=Paper=
{{Paper
|id=Vol-2338/paper5
|storemode=property
|title=Creating an Artificial Coaching Engine for Multi-domain Conversational Coaches in eHealth Applications
|pdfUrl=https://ceur-ws.org/Vol-2338/paper5.pdf
|volume=Vol-2338
|authors=Tessa Beinema,Harm op den Akker,Hermie Hermens
|dblpUrl=https://dblp.org/rec/conf/atal/BeinemaAH18
}}
==Creating an Artificial Coaching Engine for Multi-domain Conversational Coaches in eHealth Applications==
https://ceur-ws.org/Vol-2338/paper5.pdf
Creating an Artificial Coaching Engine for Multi-domain
Conversational Coaches in eHealth Applications
Tessa Beinema Harm op den Akker Hermie Hermens
Roessingh Research and Development Roessingh Research and Development Roessingh Research and Development
Telemedicine group Telemedicine group Telemedicine group
Enschede, The Netherlands Enschede, The Netherlands Enschede, The Netherlands
t.beinema@rrd.nl h.opdenakker@rrd.nl h.hermens@rrd.nl
ABSTRACT An effective approach for the prevention and control of NCDs is
In this paper the concept of an Artificial Coaching Engine is de- reducing risk factors in multiple domains. For example, for cardio-
scribed, which is being developed for a state-of-the-art multi-agent vascular diseases and diabetes, this can involve the adoption of a
multi-domain coaching application. The engine will fulfil three healthy diet and healthy exercise habits. More generally, adopting
main functions in this application. First, it will serve as a knowl- a healthier lifestyle can delay NCDs and improve quality of life.
edge base and user model, representing user, context, and artificial While the coaching approach that is most suitable varies between
coach information. Second, it will represent and select the coaching target groups, research has shown that health coaching by health
goals that the coaches will coach towards. Third, it will select the care professionals improves the management of chronic diseases
most suitable coaching strategies for reaching the selected coaching [3]. Given the size of the challenge, the implementation of health
goals. Following the description of the concept, we will discuss our coaching by human coaches for a large target audience requires
approach and the challenges in the development of the Artificial more manpower and resources from the health sector.
Coaching Engine. One solution for the limited availability of human coaches can
be found in eHealth applications for personalised coaching, or
CCS CONCEPTS e-coaching systems. These systems can be used to help people
with adopting a healthier lifestyle. Through coaching, users can be
• Human-centered computing → HCI theory, concepts and mod- informed, assisted and empowered.
els; • Computing methodologies → Knowledge representation
and reasoning; • Social and professional topics → User charac-
teristics; 1.1 E-coaching systems
We adopt the definition by Kamphorst as a description of e-coaching
KEYWORDS systems:
Artificial coaching; health behaviour change; lifestyle coaching; An e-coaching system is a set of computerized com-
user modelling; coaching goals and strategies; eHealth; tailoring ponents that constitutes an artificial entity that can
ACM Reference Format: observe, reason about, learn from and predict a user’s
Tessa Beinema, Harm op den Akker, and Hermie Hermens. 2018. Creating behaviours, in context and over time, and that engages
an Artificial Coaching Engine for Multi-domain Conversational Coaches in proactively in an ongoing collaborative conversation
eHealth Applications. In Proceedings of ACM Workshop on Intelligent Con- with the user in order to aid planning and promote
versation Agents in Home and Geriatric Care Applications (ICA-HoGeCa2018). effective goal striving through the use of persuasive
techniques [2].
1 INTRODUCTION Within e-coaching, there are many applications that focus on
The average age of the World’s population is increasing. In a 2017 a single domain or target group. Examples of domains include:
report, the United Nations reports the number of persons aged 60 healthy eating, physical activity, diabetes management or mental
or above is expected to more than double by 2050 and even triple health. Examples of target groups are: the elderly, chronic pain
by 2100 [8]. For Europe the expected increase by 2050 will be from patients or obese children. In recent years there is also an increase
25% of the population to 35%. of applications that have the focus on a combination of two domains,
With the rising average age of the population there is also an for example, healthy eating and physical activity. However, since
increasing percentage of the population that suffers from noncom- the risk factors for NCDs can be found across multiple domains,
municable diseases (NCDs) or chronic diseases. The main types of we believe that a holistic approach to healthy living is crucial and
such diseases are cardiovascular diseases, cancers, chronic respira- thus e-coaching should also focus on aiding the user in multiple
tory diseases and diabetes [6], but dementia and depression are also domains.
common. While NCDs are responsible for 70% of all deaths globally Apart from the coaching domain and target audience, other
each year (that is, 40 million people), the length of time that people important aspects for e-coaching systems that focus on health pro-
suffer from them is also very long and puts a large drain on health motion are that they should be interactive, interoperable, personally
care resources and manpower. engaging, contextually tailored and that they should be suitable to
deliver to mass audiences [4].
35
An application in development that will incorporate the aspects (see Figure 2). Dynamic knowledge relates to knowledge that can
mentioned above is the Council of Coaches [5]. The Council of quickly change, such as, for example, information available about
Coaches will be a state-of-the-art multi-agent e-coaching applica- the current user. Semi-static knowledge relates to knowledge that
tion that will combine holistic behaviour analysis, smart adaptive is quite static, but that might be updated, such as, for example,
coaching, dialogue management (using the DGEP platform [1]), and knowledge about what a healthy diet entails. Static knowledge
realistic embodied conversational agents (using the GRETA/VIB relates to knowledge that is not expected to change, such as, for
platform [7] and ASAP platform [9]) to aid its users in obtaining example, a user’s brother is someone that shares the same parents.
a healthier lifestyle. The three initial target groups will be people The represented information will be leveraged by the coaching
with Age Related Impairments, Type 2 Diabetes, or Chronic Pain. modules in the coaching engine as well as by multiple components
The Council of Coaches application will involve multiple em- of the overall system for tailoring their output.
bodied conversational coaches, each with their own expertise. The The representation of the user in the coaching engine will contain
presence of multiple coaches allows for multi-coach strategies, for information from two main sources. Firstly, information on the
example, a coach explaining something to another coach to teach user’s short-term and long-term behaviours will be represented
the user or ‘good coach - bad coach’. There are many interesting which is gathered by the system’s sensing component (for example,
facets to the project, and in this paper we focus on one of them: the physical activity or facial expressions). The sensing component
development of the Artificial Coaching Engine. will also detect changes in behaviour and it will provide context
information. Secondly, information obtained through the user’s
2 THE ARTIFICIAL COACHING ENGINE interaction with the system will be represented.
The representation of the coaches includes predefined knowl-
edge for each coach on their personalities, domains, mannerisms,
backstories, etc. It will also include a representation of the coaching
goals and available coaching strategies for each coach (which will
be elaborated on in the next two subsections). Within the coach-
ing engine, each coach has an individual knowledge base, which
contains the knowledge necessary to coach in their domain.
2.2 Automatic goal selection
The second function of the Artificial Coaching Engine is to select the
coaching goals for the coaches. For an e-coaching system to coach
people, it should be clear what the goal of the provided coaching
is. That is, there should be (intermediary) goals which are aimed
for when interacting with the user. Examples can be to achieve a
change in behaviour, the assimilation of transferred information,
or to help a user feel more empowered.
In the Council of Coaches system there are multiple embodied
conversational coaches, who will all coach in their own domain. As
can be seen in Figure 2, each of the coaches will have a personal
Figure 1: The placement of the Artificial Coaching Engine implementation of a shared goal model. That is, the model is the
in the Council of Coaches system. same, but the priorities for the goals may be different for different
coaches. In the goal model it is represented what each goal is and
how the goals can be related to each other. Each goal must at
The Artificial Coaching Engine will fulfil the role of the intelli- least include the domain or domains to which it belongs, what the
gent coaching component in the Council of Coaches system. As can prerequisites are for the goal to be allowed to be set, and if it has
be seen in Figure 1 it has a central place in the system connecting been completed or if it perhaps is irrelevant for the current user.
the components that are responsible for gathering data and those The task of the automatic goal selector is to decide which goal
generating the embodied conversational agents, the dialogues and is most relevant to pursue for the coach based on the available
their behaviour. Taking this role in the system, there are three main knowledge.
functions that the Artificial Coaching Engine will fulfil, namely, as
a knowledge base, as a goal selector and as a strategy selector. A
schematic overview of how these elements influence each other 2.3 Tailored coaching strategies
can be found in Figure 2, and an overview of process within the The third function of the Artificial Coaching Engine is to select the
Artificial Coaching Engine can be found in Figure 3. coaching strategies that are most suitable for reaching a coaching
goal (again, see Figure 2). Once a goal has been selected for a coach,
2.1 Knowledge base a coaching strategy can guide that coach’s interactions with the
The first function of the Artificial Coaching Engine is to serve as user to reach that goal. Each coach will have a set of coaching
a knowledge base, and it will contain three types of knowledge strategies available, and, while that set might partly be the same as
36
Figure 2: A schematic representation of the elements in the Artificial Coaching Engine and how these interact to fulfil each
of the coaching engine’s three main functions.
Figure 3: A schematic representation of the process in the Artificial Coaching Engine.
for other coaches, the strategies that are available to a coach are • What is the knowledge about the user and their context that
selected to be suitable for that coach’s domain. needs to be available to be able to coach?
The most suitable strategy for reaching a goal will be determined • How do we represent this knowledge?
based on the available information about the user. For example, • How do we represent and how do we select the coaching
if the goal is to inform a user about the importance of physical goals?
exercise, for a well-informed user a coach might ask that user what • How do we handle different goals for different coaches?
they know, so that they can correct flaws in their knowledge, while • Which coaching strategies are relevant for the coaching
for a very uncertain user the strategy might involve dropping subtle domains?
hints about the topic and giving compliments. • How do we represent and select the coaching strategies?
The definition of coaching strategies starts by designing strate-
gies using literature on behaviour change and e-coaching, and the 3 APPROACH AND CHALLENGES
expertise of domain experts. These strategies then need to be mod-
The realisation of the Artificial Coaching Engine brings with it
elled to create a technical representation. The modelled strategies
many challenges. In the following subsections we describe our
should contain a set of defined prerequisites and clearly structured
approach for the development of the engine’s three main functions
contents. In the end, the selected strategy (or strategies if there
and we discuss the main challenges in that process. We also discuss
are multiple suitable candidates) is sent to the dialogue manage-
evaluation, which can be challenging for a component in an e-
ment component and agent representation component to guide the
coaching system that aims for behaviour change.
coach’s dialogue actions and behaviour.
3.1 Knowledge modelling
2.4 Research questions As mentioned previously, the knowledge base will contain a rep-
resentation of information on the user and a representation of
The development of the Artificial Coaching Engine brings a number
the coaches. The design of the user model will be based on obser-
of research questions with it. The main questions are:
vations from the literature on behaviour change, the behaviours
37
and behaviour changes measured by the sensing platform and the Once a decision on the duration of strategies has been made,
information resulting from interactions with the user. the second step is to define informally what the various coaching
The first step will be to create a framework that models the strategies are that are valid options for reaching the coaching goals.
aspects of a user’s physical state, mental state, and context that Of course there can be multiple strategies that are suitable for
influence behaviour change processes. Initially the models in this reaching those goals, depending on the knowledge about the user,
framework will involve the, often abstract, concepts that can be the coaches available and other goals.
deduced from theories about behaviour change. While theories of After the informal definition of the coaching strategies, they
behaviour change describe the underlying processes of behaviour should be modelled so that this technical representation can be
change in persons, we will use these theories to infer concepts used to filter the possible dialogue actions. As can be seen in Figure
that can be build up from the data that we obtain through sensors 2 a strategy can be seen as a template that can be filled in by certain
and interaction. We will make these concepts more concrete by coaching actions. These coaching actions in turn can consist of one
mapping them to the possible means of measurement. The final or more dialogue actions, which can also be responses to replies
implementation of the knowledge for the system will be refined from the user.
from this set of measurable features based on the requirements The challenge in creating the coaching strategies lies in not only
resulting from the prerequisites of the sets of goals and strategies developing them for single coaches, but also for a joint coaching
that the coaches will have. approach between two or more coaches at the same time. An ex-
The representation of the coaches will be based on character de- ample could be the ‘good coach, bad coach’-strategy, in which one
signs and implemented following the requirements of the system’s coach might take on a very empathic role while the other might
agent representation component. That is, for example, personality enquire why the user did, for example, not reach their step goal for
and mannerisms will be implemented in such a way that they can that day.
be used in the generation of the coaches’ behaviours and adjust-
ment of their dialogue actions. We discuss the representation of the
goals and strategies, which are a part of the coaches as well, in the 3.4 Evaluation
following two subsections. Another challenge in the development of the Artificial Coaching
Engine is its evaluation, both for intermediate versions and for the
final version of the engine. That is, the aim for the Artificial Coach-
3.2 Automatic goal selection
ing Engine is to select strategies that can be deployed to change the
The construction of the automatic goal selection module brings two user’s behaviour. Since behaviour change is an effect that can only
main challenges. The first is the modelling of the goals. A single be measured after a longer period of use, this also means that the
goal will have prerequisites and can be a super- and or subgoal of possibility exists that the user’s (lack of) behaviour change might
other goals. But, and especially when dealing with multiple coaches have been caused by external factors. While measuring behaviour
that all have their own domains such as in the Council of Coaches change is challenging in itself, in addition to the external factors,
system, there is no obvious hierarchy in which goals from multiple the influence of the components layered between the coaching
domains are already related to each other. For example, a goal engine and the user, such as the dialogue management and agent
contributing to a relaxed and happy user might take some effort to representation platforms, should also be taken into account.
balance with goals on physical exercise or dieting. One possible approach for evaluation could be to artificially
The second challenge involves the selection of goals. When pre- generate users, and to, for these users, output the selected goals
sented with a hierarchical network, one might envision a manner and strategies. This output can then be evaluated on whether it
of going through that network and selecting the next goals on the indeed is the response the system should generate.
basis of being relevant or ‘to be completed’. Again, the introduc-
tion of multiple coaches with multiple domains makes this more
challenging. If in the system each coach has its own representation 4 CONCLUSION
of the goal model, this means that the goals that are selected for
In this paper we described the concept of the Artificial Coaching
the coaches should be goals that can coexist.
Engine. We have illustrated its functions and the challenges that
must be faced in the development process. In future research we
3.3 Strategy definition and representation will describe how we have tackled these challenges and report on
Once a coaching goal has been selected, the coach or a combination the implementation of the Artificial Coaching Engine itself and its
of coaches can employ a coaching strategy to reach that goal. The components.
development of these strategies brings with it some design choices.
To start, what is, for example, the ‘duration’ of a coaching strategy?
That is, can a strategy for coaching the user to move more involve
ACKNOWLEDGMENTS
simply telling the user ‘You need to move more!’, or is it a longer The Council of Coaches project, of which this research is a part, has
conversation where the coach tells the user how much they have received funding from the European Union’s Horizon 2020 research
been moving in the past few weeks, what their new step target is, and innovation programme under Grant Agreement #769553. This
why this is important, why they can do it and helps them plan to result only reflects the author’s view and the EU is not responsible
reach the behaviour? for any use that may be made of the information it contains.
38
REFERENCES
[1] Floris Bex, John Lawrence, and Chris Reed. 2014. Generalising argument dialogue
with the Dialogue Game Execution Platform. In Proceedings of the 5th International
Conference on Computational Models of Argument (COMMA 2014). 141–152. https:
//doi.org/10.3233/978-1-61499-436-7-141
[2] Bart A. Kamphorst. 2017. E-coaching systems: What they are, and what they
aren’t. Personal and Ubiquitous Computing 21, 4 (2017), 625–632. https://doi.org/
10.1007/s00779-017-1020-6
[3] Kirsi Kivelä, Satu Elo, Helvi Kyngäs, and Maria Kääriäinen. 2014. The effects
of health coaching on adult patients with chronic diseases: A systematic review.
Patient Education and Counseling 97, 2 (2014), 147–157. https://doi.org/10.1016/j.
pec.2014.07.026
[4] Gary L. Kreps and Linda Neuhauser. 2010. New directions in eHealth communica-
tion: Opportunities and challenges. Patient Education and Counseling 78, 3 (2010),
329–336. https://doi.org/10.1016/j.pec.2010.01.013
[5] Harm op den Akker, Rieks op den Akker, Tessa Beinema, Oresti Banos, Dirk Heylen,
Alison Pease, Catherine Pelachaud, Vicente Traver Salcedo, Sofoklis Kyriazakos,
and Hermie Hermens. 2018. Council of Coaches: A Novel Holistic Behavior
Change Coaching Approach. In Proceedings of the 4th International Conference
on Information and Communication Technologies for Ageing Well and e-Health
(ICT4AgeingWell 2018). 978–989.
[6] World Health Organisation. 2017. Noncommunicable diseases fact sheet. http:
//www.who.int/en/news-room/fact-sheets/detail/noncommunicable-diseases. Ac-
cessed: Apr 2018.
[7] Florian Pecune, Angelo Cafaro, Mathieu Chollet, Pierre Philippe, and Cather-
ine Pelachaud. 2014. Suggestions for Extending SAIBA with the VIB Platform.
Workshop on Architectures and Standards for IVAs, held at the 14th International
Conference on Intelligent Virtual Agents (IVA 2014) (2014), 16–20.
[8] United Nations Department of Economic and Social Affairs. 2017. World Popu-
lation Prospects: The 2017 Revision. http://www.un.org/en/development/desa/
population/events/other/21/index.shtml. Accessed: Apr 2018.
[9] Herwin van Welbergen, Ramin Yaghoubzadeh, and Stefan Kopp. 2014. AsapReal-
izer 2.0: The Next Steps in Fluent Behavior Realization for ECAs. In Proceedings of
the 14th International Conference on Intelligent Virtual Agents (IVA 2014). Springer,
449–462. https://doi.org/10.1007/978-3-319-09767-1_56
39