REFLECTIVE ANALYSIS OF THE SYNTAX AND SEMANTICS OF THE i* FRAMEWORK Jennifer Horkoff, Golnaz Elahi, Samer Abdulhadi, Eric Yu Department of Computer Science and Faculty of Information University of Toronto, Canada RIGiM’08
Dec 27, 2015
REFLECTIVE ANALYSIS OF THE SYNTAX AND SEMANTICS OF THE i*
FRAMEWORK
Jennifer Horkoff, Golnaz Elahi, Samer Abdulhadi, Eric Yu
Department of Computer Science and
Faculty of Information
University of Toronto, Canada
RIGiM’08
2
MODELING LANGUAGES: INTENTION VS. USE Modeling languages are introduced to serve a variety of
purposes, including: Facilitating communication Making tacit information explicit Storing knowledge Aiding analysis
(Sometimes the intention of a language is not explicitly expressed)
3
MODELING LANGUAGES: INTENTION VS. USE As a language is adopted and used, it may be used in ways which
the language creators did not anticipate It is useful to reflect on the intentions of the language versus its use
in practice We focus on the intentions of the syntax and semantics of a
language
Reflective Analysis of intention vs. use can help to answer:• What were the original intentions of the language syntax
and semantics?• How are the syntax and semantics being commonly used?• Why are users prone to deviate from the original language
description?• Do these variations matter?• Should the language be modified?• Or, should an effort be made to increase training?
4
REFLECTIVE ANALYSIS OF THE i* FRAMEWORK We analyze the intention vs. use of the i* Framework i* is a goal and agent-oriented framework which was
intended to be used in the early requirements stage to capture agents, their inter-relationships, and their goals
Aimed at helping to discover and compare high-level system design alternatives
i* EVOLVING A good candidate for
reflective analysis
5
The i* Framework was left open to
interpretation and modification
i* has been applied to many areas for
differing purposes
We compare i* usage to current U
of T Style
i* is used in system analysis courses
i* Adapted
Surveyed student i* assignments, research
papers and presentations
6
SURVEY METHOD Survey subjects looking for variations from U of T style:
15 student project assignments from a graduate level system analysis course
15 academic papers and presentations using i* drawn from various sources.
All models in each document were surveyed Variations were only counted once per document The models covered diverse application domains, including
health care, banking, and education systems. An analysis of the motivations behind the variations was also
performed Was the modeler confused about the syntax? Was the modeler using a syntactic shortcut? Does the variation indicate issues within i*?
7
THE i* FRAMEWORK: U OF T FLAVOUR Elements
Links between Elements
Actors and Actor Boundaries
Actor Association Links
D Help
Means-EndsDecomposition ContributionDependency
Make
Some + Unknown
Hurt Some -
Break
Hard Soft
8
THE i* FRAMEWORK: U OF T FLAVOUR Syntax Restrictions Include:
Goals decompose only with means-ends links to tasks
Decomposition links are only used from tasks
Dependums are needed
Dependency links must go outside actors
Actors are not nested
Contribution links only go to softgoals
Contribution links only inside actors
Actor Associations are restricted between certain actors
9
SURVEY RESULTS
Variations were grouped together in two ways:By syntax categoryBy perceived motivation
Decomposition Links
Dependency Links
Means-Ends Links
Contribution Links
Element Types
Other
17
108 7
35
7
14
8 97
52
Instances from Assignments Instances from Academic Work
10
SURVEY RESULTS: VARIATION INSTANCES BY SYNTAX CATEGORY
Number of Variations in
Category
8 6 8 5 4 4
DECOMPOSITION LINKS VARIATIONS
11
Decomposition links are drawn directly from goals to tasks
Decomposition links are used between goals
Softgoals are decomposed to tasks
Goals are means-ends decomposed to softgoals
Decomposition links extend outside actors' boundaries Decomposition
links are used between resources
Goals are decomposed to resources
Decomposition links are used between Softgoals
9 6
5
4
3
2
1
1
SURVEY RESULTS Variations Grouped by perceived motivations:
12
Perceived Motivation
The Nature of “Hard” Elements and Softgoals
Means-Ends vs. Decomposition
The Nature of Actor Boundaries
Potentially Ambiguous Dependencies
Understanding Associations
Intermediate Model Stages
Resource Refinement
Association Links vs. Inclusion
Dependums
Incomplete Evaluation
Total
Number of Variations
15
6
4
1
1
2
3
1
1
1
35
Instances
70
24
14
8
5
3
2
1
1
1
129
13
EXAMPLE VARIATION: NATURE OF “HARD” ELEMENTS AND SOFTGOALSGOAL (MEANS-ENDS) DECOMPOSED TO SOFTGOAL (7 INSTANCES)
Syntax not permitted in the U of T style of i* Semantics: The nature of soft and hard goals implies that a softgoal should
not be sufficient to satisfy a hard goal Something qualitative satisfies something concrete
However… In i* a task can be decomposed to a softgoal
The softgoal represents a quality the task should encompass
14
EXAMPLE VARIATION: NATURE OF “HARD” ELEMENTS AND SOFTGOALSGOAL (MEANS-ENDS) DECOMPOSED TO SOFTGOAL
Possible Responses: Discontinue decomposition of tasks to sofgoals
But how do we associate a quality with a task?
Create an alternative way to associate qualitative aspects to tasks, other than through decomposition Could also be used for goals?
Alternative Syntax
SURVEY RESULTS Variations Grouped by perceived motivations:
15
Perceived Motivation
The Nature of “Hard” Elements and Softgoals
Means-Ends vs. Decomposition
The Nature of Actor Boundaries
Potentially Ambiguous Dependencies
Understanding Associations
Intermediate Model Stages
Resource Refinement
Association Links vs. Inclusion
Dependums
Incomplete Evaluation
Total
Number of Variations
15
6
4
1
1
2
3
1
1
1
35
Instances
70
24
14
8
5
3
2
1
1
1
129
EXAMPLE VARIATION: MEANS-ENDS VS. DECOMPOSITIONGOAL DECOMPOSITION (16 INSTANCES)
Syntax: Goal decomposition is not permitted in the U of T i* style Semantics:
Tasks versus goals Goals can be accomplished in different ways Task is a particular way of doing something
Syntax restriction promotes the discovery of alternatives
16
Operate the Practice
Appointment Be
Scheduled
Medical Record Be Managed
Financial Mangement
Medication Refill Request Be Handled
Operate the Practice
Appointment Be
Scheduled
Medical Record Be Managed
Financial Mangement
Medication Refill Request Be Handled
Current Operation
Alternative Operation
….
….
17
EXAMPLE VARIATION: THE NATURE OF ACTOR BOUNDARIESCONTRIBUTION LINKS OUTSIDE BOUNDARIES (6 INSTANCES)
Syntax: Only dependency links should be used outside actor boundaries
Semantics: Emphasize actors’ autonomy Actors should not have knowledge of the internal motivations of other actors
18
POSSIBLE RESPONSES
Strict and loose versions of i* syntax: Strict syntax follows the U of T style restrictions Loose syntax relaxes a select set of rules based on common
variations: Means-ends for tasks, decomposition for goals Contribution links across actors Omitting dependums
The notion of syntactical shortcuts
Shortcut for
19
SURVEY RESULTS: STUDENTS VS RESEARCHERS Students are more likely to:
Have more difficulties understanding the difference between soft and “hard” elements
Have incomplete models (models with un-decomposed goals)
Misuse association links Researchers are more likely to adapt the Framework as
they see fit: More likely to use non-dependency links outside of actor
boundaries
20
CONCLUSIONS AND FUTURE WORK Based on an analysis of several variations we were able
to suggest: Areas where the syntax of i* could be modified
We would like to explore the notion of “loose” and “strict” versions of i* syntax
Recognized syntactic shortcuts Associating softgoals with hard elements
Areas where more training is needed Soft vs. “hard” elements Association links Consistent interpretation of standard syntax shortcuts
Future work can: Expand our survey pool and include analysis of further
variations Experiment with the utility of modified i* syntax
21
THANK YOU Jennifer Horkoff
[email protected] www.cs.utoronto.ca/~jenhork
Golnaz Elahi [email protected] www.cs.utoronto.ca/~elahi
Samer Abdulhadi Eric Yu
[email protected] www.cs.utoronto.ca/~eric
i* Wikihttp://istar.rwth-aachen.de/tiki-view_articles.php
22
THREATS TO VALIDITY The selection of academic papers and presentations was
not completely random The surveyor was less interested in papers without variations
Student assignments were longer than academic works, and had more examples But not all variations were higher for student assignments
The qualitative analysis of the variations was performed by the authors Variation intentions could be misinterpreted However, misinterpretation could indicate a general source
of confusion
23
THE NATURE OF “HARD” ELEMENTS AND SOFTGOALS
Variation # InstancesDecomposition links are used between softgoals 3
Softgoals are decomposed to tasks 2
Means-ends links are used between softgoals 1
Softgoal dependency is met by a goal 5
Softgoal dependency is met by a task 2
Goals are means-ends decomposed to softgoals 2
Goals are decomposed to softgoals 5
Contribution links are drawn from softgoals to tasks 4
Contribution links are drawn from softgoals to goals 2
Contribution links are used between goals 1
Contribution links are drawn from resources to tasks 1
Softgoal should be goal 10
Goal should be softgoal 15
Task should be softgoal 8
Softgoal should be task 7
Total 70
We select one category of perceived motivation to examine in detail
Goal (means-ends) decomposed to softgoal
24
SURVEY RESULTS: DECOMPOSITION LINKS EXAMPLE BREAKDOWN
Variation Assignments Academic Work Total Instances
Decomposition links are drawn directly from goals to tasks
5 4 9
Decomposition links are used between goals
4 2 6
Goals are means-ends decomposed to softgoals
2 3 5
Decomposition links extend outside actors' boundaries
1 3 4
Decomposition links are used between Softgoals
2 1 3
Softgoals are decomposed to tasks 2 0 2
Decomposition links are used between resources
1 0 1
Goals are decomposed to resources 0 1 1
Resources to tasks are decomposed to tasks
0 0 0
Total 17 14 31
70
24
14
85 3
211 1
15
6
4
1
1
2
31 1 1
The Nature of “Hard” Elements and Softgoals
Means-Ends vs. Decomposition
Actor Boundaries
Potentially Ambiguous Dependencies
Understanding Associations
Intermediate Model Stages
Resource Refinement
Association Links vs. Inclusion
Dependums
Incomplete Evaluation
SURVEY RESULTS Variations Grouped by perceived motivations:
25
Number of Variations in
Category
Total Instances per
Category