TOWARDS PRACTICAL TOOLS FOR MINING ABSTRACTIONS IN UML MODELS

1
TOWARDS PRACTICAL TOOLS FOR MINING ABSTRACTIONS
IN UML MODELS

• Michel Dao Marianne Huchard
• France Télécom RD LIRMM, Montpellier
• M. Rouane Hacène, C. Roume, P. Valtchev
• Université De Montréal

2
Outline

• Motivation
• Theoretical framework
• Formal Concept Analysis
• Relational Concept Analysis
• Practical issues
• Strategy for tools
• Experiments
• Conclusion / future work

3
Outline

• Motivation
• Theoretical framework
• Formal Concept Analysis
• Relational Concept Analysis
• Practical issues
• Strategy for tools
• Experiments
• Conclusion / future works

4
Motivation

• Mining abstractions in UML models
• model structuration
• Readability, classification
• reusability
• elements with higher abstraction level
• maintainability
• less duplicated elements
• model integration
• common part identification

5
Motivation An example

follows



measures
Generalization of elements (classes,
associations, operations, attributes etc.)
6
Motivation An example

follows

measures


Generalization of elements (classes,
associations, operations, attributes etc.)
7
Motivation An example
dealsWith



follows



measures
Generalization of elements (classes,
associations, operations, attributes etc.)
8
Motivation An example
dealsWith



follows



measures
Generalization of elements (classes,
associations, operations, attributes etc.)
9
Outline

• Motivation
• Theoretical framework
• Formal Concept Analysis
• Relational Concept Analysis
• Practical issues
• Experiments
• Strategy for tools
• Conclusion / future works

10
Formal Concept Analysis

• Extract abstractions from a set of entities
  described by characteristics
• Barbut Monjardet 1970 Ganter Wille 1999

Abstraction Concept
(C3,C4,f1,f3)
Extent Intent
Specialization
(C2,C3,C4,f3)
Binary Context
(C3,C4,f1,f3)
11
Formal Concept Analysis
Concept lattice
Binary Context
(C1,C2,C3,C4,)
(C2,C3,C4,f3)
(C1,C3,C4,f1)
(C3,C4,f1,f3)
(C2,f2,f3)
(C4,f1,f3,f5)
(C3,f1,f3,f4)
(,f1,f2,f3,f4,f5)
12
Formal Concept Analysis
Naive modeling with FCA
UML classes
13
Formal Concept Analysis
scaling binary context
only one concept
14
Formal Concept Analysis
One attribute abstraction without class support
balance
specializes
15
Relational Concept Analysis

• Towards exhaustive abstraction mining
• NB example restricted to classes/attributes

BankAccount balance
16
Relational Concept Analysis

• Formalization

Relational Context Family (RCF) (K, R) K set of
contexts K Kclass,Kproperty R set of
relations between entities of contexts R
type,ownedAttribute
17
Relational Concept Analysis

• Iterative lattice construction

Property lattice
Class lattice
18
Extent Intent
BA,TA,BAH,TC
CBATA
CBAHTC
BAH,TC ownedAttributeCbaltal
BA,TA ownedAttributeCbbabta
CBA
CTA
CBAH
BAH nameBasicAcHolder ownedAttributeCbal,
Cbaltal
BA nameBasicAccount ownedAttributeCbbabta,Co
TA nameTeenagerAccount ownedAttributeCbbabta,C
mw
CTC
TC nameTeenAccount ownedAttributeCtal, Cbaltal
BankAccount balance
BankClient accountListBankAccount
19
bba,bta,o,mw,bal,tal
Extent Intent
Cbaltal
bal,tal typeCBATA
Ctal
Cbal
Cbbabta
Co
Cmw
bal namebAccountList typeCBA,CBATA
bba,bta namebalance
mw namemaxWithdrawal
tal nametAccountList typeCBA,CBATA
o nameoverdraft
BankAccount balance
20
Relational Concept Analysis

• A normal form for models
• Well characterized
• Preservation of initial data (UML elements)
• Exhaustive discovery of abstractions for all UML
  elements
• Well-formed specialization/generalization links
  between UML elements
• Minimal number of constructed abstractions
• NB. The whole method includes also association
  ends, operations, etc.

21
Outline

• Motivation
• Theoretical framework
• Formal Concept Analysis
• Relational Concept Analysis
• Practical issues
• Strategy for tools
• Experiments
• Conclusion / future works

22
Practical issues - Framework

• Research project supported by RNTL
• french department for research
• call for contribution in software engineering
• Partners
• France Télécom RD
• LIRMM
• Softeam (CASE tool Objecteering)

23
Practical issues Tools
Galicia http//www.iro.umontreal.ca/galicia
Objecteering http//www.softeam.com
class diagram edition and visualization
tuning
RCA algorithms
24
Experiments

• France Télécom Projects
• Part of the general Information System
• 57 classes
• Intranet
• 1220 classes in 2 sub-systems
• User data organization
• 22 classes in a common model M
• 208613 classes in specific models from several
  departments (derived from M)

25
Experiments

• Focus on
• Part of the general Information System
• input
• 57 classes, 8 subpackages
• output
• 110 classes, no merged classes
• 9 associations, 59 properties
• 212 specialization/generalization links

26
Lessons learned and tool extension

• Hand-made Analysis by designers
• 62 useful classes among 110 new classes
• Many small classes (with 1 or 2 own properties)
• Issues
• User interface for controlling the process
• Traceability
• Unrelevant abstractions (e.g. att. with type
  String)
• Name semantics, conflicts, synonymy, hyperonymy,
  etc.
• Proposal of new tools in Objecteering

27
Outline

• Motivation
• Theoretical issues
• Formal Concept Analysis
• Relational Concept Analysis
• Practical issues
• Strategy for tools
• Experiments
• Conclusion / future works

28
Conclusion

• A theoretical framework for exhaustive
  abstraction discovery in models
• A normal form for UML models
• Practical feasability demonstrated in
  Objecteering Galicia tools
• Tested on France Télécom models
• Remaining limits
• Traceability not really achieved
• Name semantics
• Fine-tuning and generalization of the method

29
Current/Future work

• Model Driven Engineering Approach
• Explicit used metamodels
• Declarative transformation rules
• Definition of the process as a set of
  well-described small transformations
• EclipseKermeta (Univ. Rennes) tools under
  development
• Expected results
• Traceability
• Easy application of the method to any
  imput/output meta-models, ER, OWL, etc.
• fine-tuning the process

30
Current/Future work

• Natural Language processing
• In collaboration with researchers of the domain
• Research contract with France Télécom RD
• Expected results
• Better interpret words in models
• Naming constructed abstractions

31

• THANK YOU