Recommendations of Embodied Conversational Agents

to Healthcare Applications

Julio Oliveira

, Telmo Silva

Rita Oliveira

and Elizabeth Furtado

Digimedia, Universidade de Aveiro, Aveiro, Portugal

Doutorado em Informática Aplicada, Universidade de Fortaleza, Fortaleza, Brazil

Keywords: ECA, Embodied Conversational Agent, Hypertension, Older Adults.

Abstract: This paper identifies recommendations for Embodied Conversational Agents (ECA) in Healthcare

applications. The methodology employed consists of two systematic literature reviews in the fields of

conversational systems and health care. Twenty-six recommendations for ECA were categorized into four

groups: CS Interface, ECA Functionalities, Agent Behavior, ECA Customization Features, and Older Adults

Engagement. Additionally, six Healthcare dimensions were identified: Interactive Learning, Disease-Specific

Knowledge, Reinforcement, Emergency Detection, and Empathy/Rapport. These two sets of findings were

combined for evaluation by a group of experts. The impact evaluation revealed six essential, two necessary,

and eighteen desirable recommendations. The essential recommendations, derived from empirical methods,

include the following: storing encrypted information, providing secure and accurate information to patients,

facilitating interactive learning, allowing users to choose whether to enable proactive mode and ensuring ease

of installation and use. The set of recommendations is an important contribution for ECA developers as this

research presents.

1 INTRODUCTION

The use of Conversational Systems – CS in health

care is not recent. The first CS was ELIZA, developed

by Weizenbaum, and evolved from a simple chatbot

to multimodal communication (Car et al., 2020). CS

can be interactive, use active two-way

communication, and use speech as a method of

communication with the patient. The technology can

suit various populations, ranging from young children

to older people.

The use of CS can promote effective interaction,

engagement, and intervention, particularly in

managing chronic conditions, promoting healthy

behaviour, and supporting older adults (Pradhan et

al., 2020). Special Conversational Agents allow

reliable and efficient information delivery and should

exhibit social skills (Yang & Aurisicchio, 2021).

These skills include using nonverbal behaviour to

convey communicative and social signals. Ideally,

https://orcid.org/0000-0002-3516-0114

https://orcid.org/0000-0001-9383-7659

https://orcid.org/0000-0001-6041-9469

https://orcid.org/0000-0002-1584-3161

agents should be able to produce and recognize

nonverbal signals in simulations of face-to-face

actions. Agents with such capabilities are named

Embodied Conversational Agents - ECA (Barros,

2021, Bickmore & Cassell, 2005). Some features

related to ECA are facial and body expressions,

movement of parts, and even the use of lights and

colors to promote interaction (Potdevin et al., 2021;

Ruttkay et al., 2004).

From the perspective of ECA´s development,

features, recommendations, characteristics, and

emphatic aspects need to be related to Health Care.

There are a lot of existing recommendations in

Literature, but we propose a way to validate them

using a questionnaire with specialists. The result is a

list of recommended features to achieve healthcare

support and coaching. This under-progress work is a

part of PhD Thesis and intends to identify the

recommendations for the use of ECA in support of

hypertension treatment by Older Adults.

352

Oliveira, J., Silva, T., Oliveira, R. and Furtado, E.

Recommendations of Embodied Conversational Agents to Healthcare Applications.

DOI: 10.5220/0013438700003938

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 11th International Conference on Information and Communication Technologies for Ageing Well and e-Health (ICT4AWE 2025), pages 352-359

ISBN: 978-989-758-743-6; ISSN: 2184-4984

The next section presents material and methods

used to collect the recommendations and key aspects

of Health. Section three presents twenty-six ECA

recommendations grouped by their functions. The

fourth section presents six key dimensions of Health

Care as a parameter to combine with ECA´s

recommendations. Section five shows the results of

matching the two sets of Recommendations and Key

Dimensions of Health found. The last section

proposes future works.

2 MATERIAL AND METHODS

Two Systematic Literature Reviews (SLRs) and

questionnaires were used to identify the ECA

recommendations. The process was made in three

steps, as shown in Figure 1.

The first Step (S1) was a SLR with the query

“("evaluation of" OR "guidelines" OR "heuristics"

OR "recommendations”) AND ("embodied" OR

"conversational Agents" OR "voice Agents") AND

("elder*" OR “senior" OR "older adults")”.

The Scopus

database returned six articles, and the

ACM Digital Library database, 33 articles, totalling

39 sources to be analyzed. After reading the abstracts,

29 were eliminated: 12 that were not related to the

ECA in Health Care, five did not involve ECAs, 8 did

not involve older adults, two dealt with text interfaces

(chatbots), in addition to 2 repeated articles. After the

full reading of the articles, five were selected, which

will be summarized in the next section. The update

phase included more three papers considering the

actual stage of the Large Language Model (LLM),

such as ChatGPT. From the reading 26 ECA

Recommendations were listed to be used in Step 3.

Figure 1: Steps to Identify Essential Recommendations

ECA in Health Care (created by the author).

Scopus Database. Available Sep 11, 2024,

https://www.scopus.com

The SLR made in Second Step (S2) used the

query “("e-health" OR "health" OR "telemedicine")

AND ("recommendation" OR "dimension*") AND

("voice") AND ("embodied" OR "conversational

Agents" OR "Conversational System" AND "older

adults" AND NOT "mental health") excluding

publications before 2022. Scopus returned five papers

and ACM Library 35 publications. 32 were

eliminated: 10 that did not involve ECA and Health

Care, 12 did not present any recommendation for

Health use, and obtained 8 articles. Six key

dimensions were selected to play the role of

evaluating each ECA´s recommendation identified.

The final step (Step 3) investigates the ECA

Recommendations concerning the Key Dimensions

found and their importance in achieving effective

Health Care. The method used was a digital

questionnaire to eight experts in Conversational

System applied to e-Health. The results found are

detailed in section five.

3 ECA RECOMMENDATIONS

In this work, an ECA Recommendation is a guideline

related to the CS Interface, ECA Functionalities,

Agent Behavior, or ECA Customization Features.

Each author presents suggestions for the features or

characteristics of an ECA. Six design

recommendations of ECA for interaction with elderly

people with dementia were extracted from EVA

conversational agent from EVA conversational agent

(Sandoval & Favela, 2017).

ECA LOUISE presented Alternatives to

overcome the user interface restrictions linked to

cognitive impairment in the elderly, making it

possible to establish 14 recommendations (Wargnier

et al., 2018). Eleven recommendations were

identified in the results presented in a Systematic

Mapping Study of Usability and User eXperience of

CS (Guerino & Valentim, 2020).

There are 13 guidelines presented in a study that

involves older people with dementia identified in a

Voice User Interface (VUI) development (Striegl et

al., 2021a). Of the 44 recommendations A total of 18

recommendations were excluded due to the repetition

of features. The remaining 26 recommendations were

systematically numbered to reference each type. Each

recommendation is tagged according to its

corresponding group: R1?? pertains to CS Interface,

R2?? relates to ECA Functionalities, R3?? addresses

Recommendations of Embodied Conversational Agents to Healthcare Applications

353

Agent Behaviour, and R4?? corresponds to

Customization Features, with ?? representing a

sequential number. The author has proposed a

classification system for the recommendations within

each group.

3.1 Recommendations on the CS

Interface

Table 1 identifies three Recommendations for the

Conversational System (CS) Interface. These

recommendations are proposed by (Wargnier et al.,

2018) and (Guerino & Valentim, 2020) and can be

applied to most ECA for health areas or others.

Recommendation R101 refers to the need for

adjustments after the first use of ECA. At this point,

the input phrase should be indicated and, if possible,

modified (R102). Recommendation R103 highlights

the ease of use of ECA for users who are not familiar

with the technology. The configurations, such as

connection to the network, must be made by an initial

dialog.

Table 1: Recommendations on the CS Interface.

Conversational S

stem Interface

(

R101

)

Sim

le interface, no learnin

uire

(

R102

)

Indicate the A

ent's entr

ase

(R103) Agents should be ready to use immediately.

3.2 ECA Functionalities

Ten recommendations related to Conversational

System functionalities are listed in Table 2.

Recommendation R201 is related to the use of

external sensors such as the User´s temperature and

heartbeat, brain activities, and emotional responses

from the skin. Those sensors generally reduce the

usability of the equipment as they use wires to

connect the users to the ECA.

Recommendations R202 - Recognize the

environment of use and R203 - Recognize the user by

speech (Riccardi, 2014) requires a camera and

microphone connected to the ECA to assess the

environment and record the user's voice. These

recordings and further information must attend the

R204 and be encrypted to give maximum security and

privacy to the user data. Recommendations R205 -

Track the user and R206 - Locate the sound source

proposed by (Wargnier et al., 2018) are met when

more than one microphone is installed. Sound

localization in artificial systems usually uses two (or

more) microphones. By the difference in the arrival

times of a sound of the microphones, it is possible to

estimate the direction of the sound source

mathematically.

For the user to see realistic lip-synchronization

movements with speech, expressive feedback in

speech, and gestures, virtual agents need to use real

graphical animations for the user: R207 - Introduce

advanced features in avatar animation (Striegl et al.,

2021).

Recommendation R208 - Provide secure and

correct information is a feature linked to the base of

knowledge accessed by the ECA. A (medical) team

must validate the source of information. In a case of

doubt, the ECA must send the issue to a human. Once

the human team has defined the procedure to be taken

and the answer to be given, it should input it into the

knowledge base (R209: Provide learning) (Striegl et

al., 2021).

The last recommendation in this group deals with

storing the dialogues that have already been taken

with the user. Further communication can use the

previous information to provide a real human dialog.

The greatest care taken with this recommendation lies

in identifying true and doubtful information. The

process of filtering out errors must be planned to

make the result relevant (R210: Provide continuity in

dialogue).

Table 2: Recommendations for ECA Functionalities.

ECA Functionalities

(R201) Recognize embedded signals (Use of sensors)

(R202) Recognize the environment of use

(R203) Recognize the user by speech

(R204) Store encrypted information

(R205) Track the user

(R206) Locate the sound source

(R207) Introduce advanced features in avatar animation

(R208) Provide secure and correct information

(R209) Provide learning

(R210) Provide continuity in dialogue

3.3 Agent Behaviour

Recommendations

This group recommendation is related to the

Behaviour of the Conversational Agent itself. The six

recommendations are listed in Table 3 and explained

as follows. The agent must incorporate the ability to

coordinate and regulate emotions and understand the

affective sustainability of interaction with ECA

(R301: Contemplate affectivity).

To perform a (R302 Allow Sociability) “(..) Not

only must a Virtual Agent be able to understand the

user's emotion, but they must also be able to respond

accordingly" (Riccardi, 2014). This social component

is a key human skill and will be a research challenge

ICT4AWE 2025 - 11th International Conference on Information and Communication Technologies for Ageing Well and e-Health

354

for human-machine interaction and possibly for

machine-machine interaction (Riccardi, 2014).

Recommendation R303 - Contemplate the avatar's

Behavioural reaction, established by (Wargnier et al.,

2018), relates to implementing a behaviour analysis

module based on recommendations R203, R206,

R207, and R208 described in item 3.1.2.

Recommendation R304 - Assist the user as little

as necessary refers to the ability of communicate to

the user in different ways, depending on the dialog.

This feature is necessary to include people who does

not understand easily an instruction (Striegl et al.,

2021b). The R305 Track the user's progress during

the execution of a task is related to R205 but is not the

same. This recommendation refers to tracking the

user's body and face when executing a task.

The last recommendation of this group is R306 -

React autonomously to user requests. The agent needs

to be aware of a call and react not only with voice but

with non-verbal activity such as a voice change or

intonation.

Table 3: Agent Behaviour Recommendations.

ent Behaviour

(R301) Contemplate affectivity

(R302) Allow Sociability

(R303) Contemplate the avatar's Behavioural reaction;

(R304) Assist the user as little as necessary, breaking

down complex instructions based on the needs of

the users

(R305) Track the user's progress during the execution

of a task

(R306) React autonomously to user requests

3.4 ECA Customization Features

The Recommendations group related to customization

and configuration are outlined in

Table

4. The first recommendation refers to the

proactiveness of ECA (R401). Proactive engagement,

such as medication adherence and health monitoring

reminders, enhances the utility of ECAs. Interactive

elements foster continuous user participation, as

evidenced by studies on virtual agents like PACO and

Addison Care (Krutter et al., 2022). To perform user

security, the ECA must attend the R401 - provide

control and adjustments in interventions from the

ECA with the user. If the interventions of the ECA

could be made without the Entry Phrase, the user must

allow the record of all sounds around the equipment.

The interventions must be realized without user

command to permit a proactive comportment.

The second recommendation of this group is

R402, which refers to the adjustment of personality,

conversation topics, or vocabulary according to the

user. This allows the user to configure the personality

of the Agent, such as the way of abord themes, for

example. Also, the conversation topics, in our case

the Health field, and mainly a vocabulary depending

on the user's level of knowledge (Guerino &

Valentim, 2020). The recommendation numbered by

R403 provides a way to configure different faces,

voices, gestures, and embedding. Adjustment of the

ECA Agent is a key to improve empathy with the

user. The level of configuration starts from the gender

of the agent and can be personalized to a real human

characteristic. A work draws attention to the fact that

the customization of the pronunciation and

personality of the agent is not indicated (Guerino &

Valentim, 2020).

Recommendation R404 refers to adding new

character models quickly; this feature is a way to

propose a known person as the agent.

Recommendation R405 is about to describe

interaction scenarios in a dedicated syntax. A way to

guide the user to some tasks is to adopt scenarios. In

this solution, the user can describe what the ECA

should do to achieve a goal. For example, the user

should ask the ECA to measure the blood pressure 3

times a day and inform any anomaly to medical

support. The personalization of the pronunciation of

the Agent voice (R406) refers to promoting different

voice intonation and word pronunciation. It is

remarkable when the same language is spoken in

countries like Portugal and Brazil. For example, the

agent's voice must be configurable to Portuguese of

Portugal residents.

Table 4: Recommendations for Agent Customization.

ECA Personalization

(R401) Allows control and adjustments in interventions

(R402) Allows adjustment of personality, conversation

topics, or vocabulary according to the user

(R403) Allows to configure different faces, voices,

gestures, and embedding

(R404) Allows to add new character models easily;

(R405) Describe interaction scenarios in a dedicated

syntax

(R406) Personalize the pronunciation of the agent voice

(R407) Set up the entry phase, response times, speed

and speaking style

Besides, pronunciation is necessary to configure

the tone, velocity, and other voice characteristics, such

as an agitated woman's voice or an old, relaxed voice

(R407). This fine configuration of the Avatar

communication is fundamental to an effective human-

like dialog experience. The goal is to perform

comfortable and easy-to-understand communication

between the agent and the user. The configuration

Recommendations of Embodied Conversational Agents to Healthcare Applications

355

must be made easily, giving the user alternatives to

choose or recording an example sample voice to be

used.

4 HEALTHCARE DIMENSIONS

Healthcare has shown a particular interest in the use of

Conversational Agents such ECA (terStal et al., 2020).

To develop the ECA to support eHealth areas, it is

necessary to discover mandatory ECA

recommendations to achieve better patient

engagement.

Technology variability of CS is related to nine

aspects: devices, application technology, service

channel, hosting and storage, electronic health record

(EHR) integration, input/output model, intelligence

framework, sentiment detection, and privacy and

security (May & Denecke, 2024).

User-related variability is related to seven aspects:

human involvement, care flow integration, patient

profile, language and communication style, health

literacy, service duration, and personal preference

(May & Denecke, 2024).

The listed variables were mapped into six

dimensions: Interactive Learning, disease-specific

Knowledge, Reinforcement, Emergency Detection,

and Empathy/Rapport.

4.1 Interactive Learning

In an Interpretative Model of Interaction by Patient-

Physician (Gross et al., 2021) the Physician interprets

the patient's values, and the patient decides.

Using an ECA as the physician is necessary for an

Interactive Learning (IL) approach to engage and

educate patients about health topics (Egede et al.,

2021). The method should use videos, graphics, and

animations to illustrate the concepts and show

examples. The IL needs to be applied to Health

Literacy and all engaging procedures like medicine

intake, treatment procedures, alimentary suggestions,

and so on.

4.2 Disease-Specific Knowledge

Disease-Specific Knowledge (SK) refers to the ECA

performing accurate and up-to-date information on

specific. The outcome of the health system is to

provide correct information depending on the patient,

based on the EHR. (Egede et al., 2021).

Some conversational systems, such as Alexa or

Google Assistant, use the Internet as a source of

knowledge. This dimension points to a dedicated

base created by specialists.

4.3 Reinforcement

The Reinforcement (R) is a feature related to

repeating messages over time until the goal is

obtained. To follow a treatment or a medicine

administration, the patient needs to be aware several

times (Egede et al., 2021). Reinforcement learning is

a technique involving an agent who needs to decide

which actions it needs to do to perform a task that has

been assigned to it most effectively. For this, rewards

are assigned to the different actions that the agent can

take in different situations or states of the

environment.

Initially, the agent has no idea about the best or

correct actions. Using reinforcement learning, it

explores its action choices via trial and error and

figures out the best set of actions for completing its

assigned task. The basic idea behind a reinforcement

learning agent is to learn from experience. Just like

humans learn lessons from their past successes and

mistakes, reinforcement learning agents do the same

– when they do something “good,” they get a reward,

but if they do something “bad”, they get penalized.

The reward reinforces the good actions while the

penalty avoids the bad ones.

Reinforcement learning requires several key

components: Agent – This is the “who” or the subject

of the process, which performs different actions to

perform a task that has been assigned to it.

Environment – This is the “where” or a situation in

which the agent is placed. Actions – This is the

“what” or the steps an agent needs to take to reach the

goal. Rewards – This is the feedback an agent

receives after performing an action.

4.4 Emergency Detection

To allow the user a way to Emergency Detection

(ED), which refers to the ability to recognize and

respond to potential health emergencies (Sunghoon et

al., 2019). Specialized sensors can detect alterations

in the health state of the patient. Such sensors are

electrocardiogram sensor (ECG), Electromyography

(EMG), Electro dermal Activity (EDA), or

Accelerometer sensor (ACC). Cameras and

microphones can, otherwise, help to identify some

emergencies in conversational voice systems. This

feature can give the patient a way to have care even if

he/she is not conscious or cannot move.

ICT4AWE 2025 - 11th International Conference on Information and Communication Technologies for Ageing Well and e-Health

356

4.5 Empathy and Rapport

The Empathy and Rapport (ER) of a conversational

system uses a psychological approach to simulate a

human interaction. Features like mirroring verbal and

nonverbal Behaviours, providing contextually

appropriate emotional expressions, and maintaining

continuity in dialogue have improved patient

outcomes in medical systems(Salman et al., 2021,

2023).

5 RESULTS

To identify the essential recommendations that ECA

needs to contemplate in a Health Care application for

treatment support in older adults, we made an

empirical study. This holistic approach can be used to

deploy an ECA or evaluate an existing one. We start

by matching each dimension described in Section 4

with the recommendations of Section 3. This process

was made by the author and is listed in Table 5. Some

recommendations did not match any Health Care Key

Dimensions and were classified as General (G) ECA

Recommendations.

After the match, a questionnaire was elaborated in

Portuguese to identify the impact of each

recommendation on each dimension. The

questionnaire was submitted to 2 developers of ECA

to support patients, three UX specialists, a Portuguese

nurse, and experts totaling eight participants.

For each Recommendation, the participants

indicate the ECA recommendation as Unnecessary:

The recommendation does NOT need to be in the

ECA to carry out activities to support the treatment of

diseases for older adults; Desirable: The

recommendation does not need to be in the ECA to

achieve the proposed objectives but can somehow

improve the engaging process; Necessary: The

recommendation is fundamental and Cannot

Evaluate: There is no consensus, or the text of the

recommendation is unclear.

The main finding is that no recommendation was

categorized as Unnecessary (Figure 2). This points

out that, somehow, all the guidelines contribute to

performing effective Health Care. The Desirable set

was the biggest one: 70% of the total. The related

health dimensions were Emergency Detection,

Empathy and Rapport, and all general dimensions.

Special sensors related to Emergency Detection are

set as Desirable as the devices are not easy to use and

install in patients. The participants wrote about the

difficulties related to considering these features as a

second round of testing. An effective Emergency

Detection Device needs to guarantee 100% free error.

Table 5: Health Care Dimensions x ECA

Recommendations x Impact.

Health

Dimension

ECA

Recommendations

Impact

IL, SK, R R204, R208, R209 and

R210

Essential

ED R201, R202, R203,

R205, R206 and R305

Desirable

ER R103 and R401

ssential

R301 and R302

ecessar

R303 Desirable

General R101, R102, R207,

R304, R306, R402,

R403, R404, R405,

R406, R407

Desirable

Another recommendation set as Desirable was

R303, Contemplate the avatar's Behavioural reaction.

The respondents complain about the complexity of

non-verbal feedback. This recommendation is related

to the other General Recommendations, including

R402, which allows adjustment of personality,

conversation topics, or vocabulary according to the

user, and R403, which allows the configuration of

different faces, voices, gestures, and embedding.

Although the configuration set recommendations

are important to an ECA, the participants evaluate

them as desirable. This fact needs to be revisited in

future works to validate them.

There were two Necessary recommendations:

R301, contemplate affectivity, and R302, Allow

Sociability. They are concerned with the dialog

between the ECA and the user. The Agent needs to

incorporate empathic and motivational dialogues and

enhance user adherence. To confirm this, empathic

dialogue frameworks based on doctor-patient

communication inspire design principles that make

ECAs more relatable and supportive (Salman et al.,

2021). The R401 is mainly about a proactive feature

of the ECA.

Related to Interactive Learning, disease-Specific

Knowledge and Reinforcement were identified as 4

Essential Recommendations. The R204

Recommendation is related to storing encrypted

information and guaranteeing the ethical aspects of

the health field. Recommendation R208 is as

important as the previous one as it provides secure

and correct information to the patient. There are no

cues to errors when health treatments involve

recommendation R209 to provide learning from the

doctor team. The learning phase must be carefully

Recommendations of Embodied Conversational Agents to Healthcare Applications

357

planned to prevent any misunderstanding in the

dialog. R210 is concerned about the continuity of the

dialogue. This feature is necessary to prevent the user

from repeating or forgetting information.

Two essential recommendations related to

empathy and rapport were identified: R103, which is

related to the immediate use of the agents, and R401,

which allows the user to control and adjust

interventions and prevent proactiveness, for example.

6 CONCLUSION

The Recommendations encompass various aspects,

including user interface design, agent behavior, and

customization features, all aimed at improving user

experience and engagement. The findings underscore

the importance of incorporating social skills and

nonverbal communication into ECAs, which are

essential for fostering empathetic interactions and

enhancing the overall effectiveness of these systems

in healthcare settings.

Moreover, the empirical study conducted to assess

the impact of ECA recommendations on healthcare

dimensions highlights the necessity of aligning

technological capabilities with user needs. The results

indicate a set of essential recommendations

contribute to effective healthcare delivery, with a

significant proportion categorized as desirable

(Figure 2). This suggests that while certain features

may not be immediately essential, they can enhance

user engagement and satisfaction, ultimately leading

to better health outcomes.

The emphasis on proactive features and the ability

to adjust interventions further illustrates the need for

ECA to be adaptable to individual user preferences

and requirements.

In conclusion, developing and implementing

ECAs in healthcare represents a promising avenue for

improving patient support and engagement. By

adhering to the identified recommendations and

focusing on user-centered design, healthcare

providers can leverage the capabilities of ECAs to

create more effective and empathetic interactions.

Future research should continue to explore the

nuances of ECA.

7 FUTURE WORK

The SLR conducted in this study reveals a

comprehensive set of recommendations for the design

and functionality of ECAs tailored for healthcare

applications. The use of such methodology can be

complemented with a LLM search to complete the

study. Relating the ECA recommendations to health

apps mostly involved developers and UX specialists.

Figure 2: ECA Recommendations (by the author).

It is necessary to revisit these findings with doctors,

nurses, and patients to validate all finds registered.

This is the next step of the work and will be made in

Portugal and Brazil with older adults.

To complete the study, it is necessary to

investigate suggestions made by the respondents in

the questionnaire related to section five. Some of the

• Agents should be ready to use

immediately (R103)

•Store encrypted information (R204)

•Provide secure and correct information (R208)

•Provide learning (R209)

•Provide continuity in dialogue (R210)

•Allows control and adjustments in

interventions (R401)

Essencial

• Contemplate affectivity (R301)

• Allow Sociability (R302)

Necessary

• Simple interface, no learning required (R101)

• Indicate the Agent's entry phrase (R102)

•Recognize embedded signals (Use of sensors)

(R201)

•Recognize the environment of use (R202)

•Recognize the user by speech (R203)

•Track the user(R205)

•Locate the sound source(R206)

• Introduce advanced features in Agent

animation (R207)

•Contemplate the avatar's Behavioural reaction

(R303)

• Assist the user as little as necessary (R304)

• React autonomously to user requests (R306)

• Contemplate adjustment of personality,

conversation topics, or vocabulary according to

the user (R402)

• Contemplate to configure different faces,

voices, gestures, and embedding (R403)

• Contemplate to add new Agent models

easily (R404)

•Contemplate interaction scenarios in a

dedicated syntax (R405)

• Personalize the pronunciation or personality

of the Agent (R406)

• Set up response times, speed and speaking

style, user name, and reminders (R407)

Desirable

ICT4AWE 2025 - 11th International Conference on Information and Communication Technologies for Ageing Well and e-Health

358

participants wrote about the recommendations and

the health aspects. Some related to the agent of ECA

itself and some to the health issues. At this stage, the

work is intended to serve as a basis for identifying the

characteristics that the agent must have to guarantee

user engagement with the ECA.

REFERENCES

Barros, M. I. T. (2021). Operacionalização de Técnicas de

Mudança Comportamental em Agentes

Conversacionais [Master]. Universidade de Lisboa.

Bickmore, T. W., & Cassell, J. (2005). Social Dialogue with

Embodied Conversational Agents. 23–54.

https://doi.org/10.1007/1-4020-3933-6_2

Egede, J., Trigo, M. J. G., Hazzard, A., Porcheron, M.,

Bodiaj, E., Fischer, J. E., Greenhalgh, C., & Valstar, M.

(2021). Designing an Adaptive Embodied

Conversational Agent for Health Literacy. Proceedings

of the 21th ACM International Conference on

Intelligent Virtual Agents, 112–119.

https://doi.org/10.1145/3472306.3478350

Gross, C., Schachner, T., Hasl, A., Kohlbrenner, D.,

Clarenbach, C. F., Wangenheim, F. V., & Kowatsch, T.

(2021). Personalization of conversational agent-patient

interaction styles for chronic disease management: Two

consecutive cross-sectional questionnaire studies.

Journal of Medical Internet Research, 23(5).

https://doi.org/10.2196/26643

Guerino, G. C., & Valentim, N. M. C. (2020). Usability and

User eXperience Evaluation of Conversational

Systems: A Systematic Mapping Study. ACM

International Conference Proceeding Series, 427–436.

https://doi.org/10.1145/3422392.3422421

Krutter, S., Schuessler, N., Kutschar, P., Šabić, E.,

Dellinger, J., Klausner, T., Nestler, N., Beasley, M., .

(2022). Piloting of the virtual telecare technology

Addison Care’ to promote self-management in persons

with chronic diseases in a community setting: protocol

for a mixed-methods user experience, user engagement

and usability pilot study. BMJ Open, 12(9).

https://doi.org/10.1136/bmjopen-2022-062159

May, R., & Denecke, K. (2024). Conversational Agents in

Healthcare: A Variability Perspective. ACM

International Conference Proceeding Series, 123–128.

https://doi.org/10.1145/3634713.3634717

Potdevin, D., Clavel, C., & Sabouret, N. (2021). Virtual

intimacy in human-embodied conversational agent

interactions: the influence of multimodality on its

perception. Journal on Multimodal User Interfaces,

15(1), 25–43. https://doi.org/10.1007/S12193-020-

00337-9/FIGURES/6

Pradhan, A., Lazar, A., & Findlater, L. (2020). Use of

intelligent voice assistants by older adults with low

technology use. ACM Transactions on Computer-

Human Interaction, 27(4).

https://doi.org/10.1145/3373759

Riccardi, G. (2014). Towards Healthcare Personal Agents.

RFMIR 2014 - Proceedings of the 2014 ACM

Roadmapping the Future of Multimodal Interaction

Research Including Business Opportunities and

Challenges, Co-Located with ICMI 2014, 53–56.

https://doi.org/10.1145/2666253

Ruttkay, Z., Dormann, C., & Noot, H. (2004). Embodied

Conversational Agents on a Common Ground (pp. 27–

66). Springer, Dordrecht. https://doi.org/10.1007/1-

4020-2730-3_2

Salman, S., Richards, D., & Caldwell, P. (2021). Analysis

of Empathic Dialogue in Actual Doctor-Patient Calls

and Implications for Design of Embodied

Conversational Agents. Italian Journal of

Computational Linguistics, 7(1 | 2), 91–112.

https://doi.org/10.4000/ijcol.862

Salman, S., Richards, D., & Dras, M. (2023). Identifying

Which Relational Cues Users Find Helpful to Allow

Tailoring of e-Coach Dialogues. Multimodal

Technologies and Interaction, 7(10).

https://doi.org/10.3390/mti7100093

Sandoval, D. C., & Favela, J. (2017). Co-designing

ambient-assisted interventions using digital

interlocutors for people with dementia. Proceedings of

the 2017 ACM International Joint Conference on

Pervasive and Ubiquitous Computing and Proceedings

of the 2017 ACM International Symposium on

Wearable Computers, 813–821.

https://doi.org/10.1145/3123024.3125615

Striegl, J., Gollasch, D., Loitsch, C., & Weber, G. (2021a).

Designing VUIs for Social Assistance Robots for

People with Dementia. ACM International Conference

Proceeding Series, 145–155.

https://doi.org/10.1145/3473856.3473887

Striegl, J., Gollasch, D., Loitsch, C., & Weber, G. (2021b).

Designing VUIs for Social Assistance Robots for

People with Dementia. ACM International Conference

Proceeding Series, 145–155.

https://doi.org/10.1145/3473856.3473887

Sunghoon, K., Parasuraman, G. M., & Jaunbuccus, S.

(2019). Elderly Care Assistant: A Discreet Monitoring

Tool. Lecture Notes in Electrical Engineering, 561,

287–301. https://doi.org/10.1007/978-3-030-18240-

3_27

ter Stal, S., Kramer, L. L., Tabak, M., op den Akker, H., &

Hermens, H. (2020). Design Features of Embodied

Conversational Agents in eHealth: a Literature Review.

International Journal of Human Computer Studies.

https://doi.org/10.1016/J.IJHCS.2020.102409

Wargnier, P., Benveniste, S., Jouvelot, P., & Rigaud, A. S.

(2018). Usability assessment of interaction

management support in LOUISE, an ECA-based user

interface for elders with cognitive impairment.

Technology and Disability, 30(3), 105–126.

https://doi.org/10.3233/TAD-180189

Yang, X., & Aurisicchio, M. (2021). Designing

Conversational Agents: A Self-Determination Theory

Approach. Proceedings of the 2021 CHI Conference on

Human Factors in Computing Systems, 1–16.

https://doi.org/10.1145/3411764.3445445

Recommendations of Embodied Conversational Agents to Healthcare Applications

359