Slide sem ttulo

1
An Agent-Oriented Methodology for Requirements
Elicitation
Luiz Marcio Cysneiros Dept. of Math and
Stat Information Technology Program York
University
Eric Yu Department of Computer Science
University of Toronto
2
Motivation

• Nowadays we can frequently see new areas using
  computer systems to create new or more productive
  environments
• a large number of players playing many different
  roles interact among them trying to achieve
  personal and common goals

3
Motivation

• Take for example the health care domain
• Healthcare professionals, patients, and even
  family members would be supported in their
  interactions and decision making by various kinds
  of software agents personalized to meet
• Each actor would be able to personalize their
  systems eventually implicating in refusing to
  provide some services
• Agent-orientation has been emerging as a new
  paradigm to deal with these kinds of systems.

4
Motivation

• focus has been on
• Analysis
• Design
• very little coverage of
• Requirements
• what does it mean to do RE, given the AO
  characteristics?

• We need to prevent history from repeating itself
  Requirements modelling concepts being driven from
  programming!!
• Structured Analysis from Structured Programming
• OOA from OOD, OOP
• AOA from AOP ??

5
How to elicit, specify, validate requirements for
AO systems?

• behavioral requirements
• NB autonomy, pro-activeness,
• human-computer cooperation
• agent-agent cooperation
• non-functional requirements
• Privacy
• usability
• security
• reliability
• survivability
• scalability

• AO systems may require new RE techniques
• Could lead to fundamental questions about RE
  itself, e.g., due to autonomy, openness,

6
What we Aim

• Define an Agent-Oriented Methodology for
  Requirements elicitation
• This methodology
• Is based on the i Framework
• Provide methods and tools for understanding and
  modelling the current state of the domain
  (Non-Intentional Dimension)
• Provide methods and tools for understanding and
  modelling the goals that underlies the current
  process (Intentional Dimension)
• Provide methods and tools to introduce new
  alternatives and systematically evaluate them.
• This methodology intends to be well defined so it
  can be repeatedly used in a consistent manner

7
The Methodology at a glance

• Identify the common terms used in the domain
  (Using the Language Extended Lexicon)
• Identify the main Agents involved in the process
  and how they relate to each other
• Identify Main Dependencies
• Identify The Main Tasks
• Identify the Main Goals
• Identify the Softgoals
• Look For alternative Ways

Current Process (Basically Non-Intentional)
System-to-be
8
The i Framework

• Supports goal-orientation agent-orientation
• Deals with non-functional requirements
• Concepts in i graphical notation
• intentional elements
• goal, task, resource, softgoal, belief
• intentional links
• dependency, means-ends, decomposition,
  contribution, correlation
• Actors
• Agent, role and Position
• 2 Models
• Strategic Dependencies
• Strategic Rationales

9
The i framework 1. Strategic Dependencies

• Who do I depend on for what?

10
The i framework 2. Strategic Rationales

• Why? How? How else?

11
What is the LEL ?

• Aims to register the vocabulary used in the UofD
• It is based on a system of symbols composed of
  Notions and Behavioral Responses
• Notions specify the meaning of the symbol
  (denotation)
• Behavioral Responses register the results driven
  from the symbol utilization (connotation)
• Its construction is based on the minimum
  vocabulary and circularity principles

12
(No Transcript)
13

• Basically, we use Semi-Structured interviews with
  the identified actors and Document Reading
• Validate the Lexicon with Stakeholders !!
• Do it frequently !

14
Build a First Approach for the Social Context

• Can and Must happen in parallel with the LEL
  construction
• Starting Picking up all the Symbols of the LEL
  classified as subject. They are all strong
  candidates to be agents
• For each one of the symbols that you picked up to
  be an agent, look if any other agent is present
  in the notions of this symbol
• Symbols Classified as verbs can be either a
  task/goal or a role. Be careful.
• Ex

15
(No Transcript)
16
(No Transcript)
17
Building a First-Cut Dependency Model

• Identify Main Dependencies
• Identify Main Tasks
• Identify Main Resources
• Do it using a mix of semi-structured interviews,
  document reading, observation and protocol
  analysis.
• Always check to be sure that you are using LEL
  symbols in SD models
• Eventually use non-intentional process models as
  Workflow, DFD or UML diagrams
• Some Hard and Soft goal may eventually be
  elicited during this process but we will not be
  targeting it

Current Process
18
Identify Main Dependencies

• Whenever two symbols appear in the same sentence
  and both are subjects you may be facing a
  relationship among actors
• This kind of sentences may also hide tasks, goals
  and resources
• For each actor already found ask
• On whom you depend on to do your job ?
• What other people / devices do you interact with
  ?

19
Finding Main Tasks/Goals and Resources
Dependencies
B
A
20
Elaborating on Processes and Rationales

• Find the main tasks for each Actor
• Use SR Models
• Tasks are frequently easier to elicit then goals
• In many of the projects where KAOS was used the
  goals underlying the process were implicit and
  not really visible in the first place Van
  Lamsweerde 98
• Findings in the SD model can help building SR
  model and vice-versa

21
Elaborating on Processes and Rationales

• Symbols classified as verbs are candidates to be
  tasks in both SD and SR models
• Behavioral Responses in symbols that represent
  actors frequently hide a goal or a task
• Tasks can later become goals once one finds out
  that the depender does not care on how the
  dependee will perform this task

22
Elaborating on Processes and Rationales

• Common Questions
• For each actor found
• What is this person responsible for ?
• What are the processes in which this person is
  involved ?
• Note that although we are focused on getting the
  current process and not the reasons behind it,
  sometimes these reason will become quite clear
  and hence should be modeled.

23
Elaborating on Processes and Rationales

• Use scenarios to refine tasks and to expand an
  actor.
• For example
• In What possible scenarios could Nurse be
  involved ?
• Assess Patient
• Start discharging
• Coordinate discharging Process
• Describe the scenarios that are pertinent to the
  problem , validate them with the stakeholders and
  model them into SR models

24
(No Transcript)
25
Elaboration on the Intentional Dimension

• Once the current process is modelled (what) it
  is time to find out the whys.
• Try to determine not only agents goals but most
  of all organizational goals. What underlies the
  current process
• What are the reasons that led the stakeholder to
  establish them ?
• Why people do certain tasks ?

26
Identify Main Goals

• Check all the existing tasks to see if they dont
  hide any goals
• To each activity or cluster of activities ask
• Why this agent need to perform this task ?
• Why it has to be performed this way ?
• Are there alternative ways of doing that ?
• Keep asking why so you can get to high-level
  goals
• What is wrong with the process ?
• What should be changed ?
• What could be better ?
• Which goals are not satisfied ? Why ?

27
Identify Main Goals

• Use mainly semi-structured interviews and
  document readings. Eventually, protocol analysis
  and JAD Sections can be useful
• Check out for the need of privacy !
• This is not a top-down activity and neither
  bottom-up !

28
Intentional Dimension - Modeling Softgoals

• For each goal try to ask yourself and the
  stakeholder what NFR would be desirable.
• Do we need Safety ?
• Do we need Accuracy ?
• Do we need Performance ?
• Whenever the answer is yes, represent it in the
  model and refine it as in NFR Framework
• Refinement can be either top-down or bottom-up,
  most likely to be a mix of both
• Answers from the previous step could bring up
  some softgoals

29

• Evaluate models for interdependencies (both
  positive and negative)
• Evaluate different graphs for the same type of
  NFR
• Evaluate graphs involving NFR that are frequently
  conflicting (e.g. Usability and Security)
• If the number of graphs is not too large pair
  wise different graphs (excluding those previously
  compared)
• Negotiate with the stakeholders the different
  designs that could result from different ways of
  satisfying these NFR
• Agent-Oriented process naturally deals with
  different viewpoints

30
(No Transcript)
31
Exploring Further Alternatives

• Find all the goals not yet satisfied and decide
  how this goal will be achieved (choose among
  different way)
• At the end, all top goals (hard and soft) must be
  satisfied
• If we introduce new actors (both human or
  software agents) can we improve the process ?
• If we relocate responsibilities can we improve
  the process ?
• Eventually, turn the SR models into SD models or
  even into non-intentional models as WF so
  stakeholder can better visualize the final
  solution

32
(No Transcript)
33
(No Transcript)
34
The Case Study

• We conducted a case study involving 3 major
  hospital in Toronto
• Specifically, we studied the discharging process
  and the different ways it can be improved
• We got to a different solution involving the
  introduction of a new human agent supported by a
  new software agent.
• One of the hospitals is using the new process
  with good results

35
(No Transcript)
36
(No Transcript)