COGNITIVE INFLUENCES IN PRIORITIZING SOFTWARE

REQUIREMENTS

Nadina Martínez Carod and Alejandra Cechich

GIISCo Research Group, Computer Science Department, Comahue University, Neuquén, Argentina

Keywords: Requirements Engineering, Cognitive Informatics, Requirements Elicitation, Requirements Prioritization.

Abstract: In software development, the elicitation process and particularly the acquisition of software requirements

are critical success factors. Elicitation is about learning the needs of users, and communicating those needs

to system builders. Prioritizing requirements includes negotiation as an important issue, which becomes

extremely difficult, as clients often do not know exactly what they need. To overcome this situation, aiming

at improving stakeholder’s negotiation, we propose reducing the gap of misunderstanding between them by

the use of cognitive science. Particularly, we suggest using cognitive styles to characterize people from the

way their process information. In this paper, we introduce a case study showing that cognitive profiles may

affect requirement understanding and prioritization. Our controlled experiment shows that considering

cognitive profiles when performing elicitation might increase stakeholders’ satisfaction and prioritization

accuracy.

1 INTRODUCTION

In spite of the highly technical and formal nature of

software, creating it remains a very people-intensive

activity. Regardless of our cultural and educational

backgrounds, most of us have at least an intuitive

understanding that some people are more scientific,

logical, and orderly in their thought processes

whereas others are more artistic, intuitive, and

spontaneous. The system analyst (broadly defined)

also has a dual task: he or she must be a

communicator who can understand and respond to

what is found in observing and talking with those

who are commissioning a new system or who will be

the end-users of it; above all the system analyst must

be able to perceive correctly what is needed. In this

context, elicitation is about learning the needs of

users, and communicating those needs to system

builders (Hickey and Davis, 2003). And although the

literature suggests the elicitation as a simple process,

experience in real projects turns it in rather

complicated as stakeholders have different

viewpoints. As understanding involves aspects of

human processing mechanisms that are analyzed by

the cognitive sciences, we decided to look for

references into the Cognitive Informatics (CI), an

interdisciplinary research area that applies concepts

from psychology and other cognitive sciences to

improve processes in engineering disciplines like

software engineering (Wang, 2002).

Wang defines CI as “a branch of information and

computer science that studies computing by

cognitive methodologies and studies cognitive

science by informatics and computing theories”. CI

analyses how natural intelligence processes

information, using many sciences and engineering

disciplines in this work. CI can be studied from both

artificial intelligence and software engineering

viewpoints. Artificial intelligence studies the

mechanisms of natural intelligence and the

architecture of the brain often ignoring

psychological aspects of intelligence. Software

engineering (Chiew and Wang, 2003) is interested in

explaining the mechanisms and processes of

memory learning and reasoning. We focus on CI as

a software engineering area by considering

stakeholders’ characteristics, based on cognitive

psychology. From this starting point, we have

defined design and cognitive aspects as main

features to characterize different approaches of

elicitation prioritisation, aiming at identifying

possible improvements to the elicitation process

(Martínez Carod and Cechich, 2007). As related

work on the cognitive style direction, we have

already proposed a selection function for

requirements elicitation techniques according to

214

Martínez Carod N. and Cechich A. (2010).

COGNITIVE INFLUENCES IN PRIORITIZING SOFTWARE REQUIREMENTS.

In Proceedings of the 5th International Conference on Software and Data Technologies, pages 214-219

DOI: 10.5220/0003010602140219

 SciTePress

cognitive aspects of most stakeholders in a

distributed group (Aranda et al., 2005).

After analyzing varied psychological issues, we

set our interest in using some techniques called

Learning Style Models (LSMs), which may be

useful to select groupware tools and elicitation

techniques according to the cognitive style of

stakeholders. Few related works use psychological

techniques to solve problems in Software

Engineering. One work on this direction uses

cognitive styles as a mechanism for software

inspection team construction (Miller and Yin, 2004),

which describes an experiment to prove that

heterogeneous software inspection teams have better

performance than homogeneous ones, where the

heterogeneity concept is analyzed according to the

cognitive style of participants. Even when they also

used the concept of cognitive styles to classify

people, our approach is different because our goal is

choosing the best requirements specification to

improve understanding of an already given group of

people.

In this paper, we introduce a process to associate

cognitive profiles to requirements prioritization. The

last sections will present the design and results of an

interesting case study we have carried out, as well as

conclusions and guidelines for future work.

2 RELATING COGNITIVE

PSYCHOLOGY AND

SOFTWARE ENGINEERING

PROCESSES

Part of cognitive psychology theories are cognitive

styles, which classify people’s preferences about

perception, judgment and processing of information

(Miller and Yin, 2004), with the goal of analyzing

and understanding differences in human behaviour.

With the same idea, learning style models (LSMs)

classify people according to a set of behavioural

characteristics that concern the ways people receive

and process information, while their goal is

improving the way people learn a given task.

The model we have chosen as the basis for our

research is called the Felder-Silverman (F-S) Model

(Felder and Silverman, 1998). This model was

selected after studying different LSMs. The analysis

showed that the F-S model is the most complete

because it covers the categories defined by the most

famous LSMs and, additionally, the F-S model has

been widely and successfully used with educational

purposes in engineering fields (Felder and Spurlin,

2005). The F-S Model introduces four categories

(Perception, Input, Processing and Understanding),

each of them further decomposed into two

subcategories (Sensing / Intuitive; Visual / Verbal;

Active / Reflective; Sequential / Global).

Characteristics of each subcategory (Felder and

Silverman, 1998) are:

• Sensing people prefer learning facts and solving

problems by well-established methods, while

Intuitive people prefer discovering possibilities

and relationships, and dislike repetition.

• Visual people remember best what they see

(such as pictures, diagrams, flow charts, time

lines, films, and demonstrations). On the

contrary, Verbal people get more out of words,

and written and spoken explanations.

• Active people tend to retain and understand

information by doing something active with it

(discussing or applying it or explaining it to

others). In contrast, Reflective people prefer to

think about information quietly first.

• Sequential people tend to gain understanding in

linear steps, with each step following logically

from the previous one, whereas Global people

tend to work in large jumps, absorbing material

almost randomly without seeing connections,

and then suddenly "getting it".

People are classified into the different categories

by filling a multiple-choice test, available on the

WWW (Soloman and Felder, 2009), which returns a

rank for each subcategory. Depending on the

circumstances, people may fit into one category or

the other, being for instance, sometimes active and

sometimes reflective; so preference for one category

is measured as strong, moderate, or mild.

3 A PROCESS TO ASSOCIATE

COGNITIVE PREFERENCES

TO PRIORITIZATION

Previous to introducing our approach, we briefly

describe the elicitation techniques we have used, as

considered in (Hickey and Davis, 2003):

Interviewing: It is fundamental to gather new

projects’ background information, especially in new

domains. Interviews are widely used, primarily to

surface new information, to uncover conflicts or

politics.

Questionnaires: They depend on both the

analyst’ domain knowledge and the users’ written

expression. It always must be complemented with

COGNITIVE INFLUENCES IN PRIORITIZING SOFTWARE REQUIREMENTS

215

Figure 1: Cognitive-Driven Requirements Prioritization Process Graph.

another technique to obtain more details. Many

times the results are not as good as are expected.

Models: DFD diagrams, UML, state charts, Use

Cases, scenarios, storyboards. Any model specifies

different aspects and some are particular to some

applications.

Our technique complements these strategies by

focusing on the assessment of cognitive profiles,

which would help obtain a set of well-defined

requirements. For instance, starting from a goal-

oriented method (Kaiya, Horai, and Saeki, 2002),

our technique introduced in (Martínez Carod and

Cechich, 2007) extends goal graphs by using

stakeholders' cognitive characteristics.

Our process schema, shown in Figure 1, can be

divided into two sections: α

and α

. As we said, the

first one, α

, is related with preference management;

the other, α

includes specific phases for

requirements prioritization. In this paper, we have

instantiated sections α

and α

by using two

approaches – a use case-driven modelling with a

graphical notation, and a textual-based notation

following an ad-hoc elicitation.

Section α1. Phase 1 constructs statistical

predominant perceptions based on people’s

preferences about elicitation techniques. In this stage

there must be a great number of system analysts to

be convenient sampled. Each analyst must response

a questionnaire to identify his/her viewpoint about

appropriate and inappropriate elicitation techniques.

Next, these people must be classified according to

the F-S learning model. Thus, the information

gained would be the basis for the relation to each

elicitation technique. In the second phase,

stakeholders’ satisfaction levels are obtained based

on their F-S category detected by a questionnaire on

the Web. Results from this phase define the

cognitive preference profiles.

Section α2. Here, results from both previous phases

must be combined. Firstly, in Phase 3, each

stakeholder assigns an importance value to each

requirement according to his or her viewpoints. A

cognitive weight is attached to each requirement

according to relationships between the cognitive

features of the stakeholder and the elicitation

techniques’ communicational aspects. Then, in

Phase 4, results from the prioritization are analyzed

and cognitive weights and/or some elicitation

techniques might be changed reducing the

preference gap. Depending on the discrepancies, the

analyst may choose to change his/her elicitation

techniques, as it is specified in (Hickey and Davis,

2003).

3.1 Research Hypothesis

Before describing the research, it is important to

recognize that analysts’ satisfaction can be caused

by a variety of conditions from knowledge and

experience on elicitation to how comfortable they

feel with a particular elicitation technique. To

actually see whether there was a relation between

ICSOFT 2010 - 5th International Conference on Software and Data Technologies

216

Table 1: Distribution of Cognitive Preferences.

Act-Ref Sens-Int Vis-Vrb Seq-Glo

Mod- Mod- Mod- Mod- Mod- Mod- Mod- Mod-

Str Str Str Str Str Str Str Str

Act Mild Ref Sens Mild Int Vis Mild Vrb Seq Mild Glo

21,74% 47,83% 30,43% 30,43% 60,87% 8,70% 47,83% 47,83% 4,35% 17,39% 69,57% 13,04%

cognitive profiles, elicitation techniques, and

requirements prioritization, data from a survey and

from a controlled experiment were combined. In

order to do so, we surveyed 24 advanced students of

the University of Comahue, Argentina and asked

them to participate in an elicitation simulation. They

had already attended courses about requirements

engineering and data base definition; and many of

them were employed in industry. So, previous

knowledge about elicitation techniques was ensured.

Then, the following hypothesis had been

formulated for the validation:

H1: Cognitive profiles do not affect the result of the

requirements prioritization.

H2 Cognitive profiles do not affect the

understanding of software requirements.

4 A CASE STUDY

The assessment scheme is organized in three phases,

each one with a well-defined goal. The Preferences

Phase involves the detection of people’s preferences

according to the F-S learning style model as well as

their experience in applying elicitation techniques as

students and as practitioners in real projects. The

second phase (System X) studies their reaction with

a simulated case in a known domain. In this phase

we evaluate students’ satisfaction in prioritizing

requirements from a particular software

specification. Finally, in the third phase (System Y)

the study is replicated in another case, contemplating

opposite preferences. Both Phase 2 and Phase 3

constitute Stage 2.

Stage 1. The preference and knowledge section of

the first stage is made up of individual

questionnaires. To detect the student experience, we

labelled the different levels of experience as

extensive, enough, some, little and none. The results

for real cases showed that nobody had extensive

experience; only 16.66% had enough or some

experience and 83.33% had little or no experience at

all. The participants’ experience with elicitation

techniques was practically limited to interviews,

modelling cases and document analysis. As an

illustration only 4.17 % of students mentioned no

experience in interviews, 8.33 % no experience in

uses case modelling, and everyone had experience in

using models (diagrams, UML, state charts, etc). As

a statement, all of them mentioned preferring

previously known techniques.

In particular, the students were asked about how

they conducted their elicitation activities, about their

qualitative experience with the techniques and on

quantitative data in order to find out whether the

hypotheses held. Most of both questionnaires were

subjective, depending on the subjects and the

viewpoint from which they are taken. In both cases

the students contributed making a sort of judgment.

By classifying the preferences of students as

strong/moderated (values from 5 to 11 in the ILS)

and balanced (values from +3 to -3 in the ILS), we

found out the distribution of preferences shown in

Table 1. It is important to highlight the Visual

preference as the strong and moderate preference

with highest percentage 47.83 %,, in contrast to the

Verbal preference with 4.35%.

Stage 2. The next two phases involved working with

real problem simulation. We worked here with the

specification of two applications in the academic

domain. In this way, we reduced the understanding

gap between domain knowledge and working

scenarios students are familiar with. The main goal

of the first system (X) was to optimize faculty’s

classrooms and material resources. The system

managed not only the courses’ schedules but also

resource assignments. Visual SRSs (X1, Y1) were

made up of a Graphical Functional Diagram

showing system’s functions, UML Class Diagrams,

Use Case Diagrams, and UML Sequence Diagrams.

As opposite, the non-visual SRSs (X2, Y2)

described the same system domain using textual

notation. Both types of SRSs were tested by

software engineering professors to check their

similarity. The main goal of the second system (Y)

was adding new functionalities to the educational

web system support PEDCO

(http://pedco.uncoma.edu.ar); and here, SRS’s

treatments were similar to the case of system X. We

use a cross-validation experiment to obtain reliable

results. The population was divided in two groups: A

COGNITIVE INFLUENCES IN PRIORITIZING SOFTWARE REQUIREMENTS

217

and B. We used a dice to assure randomized

selection, thereby all the subjects had the same

possibility of being assigned to the group A or B.

Independently the group that each student was

assigned to, he/she was asked to accomplish the

activities individually playing the role of system

analyst. To overcome some threats, during the case

study not only data with respect to performance was

collected (time spent in prioritizing), but the student

doing process modelling in the case study was also

asked about his/her experience in the particular

elicitation technique.

Finally, as a first glance, we cannot assume our

sample as representative enough. The number of

students considered is small, and it is not possible to

perform a complex statistical analysis. Therefore, we

based our analysis on the comparison of

percentages. However, looking at the overview of

similar studies given by (Felder and Spurlin, 2005),

our results are mostly in agreement with the results

of these studies. Thus, we can suppose that our

sample is representative enough and can act as basis

for further analysis.

5 RESULTS AND SOME

LESSONS LEARNED

Although understanding was not a problem, we

found some interesting results that motivate us to

continue our research line. For instance, 81.8% of

respondents with strong visual preferences agreed on

feeling more comfortable with visual specifications,

as we expected; but 36.4% of respondents with

strong non-visual preferences felt more comfortable

with visual specifications. Another result showed

that few people had problems to understand

requirements; but 72.7% of the conflicts appeared

when the SRSs were not specified in concordance to

their preferences. Another result showed that a high

percentage (> 68%) of visual students spent more

time understanding non-visual SRSs than

understanding visual ones. This clear relation does

not appear in the other types of preferences (non-

visual). In general, except for an isolate case,

respondents did not spend much time in prioritizing

requirements; however some people were not sure

about how to assign requirements’ priorities. In this

case we could not find a clear relation with their

cognitive preferences.

The last part of stage 2 implied running a post-

experiment questionnaire, which associated

individual feelings about both applications. By

comparing satisfaction (Figure 2), we realized that

all people with strong and moderate visual

preferences felt more comfortable, spent less effort

and better understood specifications aligned to their

profiles.



Better

Less time

Less effort

More comfortable

Non-Visual

50 33 33 50

Visual

67 34 67 78

Better Less time Less effort More comfortable

Figure 2: Comparing satisfaction of visual and non-visual

people with respect to understanding a visual SRS.

More accuracy Be tte r

100

Non-Visual Visual

Figure 3: Comparing satisfaction of visual and non-visual

people with respect to prioritizing visual SRSs.

On the other hand, as we expected, verbal people

did not feel the same satisfaction with the visual

SRSs. Time showed an interesting result since both

groups felt that there was no difference at dealing

with visual or non-visual SRSs. Although further

research is needed, it seems that using visual

representations might be useful for making

understanding faster independently of the cognitive

profile (but with different effort for each case). It

seems that a better manipulation of specifications

might be related to cognitive profiles. This is clear

from the impact on prioritization (Figure 3). There is

no doubt that visual people perform better with

visual SRSs than non-visual people. Visual people

felt that accuracy of prioritization was higher when

using visual specifications. This fact let us wonder

whether a prioritization process does not reach

commitment because people consider requirements’

value differently, or just because they do not look at

the same features similarly when prioritizing.

More experimentation is needed for mild

preferences. To clarify the point, Figure 4 shows the

ICSOFT 2010 - 5th International Conference on Software and Data Technologies

218

results of prioritizing requirements of the system Y,

given by visual people using the non-visual SRS.

For example the black line represents a moderate

visual student, called MVi5, whose priorities for the

functional requirements (FR1.1 to FR1.14) and non-

functional requirements (NFR1.1 to NFR4.4) were

assigned between 5-9 (on a scale 0-10). Notice that

only 14 scores are below 6 in Figure 4. Similarly,

Figure 5 shows another group of visual people

scoring the same requirements but using the visual

SRS; and here, we found 21 scores below 6.

Considering that we had taking care of side variables

such as people’s background and context

knowledge, we might suppose there is a relation

among cognitive profiles, SRS notation and

prioritization values.



FR1.1

FR1,

FR1,10

,11

,12

FR1,13

FR1,14

NFR1,1

R1,2

R2,1

R2,2

R3,1

NFR3,2

R3,

R4,1

R4,2

NFR4,3

NFR4,4

VVi6 MVi7 MVi8 MVi9

Figure 4: Examples of prioritization given by visual

people using the non-visual specification of System Y.

Figure 5: Examples of prioritization given by other visual

people using the visual specification of System Y.

6 CONCLUSIONS

From our research, it can be said that cognitive

profiles might influence some aspects of process

elicitation, as demonstrated in the experiment.

However, controlled experiments have their

limitations as they cannot be generalized to every

situation using that technology or method. Thus,

more case studies would be helpful for the validation

of the approach. In particular, more experience with

the overall method and with concrete techniques

(such as goal-oriented graph prioritization) would be

helpful. Formulating such cases is part of our future

work in the short time.

ACKNOWLEDGEMENTS

We would like to thank the people involved in the

case study. This work is partially supported by the

UNComa project 04/E072.

REFERENCES

Aranda, G. Vizcaíno, A., Cechich, A.and Piattini. 2005. A

Cognitive-Based Approach to Improve Distributed

Requirement Elicitation Processes. Proceedings of the

4th IEEE International Conference on Cognitive

Informatics pp 322-330.

Chiew, V., Wang, Y., 2003. From Cognitive Psychology

to Cognitive Informatics. Proceedings of Second IEEE

International Conference on Cognitive Informatics, pp

114-120.

Felder, R. and Silverman, L., 1998. Learning and

Teaching Styles in Engineering Education.

Engineering Education, 78(7):674-681.

Felder, R. and Spurlin, J., 2005. Applications, Reliability

and Validity of the Index of Learning Styles.

International Journal of Engineering Education,

21(1): 103-112.

Hickey, A. and Davis, A, 2003. Elicitation Technique

Selection: How Do Experts Do It, Proceedings of the

IEEE International Engineering Conference.

Kaiya H., Horai H., and Saeki M., 2002. AGORA:

Attributed Goal-Oriented Requirements Analysis

Method. Proceedings of the IEEE International

Conference on Requirements Engineering, pp. 13-22.

Martinez Carod N. and Cechich A, 2007. A Cognitive

Psychology Approach for Balancing Elicitation Goals.

Proceedings of the 6th International Conference on

Cognitive Informatics, pp 232-241.

Miller, J. and Yin, Z., 2004. A Cognitive-Based

Mechanism for Constructing Software Inspection

Teams. IEEE Transactions on Software Engineering,

30 (11): 811-825.

Soloman B. Felder R., 2009. Index of Learning Styles

Questionnaire available at: http://www.engr.ncsu.

edu/learningstyles/ilsweb.html

Wang, Y., 2002. On Cognitive Informatics. Proceedings

of First IEEE International Conference on Cognitive

Informatics, pp 34-42.

COGNITIVE INFLUENCES IN PRIORITIZING SOFTWARE REQUIREMENTS

219