Social Adaptation

When Software Gives Users a Voice

Raian Ali

, Carlos Solis

, Inah Omoronyia

, Mazeiar Salehie

and Bashar Nuseibeh

3,4

Bournemouth University, Bournemouth, U.K.

FEXCO, Killorglin, Ireland

Lero, Limerick University, Limerick, Ireland

Open University, Milton Keynes, U.K.

Keywords:

Requirements Engineering, Requirements-driven Adaptation, Social Adaptation.

Abstract:

Adaptive systems are characterized by the ability to monitor changes in their volatile world and react to

monitored changes when needed. The ultimate goal of adaptation is that users’ requirements are always

met correctly and efﬁciently. Adaptation is traditionally driven by the changing state of the system internal

and its surrounding environment. Such state should be monitored and analyzed to decide upon a suitable

alternative behaviour to adopt. In this paper, we introduce another driver for adaptation which is the users’

collective judgement on the alternative behaviors of a system. This judgmenet should be infered from the

individual users’ feedback given iteratviely during the lifetime of a system. Users’ feedback reﬂects their main

interest which is the validity and the quality of a system behaviour as a means to meet their requirements. We

propose social adaptation which is a speciﬁc kind of adaptation that treats users’ feedback, obtained during

the software lifetime, as a primary driver in planning and guiding adaptation. We propose a novel requirements

engineering modelling and analysis approach meant for systems adopting social adaptation. We evaluate our

approach by applying it in practice and report on the results.

1 INTRODUCTION

Self-adaptive systems are designed to autonomously

monitor and respond to changes in the operational

environment and the system’s own state (Laddaga,

1997). Each change can trigger a different response to

satisfy or maintain the satisfaction of certain require-

ments (Salehie and Tahvildari, 2009). Self-adaptive

systems have basic design properties (called self-*).

Self-protection property means the ability to monitor

security breaches and act to prevent or recover from

their effects. Self-optimization means the ability to

monitor the resources availability and act to enhance

performance. Self-healing means the ability to mon-

itor faults that have occurred or could occur and cure

or prevent their effects. Finally, self-conﬁguration

means the ability to monitor changes related to all of

the other properties and add, alter, or drop upon cer-

tain software entities (Murch, 2004).

Self-adaptivity is highly reliant on the system abil-

ity to autonomously monitor the drivers of adaptation

(security breaches, the available resources, faults and

errors, etc.). In (Ali et al., 2011c), we argued that

there are adaptation drivers which are unmonitorable

by relying on solely automated means. The judgeme-

ent of users on the validity and quality of each of the

alternative behaviours of a system, is an example of

that. Such judgement is an important driver for adap-

tation which enables a system to know how the va-

lidity and quality of its alternative behaviours are so-

cially evaluated. The feedback of individual users is

the main ingredient to obtain and process as a prelim-

inary step for a collective planing of adaptation. We

deﬁne Social Adaptation as the system autonomous

ability to analyse users’ feedback and choose upon an

alternative behaviour which is collectively shown to

be the best for meeting requirements in a context.

Obtaining and processing social feedback to sup-

port adaptation decision helps to accelerate adapta-

tion. This is particularly true for highly variable sys-

tems which incorporate a large number of alterna-

tives. For such systems, a large number of users will

be required to validate all of the system alternatives.

Establishing such validation as a design time activ-

ity which is directed by designers, as often done in

usability testing and user-centred design (Dumas and

Redish, 1999; Vredenberg et al., 2001) is highly ex-

pensive, time-consuming and hardly manageable. So-

cial adaptation allows the actual users to act as val-

idators and give feedback at runtime so that the sys-

Ali R., Solis C., Omoronyia I., Salehie M. and Nuseibeh B..

Social Adaptation - When Software Gives Users a Voice.

DOI: 10.5220/0003991900750084

In Proceedings of the 7th International Conference on Evaluation of Novel Approaches to Software Engineering (ENASE-2012), pages 75-84

ISBN: 978-989-8565-13-6

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

tem can analyse the set of users’ feedback and vali-

date upon each system alternative by analysing how

it is collectively judged. This is particularly impor-

tant when there is a large space of system alterna-

tives already implemented and enacted and their qual-

ity judgement is likely to change frequently over time.

Social adaptation advocates the repetitive obtain-

ing and analysis of social feedback to keep adaptation

up-to-date. Due to the dynamic nature of the world,

which includes users trends and experience, compet-

itive technologies, and context of use, users’ judge-

ment on the system is sometimes dynamic. Thus, a

traditional one step design-time system validation and

customization might lead to judgements which are

correct but only temporarily. For example, users’ who

currently like interaction via touch screens, might dis-

like it in the future when a better interaction technol-

ogy is devised. Users’ evaluation of a system alter-

native is not static and what is shown to be valid at

certain stage of validation may become eventually in-

valid. Thus, the repetitive nature of social adaptation

allows to accommodate the volatile nature of users

judgement on a system which would not be easy to

accommodate in traditional usability testing sessions.

Requirements are the natural subject of social

feedback and, thus, social adaptation should be in the

ﬁrst place a requirements-drivenadaptation. Upon the

execution of a system alternative, users will be pri-

marily concerned whether their requirements are met

correctly. Users would also care about the degree of

excellence of an execution against certain quality at-

tributes. Consequently, social feedback concerns the

validity and the quality of a system alternative as a

way to meet users’ requirement. This judgement is af-

fectable by context which, thus, has to be monitored.

The same system alternative could be judged differ-

ently when operating in different contexts. When a

context C occurs, social adaptation analyses the feed-

back provided for each alternative when it was exe-

cuted in C and choose the alternative which is col-

lectively judged to suit C. However, as we proposed

in (Ali et al., 2010), some contexts have clear inﬂu-

ence on the applicability and quality of systems be-

haviour regardless the users’ feedback.

In this paper, we propose social adaptation which

relies on the users’ collective judgement on the alter-

native behaviours of a system as a ﬁrst-class driver for

adaptation. We discuss the foundations and motiva-

tions of social adaptation. We provide a conceptual-

ization of the main artefacts needed in a requirements

model for systems adopting social adaptation. We de-

velop analysis techniques to process social feedback

and choose upon the system alternative which is col-

lectively shown to be the most appropriate to meet a

requirement. We evaluate our approach by applying

it on a messenger system looking to convey messages

in a way collectively shown to ﬁt a certain context.

The paper is structured as follows. In Section 2

we deﬁne and motivate and discuss main principles

of social adaptation. In Section 3 we present a set

of core modelling artefacts for social adaptation from

a requirements engineering perspective. In Section 4

we develop analysis algorithms to process social feed-

back and adapt the system at runtime. In Section 5

we evaluate our approach in practice. In Section 6 we

discuss related work and in Section 7 we present our

conclusions and future work.

2 SOCIAL ADAPTATION

Social adaptation is a speciﬁc kind of adaptation

which responds to the social collective judgement on

the correctness and efﬁciency of a system behaviour

in meeting users’ requirements. Social adaptation

treats social feedback as a primary driver for adap-

tation. Social feedback allows a runtime continuous

evaluation of each of the alternative behaviours of a

system. The alternative which is socially proved to

be correct and more efﬁcient in a certain context will

be the one to apply when that context occurs. That

is, social adaptation allows for inferring the applica-

bility and the quality of each system behaviour over

a space of different contexts. In Figure 1, we outline

the loop of social adaptation where the system anal-

yses social feedback and decide upon which alterna-

tive behaviour to apply, applies it, and ﬁnally gets and

stores the feedback of the users of that operation.

Analyse and

choose alternative

Social

Feedback

Apply

alternative

Obtain and store

feedback

Figure 1: Social adaptation loop.

The key characteristic of social adaptation is the

incorporation of users’ perception and judgement as a

part of the system monitor. In self-adaptation, the sys-

tem has to monitor changes in its internal and opera-

tional environment autonomously. Monitored events

and states could indicate security breaches trigger-

ing a self-protecting action, or new settings of the

available resources triggering a self-optimizing ac-

tion, or faults triggering a self-healing action. In

social adaptation, the system has to monitor the so-

cial judgement about its role in meeting requirements.

ENASE2012-7thInternationalConferenceonEvaluationofNovelSoftwareApproachestoSoftwareEngineering

Such judgements concern humans opinions and con-

clusions rather than events and states in the system

environment. In self-adaptation, the system enjoys an

autonomous ability to monitor all information which

is necessary for taking adaptation decisions which

means that the monitor is a fully automated com-

ponent. Social adaptation, on the other hand, re-

quires capturing its own kind of drivers, the users’

judgement, which is un-monitorable by relying on

solely automated means as we discussed in (Ali et al.,

2011c). Social adaptation requires a socio-technical

monitor which involves users’ perception as an inte-

gral part of the system monitor.

Social adaptation includes users in the activity of

shaping adaptation and evolving it continuously dur-

ing the lifetime of a software. Designers should plan

adaptation at design time and leave certain adaptation

decisions to be taken collaboratively by the commu-

nity of users at runtime. Users are enabled to pro-

vide their feedback and the system will analyse users

feedback and adapt accordingly. Self-adaptive sys-

tems follow an evolution rationale, which is planned

at design time, until the system apparently fails in

taking correct adaptation decisions where a mainte-

nance evolution has to take place. Social adaptation

enriches this evolution paradigm by allowing users-

driven runtime evolution. A socially-adaptive system

responds to the users feedback which is itself likely to

evolve over time, and thus the adaptation rationale it-

self evolves without the need to re-design the system.

As a running example to illustrate the rest of this

paper, we consider the case of a gallery-guide system

designed to support the visitors of an art gallery. The

system has two alternatives to convey information

about art pieces to the visitors. PDA-based system

alternative: by which the system will use the visitor’s

PDA (personal digital assistant) to explain, possibly

interactively, about an art piece. Staff-supported sys-

tem alternative: by which the system will try to ﬁnd

a mentor to meet with the visitor and explain the art

piece and also facilitate the meeting. Each of the two

alternatives ﬁts a set of different contexts. For certain

contexts, designers can not be fully certain of which

alternative (PDA-based, Staff-supported) is more ap-

propriate. As a solution, they leave this decision to

be taken by the users collectively. The system will try

these two alternatives in the different context where

designers are uncertain of their quality and validity

and get user feedback and infer the ﬁtness between

each system alternative and a context variation. This

process should be iterative to cope with the aforemen-

tioned fact that users judgement is not static and likely

to change over time.

Users can judge and give feedback whether a sys-

tem alternative is a valid means to reach their require-

ments and its quality. We classify social feedback into

two kinds:

• System Validity which concerns whether a certain

applied system alternative succeeds in meeting a

certain requirement in a certain context. Users are

able to give feedback using requirements-related

terms rather than software-related ones. For ex-

ample, upon executing the PDA-based system al-

ternative, a gallery visitor who is not really famil-

iar with the use of PDAs would give a feedback

as a Boolean answer saying “I could not ask for

and get information”. The visitors cannot gener-

ally explain how the complexity or simplicity of

the HCI design prevented or facilitated the role of

the system in the process of asking for and getting

information.

• System Quality which concerns the degree of ex-

cellence of a system alternative. For example, a

visitor who is familiar with the use of PDAs but

had to wait for a long time to meet a mentor,

would admit the delivery of information upon the

execution of the PDA-based system alternative,

i.e. the requirement is reached, but would prob-

ably give a negative assessment of some quality

attributes like “comfort” and “quick response”.

3 MODELLING REQUIREMENTS

FOR SOCIAL ADAPTATION

In the last section we discussed the deﬁnition and mo-

tivations of social adaptation and its main character-

istics. It relies on users to give feedback about the

validity and the quality of each of the different sys-

tem alternatives. This feedback could be inﬂuenced

by certain contextual attributes. The ultimate goal of

social adaptation is to identify the best alternative to

adopt in each settings of contextual inﬂuencers. In

this section, we relay on our discussion in the last

section and propose a set of fundamental artefacts a

requirements models should include when designing

systems to enact social adaptation. Figure 2 shows

the metamodel of these artefacts and Figure 3 shows

an instance of it. It is worth pointing out that this

model is meant to operate on the top of established

requirements models which capture the relation be-

tween requirements, quality attributes and system al-

ternatives. It is meant to extend such requirements

models by the social feedback about this relation and

the context inﬂuence on it. We will show an example

of the application of our metamodel on a main-stream

requirements model, the goal model, in Section 5.

SocialAdaptation-WhenSoftwareGivesUsersaVoice

Requirement System Alternative Quality Attribute

m eant to m eet

1..*

quality refied by

0..*

1..*

Validity I nfluencer

influences ability of

0..*

1..*

in m eeting of

1..*

0..*

Quality I nfluencer

influences excellence of

0..*

1..*

against

1..*

0..*

Context Attribute

Operation

execution of

0..*

Quality FeedbackValidity Feedback

evaluating

0..*

regarding

0..*

evaluating

0..*

regarding

0..*

Figure 2: Requirements modelling for social adaptation.

Instance of the Model

Requirement

R: visitor can ask for and get information about an art piece

System

Alternatives

: PDA-based (input handling, interactive presentation, video, etc.)

: Staff-supported (estimating time to meet, notifying staff, etc.)

Quality

Attributes

: visitor is well-informed

: visitor comfort

Validity

Influencers

: visitor’ age, C

: visitor’ technology expertise level, influence

the ability of S

to meet R

: estimated time for a staff to meet visitor, C

: visitor has

companions? Influence the ability of S

to meet R

Quality

Influencers

, C

, and C

: complexity of art piece information, influence the

quality of S

with regards to Q

, C

, and C

: user movement status (moving, standing, sitting),

influence the quality of S

with regards to Q

: staff expertise and C

: staff ability to speak the visitor’s native

language, influence the quality of S

with regards to Q

and C

: existence of a free seat closed to the visitor’s location,

influence the quality of S

with regards to Q

Runtime Feedback Example

Operations

Operation

: execution of S

. The values of its relevant context

attributes are C

= >65 years, C

=low, C

: low, C

= standing.

Operation

: execution of S2. The values of its relevant context

attributes are C3 = <5 min, C

= alone, C

= medium, C

= fair, C

= no

Validity

Feedback

Operation

.Validity_feedback= False (R is not reached)

Operation

.Validity _feedback= True (R is reached)

Quality

Feedback

Operation

. Quality_feedback is irrelevant (R was judged unreached)

Operation

.Quality_feedback(Q

)= medium

Operation

.Quality_feedback(Q

)= high

Figure 3: An instance of the model of requirements artifacts for social adaptation shown in Figure 2.

The central horizontal part of the metamodel of

Figure 2 captures the relation between the require-

ments and quality attributes on one side, and the space

of system alternatives one the other. Requirement

indicates an intention to reach a certain state of the

world. System Alternative is a synthesis between hu-

man and automated activities designed to meet a re-

quirement. A requirement could be reached via mul-

ENASE2012-7thInternationalConferenceonEvaluationofNovelSoftwareApproachestoSoftwareEngineering

tiple system alternatives. Quality Attribute is a distin-

guished characteristic of the degree of excellence of a

system alternative. In our framework,these three arte-

facts and the relations between them are speciﬁed by

the designers at design time and are static and, thus,

are not subject of monitoring at runtime.

The lower part of the metamodel of Figure 2

stands for the context inﬂuence on the relation be-

tween the system alternatives on one side and the re-

quirements and quality attributes on the other. Con-

text Attribute is a distinguished characteristic of the

environment within which the system operates. Va-

lidity Inﬂuencer is a context attribute that inﬂuences

the ability of a system alternative to meet a require-

ment. Quality Inﬂuencer is a context attribute that

inﬂuences the degree of excellence of a system alter-

native with respect to a quality attribute. The context

attributes, of both categories, are speciﬁed by design-

ers at design time and monitored at runtime, i.e. the

real values are obtained and stored at runtime.

The upper part of the metamodel of Figure 2

stands for the social feedback that expresses users’

judgements about each operation of a system alter-

native. Operation is a single execution of a system al-

ternative. Validity Feedback is a Boolean judgement

given by a user concerning his evaluation whether an

operation has led to meet his requirement. Quality

Feedback is an assessment, reiﬁed by a numericvalue,

given by a user concerning his evaluation of an op-

eration against a quality attribute. The social feed-

back, of both categories, is speciﬁed at design time

by designers. The value of the feedback relevant to a

speciﬁc system alternative is obtained from users and

stored at runtime after an operation of that alternative

is executed.

In this work, and to enable the analysis which we

propose in Section 4, we restrict the values of each

context attribute to belong to an enumeration speci-

ﬁed by the analyst. For example, the visitor age could

be speciﬁed to be in {“<18”, “between 18 and 25”,

“between 25 and 65”, “>65”}. The validity feedback

is already restricted to be a Boolean value. We also re-

strict the quality feedback value to be within a range

of integers [0..n] where 0 stands for “the alternative

has very bad quality against the quality attribute q”

and n stands for “the alternative is excellent against

the quality attribute q”.

4 SOCIAL ADAPTATION

ANALYSIS

The main goal of obtaining social feedback is to sup-

port the system decision about the best alternative

to apply for reaching users’ requirements and over-

come the designers’ uncertainty about this decision

and cope with the changing trends of users and the

world over time. When the system has to meet a re-

quirement, it has to choose an alternative to apply.

The system has to assess the collective judgement of

each alternative of being a valid and a good-quality

means to meet requirements. We propose to take into

consideration basic factors that together help for in-

ferring the collective judgement on the validity and

quality of a system alternative. We emphasize here

that this set is not restricted and more research would

lead to discover yet other factors. In what follows,

we discuss our proposed factors taking examples from

Figure 3.

• Feedback Value. This factor stands for the val-

ues given by the users when evaluating the va-

lidity and quality of each operation of a system

alternative. Users provide the validity feedback

by giving a Boolean answer reﬂecting their judge-

ment whether the operation led to reach their re-

quirements. Users evaluate the quality of each op-

eration of a system alternative against each qual-

ity attribute by giving a value within a designated

rank [0..n] where 0 means the lowest quality and

n means the highest. These values are the basic

factor in assessing the validity and the quality of a

system alternative.

• Feedback Relevance. This factor stands for the

meaningfulness of each of the users’ validity and

quality feedback when assessing a system alter-

native. This relevance will be interpreted as a

weight for the feedback value which reiﬁes the

user’ judgement of the validity and quality of a

system alternative. We consider two sub-factors

which inﬂuence the relevance:

– Feedback Context. This factor stands for the

match between the context of a particular oper-

ation of a system alternative for which the feed-

back was given, and the current context where

a decision about that alternative has to be taken.

The validity of a system alternativeand its qual-

ity against each quality attribute are affected by

a set of context inﬂuencers as we explained ear-

lier. The more the match between the values

of these context inﬂuencers when the feedback

was given and their values at the assessment

time, the more relevant the feedback is. For

example, suppose the system is assessing the

validity of the PDA-based system alternative

and that the current values for its validity inﬂu-

encers are (C

: visitor’ age) = “> 65”, (C

: vis-

itor’s technology expertise level) = “low”. Sup-

pose we have two validity feedback F

= “valid”

SocialAdaptation-WhenSoftwareGivesUsersaVoice

and F

= “invalid” and the values of contexts

for F

and F

are C

= “>65”, C

= “medium”,

and C

= “>65”, C

= “low”, respectively. Then

the relevance of F

is higher than the relevance

of F

because more context inﬂuencers match

in F

than in F

. Thus, and according to the

feedback context factor, the alternative will be

judged closer to “invalid” than “valid”.

– Feedback Freshness. This factor stands for the

recentness of the feedback. Feedback relevance

is proportional to its freshness. That is, the

more recent the feedback is, the more mean-

ingful. There could be several ways to compute

feedback freshness. One design decision could

compute it by dividing its sequential number

by the overall number of feedback of its kind.

For example, suppose that PDA-based system

alternative got two validity feedback, the ear-

lier (with a sequential number 1) F

= “invalid”

and the later (with a sequential number 2) F

“valid”. Thus, and according to the feedback

freshness factor, the alternative will be judged

closer to “valid” than “invalid”.

• User’s Preferences. This factor stands for the

preferences of a user while assessing the overall

quality of a system alternative. The analysis of the

social feedback results in giving an overall assess-

ment of each system alternative against each qual-

ity attribute. However, the assessment of the over-

all quality, i.e., the aggregated quality, of a system

alternative may consider how the user appreciates

each of these quality attributes. Similarly to the

work in (Hui et al., 2003), we allow users to ex-

press their preferences by ranking the degree of

importance of each of the quality attributes. For

example, suppose that by analysing the histori-

cal quality feedback, the PDA-based alternative

quality against Q

= “visitor is well-informed”was

assessed to 3 and against Q

= “visitor’s com-

fort” was assessed to 2. Suppose that the Staff-

supported alternative quality against Q

was as-

sessed to 2 and against Q

to 3. If a user appreci-

ates Q

more than Q

then PDA-based alternative

overall quality will be assessed higher than that

of the Staff-supported system alternative, and vice

versa.

The algorithm Assessing Validity shown in Fig-

ure 4 computes the collective validity judgement of

a system alternative based on the validity feedback

users have provided in the past. It takes as input a

system alternative s, the set of validity inﬂuencers C

which affect the ability of s to meet the requirement it

is designed to reach and the actual values of these in-

ﬂuencers at the time of assessment C.Values. It gives

as output an assessment of the validity of the state-

ment “s is a valid means to meet the requirement it

is designed for”. First, the algorithm identiﬁes the

operations OP of the system alternative s which got

validity feedback from users (Line 1). If the system

alternative has no validity inﬂuencers then the con-

text match factor is irrelevant and the algorithm re-

turns simply the proportion of valid operations over

the overall number of operations | OP | multiplied

(weighted) by the average freshness of the operations

set OP (Lines 2-3).

When the system alternative has validity inﬂu-

encers, the algorithm iterates for each possible par-

tial or complete match of the context validity inﬂu-

encers at the feedback time and the assessment time

(Lines 7-17). For each combination of validity in-

ﬂuencers C

with a cardinality i, the algorithm iden-

tiﬁes the set of operations OP C

whose validity in-

ﬂuencers values (o.C

.Values) match with the validity

inﬂuencers values at the assessment time (C.Values)

(Lines 9-11). The algorithm then computes the va-

lidity probability concerning the context matches of

the cardinality i by dividing the number of valid op-

eration of Op.C

by | Op.C

| (Line 12). The rele-

vance of this probability is decided by both the car-

dinality of context match (i/| C |) and the value of

the freshness factor (Avg Freshness(Op.C

)), com-

puted as we explained earlier (Line 13). The al-

gorithm then multiplies the relevance with the com-

puted validity to get the relevant (i.e., the weighted)

validity (Line14). The algorithm then (Lines 15-

16) accumulates the relevance and the relevant va-

lidity into the variables relevant validty sum and

relevance sum (initiated at Lines 5-6). After the it-

eration goes through all partial and complete con-

text matches, the algorithm gives the overall assess-

ment by dividing the relevant validty sum by the

relevance sum (Line 18).

Similarly to this algorithm, we have developed the

algorithm Assessing Quality to calculate the collec-

tive judgement of the quality of a system alternative

against a quality attribute. The main difference with

regards to Assessing Validity algorithm is the consid-

eration of quality feedback of only the operations with

positive validity feedback as negative validity feed-

back makes any quality feedback irrelevant. More-

over, Assessing Quality deals with the average value

of the quality feedback provided for an alternative

against a quality attribute. Assessing the overall qual-

ity of an alternative considers also the users’ prefer-

ences expressed via ranking the importance of each

quality attributes. For the limitation of space, we have

not included neither this algorithm nor the examples

which explains both algorithms. For details, please

ENASE2012-7thInternationalConferenceonEvaluationofNovelSoftwareApproachestoSoftwareEngineering





ǣ





ǣǣ

ǣሼǡሽ

Ǥǣሼሺǡሻǡൌǯሽ





ǣሺǡሻ

1. ǣൌሼאǤǡሽ

2. ȁȁൌͲ

3. ̴ሺሻȗሺȁሼǡǤ̴ൌǮǯሽȁȀȁȁሻ

4. 

5. ̴̴ǣൌͲ

6. ̴ǣൌͲ

7. ൌͳȁȁ

8. ̴



ǣൌ]

9. א

ʹ





ȁȁൌ

10. ̴



ൌ̴ሼǢሺǤ



ǤǡǤሻሽ

11. 

12. ̴ǣൌȁሼ̴ǢǤ̴ൌǮǯሽȁȀȁǤȁ

13. ̴ǣൌሺȀȁȁ൅̴ሺǤሻሻȀʹ

14. ̴̴ǣൌ̴ȗ̴

15. ̴̴ǣൌ̴̴൅̴̴

16. ̴ǣൌ̴൅̴





17. 

18. ̴̴Ȁ̴

19. 

Figure 4: Assessing validity algorithm.

see our technical report (Ali et al., 2011b).

5 EVALUATION

To evaluate our framework, we have organized a lab

session and invited 5 researchers specialized in re-

quirements engineering and their research is in the

area of requirements-driven adaptive systems engi-

neering. We have explained our design principles

of modelling requirements for socially-adaptive sys-

tems. We then explained the scenario of an adaptive

messenger system which is able to deliver messages

in different ways and asked the participants to draw

a goal model, namely Tropos goal model (Bresciani

et al., 2004), presenting its requirements together with

the validity inﬂuencers (on the decompositions and

means-end) and the quality inﬂuencers on the contri-

bution links between any goal/task and a set of soft-

goals which evaluates all alternatives (please refer to

our report in (Ali et al., 2011b) for more details). The

left part of Figure 5 shows a goal model built during

the session; in the other part we extract one alternative

and show its validity and quality inﬂuencers.

The main issues which were raised by the partici-

pants concerned the following. Context monitorabil-

ity: besides the validity and quality judgement, the

values of certain context attributes might not be mon-

itorable by relying solely on automated means and

may require users to act as monitors. Quality and con-

text attributes identiﬁcation: users should be giventhe

chance to deﬁne extra quality and context attributes

which were not considered by the designers. Context

inﬂuence (un)certainty: the design should also con-

sider that some context inﬂuences are already known

and our approach should accommodate both certain

and uncertain context inﬂuences. Runtime evolu-

tion of the requirements model: the analysis should

be enriched to also decide parts of the model which

should be removed when collectively proved to be in-

valid or having very low quality and also help the an-

alyst by indicating loci in the model where an evolu-

tion should take place. Feedback relevance: we need

mechanisms to exclude feedback which are not signif-

icant or inconsistent taking into consideration several

factors like the user’ history and pattern of use. We

need to handle these limitations of our approach to

maximize its applicability and operationsand broaden

and optimize its automated analysis.

To evaluate the social adaptation analysis pro-

posed in Section 4, we have asked 15 users to pro-

vide validity and quality feedback about 6 alternatives

of a prototype messenger in 3 different contexts. For

the quality feedback we have considered 2 quality at-

tributes (Q

: less distraction, Q

: less effort). Then

we have run the validity and the quality assessment

analysis for the different alternatives in the different

contexts taking as input the set of obtained users’

feedback. Then, we have surveyed a set of 3 other

users (testing users) and compared their feedback to

SocialAdaptation-WhenSoftwareGivesUsersaVoice

Handle incoming

messages

Instant

messaging

Offline

messaging

Notify about

new message

Display message

AND

Reply to

message

Popup alert +

sound alert

Sound

alert only

Popup

alert only

Less distraction

Automati

Interactive

Visual

window

Translate

text to sound

User clicks

on messages

box

User presses

shortcut

bottons

Voice

message

Typed

message

Message

readability

Less

distraction

Less effort

C4C4

Less effort

Forward

to email

Handle incoming

messages

Instant

messaging

Notify about

new message

Display

message

AND

Reply to

message

Sound

alert only

Less

distraction

Automatic

Visual

window

Voice

message

Message

readability

Less

distraction

Less

effort

C1= Involved in other communication?

C2= Urgent message?

C3= Listening to audio?

C4= Currently editing?

C5= Long message?

Validity influencers: C1, C2, C3, C4

Quality influencers on

Less distraction: {C3, C4, C5}

Message readability: {C4}

Less effort: {C4}

Figure 5: A goal model (left) and an alternative of it with quality and validity inﬂuencers (right).

the collective judgements obtained by running the va-

lidity and quality analysis algorithms.

To evaluate the validity assessment analysis, we

have considered 3 alternativesand 3 different contexts

where the validity assessment algorithm has given

high probability for validity (>85 %). We have asked

the 3 users to provide their validity feedback (thus 27

feedback were provided in total) and out of which 23

feedback had the value “valid” and 4 had the value

“invalid”. This shows a good match between the col-

lective validity judgement of the 15 initial users, com-

puted by the validity assessment algorithm, and the

judgement of each of the testing users. It is worth

pointing out that the consensus of users about the va-

lidity is more likely to be achieved than the quality

due to the nature of the decision about validity which

has a clear-cut criteria to be judged upon.

To evaluate the quality assessment analysis, we

have considered 3 other alternatives and 3 different

contexts and asked the 3 testing users to provide the

quality feedback of each alternative in each of the

contexts against Q

and Q

. Table 1, show the ranking

of the automated analysis assessment and the testing

users assessment of the 3 different messenger system

alternatives (A

, A

) against the quality attribute

: “Less Distraction” (LD), in 3 different contexts

, C

). The acronym “Sys” stands for the as-

sessment given by the algorithm Assess Quality and

stand for each testing user. For example, and tak-

ing the ﬁrst data column, the automated analysis of

the users’ feedback to obtain the quality assessment

of A

, A

, and A

alternatives against the quality at-

tribute “Less Distraction” (LD) within the context C

indicated that A

has the best quality, and A

has the

second and A

has the lowest quality. The ranking

the automated analysis gave in this case, matched the

ranking made based on the quality assessment each

testing user gave. In the context C

, the user U

gave a ranking different from the one given by the

automated analysis and we highlight the mismatch-

ing results. The same for U

and U

for the context

. As shown in the table, the matching between the

collective quality judgement computed by the quality

assessment algorithm and the testing users feedback

was good enough (21 out of 27). For the quality at-

tribute Q

= “Less Effort”, the number of matches be-

tween the automated collective judgement and testing

users was also good (18 matches out of 27 compar-

isons).

There are several threats to validity concerning

our evaluation. The ﬁrst threat concerns the small size

scenario which we used (the messenger system) and

the relatively small number of participants who gave

feedback. Larger scale experiment would maximize

the credibility of our results. The second is the kind of

practitioners who modelled the scenario who already

had a good expertise in requirements modelling and

self-adaptive systems. Communicating our princi-

ples and guidelines to novice practitioners might raise

other concerns related to the practitioners understand-

ability and acceptability of our framework. The third

is the relatively short period of time for getting feed-

back which makes it hard to evaluate certain things

such as the evolution of social trends. The fourth is

that our users committed to provide feedback which

might not be the case with real users. Thus, we still

need to study users’ acceptance of our approach.

6 RELATED WORK

There are several research areas highly related to So-

cial Adaptation. We are mainly concerned about

ENASE2012-7thInternationalConferenceonEvaluationofNovelSoftwareApproachestoSoftwareEngineering

Table 1: For each context, the rank of the different alternatives (mismatches are in bold font).

: C

LD Sys U

Sys U

2 2 2 2 2 2 1 2 2 2 2 2

1 1 1 1 1 1 2 1 1 1 3 3

3 3 3 3 3 3 3 3 3 3 1 1

addressing challenges in the research in software-

engineering for adaptive systems and, particularly,

requirements-driven adaptation. Our proposed ap-

proach enriches requirement-driven adaptation re-

search by a systematic approach to incorporate users’

collective judgement of systems behaviours, per-

ceived as means to meet requirements, as a primary

adaptation driver. This helps for a more holistic adap-

tation which overcomes the limitation of automated

means to judge if requirements (functional and non-

functional) are being met (Ali et al., 2011c). It also al-

lows for reﬂecting the collective voice of users’ rather

than relying on designers’ judgements which could

be, or eventually become, invalid (Ali et al., 2011a).

This systematic approach is inline with an increasing

trend to involve users in the adaptation loop (Cheng

et al., 2008). We concretely position users as feed-

back providers and specify their feedback structure

and provide mechanisms to process it and assess the

collective judgement of system behaviours.

The seminal work of (Fickas and Feather, 1995)

highlights the importance of requirements monitoring

at runtime as a basic and essential step for planning

and enacting adaptation. This is a fundamental prin-

ciple of our approach. Cheng et al. note that in re-

quirement models uncertainty have not been explic-

itly addressed in traditional requirements engineer-

ing (Cheng et al., 2008). We address uncertainty by

involving users in evaluating the system alternatives

against their capability to meet requirements so that

certainty is achieved based on the perception of users

regarding the actual operation. In (Silva Souza et al.,

2011) the authors note that the (partial) un-fulﬁlment

of requirements triggers adaptation. They introduce

awareness requirements to refer to success, failure,

performance and other properties of software require-

ments (i.e. meta-requirements) and propose that the

system should monitor changes in these properties

and decide upon when and what adaptation should

take place. We argued that certain information can

not be monitored by the system and require an ex-

plicit users’ intervention via feedback and provided a

systematic way to realize that.

The work in (Baresi et al., 2010) proposes FLAGS

(Fuzzy Live Adaptive Goals for Self-adaptive sys-

tems) for requirements-driven adaptation at runtime.

FLAGS extends goal models mainly with adaptive

goals which incorporate countermeasures for adap-

tation. When goals are not achieved by the current

course of execution, adaptation countermeasures are

triggered. This approach can be potentially integrated

with ours so that when social collective judgement

indicates some failures, a certain adaptation goals

should be activated. The work in (Qureshi and Perini,

2010) emphasizes on ﬂexibility of requirements re-

ﬁnement and provide a method that supports the run-

time reﬁnement of requirements artifacts as a repet-

itive activity performed collaboratively between the

users and the application itself. We plan to beneﬁt

from this approach to incorporate users in the mod-

elling process at runtime and not only the judgement

of the system different behaviours.

7 CONCLUSIONS

Adaptation is a demand to maintain the validity and

quality of software over time. Adaptation is driven

by certain categories of changes in the system in-

ternal state and its operational environment (security

breaches, faults and errors, available resources, con-

text, etc.). We advocated another driver for adapta-

tion; the collective judgement of users on the alterna-

tive behaviours of a system. We proposed social adap-

tation to refer to the repetitive process of analysing

it to choose upon the behaviour which is collectively

judged to best ﬁt a certain context. The main ingredi-

ent to get the collective judgement is the feedback of

individual users. Feedback reﬂects users’ main inter-

est which is the validity and quality of a behaviour in

meeting requirements. Feedback is inﬂuenced by cer-

tain contextual attributes which should be monitored

when obtaining the users feedback. We haveproposed

and tested analysis mechanisms to infer the collective

judgement on the validity and quality of different al-

ternatives of a messenger system and discussed the

results. We have also reported several observations

on modelling requirements for social adaptation and

presented them as research challenges to face.

Social adaptation is meant to enrich self-

adaptation by accommodating users perception as a

part of the system computation and their collective

judgement as an adaptation driver. Our inspiring prin-

SocialAdaptation-WhenSoftwareGivesUsersaVoice

ciple is the wisdom-of-crowds (Surowiecki, 2005)

which keeps the door open for decisions formulated

collectively by users. The power of the crowd is not

easy to manage and social adaptation would work

when users are wiling to collaborate with the system

and moreover when the system already has imple-

mented different alternatives and is running them. So-

cial adaptation helps the system to tune the distribu-

tion of its alternative behaviours over the space of the

different contexts of use. Social adaptation does not

replace personalization; it only helps to know the col-

lective judgement of users which is speciﬁcally use-

ful for novice users as a recommendation system. It

also helps developers to know what improvement and

maintenance to do during the software lifetime.

As a future work, we are going to address the

research challenges which we listed in Section 5.

Mainly, we plan to enrich our modelling framework

to capture constructs supporting a better judgement

of the relevance of a feedback like the pattern of use

and the learning history of a user. We also need to

devise techniques to involve users in collectively tak-

ing harder decisions at runtime such as altering the

requirements model itself by adding (removing) re-

quirements, context and quality attributes. However,

our main challenge is to ﬁnd ways for balancing be-

tween users effort and computing transparency which

is the essence of adaptive systems. We also need to

work on incentives such as rewarding active users as

a part of the social adaptation engineering. We need

to further validate our approach on more complex sys-

tems and settings and develop a CASE tool to better

facilitate social adaptation modelling and analysis.

ACKNOWLEDGEMENTS

This work was supported, in part, by Science Foun-

dation Ireland grant 10/CE/I1855 to Lero - the Irish

Software Engineering Research Centre (www.lero.ie)

and by the European Union Seventh Framework Pro-

gramme (FP7/2007-2013) under grants no 258109

(FastFix).

REFERENCES

Ali, R., Dalpiaz, F., and Giorgini, P. (2010). A goal-based

framework for contextual requirements modeling and

analysis. Requir. Eng., 15:439–458.

Ali, R., Dalpiaz, F., Giorgini, P., and Souza, V. E. S.

(2011a). Requirements evolution: from assumptions

to reality. In the 16th International Conference on

Exploring Modeling Methods in Systems Analysis and

Design (EMMSAD 11).

Ali, R., Solis, C., Omoronyia, I., Salehie, M., and Nuseibeh,

B. (2011b). Social adaptation: When software gives

users a voice. Technical Report Lero-TR-2011-05,

Lero. University of Limerick. Ireland.

Ali, R., Solis, C., Salehie, M., Omoronyia, I., Nuseibeh,

B., and Maalej, W. (2011c). Social sensing: when

users become monitors. In Proceedings of the Euro-

pean Software Engineering Conference and the ACM

SIGSOFT Symposium on the Foundations of Software

Engineering, ESEC/FSE ’11, pages 476–479. ACM.

Baresi, L., Pasquale, L., and Spoletini, P. (2010). Fuzzy

goals for requirements-driven adaptation. In Proceed-

ings of the 2010 18th IEEE International Require-

ments Engineering Conference, RE ’10, pages 125–

134.

Bresciani, P., Perini, A., Giorgini, P., Giunchiglia, F., and

Mylopoulos, J. (2004). Tropos: An agent-oriented

software development methodology. Autonomous

Agents and Multi-Agent Systems, 8(3):203–236.

Cheng, B. H. C., Giese, H., Inverardi, P., Magee, J., and

de Lemos, R. (2008). Software engineering for self-

adaptive systems: A research road map. In Software

Engineering for Self-Adaptive Systems, pages 1–26.

Dumas, J. S. and Redish, J. C. (1999). A Practical Guide

to Usability Testing. Intellect Books, Exeter, UK, UK,

1st edition.

Fickas, S. and Feather, M. S. (1995). Requirements moni-

toring in dynamic environments. In Proceedings of the

Second IEEE International Symposium on Require-

ments Engineering, RE’95.

Hui, B., Liaskos, S., and Mylopoulos, J. (2003). Require-

ments analysis for customizable software goals-skills-

preferences framework. In Proceedings of the 11th

IEEE International Conference on Requirements En-

gineering, pages 117–126.

Laddaga, R. (1997). Self-adaptive software. Technical Re-

port 98-12, DARPA BAA.

Murch, R. (2004). Autonomic computing. IBM Press.

Qureshi, N. A. and Perini, A. (2010). Requirements en-

gineering for adaptive service based applications. In

Proceedings of the 2010 18th IEEE International Re-

quirements Engineering Conference, RE ’10, pages

108–111.

Salehie, M. and Tahvildari, L. (2009). Self-adaptive soft-

ware: Landscape and research challenges. ACM

Transactions on Autonomous and Adaptive Systems,

4:14:1–14:42.

Silva Souza, V. E., Lapouchnian, A., Robinson, W. N., and

Mylopoulos, J. (2011). Awareness requirements for

adaptive systems. In Proceeding of the 6th interna-

tional symposium on Software engineering for adap-

tive and self-managing systems, SEAMS’11, pages

60–69. ACM.

Surowiecki, J. (2005). The Wisdom of Crowds. Anchor.

Vredenberg, K., Isensee, S., and Righi, C. (2001). User-

Centered Design: An Integrated Approach. Prentice

Hall PTR.

ENASE2012-7thInternationalConferenceonEvaluationofNovelSoftwareApproachestoSoftwareEngineering