Content uploaded by Florian Onur Kuhlmeier
Author content
All content in this area was uploaded by Florian Onur Kuhlmeier on May 20, 2022
Content may be subject to copyright.
This is the author’s version of a work that was published in the following source:
Kuhlmeier, Florian Onur, Gnewuch, Ulrich, Lüttke, Stefan, Brakemeier, Eva-Lotta,
Mädche, Alexander. ‘A Personalized Conversational Agent to Treat Depression in Youth
and Young Adults – A Transdisciplinary Design Science Research Project’. International
Conference on Design Science Research in Information Systems and Technology.
Springer, Cham, 2022.
DOI: 10.1007/978-3-031-06516-3_3
Please note: Copyright is owned by the author and / or the publisher. Commercial
use is not allowed.
© 2017. This manuscript version is made available under
the CC-BY-NC-ND 4.0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/
A Personalized Conversational Agent to Treat Depression
in Youth and Young Adults – A Transdisciplinary
Design Science Research Project
Florian Onur Kuhlmeier1,2, Ulrich Gnewuch2, Stefan Lüttke1, Eva-Lotta Brakemeier1,
Alexander Mädche2
1 Department of Psychology, University of Greifswald, Germany
{stefan.luettke, eva-lotta.brakemeier}@uni-greifswald.de
2 Institute of Information Systems and Marketing, Karlsruhe Institute of Technology, Germany
{florian.kuhlmeier, ulrich.gnewuch, alexander.maedche}@kit.edu
Abstract. Depression is a large-scale and consequential problem in youth and
young adults. Conversational agents (CAs) can contribute to addressing current bar-
riers to seeking treatment, such as long waiting lists, and reduce the high dropout
rates reported for other digital health interventions. However, existing CAs have
not considered differences between youth and adults and are primarily designed
based on a ‘one-size-fits-all’ approach that neglects individual symptoms and pref-
erences. Therefore, we propose a theory-driven design for personalized CAs to treat
depression in youth and young adults. Based on interviews with patients (i.e., peo-
ple diagnosed with depression), we derive two design principles to personalize the
character of the CA and its therapeutic content. These principles are instantiated in
prototypes and evaluated in interviews with experts experienced in delivering psy-
chotherapy and potential nondiagnosed users. Personalization was perceived as cru-
cial for treatment success, and autonomy and transparency emerged as important
themes for personalization. We contribute by providing design principles for per-
sonalized CAs for mental health that extend previous CA research in the context
of mental health.
Keywords: Conversational Agent, Mental Health, Personalization, Transdiscipli-
nary Research.
1 Introduction
Depression is one of the most common mental disorders in adolescence and early adult-
hood. Approximately 5.6% of young people worldwide are affected by depression [1].
The individual and social consequences are enormous. Affected individuals are more
likely to exhibit physical impairment and substance abuse, have poorer academic results,
and have an elevated risk of suicide [2–4]. Furthermore, depression causes high health
economic costs [3]. Psychotherapy, delivered by human therapists, is an effective treat-
ment and often the first choice to mitigate the individual and social consequences associ-
ated with depression [5, 6]. However, treatment resources are scarce: On average, people
seeking help have to wait almost five months to start psychotherapy treatment [7]. In
2
addition, young people experience two additional barriers when seeking treatment: First,
they are significantly less likely to use professional support [8] due to feelings of shame,
insecurity, and a greater desire to solve problems themselves [8]. Second, weekly in-per-
son sessions with an adult therapist may not match the technology-driven lifestyle of
youth and young adults. Although digital health interventions (DHI) are available and
effective, studies have shown high dropout rates [5, 9]. Using a conversational agent (CA)
may have great potential to tackle this problem. CAs are software systems that mimic
human conversational behavior [10]. In contrast to other DHI, CAs can not only realize
(1) the specific effects of therapy [11] by delivering therapeutic content, such as providing
information on depression and working through exercises but also (2) the common factors
of therapy [11], such as the alliance between patient and therapist, because CAs offer an
interactive, conversational format that mimics human-delivered therapy [12–14]. By add-
ing the realization of common factors, CAs seem thus promising to increase engagement
and reduce dropout rates to match human-delivered therapy and ultimately improve treat-
ment success. CAs in the context of mental health, such as the highly cited [13, 14] and
successful commercial apps Woebot (woebothealth.com) and Wysa (wysa.io), provide
self-guided therapy based on the principles of cognitive behavioral therapy (CBT), inter-
personal therapy (IPT), or dialectical therapy and have shown promising effectiveness in
reducing symptoms of depression [13, 14]. Moreover, users of mental health CAs report
experiencing relationship building [15] and feelings of social support [16], which sup-
ports the argument that mental health CAs can also realize common factors of therapy
and may thus be better suited than other DHI to treat mental health problems. Although
preliminary evidence shows promising potential for CAs to reduce depressive symptoms,
there are several limitations. First, the majority were tested in pilot studies with a focus
on adults. However, youth differ from adults in terms of cognitive and emotional devel-
opment, social relationships, and problem behavior [17]. In addition, neither the develop-
ment nor the evaluation included participants diagnosed with clinical depression. Thus,
the development and evaluation of CAs for youth (13-17 years) and young adults (18-25
years) must consider these aspects. Second, existing CAs are designed primarily based on
a ‘one-size-fits-all’ approach that neglects individual symptoms and preferences [18].
This is particularly important for youth and young adults because they are used to person-
alizing the content and appearance of digital applications according to their own needs
and preferences. Therefore, it is necessary to consider how CAs can be designed in a way
that allows for personalization.
Against this backdrop, our research focuses on the question of how to design a person-
alized CA to treat depression in youth and young adults. To address this research question,
we are conducting a comprehensive transdisciplinary design science research (DSR) pro-
ject [19, 20]. In the first cycle, we first conducted interviews with youth suffering from
depression to gain an in-depth understanding of the problem, their needs, and preferences.
Based on the interviews, CBT and IPT, and theories of personalization [18, 21], we de-
rived two initial design principles (DPs) for personalized CAs to treat depression. Next,
we instantiated these two initial design principles in four prototypes, which were evalu-
ated in interviews with five experts and five potential users. Our results suggest that per-
sonalizing character and content is crucial to designing effective CAs to treat depression.
3
In addition, transparency and agency are the most important aspects to consider when
implementing personalization.
2 Related Work
2.1 Conversational Agents for Mental Health
The use of CAs to provide self-help psychotherapy interventions has been explored in
several studies [22]. For example, a 2-week use of Woebot, a CA developed based on the
theoretical foundations of CBT to work on depression-typical, dysfunctional thoughts or
behaviors of depression, significantly reduced symptoms of depression [13]. Symptom
reduction was also shown after using Wysa [14]. Recent reviews of mental health CAs
reported high user satisfaction, sufficient effectiveness, and safety to conduct research
with clinical populations [22]. In summary, CAs seem more suitable than other DHI, as
users have reported experiencing social support [16] and a stronger working alliance [15].
2.2 Personalization
In the context of information technology, personalization has been defined as a 'process
that changes the functionality, interface, information access, and content, or distinctive-
ness of a system to increase its relevance to an individual or a category of individuals [12,
p. 183]. Users appreciate personalization features because they can improve ease of use,
efficiency, and provide users with a feeling of being in control [23]. Our work draws on
the frameworks of personalization approaches of Fan and Poole [21] and Kocaballi et al.
[18]. Depending on the specific field of research and discipline, personalization is often
used synonymously with adaptation, customization, and tailoring [21]. We decided to use
the term personalization because it is commonly used in the medical and health literature
[17]. Fan and Poole [21] conceptualize personalization along three dimensions: (1) what
is personalized, i.e. the elements of the system that are being changed, (2) for whom is
the personalization, i.e., the target: individual vs. group, and (3) who is in control of per-
sonalization, i.e. the user or the system. Within dimension (3), the authors differentiate
between implicit (i.e., executed by the system) and explicit personalization (i.e., executed
by the user), Kocaballi et al. [18] extended Fan and Poole’s framework with (4) the pur-
pose of personalization. Table 1 below illustrates the dimensions of personalization that
serve as the basis for our proposed design.
Table 1. Dimensions of Personalization (based on [18, 21])
Dimension
Question
Values (examples)
Purpose
What is the purpose of personalization?
- Increased user motivation
Elements
What is personalized?
- Content
- Functionality
Target
To whom is personalized?
- Single-User vs. Group of Users
Agency
Who is in control of personalization?
- System: implicit/adaptive
- User: explicit/adaptable
- Mixed initiative
4
In their review of personalization features in health CAs, Kocaballi et al. [18] pointed out
that several CAs implemented personalization, such as tailoring content or interaction
styles to individuals. However, they also identified a lack of investigating personalization
within a theoretically grounded and evidence-based framework [18]. In our work, we
mainly focus on the dimensions of purpose, elements, and agency.
3 Methodology
Our research project follows DSR approach [19] to solve an important real-world prob-
lem and design a personalized conversational agent to treat depression in youth and young
adults. We chose this research approach because it allows iterative design [19, 25] and
the participation of users and experts in the design and evaluation phases [19]. We con-
duct a transdisciplinary project due to (1) the focus on a complex problem, (2) the inclu-
sion of an interdisciplinary team consisting of researchers from information systems, clin-
ical psychology, and psychotherapists, and (3) involving societal actors (i.e., patients) as
process participants [20]. A transdisciplinary approach is particularly important given that
poorly designed mental health interventions can have fatal consequences. The DSR pro-
ject is based on the well-established approach suggested by Kuechler and Vaishnavi [25]
and divided into three design cycles to incrementally improve the functionality and im-
pact of our artifact. In this paper, we report the results of the first design cycle, which
focused on understanding the problem space (i.e., treating depression in youth and young
adults using CAs) and exploring personalization to improve treatment success.
Table 2. Overview of our DSR Approach
DSR Project Phases
1. Design Cycle
2. Design Cycle
3. Design Cycle
Awareness of Problem
Interviews with patients
Analysis of Initial Evalua-
tion
Analysis of prior evaluations
Suggestion
Formulation of the initial
design principles
Refinement of
DPs
Refinement of
DPs
Development
Implementation of
first prototype
Implementation of a fully
functional prototype
Implementation of final soft-
ware artifact
Evaluation
Interviews with experts and
potential users (N=10)
Online experiment
with potential users.
Field experiment
with patients
Conclusion
Reflection of initial design
and evaluation results
Reflection of fully functional
prototype and evaluation re-
sults
Formulation of
nascent design theory
In the problem awareness phase, we reviewed the literature on mental health CAs in
clinical psychology and conducted interviews with 15 youth diagnosed with depression,
which we analyzed by first creating a coding scheme and then deriving higher-order
themes. In the suggestion phase, we drew upon frameworks of personalization approaches
[18, 21] as well as CBT and IPT to propose two design principles on how to personalize
mental health CAs for the treatment of depression. Subsequently, we instantiated design
principles in four different prototypes of text-based mental health CAs (i.e., chatbots) de-
veloped with Figma (figma.com) and Botsociety (botsociety.io). These prototypes were
evaluated in interviews with five experts, experienced in clinical psychology and psycho-
therapy, and five potential users. For the evaluation, we selected the technical risk and
efficacy strategy [26] due to the sensitive context of depression: We decided to first
5
evaluate the proposed DPs with a group of experts and potential users to get feedback and
improve our design before evaluating a fully functional prototype in a more naturalistic
setting.
As shown in Table 1, we plan two more design cycles. We will first use the open-
source conversational AI framework Rasa to develop a fully functional prototype. Subse-
quently, we will refine the DPs and improve the prototype based on studies in an online
and naturalistic setting.
4 Design Science Research Project
4.1 Problem Awareness
To improve our understanding of the problem space, we first conducted interviews with
youth diagnosed with depression. We recruited 15 participants between 14 and 17 years
of age, all female, through local clinical psychologists and psychiatrists. The previous
experience of the participants with psychotherapy varied. In line with the literature [7],
all participants previously struggled to find professional treatment due to long waiting
lists. Some participants were frustrated by the lack of interventions to bridge the waiting
time. One participant stated: ‘[I] signed up for this study, because there were no other
forms of treatment when I was on a waiting list. So, [I] wanted to help creating one’.
Another participant expressed her dissatisfaction with a self-help book she had tried. Add-
ing to the literature [8], multiple participants reported feelings of insecurity, stigma, and
the desire to solve their problems on their own as barriers to seeking treatment. The par-
ticipants also identified several advantages of CAs compared to face-to-face psychother-
apy. For example, participants mentioned that CAs would be neutral, non-judgmental,
and anonymous, which facilitates sharing sensitive information. In addition, they appre-
ciated that they could rely on CAs being continuously available and not limited to a single
therapy session per week. In summary, there is evidence that CAs can address some of
the issues raised in the introduction, particularly bridging waiting times.
Regarding the design, the participants expressed a wide variety of needs and prefer-
ences, revealing the importance of personalization. Some participants desired CAs to be
like a friend, that uses similar language. Yet, others wanted the CA to resemble a human
therapist due to the distant, professional relationship, which facilitates conversations
about sensitive topics. Another frequently mentioned topic was the usage of emojis.
While some participants wanted the mental health CA to include emojis (and gifs) in its
messages, others stated that this would look unprofessional and counteract the seriousness
of depression. While some preferred to access the CA through instant messaging apps
such as WhatsApp, others suggested a standalone app. For a standalone app, the design
preferences ranged from a very colorful appearance to a 'professional' black-grey-white
appearance, which was associated with professionalism. Yet, current mental health CAs
do not accommodate the wide-ranging needs and preferences mentioned by our partici-
pants [18]. In addition, our participants explicitly requested personalization features re-
garding the character and the content: ‘I would like to choose a name, change the avatar
and select the topics I want to work on’. One participant wanted the CA to automatically
6
adapt to her therapeutic needs and language style. Taken together, our findings suggest
that a 'one-size-fits-all' approach to designing CAs to treat depression may not be able to
reach its full potential. Although our interviews revealed potential advantages of CAs
compared to human therapists and other interventions, they also emphasized the crucial
role of personalization to improve the user experience and subsequently improve therapy
outcomes.
4.2 Suggestion
From the interviews, we obtained substantial evidence for the importance of personaliza-
tion. However, personalization is complex due to its elusive and multifaceted nature and
the variety of definitions assigned to it by scholars from different fields (e.g., information
systems, health, computer science). To guide our design, we, therefore, drew upon estab-
lished frameworks of personalization [18, 21] that were introduced in Section 2.1. Ac-
cording to these frameworks, the fundamental dimension of personalization is the element
of personalization (i.e. what is being personalized). In the context of CAs, these elements
primarily include the CA’s character (i.e., gender, age, social role etc.) and the content
(i.e., the content of the messages, knowledge base, etc.) [27]. In the interviews, 8 out of
15 participants expressed the desire to personalize the name, gender, and social role of a
CA, suggesting that personalizing the character should represent a major design principle
(DP). Therefore, we propose DP1: To improve treatment outcomes for depressed youth
and young adults, provide the conversational agent with the capability to personalize its
character to match user needs and preferences because a personalized character helps
users to form a stronger relationship with the CA. The second key element of personali-
zation is the CA’s (therapeutic) content. According to the health literature, personalized
content improves the use [28] and the perceived helpfulness of DHI [29]. Thus, we pro-
pose DP2: To improve treatment outcomes for depressed youth and young adults, provide
the conversational agent with the capability to personalize the therapeutic content to
match user needs and preferences because personalized content increases the relevance
and efficiency of the CA. As introduced above, the second dimension of personalization
is agency (i.e., who controls the personalization). As our participants expressed their in-
terest in both adaptable CAs, in which they are in control of personalization, and adaptive
CAs, in which CAs control personalization, we integrate adaptable, adaptive, and mixed-
initiative personalization into our DPs. By instantiating prototypes that demonstrate all
these approaches, we aimed to evaluate and prioritize these approaches and then refine
the DPs accordingly.
4.3 Development
To instantiate our initial DPs, we developed four prototypes. As the participants’ prefer-
ences varied substantially, we aimed to explore different elements and degrees of agency
of personalization in our prototypes. Based on the evaluation results, we aim to find the
most important features and refine the DPs accordingly. The first two prototypes instan-
tiated the personalization of the CA’s character (DP1). The first prototype provided the
user with the opportunity to personalize the name, gender, typing speed, avatar, and social
7
role. These characteristics were selected based on our findings from the interviews with
patients. The second prototype showcased the possibility for the CA to automatically
adapt to the users’ use of emojis, since the use of emojis emerged as a polarizing element
during the interviews.
Fig. 1. DP1 – Personalization of Character: Prototypes 1 (left) and 2 (center and right).
The other two prototypes instantiated the personalization of the content (DP2). In CBT and
IPT, content comes in the form of modules (e.g., behavioral activation, sleep hygiene). We in-
stantiated two prototypes that reflect the personalization of these modules in different ways.
Prototype three contained the task to respond to items from a
Fig. 2. DP2 – Personalization of Content: Prototypes 3 (left) and 4 (center and right).
depression scale and the relevant modules were selected based on their responses.
For instance, the module on sleep improvement is only integrated if a user reports sleep
problems. Prototype four instantiated a more flexible version of the second design prin-
ciple. Here, instead of personalizing the content once in the beginning, a matching mod-
ule is suggested when users report specific issues on a particular day. For example,
CADY suggests the module sleep hygiene if users report sleep problems during daily
check-in.
8
4.4 Evaluation
To evaluate our prototypes, we conducted interviews with five independent experts with
experience in delivering psychotherapy (3 female, Mage=29) and five potential users (3
female, Mage=24). By including experts, our objective was to understand whether our pro-
posed design is consistent with established principles of psychotherapy. We decided to
recruit non-diagnosed individuals as potential users to first ensure the safety of the proto-
types before including young people diagnosed with depression. In each interview, we
first explained the concept of CAs and introduced our research project. Subsequently, we
explained the DPs and demonstrated their instantiations. During the presentation and af-
terwards, participants were asked to evaluate the prototypes and to provide ideas for fur-
ther personalization. The interviews lasted 40 minutes on average. All interviews were
recorded and transcribed. To analyze the feedback from the participants, we used a bot-
tom-up approach to synthesize the interviews into higher-order themes.
Results and Discussion.
All participants appreciated the personalization of the CA to suit their own needs and
preferences (or those of their clients), providing evidence of the utility of both DPs.
Moreover, all participants emphasized personalization as a crucial feature for the suc-
cess of mental health CAs. In terms of DP1 and prototype 1, every participant supported
the idea of personalizing the agent’s name, gender, and avatar as a mechanism for rela-
tionship building. Especially gender was identified as an important characteristic for
users to feel safe and comfortable in case they’ve had negative experiences regarding
one gender in the past. Using a robot or an animal avatar was suggested as an additional
gender-neutral and nonhuman version to satisfy users who prefer to talk with a robot
instead of a human. The participants also suggested adding age as a variable to choose
from. Instead of personalizing each aspect separately, multiple participants suggested
combining variations of gender, avatar, age, and social role into 3-4 different characters,
from which users can choose. They argued that presenting a few characters instead of
each characteristic separately would decrease the variables to choose from, which could
otherwise be overwhelming and result in annoyance or dropout. In addition, partici-
pants suggested comprehensive information (e.g., brief introductory videos) about each
character, so users can imagine what interacting with them would feel like. In terms of
the specific social role, participants expressed interest in a non-human, agender robot,
an older therapist-like role and a younger coach-like role. Most experts advised against
implementing a friend-like role (like in prototype 1) as they feared that the lack of a
professional relationship could endanger the therapeutic process. Therefore, they sug-
gested that one should be able to choose between professional roles that encompass
different personality traits: ‘For example, I would suggest that social roles differ be-
tween warm, understanding, empathic versus rather cool, rational, direct.’ Regarding
prototype 2, experts and users generally valued the idea of providing the CA with the
agency to adapt to their use of emojis and language more generally, as experts explained
that adapting to the clients’ language resembles therapist-client relationship building in
the context of psychotherapy. In addition, potential users indicated that they regularly
adapt the emoji and language use to their friends and that this could improve the human-
9
chatbot relationship. However, some participants were concerned with implementing
the feature before it had reached sufficient accuracy. They stated that an insufficient
automated adaptation would be worse than a non-adaptive system. Participants also
requested the feature to turn off the automated adaption and information on how the
CA adapts to them. Instead of automatically regulating emoji and language usage, one
participant suggested integrating different language styles and emoji use into the dif-
ferent characters to give users control and counter potential technical limitations.
In terms of DP2, experts and potential users perceived the personalization of the
therapeutic content, i.e. the purpose of the personalization, to be crucial for the success
of a CA to treat depression and more important than DP1. Regarding prototype 3, ex-
perts and potential users liked the idea of personalizing content at the beginning based
on responses to a depression scale: ‘I think it is important that the agent asks about the
symptoms of depression. And it's also important that it's highly structured because most
of the time it's very, very difficult for my clients to verbalize their issues’. One expert
suggested an extension of prototype 3: ‘In addition to the depression scale, it should be
possible for a user to openly state the most pressing issue. If users feel that the agent
listens and prioritizes this issue, it will increase their motivation, which is crucial for
the treatment success.’
When evaluating prototypes 3 and 4, a trade-off between flexible personalization
and a structured plan emerged. On the one hand, experts and potential users emphasized
the need for autonomy, i.e., the ability to flexibly choose or change a module instead of
a fixed schedule, and its potential to increase motivation and engagement. On the other
hand, experts emphasized the importance of a plan with compulsory modules and a
fixed sequence. The fixed sequence was deemed important because some modules can
be tiring and difficult but play a crucial role in achieving treatment success and there-
fore need to be completed. Experts mentioned that a structured plan also provides users
with certainty and transparency, which makes CAs more reliable and the treatment
goals more visible. However, an inflexible plan, which does not sufficiently integrate
individual needs and preferences, could reduce motivation, user engagement, and thus
lead to dropout. Consequently, the challenge is a compromise between personalizing
therapeutic content flexibly and maintaining a structured program, which one expert
summarized: ‘Some content should be fixed, but users should still feel that they can
decide for themselves. But not only depending on the momentary mood. If users only
choose based on the momentary mood, then there will probably not be much change.
You will have to build some feature that makes sure users are also doing the exercises
and consume the information no matter what their mood is like.’ A possible solution
emerged from combining prototypes 3 and 4: Experts suggested keeping the personal-
ization of the therapy modules in the beginning based on psychometric data and pre-
senting these results as a personalized structured program while being able to deviate
when a specific issue (like sleep problems or low energy) arises. However, when devi-
ating, it should be explicitly framed as a deviation from the personalized structured
treatment plan. In prototype 4, the CA suggested a module because it recognized sleep
problems in the users’ text messages during daily check-in. Although participants ap-
preciated that the CA was able to handle an acute problem, experts reiterated that young
people often cannot verbally express their problems. Therefore, one expert suggested
10
personalizing the daily check-in: ‘Maybe it is helpful to ask ‘how are you today’ in
different ways because there are people who just never know an answer to this question.
You could work with something like a thermometer or emojis. So, the agent could first
ask 'I would like to know how you are doing, in what way do you want to tell me today?’
and then the user can select a thermometer, choose an emotion from a list, or select to
write a text message.'
Based on feedback from our participants, we identified several opportunities to im-
prove the prototypes. While both DPs received positive feedback, the feedback also
revealed that the automatic personalization of the character may be less promising than
initially expected. Combining this feedback with the technical challenges of making the
CA’s character adaptive, we have decided to no longer pursue automatic adaptation.
Regarding DP1, we will focus on user-controlled personalization of the mental health
CA’s character and regarding DP2, we will implement explicit personalization and
mixed-initiative. This refinement and the suggested improvements for the prototypes
serve as the entry point into the second cycle. In general, participants discussed two
themes the most: (1) autonomy, i.e., giving user control over personalization features,
and (2) transparency, i.e., being transparent about what is being personalized and how
it is done.
5 Conclusion
This paper presents insights from our ongoing transdisciplinary DSR project to design
a personalized CA to treat depression in youth. Based on interviews with our target
group, we corroborated the need to integrate personalization features into the design
process. We proposed two DPs to guide the design of a personalized CA and instanti-
ated the DPs in four prototypes. We evaluated the prototypes in interviews with experts
and potential users. Overall, the feedback was positive, and the importance of person-
alization was confirmed. However, participants also expressed concerns about auto-
mated personalization performed by a CA since they were sceptical of the technical
feasibility and emphasized the loss of control. In general, autonomy and transparency
emerged as important themes guiding the design of personalization efforts. Finally, our
participants gave valuable feedback for (1) refining and extending the proposed per-
sonalization features and (2) suggesting additional personalization features (e.g. per-
sonalized reminders), which we will incorporate into our next DSR cycle. In summary,
our results show that personalized mental health CAs are a promising approach to ac-
commodate users’ symptoms and preferences. However, to comprehensively evaluate
the impact of personalization, more research is needed that compares CAs with and
without personalization features. Although our research follows established guidelines
for conducting DSR [19, 25], we need to highlight some limitations. First, the samples
for the problem awareness and the evaluation interviews were relatively small. In addi-
tion, the evaluation interviews included only nondiagnosed individuals. Consequently,
for the results to be more comprehensive and generalizable, larger sample sizes are
necessary. Second, we used an interactive prototype and brief prototype videos to
demonstrate our proposed design. Although we argue that this approach is appropriate
11
for a first DSR cycle, further research based on a fully functional prototype is crucial.
Therefore, in our second DSR cycle, we will implement the most important personali-
zation features in a fully functional prototype. Evaluating our DPs again in the second
DSR cycle will also contribute to further refining and validating our DPs, which is a
crucial next step. With our research presented in this article, we contribute valuable
design knowledge that serves as a starting point for future research on the design of
personalized mental health CAs.
References
1. Jane Costello, E., Erkanli, A., Angold, A.: Is there an epidemic of child or adolescent de-
pression? J. Child Psychol. Psychiatry. 47, 1263–1271 (2006).
2. Ellsäßer, G.: Unfälle, Gewalt, Selbstverletzung bei Kindern und Jugendlichen 2017. Ergeb-
nisse der amtlichen Statistik zum Verletzungsgeschehen 2014. Fachbericht. (2017).
3. Greiner, W., Batram, M., Witte, J.: Kinder- und Jugendreport 2019. Gesundheitsversorgung
von Kindern und Jugendlichen in Deutschland. Schwerpunkt: Ängste und Depressionen bei
Schulkindern, in Beiträge zur Gesundheitsökonomie und Versorgungsforschung. , Bielefeld
und Hamburg (2019).
4. Thapar, A., Collishaw, S., Pine, D.S., Thapar, A.K.: Depression in adolescence. Lancet.
379, 1056–1067 (2012).
5. Oud, M., de Winter, L., Vermeulen-Smit, E., Bodden, D., Nauta, M., Stone, L., van den
Heuvel, M., Taher, R.A., de Graaf, I., Kendall, T., Engels, R., Stikkelbroek, Y.: Effective-
ness of CBT for children and adolescents with depression: A systematic review and meta-
regression analysis. Eur. Psychiatry J. Assoc. Eur. Psychiatr. 57, 33–45 (2019).
6. Cuijpers, P., Noma, H., Karyotaki, E., Vinkers, C.H., Cipriani, A., Furukawa, T.A.: A net-
work meta-analysis of the effects of psychotherapies, pharmacotherapies and their combi-
nation in the treatment of adult depression. World Psychiatry. 19, 92–107 (2020).
7. Bundespsychotherapeutenkammer: Ein Jahr nach der Reform der Psychotherapie-Richtli-
nie. (2018).
8. Gulliver, A., Griffiths, K.M., Christensen, H.: Perceived barriers and facilitators to mental
health help-seeking in young people: a systematic review. BMC Psychiatry. 10, 113 (2010).
9. Leech, T., Dorstyn, D., Taylor, A., Li, W.: Mental health apps for adolescents and young
adults: A systematic review of randomised controlled trials. Child. Youth Serv. Rev. 127,
106073 (2021).
10. Dale, R.: The return of the chatbots. Nat. Lang. Eng. 22, 811–817 (2016).
11. Cuijpers, P., Reijnders, M., Huibers, M.J.H.: The Role of Common Factors in Psychother-
apy Outcomes. Annu. Rev. Clin. Psychol. 15, 207–231 (2019).
12. Ahmad, R., Siemon, D., Gnewuch, U., Robra-Bissantz, S.: Designing Personality-Adaptive
Conversational Agents for Mental Health Care. Inf. Syst. Front. (2022).
13. Fitzpatrick, K.K., Darcy, A., Vierhile, M.: Delivering Cognitive Behavior Therapy to
Young Adults With Symptoms of Depression and Anxiety Using a Fully Automated Con-
versational Agent (Woebot): A Randomized Controlled Trial. JMIR Ment. Health. 4, e19
(2017).
12
14. Inkster, B., Sarda, S., Subramanian, V.: An Empathy-Driven, Conversational Artificial In-
telligence Agent (Wysa) for Digital Mental Well-Being: Real-World Data Evaluation
Mixed-Methods Study. JMIR MHealth UHealth. 6, e12106 (2018).
15. Darcy, A., Daniels, J., Salinger, D., Wicks, P., Robinson, A.: Evidence of Human-Level
Bonds Established With a Digital Conversational Agent: Cross-sectional, Retrospective
Observational Study. JMIR Form. Res. 5, e27868 (2021).
16. Brandtzaeg, P., Skjuve, M., Dysthe, K., Følstad, A.: When the Social Becomes Non-Hu-
man: Young People’s Perception of Social Support in Chatbots Social Support in Chatbots.
Presented at the April 3 (2021).
17. Lohaus, A. ed: Entwicklungspsychologie des Jugendalters. Springer-Verlag, Berlin Heidel-
berg (2018).
18. Kocaballi, A.B., Berkovsky, S., Quiroz, J.C., Laranjo, L., Tong, H.L., Rezazadegan, D.,
Briatore, A., Coiera, E.: The Personalization of Conversational Agents in Health Care: Sys-
tematic Review. J. Med. Internet Res. 21, e15360 (2019).
19. Hevner, A.R., March, S.T., Park, J., Ram, S.: Design science in information systems re-
search. MIS Q. 75–105 (2004).
20. Lawrence, M.G., Williams, S., Nanz, P., Renn, O.: Characteristics, potentials, and chal-
lenges of transdisciplinary research. One Earth. 5, 44–61 (2022).
21. Fan, H., Poole, M.S.: What Is Personalization? Perspectives on the Design and Implemen-
tation of Personalization in Information Systems. J. Organ. Comput. Electron. Commer. 16,
179–202 (2006).
22. Vaidyam, A.N., Linggonegoro, D., Torous, J.: Changes to the Psychiatric Chatbot Land-
scape: A Systematic Review of Conversational Agents in Serious Mental Illness: Change-
ments du paysage psychiatrique des chatbots: une revue systématique des agents conversa-
tionnels dans la maladie mentale sérieuse. Can. J. Psychiatry. 0706743720966429 (2020).
23. Blom, J. o, Monk, A.F.: Theory of Personalization of Appearance: Why Users Personalize
Their PCs and Mobile Phones. Human–Computer Interact. 18, 193–228 (2003).
24. Huibers, M.J.H., Lorenzo-Luaces, L., Cuijpers, P., Kazantzis, N.: On the Road to Personal-
ized Psychotherapy: A Research Agenda Based on Cognitive Behavior Therapy for Depres-
sion. Front. Psychiatry. 11, (2021).
25. Kuechler, B., Vaishnavi, V.: On theory development in design science research: anatomy
of a research project. Eur. J. Inf. Syst. 17, 489–504 (2008).
26. Venable, J., Pries-Heje, J., Baskerville, R.: FEDS: a framework for evaluation in design
science research. Eur. J. Inf. Syst. 25, 77–89 (2016).
27. Diederich, S., Brendel, A., Morana, S., Kolbe, L.: On the Design of and Interaction with
Conversational Agents: An Organizing and Assessing Review of Human-Computer Inter-
action Research. J. Assoc. Inf. Syst. 23, 96–138 (2022).
28. Radomski, A.D., Wozney, L., McGrath, P., Huguet, A., Hartling, L., Dyson, M.P., Bennett,
K., Newton, A.S.: Design and Delivery Features That May Improve the Use of Internet-
Based Cognitive Behavioral Therapy for Children and Adolescents With Anxiety: A Realist
Literature Synthesis With a Persuasive Systems Design Perspective. J. Med. Internet Res.
21, e11128 (2019).
29. Garrido, S., Cheers, D., Boydell, K., Nguyen, Q.V., Schubert, E., Dunne, L., Meade, T.:
Young People’s Response to Six Smartphone Apps for Anxiety and Depression: Focus
Group Study. JMIR Ment. Health. 6, e14385 (2019).