to reason about
  goals and structure decision processes
• Prioritization of goals

49
Using i modeling for the multidimensional design
of data warehouses

• Jose-Norberto Mazón, jnmazon_at_dlsi.ua.es
• Juan Trujillo, jtrujillo_at_dlsi.ua.es
• Toronto, 17th July 2008