COACH BOT

Modular e-Course with Virtual Coach Tool Support

Ilaria Mascitti, Mikail Feituri, Federica Funghi, Susanna Correnti and Luca Angelo Galassi

Consorzio For.Com. – Formazione per la Comunicazione, Via Virginio Orsini 17/a, Rome, Italy

Keywords: Intelligent Agent, E-learning, Health care training.

Abstract: The COACH BOT project aims at designing and testing an innovative e-learning methodology for adult

education that combines Conversational Agent Technology (chatbot) with an ad hoc designed modular

learning path. The pilot e-course addresses a target group of home health care professionals (e.g. home

caretakers, nurses, etc). The project’s innovation consists of the development of a collaborative e-learning

environment featuring a “chatbot” or “Virtual Coach” who interacts with users through a human-like

interface. The “Virtual Coach” acts as a teacher, coach and tutor, who supports learners “individually”

during the modular e-course by providing in-depth information, assessment, case studies, technical and

methodological support. The e-course curriculum is based on a personalised approach allowing learners,

with help from the COACH BOT, to customize their own training path and benefit from a suitable training

path that is relevant to their profession and based on their own specific needs, knowledge and skill

requirements.

1 INTRODUCTION

The “COACH BOT” project is financed by the

European Commission within the framework of the

Lifelong Learning Programme. The two year project

started in November 2008 and will end in October

2010. The project aims at designing a new model of

adult distance education, addressing home health

care service professionals, based on an e-learning

methodology that combines the Conversational

Agent Technology (chat-bot) with an ad hoc

designed modular learning path.

The multi-actor partnership consists of seven

organizations, six from the European Commission

and one from Switzerland, from six different

countries, namely:

1. Consorzio FOR.COM. Formazione per la

Comunicazione (Italy)

2. Aarhus Social and Health Care College

(Denmark)

3. Gruppo Pragma S.r.l. (Italy)

4. Romanian Society for Lifelong Learning

(Romania)

5. Secondary school of nursing Ljubljana

(Slovenia)

6. Norton Radstock College (United Kingdom)

7. Seed Association (Switzerland)

This article will explore the project’s main

objectives, potential results and e-learning

methodology that will provide significant training

benefits to home heath care professionals.

2 THE PROJECT OBJECTIVES

Today, most professionals do not have extra time to

study or enroll in training courses, therefore on-line

vocational training courses are an ideal solution. It is

essential for an e-learning course to be able to tailor

to learners’ personal and professional training needs

offering valuable content to enhance their work

experience. Online training also needs to provide

learners with assistance, feedback and

encouragement in order to customise their learning

experience according to their needs and alleviate any

feelings of detachment during the course. An online

teacher/tutor can provide a sense of “online

presence” which is a critical element to enhance

distance learning that can also have a direct effect on

establishing interpersonal relationships and trust

during online communication (Craig S.D. et al.,

2000).

115

Mascitti I., Feituri M., Funghi F., Correnti S. and Angelo Galassi L. (2010).

COACH BOT - Modular e-Course with Virtual Coach Tool Support.

In Proceedings of the 2nd International Conference on Agents and Artiﬁcial Intelligence - Agents, pages 115-120

DOI: 10.5220/0002589901150120

 SciTePress

In response to these aspects, the COACH BOT

project will design and test an online training course

for home health care professionals using the chat-bot

methodology (a virtual tutor) that will allow trainees

to personalize their own training paths and access

different learning contents according to their own

needs, knowledge and skills.

The health care sector is a complicated system

that demands extensive resources and consists of a

set of integrated services and inter-collaborative

health teams that require a broad skills base.

Despite growing training demands in the field,

current training systems are too slow and inefficient

to cope with new changes. The “COACH BOT”

chatbot and e-course addresses these issues by

providing health care professionals with the

opportunity to renew and improve their skills

through a flexible distance learning approach.

3 THE E-LEARNING PLATFORM

The E-Learning platform is based on the open

source LMS Claroline that allows teachers to create

and administer course websites through a WEB

browser. This LMS is worldwide used and the vast

community guarantees to solve any problems the

platform administrators or users might have.

The project’s technological team selected the

LMS Claroline, among other possibilities e.g.

Moodle, for its very clean and comprehensible

source code, allowing developers to easily

implement new modules to create highly

personalized learning paths and embed the virtual

agent into the LMS.

4 PEDAGOGICAL AGENTS

The COACH BOT project methodology is based on

Pedagogical Agents that are autonomous software

systems, realized with Artificial Intelligence

methods that can operate in the training environment

as tutors who adaptively assist users in performing

training tasks (Craig S.D. et al., 2000). They can

intervene in case of suboptimal performance,

demonstrate skills, provide explanations, answer

questions, and play the role of team members in

multi-person tasks. Agents can be represented either

as abstract pointers, disembodied hands, or as virtual

humans with articulated bodies. Experiments have

shown that ECAs can increase the motivation of a

student interacting with the system. Jonhson

(Jonhson et al., 2000) showed that a display of

involvement by an ECA motivates a student in doing

his or her learning task. Pedagogical Agents are

therefore virtual facilitators gifted with great

reactive and interpretative skills promoting learning

that is based on a knowledge transfer and the student

is followed “step by step” by his/her own

agent/trainer. This new learning methodology is

highly experiential which allows real time testing

and interaction.

Intelligent Agents make the content delivery

highly interactive and personalized, articulating

along individual paths following the learners’

natural inclinations and respecting the different

times of knowledge acquisition from individual to

individual (Monova-Zheleva M. et al., 2008).Virtual

teachers may use methods of Artificial Intelligence

for evaluating the student’s performance and

reactions, and mainly for adapting teaching

according to specific needs and particular

environments. They can show the student how to

perform a rather complex task; taking advantage of

non-verbal behaviours, in order to capture the

student’s attention during the crucial moments of

learning. Thanks to anthropomorphic features,

virtual teachers make the interaction between

student and learning system more involved and

effective, allowing the acquisition of new contents to

be improved and considerably implementing the

learning level of the student, who learns with an

active experiential participation. The methodology

of Intelligent Agents as virtual professors/facilitators

interacting with the student activates a strong

emotional response on one side, and a real know

how capability on the other. In the first case, it is

important to underline the fact that training has a

major impact if the person involved in the process is

stimulated, not only by the cognitive-rational

component, but also through the emotional

component.

5 THE CONVERSATIONAL

AGENT TECHNOLOGY

Today, many AI researchers have created domain-

oriented chatbots, able to understand a specific

knowledge domain with realistic, multi-purpose

initiatives and human-like behaviour. The main

challenging function of the agent is natural language

analysis where the COACH BOT must "understand"

what the user wants to know by analyzing the input

phrase. In order to create the “brain” and personality

ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence

116

of the virtual agent, the project team decided to use

AIML (Artificial Intelligence Markup Language),

which is a dialect of the popular XML. AIML files

consist of simple stimulus-response modules called

categories. Each <category> contains a <pattern>, or

"stimulus," and a <template>, or "response." AIML

software stores the stimulus-response categories in a

tree managed by an object called the Graphmaster.

When a bot client inputs text as a stimulus, the

Graphmaster searches the categories for a matching

<pattern>, along with any associated context, and

then outputs the associated <template> as a

response. These categories can be structured to

produce more complex humanlike responses with

the use of a very few markup tags. AIML bots make

extensive use of the multi-purpose recursive <srai>

tag, as well as two AIML context tags, <that> and

<topic>. Conditional branching in AIML is

implemented with the <condition> tag (Wallace R.,

2003). Bot personalities are created and shaped

through a cyclical process of supervised learning

called Targeting. Targeting is a cycle incorporating

client, bot, and botmaster, wherein client inputs that

find no complete match to the categories are logged

by the bot and delivered as Targets to the botmaster,

who then creates suitable responses, starting with the

most common queries. The Targeting cycle produces

a progressively more refined bot personality. The art

of AIML writing is most apparent in creating default

categories, which provide noncommittal replies to a

wide range of inputs.

To provide academic and technical assistance

and emotional support to users, the COACH BOT is

present in the e-learning interface and possess

human features. The learning process occurs inside

the Learning Management System (LMS) that offers

the didactic framework in which the user can follow

the course and interact with other LMS didactic

functionalities. Since the upper banner is almost

always constant in all e-learning applications, it is

the best place to put the interface. In this position the

COACH BOT is always "near" the user and can

always interact.

6 THE WORKFLOW OF THE

COURSE

The virtual agent’s main semantic structure is

described by the following graph:

All these sections are embedded in the same

virtual assistant, so the user does not see this

segmentation in the dialogue software. Each area is

Figure 1: Virtual agent’s conceptual map.

then structured into sub areas that relate to the

COACH BOT’s different tasks and the course

contents it will cover.

6.1 Guidance Interview

The guidance interview can be considered the first

contact between the conversational agent and the

student. The student first becomes familiar with the

virtual agent who starts a friendly conversation with

him/her in order to create a sort of empathetic

relationship. The ultimate goal of this interaction is

for the agent to define a professional profile and

consequently a learning path that best fits the

student. The conversational agent also asks more

general questions concerning the student’s

expectations and professional ambitions. Here the

agent behaves more like a mentor who tries to

understand the user’s emotional aspects. The user’s

information is stored for further statistics and to

enrich and personalize future conversations between

the student and the agent.

Figure 2: Student profiling.

6.2 Start up Quizzes

Following the guidance interview that defines the

student’s professional profile and consequently

learning path, the student is allowed to access the

exercise area on the E-learning platform. The

system presents the user with as many quizzes as the

number of modules foreseen for his / her learning

path. These exercises are considered start up quizzes

that test the initial knowledge of the student on

topics presented in the course. There are a total of

three learning objects (i.e. audio/video lessons) for

Learning path 1

Learning path 2

COACH BOT - Modular e-Course with Virtual Coach Tool Support

117

each module and for each learning object the system

asks the student three questions. The student

therefore must answer 9 questions per module. If the

student answers all 3 questions concerning a

particular topic correctly the corresponding learning

object is optional, otherwise it is mandatory for the

student to continue with the subsequent learning

objects.

Based on the results of the start up quiz, a

customized learning path is then created considering

the previous knowledge of the student. The learning

path adapts a constructivist approach which

indicates when course contents are mandatory for

study concerning topics that scored low and are not

well known by the user and contents that are

optional for topics that the user shows sufficient

knowledge. The virtual agent will present the

quizzes, explain their meaning and comment on the

user’s results. From a student’s viewpoint, the

virtual agent behaves like a teacher who tests his/her

knowledge and assigns relevant learning objects to

be studied.

6.3 Customized Learning Path

The student can find their customized list of course

modules in the learning path section of the LMS.

Unlike the standard version of the LMS, this area

highlights the list of relevant modules that

correspond to the user’s start up quiz results. Within

each module, learning objects are highlighted in two

different colours indicating if they are compulsory or

optional. The standard tracking system has been

enhanced in order to allow the virtual agent to

memorize, according to each user, what learning

objects are compulsory and the number of times and

the duration he or she studies a particular learning

object. This kind of information is useful for the

virtual agent to track the student’s progress and

when delays occur, the agent can invite him or her to

study compulsory learning objects. In this situation,

the virtual agent behaves like a tutor.

6.4 Help Desk

The goal of the Help Desk area is to provide users

with extensive help concerning LMS functionalities

and tools. By using only natural language, the

COACH BOT can explain the meaning of the

different functions in a way that is easy for the

student to understand when asking for help. This

user-friendliness allows learners to focus more on

studying course contents and permits users who are

not computer or LMS experts to use the system

effectively. Each tool featured in the LMS is

explained in detail to cover both the generic and

specific (interface related) questions that a user can

do. The forum area, for example, can be "exploded"

to cover these questions. The virtual agent has to

understand and answer questions related to the

"generic" forum usage (e.g. "Who can participate?",

etc.). In addition to all the questions that can arise

from the specific LMS user interface such as "What

is a reply?" and so on. The virtual agent behaves in

this case as a technical tutor. It is important to

underline that the Coach Bot is not a normal FAQ

but rather an agent who uses a natural language

interface. This means that the user can write and ask

certain question in different ways. Conceptually it is

possible to “explode” this semantic area by the

following graph:

Figure 3: Virtual agent’s conceptual map: help desk.

All the lines under the "Forums" are "exploded"

such as the "What is" line and all the areas of the

Coach Bot are "exploded" like the "Help Desk -

Forums – “What is" one. This natural language

capture technique applies to the Coach Bot’s entire

structure and not only to the Help Desk semantic

area.

6.5 Suggestions

The virtual agent is able to help the user concerning

specific course topics, which is referred to as the

Suggestions area and is illustrated in the figure

below.

The lines Content 1, Content 2, etc. are related to

keywords specific to the course that the LMS offers.

The virtual agent’s aim in this area is to act as a

teacher and not, like in other areas, as a tutor. Each

content keyword is related to a set of questions that

in turn are related to particular aspects of the

ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence

118

Figure 4: Virtual agent’s conceptual map: suggestions.

content. If, for example, Content 1 is "Catheter", the

user can ask for the definition, what is its usage, pros

and cons, warnings, etc. Similar to the Help Desk

area, many different syntaxes that a user can supply

are collected. For example the answer related to the

definition can be activated by the sentences

illustrated in the figure. This virtual agent’s area can

be seen as an interactive glossary or a "quick answer

teacher" but does not have to substitute the main

learning sources that are, and remain, the learning

objects. When the user wants more information, the

virtual agent will refer him/her to the learning object

that discusses the requested content.

6.6 Ongoing Presence

In order to keep learners motivated, the COACH

BOT interacts with each learner throughout the

course in different ways by providing ongoing

verbal feedback during the learning process.

Learners can receive positive feedback when they

are proceeding well or notified when they are

studying too slowly or when they do not study

specific fundamental lessons.

• The project’s didactic goal for the course is to

ensure that each student completes at least 80 % of

the compulsory learning material. This in turn

depends on the performance results of the student’s

start up quizzes. Two thresholds have been fixed in

order to trigger virtual assistant interventions: 40 %

of the learning material completed by the student at

mid course and 60 % of the learning material

completed at three quarters of the course.

6.7 Final Assessment

The aim of the final assessment is to determine what

the student has learned from the e-course with the

help of the virtual agent. Ideally students should

obtain better results than the startup quiz, thanks to

the e-course curriculum. Each learning path will

feature the same number of final quizzes as start up

quizzes and consequently modules. A singular final

quiz will be presented in the exercise section on the

E-learning platform, only if the student has

answered all the different startup quiz questions

correctly at the beginning of the course and if he/she

has studied the corresponding module ‘enough’

according to the tracking results.

Technically, the concept of studying ‘enough’

signifies that a student has accessed all compulsory

learning objects within a module within the expected

time predefined by the didactic development team.

When a student answers some of the startup quiz

questions incorrectly, he/she should ideally study the

respective module and repeat the same process for

the next modules. The virtual agent suggests this

same study tip concept throughout the course. In

case the student answers all the questions in the

different startup quizzes correctly, he/she will have

no questions to answer in the final quizzes.

6.8 Case Study

Case studies aim at enhancing the student’s

knowledge by showing a practical case study and

then testing the student’s ability to implement the

acquired skills. The COACH BOT will present three

case studies at the end of the course through the

Machinima technique. Machinima is the use of real-

time three-dimensional (3-D) graphics rendering

engines to generate computer animation. In

particular, Linden lab’s Second Life will be used to

create these highly engaging 3-D animations.

7 THE EXPERIMENTATION

The experimentation methodology consists of two

test groups that will follow the same e-course:

• the experimental group: trainees who take the

e-course with the support of the virtual COACH

BOT;

• the control group: trainees who take the e-

course without the support of the virtual COACH

BOT.

These two different groups will offer meaningful

data for analyzing the effectiveness of the virtual

coaching facility. The two different questionnaires

will be given to both the control and

experimentation groups:

- “Expectations questionnaire” delivered at the

beginning of the e-course.

- “Customer satisfaction questionnaire” delivered

at the end of the e-course.

COACH BOT - Modular e-Course with Virtual Coach Tool Support

119

The questionnaire results will be compared in

order to assess different self-perceptions, the level of

satisfaction of the e-course contents, structure and

virtual support and to compare the achieved results

with users’ initial expectations. Moreover, a follow

up session will be held two months after the end of

the e-course to assess the pilot application usability

and transferability through the arrangement of three

focus groups, namely:

1° experimental group users

2° control group users

3° mixed focus group

This method will provide qualitative data from

different points of view incorporating all three types

of group sessions.

8 CONCLUSIONS

The real innovation of the COACH BOT project is

the embedding of a pedagogical agent in an open

source and SCORM compliant learning management

system. The virtual agent becomes a mentor, a

coach, a teacher, a didactic or technical tutor

depending on the student’s type of learning activity.

This project provides the premises to provide the

distance learning community with a multiple

purpose pedagogical agent that is easy to integrate in

any open source LMS like Moodle, ILIAS, Dokeos,

Atutor, etc. This is true also because the pedagogical

agent has been implemented as an independent

module even if the student is perceiving it as if it

was completely embedded in the E-Learning

platform.

Considering these aspects, the COACH BOT

project is therefore in line with the priorities of the

European Lifelong Learning Programme who

financed the project based on its implementation of a

new methodology in the E-learning field.

From a technological viewpoint, different

improvements can be made to the project as a follow

up in the future, such as:

• The guidance interview that creates a profile

(cluster) of the students can be further

developed by means of adaptive neural

networks that are very appropriate for

classification purposes.

• The start-up quizzes that test student’s prior

knowledge can be processed by means of Fuzzy

logic inference engines to guarantee more

flexibility in understanding student’s didactic

needs.

• Classes could be conducted in virtual worlds by

a pedagogical intelligent agent that would

always be present and available for students

independently of when the real teacher was

available or online.

• Additional conversational agent technologies

e.g. Autotutor with The DAN (Dialog Advanced

Network) and the LSA (Latent Semantic

Analysis) systems can be experimented.

REFERENCES

Craig, S.D., Gholson, B., Driscoll, D., 2000. Animated

Pedagogical Agents in Multimedia Educational

Environments: Effects of Agent Properties, Picture

Features, and Redundancy. Journal of Educational

Psychology, Vol. 94, pp. 95-102.

Graesser, A.C., Lu, S., Jackson, G.T., Mitchell, H.,

Ventura, M., Olney, A., & Louwerse, M.M. (2004).

AutoTutor: A tutor with dialogue in natural language.

Behavioral Research Methods, Instruments, and

Computers, 36, 180-193.

Johnson, W. L., Lewis, W., Rickel, J., 2000. Animated

Pedagogical Agents: Face-to-Face Interaction in

Interactive Learning Environments. International

Journal of Artificial Intelligence in Education, 11, pp.

47-78.

Monova-Zheleva, M., Zhelev, Y., Mascitti, I., 2008. E-

learning, e-practising and e-tutoring: an integrated

approach. Methodologies and Tools of the Modern (e-)

Learning, pp. 84-90.

Nijholt, A., 2004. Where computers disappear, virtual

humans appear. Computers & Graphics. N.28, pp.

467- 476.

Wallace, R., 2003. The Elements of AIML Style. ALICE A.

I. Foundation.

ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence

120