An Application Framework for Personalised and Adaptive

Behavioural Change Support Systems

Ulrich Reimer and Edith Maier

Institute for Information and Process Management, University of Applied Sciences St. Gallen, St. Gallen, Switzerland

Keywords:

Behavioural Change Support System, Systems to Encourage Healthy Behaviour, Nudging, Mobile Health,

Personalisation, User Modelling, User Adaptation, Meta Modelling.

Abstract:

The paper analyzes current weaknesses of behavioural change support systems (BCSS) such as the failure of

adequately taking into account the heterogeneity of target users. Based on this analysis the paper presents

an application framework that comprises various components to accommodate user preferences and to adapt

system interventions to individual users. Among these components is a goal hierarchy which can be set up to

represent the goals a user wants to achieve. The higher-level goals can be broken down into more speciﬁc goals

that can be measured and associated with appropriate activities. Furthermore, our BCSS framework includes

components for adapting its interactions according to a user’s observed behavioural preferences as well as

his or her previous reactions to system interventions. User adaptation also takes into account the preferences

of similar users by employing a collaborative ﬁltering approach. Thus, overall user acceptance should be

improved and motivation for behavioural change sustained. The framework is currently being implemented

and will subsequently be evaluated.

1 INTRODUCTION

One of the greatest challenges to health systems all

over the world is the growing number of people

with (multiple) chronic conditions such as diabetes,

asthma, cardiovascular disease and obesity. Accord-

ing to the WHO chronic diseases nowadays account

for about 80 % of the burden of disease (World Health

Organization, 2012). Most of these are lifestyle-

related and the risk factors are well-known, including

the lack of physical exercise, smoking, a diet rich in

fat and sugar, and the excessive consumption of al-

cohol. Although people are generally aware of the

long-term negative consequences, they often lack the

motivation as well as the social and emotional sup-

port that is required for changing one’sbehaviour. Be-

sides, we tend to discount long-term gains such as a

higher life expectancy and better quality of life in the

long run in favor of short-term rewards like the one

offered by some delicious cookies. Whilst the ma-

jority of the chronically ill may well agree with their

doctors’ or caregivers’ recommendations and fully in-

tend to adhere to them, e.g. engage in regular exercise

and change their diets, they fail to do so.

Behavioural economics is an approach that

promises to ameliorate the shortcomings of tradi-

tional healthcare management, especially with regard

to chronic disease (see e.g. (Cabinet Ofﬁce, 2010; Eu-

ropean Commission, 2014)). Behavioural economists

use knowledge from behavioural science as well as

motivational psychology and neuroscience to study

how individuals make decisions which are often non-

rational, and biased by a series of mental shortcuts,

for instance, the so-called “status quo bias” (Kahne-

man, 2011). Apart from the status quo bias, people’s

behaviour is also susceptible to the inﬂuence of de-

fault rules, framing effects and starting points. Con-

sequently, persuasion strategies can involve changing

the way options are presented, e.g. by adapting the

rules that drive user interaction.

The philosophy of behavioural economics is also

called “libertarian paternalism”, namely that people

should not be forced to act in certain ways, but rather

encouraged to act in ways that are better for them or

help them stopping bad habits formed over time. This

idea of a “gentle push”, or “nudge” favours invitations

to change behaviours, rather than the introduction of

constraints and sanctions to obtain behaviour change

(Thaler and Sunstein, 2009).

It has been shown that frequent and immediate

feedback is very helpful to nudge people towards

healthy behaviour (Loewenstein et al., 2013; Maier

152

Reimer, U. and Maier, E.

An Application Framework for Personalised and Adaptive Behavioural Change Support Systems.

In Proceedings of the International Conference on Information and Communication Technologies for Ageing Well and e-Health (ICT4AWE 2016), pages 152-159

ISBN: 978-989-758-180-9

and Ziegler, 2015). Mobile devices including smart-

phones and wearables such as smartwatches offer

great opportunities because they can be used for mea-

suring vital parameters such as heart rate, skin con-

ductance or blood pressure but also the number of

steps or sleep patterns. Most mobile health solutions,

i.e. mobile devices connected to medical applications

or sensors, as well as pure lifestyle apps actually in-

clude some kind of support for the users to achieve

their goals. However, these nudges tend to be hard-

wired, i.e. they do not adapt to user preferences and

needs and on the whole they are not grounded in be-

havioural change theory (see e.g. (Lister et al., 2014;

Patel et al., 2015)).

Our recent research therefore focuses on how

to use digital technologies to support behavioural

change in a systematic way and to allow adaptation

to what works best for an individual user.

2 CHALLENGES OF BEHAVIOUR

CHANGE SUPPORT SYSTEMS

Nudging for healthy behaviour using mobile technol-

ogy has to be viewed in the larger context of so-called

behaviour change support systems (BCSS) as intro-

duced and deﬁned by (Oinas-Kukkonen, 2010):

“Behavior change support systems (BCSS)

are information systems designed to form, al-

ter, or reinforce attitudes or behaviours or

both without using coercion or deception.”

The persuasivesystems design (PSD) model, a frame-

work for designing a BCSS introduced in (Oinas-

Kukkonen and Harjumaa, 2009), draws from the sem-

inal work by Fogg on persuasive technology (Fogg,

2002). It distinguishes two major design steps: ﬁrst,

analyzing the persuasion context, second, designing

the BCSS features. The persuasion context is deﬁned

by the intent, the type of change to be achieved, e.g.

if it is a one-time or a permanent change, the event,

which includes the use context as well as the user’s

goals, and the strategy, which determines what kinds

of message are to be delivered via which route to the

user. The BCSS design features consist of four cate-

gories:

Primary Task Support. distinguishes various prin-

ciples of how to support the user, e.g. by reducing

complex behavioural goals to smaller goals that

can be achieved by simple tasks, or by personaliz-

ing the system to the user’s speciﬁc behaviour and

preferences;

Dialog Support. deals with how to set up the dialog

with the user;

System Credibility. addresses the issue how to make

the system credible for the user;

Social Support. deals with how to improve motiva-

tion and adherence by including social inﬂuence,

e.g. via the peer group, into the system.

While the model suggested by Oinas-Kukkonen

already mentions personalisation as one of many de-

sign principles, it plays a more important role in

more recent work on persuasive systems. Target user

groups are typically very heterogeneous so that it is

nearly impossible to design a “one-size-ﬁts-all” sys-

tem. (Kaptein et al., 2010) examines individual dif-

ferences in persuadability in the health domain and

concludes that the intervention of a persuasive sys-

tem needs to be tailored to the persuasion proﬁle

of the speciﬁc user. For example, some users re-

act best to strongly persuasive messages while other

users respond adversely to too strong an intervention

and would require a more low-key suggestion. (Prost

et al., 2013) build upon these results and describe a

system that employs personalisation based on factors

such as persuadability of the user, social-emotional

attitude and behaviour history. The results of an em-

pirical study on the relationship between personal-

ity and the effectiveness of persuasive technologies is

presented in (Halko and Kientz, 2010).

Laverman and his colleagues (Laverman et al.,

2014) present an approach to personalizing communi-

cation in a BCSS (which they call “self-management

support system”). The authors argue that the system

should provide information in a way that is “relevant

to the user’s situation and match[es] the user’s pref-

erences and abilities to understand and be persuaded

by [it]”. The effect of personalising short text mes-

sages to reduce snacking behaviour was investigated

by Kaptein and his colleagues and the results reported

in (Kaptein et al., 2012). A more general overview of

the possible roles personalisation can play in persua-

sive systems can be found in (Berkovsky et al., 2012).

Behavioural change starts with motivation and in-

tent and requires the setting of clear and measurable

goals which direct attention and effort toward goal-

relevant activities (Locke and Latham, 2002). There-

fore, BCSS include mechanisms for goal setting as

well as measuring goal achievement to give appro-

priate feedback. Many of the smartphone apps that

have come into existence as part of the quantiﬁed self

movement for tracking and measuring all kinds of ac-

tivities, support goal setting and typically offer sup-

port for achieving these goals, e.g. by giving feed-

back on current goal achievement, by drawing on peer

group support, or by playful competition. In these

cases, while goal setting is supported, the types of

goals a user can set are very limited due to the spe-

An Application Framework for Personalised and Adaptive Behavioural Change Support Systems

153

ciﬁc focus each of these apps has, e.g. on measuring

physical activity, calorie intake, or stress level.

Consequently, while there are many theoretical

models available for guiding the proper design of a

BCSS and for designing mobile systems for support-

ing behavioural change in particular, in the end each

application has to be hand-crafted and tailored to a

speciﬁc domain and application scenario. When de-

signing a system, developers make assumptions about

what will work for the target user group, but once

the app has been completed, maybe even evaluated

with a focus group, one cannot but hope that the app

will be effective in supporting the intended behaviour

changes. Should this not be the case, it will be very

difﬁcult to identify the reasons. Thus, despite the the-

oretical guidelines available, the actual task of creat-

ing a BCSS is more an art than a systematic develop-

ment process.

One way to tackle this challenge is to devise a

more generic BCSS which can be easily conﬁgured to

meet the needs of a speciﬁc user. Ideally, the users can

carry out the necessary conﬁgurations themselves. In

this way, fewer assumptions need to be made about

the functions that a user actually wants to have.

To this end, we propose mapping the existing the-

oretical concepts to an application framework for cre-

ating mobile persuasive systems that can be easily

conﬁgured to accommodate a wide variety of user re-

quirements without the need to reimplement parts of

the system. Additionally, we propose that the frame-

work includes components for automatic user adap-

tion during system runtime.

In our current research we focus on those aspects

of the framework which we deem most important for

a mobile BCSS. Our framework will

• remedy the limited goal setting capabilities of ex-

isting apps by including a goal network that can

be set up and edited by a user according to his or

her speciﬁc needs (maybe together with a person

acting as a coach or therapist for the user);

• distinguish between the ultimate goals a user

wants to achieve and more concrete operational-

ized sub-goals whose achievement can be mea-

sured, e.g. with sensors;

• offer a variety of persuasive interventions (nudge

types) a user can choose from according to his or

her preferences;

• include automatic adaptation mechanisms that

monitor user behaviour, correlate system inter-

ventions with user behaviour and determine which

kinds of system interventions work best for a spe-

ciﬁc user and then adapt its intervention strategy;

• permit users to conﬁgure a system according to

their needs.

Throughout the paper we use the term “nudge” in the

following meaning:

Deﬁnition:

A nudge is a brief persuasive intervention that

encourages a speciﬁc behaviour.

In the following section we will describe our frame-

work in more detail.

3 APPLICATION FRAMEWORK

FOR BEHAVIOURAL CHANGE

SUPPORT SYSTEMS

3.1 Goal Hierarchies

At the heart of any BCSS, which also includes mo-

bile health apps, are the goals a user wants to achieve.

An application framework for creating a BCSS there-

fore needs to include some mechanism for specifying

goals or target behaviours. Many of today’s mobile

health apps support the setting of user-speciﬁc goals

but fail to consider the larger context within which

these goals are embedded, i.e. what the higher-level

goals are. For example, an app might allow users to

specify the number of steps per day to make. The

higher-level goal behind walking a certain amount

of steps per day could be to keep healthy or to lose

weight. Walking 10000 steps per day is only one pos-

sible way to achieve this; other possibilities could be

to go swimming or cycling. Consequently, in order

to give users more ﬂexibility in how to achieve their

goals, a goal hierarchy is needed which represents

the users’ higher-level goals as well as how to reach

them. This enables a user to achieve a higher-level

goal via (a combination of) alternative sub-goals, e.g.

a combinationof walking, running, cycling and swim-

ming. Goal hierarchies originally go back to cognitive

psychology (e.g. (Schank and Abelson, 1977)) and

play an important role e.g. in interactive systems that

create and maintain models of their users’ goals and

plans.

Our BCSS application framework therefore in-

cludes a construct for specifying one or more goal

hierarchies for the targeted application domain of a

BCSS. A goal hierarchy starts with a top goal which

represents a target user’s primary goal. The top goal

tends to be long-term, it may be measurable, e.g.

“body mass index of below 30”, or be more generic,

e.g. “keep healthy”. It can usually be achieved by a

variety of different ways, e.g. by engaging in physi-

cal activity, lowering the stress level, by eating regular

ICT4AWE 2016 - 2nd International Conference on Information and Communication Technologies for Ageing Well and e-Health

154

Figure 1: Example of a goal hierarchy (application model level).

meals or a combination thereof. Each option is repre-

sented by a sub-goal, together with an indication if

the sub-goal is sufﬁcient for reaching the higher-level

goal or if several sub-goals need to be reached. Sub-

goals can be broken down into further sub-goals un-

til these can be associated with a measurable activity.

We call such goals operationalized:

Deﬁnition:

An operationalized goal is short-term and is

associated with a measurable activity to reach

the goal.

Figure 1 shows some examples of operationalized

goals. Activities associated with operationalized

goals can e.g. be measured via sensors or diary en-

tries. An activity detection module using a 3D ac-

celerometer and state-of-the-art algorithms can auto-

matically determine if the user is e.g. walking, run-

ning, cycling, or climbing stairs, and thus can help to

keep track of the achievement of alternative goals for

physical activity (see e.g. (Slim et al., 2016)).

Operationalized goals are parameterized so that

they are not yet goals but rather goal types. An ex-

ample of such a goal type is “physical activity > x

steps per day” (cf. Fig.1). The user, possibly together

with a coach or a therapist, selects goal types and sets

values for the goal parameters to obtain speciﬁc (op-

erationalized) goals.

We will implement the application framework by

adopting a meta modelling approach (Karagiannis

and K¨uhn, 2002; Atkinson and K¨uhne, 2003). The

meta model deﬁning our application framework (cf.

Fig.2 for its main part) introduces all constructs that

can be used to create a speciﬁc BCSS. One example of

such a construct is the goal hierarchy we have already

introduced. A speciﬁc BCSS application is then rep-

resented by an application model that is an instance

of the meta model. A speciﬁc goal hierarchy with pa-

rameterized goal types would be on this model level

(cf. Fig.1). The parameters of the goal types are set by

a user during runtime to create concrete goals. Thus,

the runtime system is essentially an instance of the ap-

plication model.

Users are not only allowed to set the parameters

in the goal types but they can also delete parts of the

goal hierarchy so that only the goals they wish to pur-

sue are left. For example, a user who does not like

running would delete the associated sub-goal. How-

ever, we do not permit a user to add sub-goals be-

cause this would essentially lead to a new application

model, which then would require additional imple-

mentation effort, e.g. to provide the means to mea-

sure the achievement of the added goal. Therefore the

goal hierarchy in the application model provides the

set of all possible goals for the intended application,

from which users pick the ones they like. Modifying

the goal hierarchy is a task for the developers.

3.2 Adapting Nudges to Users

In the course of our research we have conducted

extensive interviews with potential end-users which

conﬁrm the ﬁndings of other researchers (Halko and

Kientz, 2010; Kaptein et al., 2010; Prost et al., 2013)

namely that behaviour is inﬂuenced by a variety of

factors, e.g. age, sex, socio-economic status, atti-

tudes, personality, social environment and peer group.

Thus, the ﬁxed set of interventions (or nudges) that

existing BCSS have implemented do not take into ac-

count the heterogeneity of target users. This results

An Application Framework for Personalised and Adaptive Behavioural Change Support Systems

155

Figure 2: Core fragment of the BCSS meta model.

in low intervention efﬁcacy and low user acceptance.

In our framework we therefore provide for a variety

of nudge types, such as suggestion, praise, reminder,

reward, and we devise different means to control the

user-speciﬁc selection of nudge types and their tim-

ing. The corresponding constructs of our framework

are deﬁned in the meta model (cf. Fig.2), just like the

construct of goal hierarchies.

The simplest way to adapt nudges to users is to

have them select the preferred nudges. However, as

shown by our interviews the average user will shun

the additional effort and ﬁnds it difﬁcult deciding

which choices to make. We therefore aim at enabling

the system to choose the nudge types using a threefold

approach – a) user modelling, b) collaborative ﬁlter-

ing, and c) progress-dependent adaptation.

3.2.1 User Modelling

Our framework includes a user modelling compo-

nent (Kobsa, 2001) which allows to monitor user

behaviour and keep track of the user’s preferences

with regard to nudge types, choice between alter-

native goals, and schedule of activities. To de-

duce these preferences, the system starts by selecting

nudge types randomly and monitoring which of them

work best for a user. Once the system has identiﬁed

the nudge types that are more successful in terms of

triggering intended behaviour, the system uses those

types more often. However, according to our frame-

work the system does not completely stop using the

other nudge types so that possible future changes in

user behaviour can be detected and the user model

adapted accordingly.

For example, if a user often follows a suggestion

made by the system, this is a good indicator that the

user responds well to suggestion nudges. Also, whilst

some users might respond well to reminders or feed-

backs that they are falling behind their peer group,

other users might simply ignore such messages.

Additionally, the system keeps track of the tim-

ing of different kinds of activities performed by the

user and adjusts the timing of nudges, such as sug-

gestions and reminders accordingly. The user model

also keeps track of the alternatives that a user prefers

to achieve a higher-level goal, e.g. cycling instead of

walking.

Although the user preferences derived by the user

modelling approach can never do justice to the enor-

mous complexity of human behaviour, we neverthe-

less expect our approach to result in superior system

performance with higher user acceptance rates.

3.2.2 Collaborative Filtering

For the purpose of user adaptation our framework will

also make use of collaborative ﬁltering (Adomavicius

and Tuzhilin, 2005). This approach implies that the

system primarily uses those nudges that work best for

similar users. Similarity is determined with respect to

the user proﬁle, i.e. what the user initially said about

himself or herself, and the user model. For exam-

ple, collaborative ﬁltering might indicate that show-

ing the user’s achievements compared to those of the

peer group works poorly for female users in a certain

age range and with a low activity proﬁle.

With collaborative ﬁltering we are able to identify

user preferences much faster than with user modelling

because it can draw on the collected evidence from all

the other users of the system, provided this evidence

has already been collected. Within our BCSS frame-

work initial preferences are determined by collabo-

ICT4AWE 2016 - 2nd International Conference on Information and Communication Technologies for Ageing Well and e-Health

156

rative ﬁltering and are then continuously ﬁne-tuned

using the user modelling approach described above.

3.2.3 Progress-Dependent Adaptation

As already described above, in our BCSS framework

the user modelling component keeps track of a user’s

typical activities as well as their timing and adjusts

the timing of suggestions and reminders accordingly.

Another important mechanism to adapt the time and

type of nudges consists in monitoring the progress

of achieving (operationalized) goals. Plotting goal

achievement along the time axis produces a rectan-

gular area which we divide into the three sub-areas A,

B and C (see Fig.3): Area A signiﬁes good progress,

area B indicates slow progress, while area C indicates

that considerable effort is required to achieve the cor-

responding goal.

For example, a ﬁrst, still rather simplistic selec-

tion of nudges could work as follows: As long as a

user’s progress lies in area A, no nudges are generated

because the user is performing well. When progress

falls within area B, unobtrusive, low-key nudges are

appropriate, e.g. a suggestion or a discreet reminder.

When goal achievement moves into area C, the user

is at risk to miss the goal and stronger nudges may

be called for than in area B. On the other hand, if the

user catches up and moves back into area B or from

area B to area A, a praise message could be generated.

More elaborate algorithms could ﬁne-tune the system

responses and e.g. consider if the user just crossed

into a new area or is already deep inside it.

In our application frameworkeach operationalized

goal can be associated with nudge types and corre-

sponding trigger conditions (cf. the meta model in

Fig.2):

• A nudge is generated when progress crosses from

one area into the neighbouring one, indepen-

dent from the percentage of achievement and the

elapsed time, e.g. from A to B or from B to C or

back.

• A nudge is generated when a predeﬁned amount

of time has elapsed and the user’s progress is lo-

cated in area A, B or C. For example, when 70%

of the time available for reaching a goal is over

a praise message is generated if progress is in

area A, a reminder if progress is in area B, and

a prompt or a challenge is generated if progress is

in area C.

• Similarly, a nudge is generated when a predeﬁned

percentage of achievement has been reached and

the user’s progress is in a speciﬁc area.

The boundaries between the areas in the ABC

graph are initially deﬁned according to the user pro-

ﬁle, e.g. taking into account if a user is an early riser

or a night owl. The system monitors a user’s achieve-

ment pattern for each goal and adapts the boundaries

to ﬁt with his or her typical daily routine. For example

for a user who likes to go running early in the morn-

ing, area A would be far to the left and small. On the

other hand, for a user who usually goes running in the

afternoon, area A would stretch far to the right.

Figure 3: An ABC graph for progress-dependent nudges.

4 EVALUATION

We are currently implementing the BCSS framework.

Subsequently we will conduct an evaluation with peo-

ple who wish to reduce their weight or maintain their

previously achieved weight loss. To this end, we

will develop a smartphone app which is based on

our framework and offers a variety of weight-related

goals, including physical activity, self-weighing, eat-

ing behaviour and calorie intake. We will distribute

the app via social media platforms and utilize the Ap-

ple ResearchKit to take care of issues such as seeking

informed consent and giving participants control over

what data they want to share.

The study will randomly divide participants into

two groups. One group will get the version that al-

lows customizing the goal hierarchy and which au-

tomatically adapts to the individual user. The other

group will get a version with predeﬁned goals and

with ﬁxed interventions. Our hypothesis is that the

former version, which makes full use of the capabil-

ities of our framework, is more effective in terms of

achieving one’s goals.

5 CONCLUSIONS AND

OUTLOOK

In this paper, we have presented an application frame-

work for behavioural change support systems (BCSS)

An Application Framework for Personalised and Adaptive Behavioural Change Support Systems

157

that comprises various components for tailoring a

BCSS to users’ needs and preferences. One of these

components is a goal hierarchy which can be set up

to represent the goals a user wants to achieve. The

higher-level goals (e.g. reaching a BMI below 30)

can be broken down into more speciﬁc goals that are

operationalized, i.e. can be achieved by associated

measurable activities. Furthermore, our BCSS frame-

work includes components for adapting its interac-

tions in line with a user’s previous reactions to system

interventions (nudges) and with observed behavioural

preferences, whilst at the same time taking into ac-

count the preferences of similar users by employing a

collaborative ﬁltering approach. The adaptation com-

ponents do justice to the heterogeneity of target user

groups, which tends to be ignored by most of the cur-

rent systems.

We are planning to embed our framework into a

more general approach where the activities associated

with operationalized sub-goals are chosen from the

predictors of a predictive model. For example, long-

term studies have shown that regular self-weighing

and having breakfast regularly are strong predictors

for weight loss and weight-loss maintenance (see e.g.

(Feller et al., 2015)). Such predictors would therefore

be included as evidence-based goals in the goal hier-

archy. In this scenario, a therapist familiar with such

predictors co-decides with a user which goals to set.

Furthermore, we are planning to develop a deci-

sion support system which utilizes a combination of

predictive models to predict a person’s health out-

comes in the near future based on various input data,

such as genetic, physiological, psychological, and be-

havioural data. The decision support system will

be able to analyse the inﬂuence of each predictor in

the predictions. The predictors shown to have the

most impact on the desired health outcomes are good

candidates for evidence-based goals because they are

likely to be most effective in terms of improving the

corresponding person’s future health.

Whilst we consider our approach very promising

and conducive to adopting and maintaining a healthy

lifestyle, we are aware of the ethical issues involved

(see e.g. (McGrady and Nelms, 2010)). Above all, it

is important that users should always stay in control,

participate in all decisions, and can grasp the impli-

cations of a prediction, on what it is based and that it

is only a likely but not a certain outcome. Besides,

it is absolutely essential to guarantee the privacy and

conﬁdentiality of health-related data. We will develop

measures and guidelines that assure users’ autonomy,

privacy and right of control whilst making sure that

the decision support system really supports and moti-

vates them in their every-day lives.

Furthermore, we are considering applying our

BCSS framework to domains other than health, e.g.

to mobility. By encouraging users to use public trans-

port or cycle to work, they will achieve the predeﬁned

goal of causing a smaller ecological footprint (Maier,

2012).

ACKNOWLEDGEMENTS

The research presented in this paper has been made

possible by a grant from Gebert R¨uf Foundation. Our

thanks go also to the members of the student group

for their contributions to this research: M. Eggen-

schwiler, S. Frigg and R. Zuberb¨uhler. We further

thank the reviewers of an earlier version of this pa-

per whose feedback helped to improve it.

REFERENCES

Adomavicius, G. and Tuzhilin, A. (2005). Toward the

next generation of recommender systems: A survey

of the state-of-the-art and possible extensions. IEEE

Transactions on Knowledge and Data Engineering,

17(6):734–749.

Atkinson, C. and K¨uhne, T. (2003). Model-driven develop-

ment: A metamodeling foundation. IEEE Software,

20(5):36–41.

Berkovsky, S., Freyne, J., and Oinas-Kukkonen, H. (2012).

Inﬂuencing individually: Fusing personalization and

persuasion. ACM Transactions on Interactive Intelli-

gent Systems, 2(2):9:1–9:8.

Cabinet Ofﬁce (2010). Applying behavioural insight to

health. Institute for Government, UK.

European Commission (2014). Green paper on mobile

health.

Feller, S., M¨uller, A., Mayr, A., Engeli, S., Hilbert, A.,

and Zwaan, M. d. (2015). What distinguishes weight

loss maintainers of the german weight control registry

from the general population? Obesity, 23(5):1112–

1118.

Fogg, B. J. (2002). Persuasive Technology: Using Com-

puters to Change What We Think and Do. Morgan

Kaufmann, 1 edition.

Halko, S. and Kientz, J. A. (2010). Personality and per-

suasive technology: An exploratory study on health-

promoting mobile applications. In Proceedings of the

5th International Conference on Persuasive Technol-

ogy, PERSUASIVE’10, pages 150–161, Berlin, Hei-

delberg. Springer-Verlag.

Kahneman, D. (2011). Thinking, fast and slow. Penguin

Books, New York, Toronto, London.

Kaptein, M., De Ruyter, B., Markopoulos, P., and Aarts,

E. (2012). Adaptive persuasive systems: A study of

tailored persuasive text messages to reduce snacking.

ICT4AWE 2016 - 2nd International Conference on Information and Communication Technologies for Ageing Well and e-Health

158

ACM Transactions on Interactive Intelligent Systems,

2(2):10:1–10:25.

Kaptein, M., Lacroix, J., and Saini, P. (2010). Individual

differences in persuadability in the health promotion

domain. In Ploug, T., Hasle, P. F. V., and Oinas-

Kukkonen, H., editors, PERSUASIVE, volume 6137

of Lecture Notes in Computer Science, pages 94–105.

Springer.

Karagiannis, D. and K¨uhn, H. (2002). Metamodelling plat-

forms. In Bauknecht, K., Tjoa, A. M., and Quirch-

mayr, G., editors, E-Commerce and Web Technolo-

gies, Lecture Notes in Computer Science, page 182.

Springer.

Kobsa, A. (2001). Generic user modeling systems. User

Modeling and User-Adapted Interaction, 11(1-2):49–

63.

Laverman, M., Neerincx, M. A., Alpay, L. L., R¨ovekamp,

T. A., and Schonk, B. J. (2014). How to Develop Per-

sonalized eHealth for Behavioural Change: Method

& Example. Technical Report TNO 2014 R10758.

Lister, C., West, J. H., Cannon, B., Sax, T., Brodegard, D.,

and Eysenbach, G. (2014). Just a fad? Gamiﬁcation in

health and ﬁtness apps. JMIR Serious Games, 2(2):e9.

Locke, E. and Latham, G. (2002). Building a practically

useful theory of goal setting and task motivation: A

35-year odyssey. American Psychologist, 57(9):705.

Loewenstein, G., Asch, D. A., and Volpp, K. G. (2013).

Behavioral economics holds potential to deliver better

results for patients, insurers, and employers. Health

Affairs, 32(7):1244–1250.

Maier, E. (2012). Smart mobility - encouraging sustain-

able mobility behaviour by designing and implement-

ing policies. JeDEM-eJournal of eDemocracy and

Open Government, 4(1):115–141.

Maier, E. and Ziegler, E. (2015). Sanfte Stupser f¨ur

gesundheitsf¨orderliches Verhalten – oder Nudging im

Gesundheitswesen. Clinicum, 3-15:76–81.

McGrady, E. and Nelms, L. W. (2010). Ethical issues of

health management predictive modeling. In Pease, W.,

Cooper, M., and Gururajan, R., editors, Biomedical

Knowledge Management: Infrastructures and Pro-

cesses for E-Health Systems. IGI Global.

Oinas-Kukkonen, H. (2010). Behavior change support sys-

tems: The next frontier for web science. In Proceed-

ings of the Web Science Conference 2010.

Oinas-Kukkonen, H. and Harjumaa, M. (2009). Persuasive

systems design: Key issues, process model, and sys-

tem features. Communications of the Association for

Information Systems, 24(1):28.

Patel, M., Asch, D., and Volpp, K. (2015). Wearable devices

as facilitators, not drivers, of health behavior change.

JAMA, 313(5):459–460.

Prost, S., Schrammel, J., R¨oderer, K., and Tscheligi, M.

(2013). Contextualise! personalise! persuade!: A

mobile HCI framework for behaviour change support

systems. In Proceedings of the 15th International

Conference on Human-computer Interaction with Mo-

bile Devices and Services, MobileHCI ’13, pages

510–515, New York, NY, USA. ACM.

Schank, R. C. and Abelson, R. P. (1977). Scripts, Plans,

Goals and Understanding: an Inquiry into Human

Knowledge Structures. Lawrence Erlbaum, Hillsdale,

NJ.

Slim, S., Atia, A., and Mostafa, M.-S. (2016). An exper-

imental comparison between seven classiﬁcation al-

gorithms for activity recognition. In Gaber, T., Has-

sanien, A. E., El-Bendary, N., and Dey, N., editors,

The 1st International Conference on Advanced Intel-

ligent System and Informatics (AISI2015), November

28-30, 2015, Beni Suef, Egypt, volume 407 of Ad-

vances in Intelligent Systems and Computing, pages

37–46. Springer International Publishing.

Thaler, R. H. and Sunstein, C. R. (2009). Nudge: improving

decisions about health, wealth, and happiness. Pen-

guin Books, New York, Toronto, London.

World Health Organization (2012). Global Health Esti-

mates: Deaths, disability-adjusted life year (DALYs),

years of life lost (YLL) and years lost due to disability

(YLD) by cause, age and sex, 2000–2012.

An Application Framework for Personalised and Adaptive Behavioural Change Support Systems

159