J. Software Engineering & Applications, 2010, 3, 869-874
doi:10.4236/jsea.2010.39101 Published Online September 2010 (http://www.SciRP.org/journal/jsea)
Copyright © 2010 SciRes. JSEA
869
Requirements Analysis: Evaluating KAOS Models
Faisal Almisned, Jeroen Keppens
King’s College, London, UK.
Email: faisal.almisned@kcl.ac.uk, jeroen.keppens@kcl.ac.uk
ABSTRACT
Wigmores charts and Bayesian networks are used to represent graphically the construction of arguments and to evalu-
ate them. KAOS is a goal oriented requirements analysis method that enables the analysts to capture requirements
through the realization of the business goals. However, KAOS does not have inbuilt mechanism for evaluating these
goals and the inferring process. This paper proposes a method for evaluating KAOS models through the extension of
Wigmores model with features of Bayesian networks.
Keywords: KAOS, Requirements Evaluation , Wigmores Chart, Bayesian Networks
1. Introduction
The alignment of requirements analysis to business goals
and objectives is essential for th e return of investment of
any project. KAOS is a goal driven requirements analysis
method that defines a goal tree with parent and sub goals.
KAOS assumes that achieving all sub goals of a parent
goal will guide to the achievement of the parent goal.
The inferring process in KAOS is informal, due to the
nature of deduction in KAOS, which is based on the as-
sumption that the completion of sub goals leads deci-
sively to the parent goal. However, there is no guarantee
that the previous assumption is always valid. The lack of
precise assessment for KAOS goals requires further con-
sideration. Usually, in realty some sub goals does not
lead to the parent goal due to some contextual knowledge
that was not measured completely in KAOS representa-
tion. Another cause of the uncertainty of goals originates
from the possibility of assigning multiple values to one
goal rather than only two possible values (true or false),
which is the only option taking into account in the cur-
rent features of KAOS. For instance, if one of the sub
goals was completed partially, there is no feature to
measure the impact of this sub goal to the parent goal.
This paper takes into account the possibility of failures
in achieving the ultimate goals in KAOS models. This
paper will propose a new graphical representatio n model,
which can absorb KAOS models to be represented
through it. The new model enables analysts to provide
measurable ultimate goals accompanied with probability
to give analysts statistical results. These results will fa-
cilitate the evaluation process of the whole KAOS model.
The new Model will formalize the inferring process to be
mathematically valid.
2. KAOS
KAOS is a goal oriented requirements analysis method,
developed by University of Oregon and university of
Louvain. KAOS stands for Knowledge Acquisition in
automated Specification [1]. The main advantage of
KAOS over other requirements analysis methods, which
are not part of the goal analysis family, is its ability to
align requirements to business goals and objectives. This
alignment increases the chances that the new develop-
ment will add value to business.
KAOS focus on realizing and indicating the business
goals, then specifying the requirements that infer to the
business goals. “Each goal (except the leaves, the bottom
goals) is refined as a collection of sub goals describing
how the refined goal can be reached” [2]. The structure
of the various connected requirements and goals is rep-
resented hierarchically in graphical notation in an up-
wards direction. The top go als are strategic objectives for
the business. As low as the diagram level reaches as
closer to the low level requirements. The root of the dia-
gram is the ultimate business goals. Then, the analysts
must identify the penultimate goals followed by the
lower goals and so on. The previous step is recurring
until the analysts reach the basic goals. The lower goals
are linked with the parent goals through union. The union
indicates that the completion of the lower goal success-
fully will definitely cause the completion of their parent
goal. Figure 1 shows an example of a simplified KAOS
model. KAOS main focus is on the business require-
ments, disregarding if this requirement is part of the
Requirements Analysis: Evaluating KAOS Models
Copyright © 2010 SciRes. JSEA
870
Figure 1. An example of KAOS.
computer system requirements or not. Each goal is ac-
companied with obstacles and the stakeholders involving
in this goal. A limitation of KAOS is the lack of any in-
ference evaluation capabilities. The achievement of sub
goals does not imply the achievement of their parent
goals in all cases. The next section presents a review of
two candidate approaches to solve this issue.
3. Related Work
In this section, two graphical representation models will
be studied as possible methods to evaluate KOAS models.
The features of these approaches will be examined to
check the suitability of th em to enclose KAOS models.
3.1. Bayesian Networks
Bayesian Network (BNs) is a general statistical tool that
can be applied to various applications. BNs are helpful to
assess the weight or the influences of premises, to deter-
mine the strong inference links. [3] Bayesian Network is
a graphical representation tool using symbols, numbers
and arrows to enable analysts to reason logically far from
doubt. It is an appropriate tool to gather and analyze evi-
dences, in order to produce strong arguments. There are
two components to construct BNs. First, nodes are rep-
resenting the noticed evidential facts, propositions and
variables. Second, arrow that connects between various
nodes in the diagram. These arrows indicate the depend-
ency probabilities. The value or the weight of each node
is affected by the value of the nodes influencing this
node and linked with it. The final conclusion of the net-
work is affected by the probabilities of each proposition
and inference. (See Figure 2 )
Bayesian Network is a method to reason logically and
rationally using probabilities. Th e simplest way to under-
stand the goals of BNs is to think of a circumstance you
need to “model a situation in which causality plays a role
but where our understanding of what is actually going on
is incomplete, so we need to describe things probab ilisti-
cally” [4]. There, BNs allow analysts to compute the
overall probability of the final conclusion . By, computing
the probability of propositions connected directly, then
the higher connections, then the higher and so on. The
benefits from BNs are obvious in the prediction of out-
comes in doubtful cases. Also, the benefits are apparent
in the detection of the causes of certain results. The in-
fluencing relations are not decisive but probabilistic; the
precise probability is assigned for each node and relation.
BNs are a Directed Acyclic Graph. BNs are constructed
from nodes and directed links. Arrows that connect vari-
ous propositions are accompanied with the probabilistic
information required to define the prob ability d istribu tion
all over the network. To achieve that, initial probability
value should be assigned to the nodes with no earlier
nodes. Then, calculate the provisional probability for the
Figure 2. Simple bayesian network.
Requirements Analysis: Evaluating KAOS Models
Copyright © 2010 SciRes. JSEA
871
rest of the nodes and for all possible combinations of
nodes and their antecedents. BNs permit the computation
of the provisional probabilities of every node, bearing in
mind that the value of some of the nodes has been speci-
fied before that computation took place. The diagrams’
direction of Bayesian Network is downwards. In brief,
the strength of the final argument is affected by the
probability calculations of the supporting evidences and
facts. The connections in the network represent the direct
inference probabilities. The structure of the network il-
lustrates the probabilistic dependency between various
variables in a case. Each node is accompanied with a
conditional probabilistic table of that node. The mixture
of values for the nodes' ancestors will be provided.
[5].The main incompatibility between Bayesian Net-
works and KAOS modeling is the fact that the direction
of BNs is downwards which contradicts with the deduc-
tion process of KAOS. However, the probabilities feature
is an important aspect to be added to the evaluation
process.
3.2. Wigmore’s Chart
Wigmore’s chart (WC) was created by John H enry Wig-
more (1913) to help lawyers. [6] Wigmore’s chart acts as
a legal reasoning diagramming method. Wigmore’s chart
considered as an argument diagramming techniques to
demonstrate the structure of reasoning and inferring for
an argument in a legal case. The diagram as a whole
identifies the logic, structure and grounds behind the
reasoning of arguments in legal cases. WC is a tool
which enables the creation of arguments followed by the
examination of those arguments, then the recreation of
those arguments. WC is valuable in cases surrounded
with doubt and uncertainty. In order to create WC, ana-
lysts of legal cases must identify the connections in all
steps of the arguments. Then, the analysts should break-
down the argument in to propositions an d facts. After that,
the analysts should connect these facts and propositions
together towards inferring the final conclusion of that
argument. The chart method of Wigmore has a number
of symbols to represent the different types of proposi-
tions and evidences. These symbols are connected with
arrows to specify the direction, influence and weight of
the inference. The final conclusions of the chart illu strate
the logical deduction of the propositions and facts that
assemble the inference. One of the main characteristics
of WC is the production of key lists. The key list con-
tains a list of all propositions, facts, evidences and as-
sumptions, which are used to build the final conclusion
of the arguments presented. In addition, inference maps
show the gathering and lin king process of evidences, this
validates the argument construction procedure. The chart
direction is upwards from facts to assumptions. The chart
contains symbols, numbers and arrows only, but, will be
accompanied with a key list clarifying the statement of
each proposition or evidence (see Figure 1). There are
five main symbols required for the construction of the
Chart Method of Wigmore according to Schum [7] (See
Figure 3).
Wigmore’s chat properties can be used to evaluate the
deduction process of KAOS models. But, the lack of
measurable results affects the reliability o f the evaluation
process of KAOS models.
3.3. Comparison
Bayesian networks and Wigmore’s chart have valuable
features, which can aid the needed evaluation of KAOS
models. However, their weakness does not provide a
sufficient method for evaluation. The following table
compare the two models.
Bayesian Networks Wigmore Chart
Based on statistics, using prob-
abilities calculation for prem-
ises and relations
Based on the natural logic of
rea-soning. In addition to the
skills and knowledge of the chart
creators
The network direction is down-
wards The chart direction is upwards
Not extendable notations, BN is
a Directed Acyclic Graph
Extendable notations, richer se-
mantics and it has some under-
standing of what it represents
Applicable to wide range of
domains, used in various ap-
plications
Designed for law domain, but
can be applied to other domains
only if it can be extended
Produce supportive probabilis-
tic arguments for the final con-
clusion
Enable the production of argu-
ment in favour and disfavour of
the desired outcome
More Complex generation Less complex generation
Top down approach Enable Top down and Bottom
up approaches
Measurable results, not decisiveEither for or against the intended
outcome
The perspective of the creators
does not play any role in the
outcome of the network
The Chart Method of Wigmore
allows the occurrence of multi-
ple evaluations and considera-
tions of same evidences in legal
cases, from various perspec-
tives.
The information flow from the
basic fact or v ariable to the final
outcome or goal
The information flow from the
final outcome or g oal to the basic
fact or variable
As shown in the e arlier comp arison, the need to combine
selected features from these two approaches could prove
to be beneficial in terms of producing valuable method to
evaluate KAOS models, as explained in the next section.
Requirements Analysis: Evaluating KAOS Models
Copyright © 2010 SciRes. JSEA
872
Figure 3. An example of wigmore chart.
4. Extending Wigmore’s Chart
Bayesian networks and Wigmore’s chart have number of
practical and valuable features. The integration of some
of these features will offer a model with superior capa-
bilities and usage. The offered model will encapsulate
several characteristics from both earlier models that do
not contrast with each other. The extended model has to
capture the properties, which are compatible with each
other. This will allow the production of useful model,
which facilitates the graphical representation of various
tasks.
The suggested model extracts most of its properties
from the chart method of Wigmore with the inclusion of
one property of Bayesian networks and other external
aspects. The model has to include additional aspects in
order to address the gaps, which are not fulfilled com-
pletely by BNs and WC. The new model has several fea-
tures. First, it enables both Top down and Bottom up
Approaches, in order to facilitate the generation of mod-
els starting from the basic premises or starting from the
desired conclusion. Second, the new model allows the
production of measurable results to provide more accu-
rate and reliable representation, through the introduction
of probabilistic calculations. The third feature states that
the new model should be extendable to be applicable to
various domains. This is related to the notations of the
models and the observation of contextual knowledge.
Fourth, the new model eases the creation of representa-
tion supporting the desired goal, and against the desired
goals. Finally, the new model will eliminate the com-
plexity and ambiguity raised from representing the mul-
tilayered nature of cases or similar repetitive patterns.
This multilayered nature could cause high complexity as
stated by Hepler “If all these features are represented in
one diagram, the result can be messy and hard to inter-
pret” [8]. Another cause of complexity is the reappear-
ance of a similar pattern of evidences and relations be-
tween facts and propositions, within the same case or in
similar cases. And it would be “wasteful to model these
all individually” [8]. It allows any network to contain an
instance of another network without showing the detailed
structure until requested. Moreover, it authorizes the
creation of general networks that contains repeated pat-
terns of evidences and relations, which can be reused
after few amendments to customize the structure to the
current case. This feature can be represented in the dia-
gram as a special symbol. This model aims to simplify
the creation of probabilistic graphical models and to
convert the presentation into more efficient and under-
standable form.
There are six main symbols required for the construc-
tion of the new model. The five foremost symbols are
derived from Wigmore’s chart in a generalized manner to
make them applicable to various domains. The last sym-
bol is to represent the new feature of representing a re-
peated pattern or inclusive diagram. First, white circles
are representing the directly related proposition s or goals.
Second, the black circle is a symbol of the directly re-
lated facts. Third, white squares stand for the subsidiary
propositions. Fourth, the Black squares, which corre-
sponds to the subsidiary facts. Fifth, arrows are showing
the flow of relations between propositions and facts. Ar-
rows are used to clarify the inference logic or flow in the
arguments. Finally, the black rectangle illustrates the
presence of repetitive pattern or another inclusive dia-
gram. The number inside the rectangle will refer to the
final conclusion of the repeated pattern or inclusive dia-
gram. The diagram direction is upwards. Every symbol
will be accompanied with the probability calculation
function, which will calculate the provisional probability
of each node based on the probabilities of the precedent
directly connected nodes. The black circles and squares
will be assigned with initial probability values.
There are a series of sequential steps to construct the
new model representation. First, the analysts should start
by identifying the ultimate goal from the analysis. Sec-
ond, the analysis tea m should realize and assign the final
conclusion of the model usage, the penultimate proposi-
tions which support the final conclusion and the middle
propositions that support the higher propositions. The
previous step could be repeated recursively. Third, the
analysis team have to define the provisional facts and
evidences that support all of the propositions in the chart.
This can happen by indicating the scenario behind the
construction for or against the goal of the analysis team
in this case. Fourth, the analysts have to list all key
premises and inference links to simplify the construction
process. Fifth, analysts should commit to the appropriate
construction of the model, by using the accurate symbols
Requirements Analysis: Evaluating KAOS Models
Copyright © 2010 SciRes. JSEA
873
and right features. Numbers will be assigned to each
symbol indicating the correct proposition or fact from the
key list. Finally, after the existence of real arguments, the
evaluation process sh ould start. The analysis team should
assess the arguments and evidences behind them. The
analysts have to assign the initial probability values then
calculate the provisional probabilities for the whole dia-
gram. Afterwards, the joint probability for the whole
diagram must be calculated, according to the probability
computations rules. By this, the evaluation process could
be emphasized. This will help to generate measurable
outcomes to solve various issues. Figure 4 presents the
usage of the new models symbols. The next section will
provide a glimpse about the significant of using our new
model.
5. Evaluation KAOS
This section will show how the new extended model
could be used to evaluate KAOS. The extended model
will enclose KAOS goals and provide a measurable
evaluation of the possibility of achieving the final out-
come. Figure 5 shows a basic KAOS model with three
goals.
The constructed KAOS model could be evaluated by
transferring the current goals and requirements in this
KAOS model to the new models’ graphical representa-
tion. This step is quite simple. The new model is simi-
larly upwards. Each goal will be in the same position in
the diagram as it was. The direct goals and requirements
are represented as white and black circles. All nodes in
KAOS tree will be represented as circles, the basic re-
quirements with no earlier nodes are black and the goals
are white. The accessory goals and requirements corre-
spond to white and black squares. In the new model,
analysts will be allowed to represent partially related
goals and their requirements as squares, the basic related
requirements are black and the related goals are white.
Unions inside the KAOS model can be represented as
arrows in the model. The constraints within KAOS
model can be represented as a rectangle, which can
symbolize the contextual knowledge or any compound
model that involves repeated pattern or another model
structure.
Figure 6 shows how to enclose the previous KAOS
model into the new model. Then, the initial probability
values have to be assigned to the nodes in the diagram.
After that, analysts have to calculate and assign the pro-
visional probability to all goals. These provisional prob-
abilities will be produced by computation functions as-
signed with the inference process, which will calculate
the provisional probability of each node based on the
probabilities of the precedent directly connected nodes.
This function should be following the acknowledged
probability rules.
6. Conclusions
This paper proposed an extension of Wigmore’s chart
model, intended for evaluating the inference process
among goals in KAOS models. Additionally, it provided
a mechanism to measure the possibility of achieving a
parent goal if its sub goals are achieved.
Figure 4. An example of the new model with sample prob-
abilities.
Figure 5. A basic KAOS model.
Figure 6. An example of the new model with the accompa-
nied provisional probabilities.
Requirements Analysis: Evaluating KAOS Models
Copyright © 2010 SciRes. JSEA
874
Both Wigmore’s chart and Bayesian networks were
reviewed before an extended Wigmore’s chart could be
proposed. The new model provides a mathematical eva-
luation of KAOS, increasing the chances of constructing
the right model. The new model presents a method for
producing measurable results of the overall goals.
The main obstacle of the proposed evaluation ap-
proach is that it is not always feasible to know and assign
the possibility of the inference from the leaves to the
parent node. The proposed model suggested the use of a
new separate model rather than extending KAOS model.
This is to avoid adding complexity to KAOS models, in
addition to the standardization grounds.
This work can be extended by building on the mathe-
matical properties of the extended Wigmore’s chart and
by identifying advanced means for assigning the initial
probability values.
REFERENCES
[1] A. Dardenne, A. van Lamsweerde and S. Fickas, “Goal-
Directed Requirements Acquisition,” Science of Com-
puter Programming, Vol. 20, No. 1-2, April 1993, pp.
3-50.
[2] Respect IT, A KAOS Tutorial, Objectiver, 2007.
[3] F. Taroni, C. Aitken, P. Garbolino and A. Biedermann,
“Bayesian Networks and Probabilistic Inference in Fo-
rensic Science,” John Wiley and Sons, Chichester, 2006.
[4] E. Charniak, “Bayesian Networks without Tears,” AI
Magazine, Vol. 12, No. 4, 1991, pp. 50-63.
[5] R. Neapolitan, “Learning Bayesian Networks,” Prentice
Hall, New Jersey, 2004.
[6] C. Reed, D. Walton and F. Macagno, “Argument Dia-
gramming in Logic, Law and Artificial Intelligence,” The
Knowledge Engineering Revie w, Vol. 22, No. 1, 2007, pp.
87-109.
[7] D. A. Schum, “A Wigmorean Interpretation of the
Evaluation of a Complicated Pattern of Evidence,” Tech-
nical Report, 2005. http://tinyurl.com/2a3elq
[8] A. Hepler, P. Dawid and V. Leucari, “Object Oriented
Graphical Representations of Complex Patterns of Evi-
dence,” Law, Probability and Risk, Vol. 6, No. 1-4, 2007,
pp. 275-293.