Collaborative Agents in Adaptative VLEs: Towards an Interface

Agent for Interactivity and Decision-making Improvement

Karize Viecelli

, Aluizio Haendchen Filho

, Hércules Antonio do Prado

, Edilson Ferneda

Jeferson Miguel Thalheimer

and Anita Maria da Rocha Fernandes

Laboratory of Applied Intelligence, University of the Itajaí Valley (UNIVALI), Rua Uruguay, 458, Itajaí, Brazil

Catholic University of Brasilia (UCB), Brasilia, Brazil

anita.fernandes@univali.br

Keywords: Interface Agent, Adaptative e-Learning, Virtual Learning Environments, Microservices.

Abstract: This paper presents an Interface Agent (IAg) in the context of collaborative software agents aiming at

improving the interaction, interactivity and decision-making processes in Virtual Learning Environments

(VLE). Working collaboratively in a multi-agent system, IAg receives notifications about situations that

require interaction with students to assist and motivate them in the processes of navigation and use of VLE.

In order to assist decision-making processes, it provides dashboards that enable the human tutor and VLE

coordinators to make real-time decisions about non-normal situations. In addition, it monitors the actions of

students seeking for clarifying doubts, utilizing a knowledge based on past situations. With this approach it

is expected to enable a more attractive environment to students by reducing feelings of demotivation and

isolation, and helping to reduce student dropout.

1 INTRODUCTION

The Distance Learning (DL) market is the fastest

growing modality in Brazil, and already represents

¼ of enrolments, according to the Brazilian

Association of University Education Maintainers

(ABMES, 2019). This growth tends to continue.

ABMES predicts that distance education will

surpass presential university education by 2023.

However, there are major challenges.

In the 10

Distance Learning Census, the

Brazilian Association of Distance Learning (ABED,

2018), pointed out a high dropout rate and a low

graduation completion rate. According to ABMES

(2019), the indicators of the completion rate of

distance learning students in 2016 was 35% and the

dropout rate reached 62%, tending to increase. For

Open University (UK), this is a globalized scenario,

as internationally graduation rates would be close to

10% and tending to decline (Woodley, Simpson,

2014).

It was observed that the main reasons for high

dropout rates are the feeling of demotivation and

isolation of students (ABED, 2018). From this

perspective, the suggested actions to reduce these

rates are similar for both Open University

researchers and ABED and ABMES. They are

unanimous in recommending the development of

proactive motivational support from institutions for

student retention.

In contrast to this idea, it is clear that the vast

majority of institutions are reactive, that is, they wait

for students to contact them for help (Woodley,

Simpson, 2014).

In this sense, research seeks to approach the

issue of isolation and motivate the student by means

of interaction and interactivity actions in the context

environment / tutor / student. For achieving quality

in student teaching and learning, it is clearly

necessary to monitor the Virtual Learning

Environment (VLE). However, one cannot ignore

that, by supporting large numbers of students and

proactive supportive actions, mentoring quality can

be compromised.

In this context, Simbine et al. (2018) implement

a model of visualization of student interactions

based on their learning trajectory. By monitoring the

way the student interacts in the VLE, the model

generates graphical information with their

interaction characteristics.

Viecelli, K., Filho, A., Antonio do Prado, H., Ferneda, E., Thalheimer, J. and Fernandes, A.

Collaborative Agents in Adaptative VLEs: Towards an Interface Agent for Interactivity and Decision-making Improvement.

DOI: 10.5220/0009415006910702

In Proceedings of the 22nd International Conference on Enterprise Information Systems (ICEIS 2020) - Volume 1, pages 691-702

ISBN: 978-989-758-423-7

691

The environment adaptation to the student profile

is another important issue. Vaidya and Sajja (2016)

proposed an adaptive learning agent platform that

produces and assesses learning contents. It applies

analytical reasoning about content before and after

presenting it to the student in VLE. Student

interaction monitoring data is used for planning and

organizing the content of the learning environment.

This paper presents an intelligent software agent

for making a VLE more interactive, intelligent,

enjoyable, and student-adaptive environments. This

agent, herein called Interface Agent (IAg), is

responsible for interaction and interactivity in the

environment. IAg operates in a collaborative

organization of agents, playing its role proactively

and autonomously to take action and reduce the

work of human actors. The goal is to develop

proactive motivational support resources to reduce

the sense of isolation and abandonment while

making the student a protagonist of their learning.

The paper is structured as follows: Section 2

presents the background, describing the agent

technologies, data structure, and the Artificial

Intelligence technique adopted for modelling IAg,

Section 3 discusses some related works found in the

literature, Section 4 explains the methodological

approach, Section 5 discusses some preliminary

results, and Section 6 concludes with some remarks

and future work.

2 BACKGROUND

2.1 Microservice-oriented Multi-agent

System

Despite the drawbacks reported about Service-

Oriented Architecture (SOA), it remains the best

option available for system integration and leverage

of legacy systems (Alencar et al., 2013) due to its

inherent ability to compose applications, processes,

and assemble new functionalities from existing

services. Inside the industry segment, the SOA

principles have evolved in the form of

microservices, an architectural paradigm that is

based on fine-grained and independent software

components that interact to build highly scalable

distributed systems (Dragoni et al., 2017).

Although not so spread in the industry realm

(Collier et al., 2015), several principles of MAS, as

decentralization, distributed environments, amongst

others, have been observed in the microservices

model (Burkhardt, 2018). Several authors have

studied the applications of microservices paradigm

as a framework for building modern MAS, in an

attempt to shorten this gap between industry and

academic efforts (Burkhardt, 2018; Collier et al.,

2015; Higashino et al., 2018). To assure that

microservices can meet these expectations, multiple

specifications and standards have been proposed and

created, and middleware products are becoming

more robust (Alencar et al., 2013). MIDAS

(Haendchen Filho, 2017) is a platform that relies on

microservices as the basis for the development of

distributed MAS. Its architecture is composed by a

front-end server (MIDAS Server) and one or more

agent containers (MIDAS Container), as shown in

Figure 1.

Figure 1: MIDAS Generic Architecture.

MIDAS Server is responsible for the platform

integration rules, synchronizing the containers and

interoperating with external applications. It contains

three main interfaces: (i) an HTTP interface for

intra-platform communication between MIDAS

Server and the MIDAS Containers; (ii) a REST

interface that allows communication with external

applications; and (iii) Web interface for human

management and configuration.

Each MIDAS Container is a lightweight

container that houses software agents and/or

microservices. It is capable of cataloguing the

interface of its own services, and to break the

conversation between its own agents and foreign

agents. The containers may register themselves on a

MIDAS Server, exposing their services and agents,

allowing for distributed collaboration with other

containers within the same server domain.

Also, the MIDAS Server performs the

integration and discovery on its child containers,

eliminating the complexity of service lookup and

remote requests between containers. The application

agents are instantiated in containers and developed

by extending the abstract class, from which specific

application behaviour can be implemented.

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

692

Placed in the MIDAS Server and also in the

MIDAS Containers are middleware agents: Broker,

Proxy, Catalog, Blackboard and Manager. They

provide infrastructure services, playing in a

collaborative and pro-active way the roles defined

by the reference architecture. The introduction of the

agent concept to play these roles complies with the

current tendency and non-functional requirements

for microservices-oriented architectures: flexibility,

dynamic behaviour, pro-activity, and adaptability.

They completely abstract the standard code required

to implement those characteristics, such as

communication protocols, concurrency control,

lifecycle management, and services discovery and

interoperability, enabling the developer to focus only

in specific characteristics of the application business.

The Broker agent focuses on the architectural

aspects related to the message transport:

send/receive, pack/unpack, and managing

exceptions. It translates agents and services request

in HTTP streams.

The Catalog agent is responsible for the relevant

aspects related to the resources concept of a

resource-oriented model. A resource description is a

machine runnable metadata representation that

makes possible for a human or software program to

locate services and agents within the ecosystem.

The Proxy agent plays the role defined by the

service-oriented model, which focuses on the

architectural aspects related to the messages

processing. It acts as a service provider

representative, being responsible for the dynamic

configuration and creation of instances. Dynamic

configuration focuses on the capacity of redirecting

messages to different providers during runtime,

whenever the Catalog agent updates the resource

model.

The Manager is the most complex agent in the

architecture, playing the roles defined at the

management and policy levels. It involves a set of

tasks that enable the control over the platform, such

as the life cycle management, checking activities,

statistics, QoS (Quality of Services) reporting, and

GUI (Graphical User Interface) wizards.

Finally, the Blackboard takes responsibility for

information exchange in symbolic cognitive MAS.

Its structure follows the basic blackboard pattern: the

knowledge sources represent the agents, the data

structure is visible to all agents, and the controller is

responsible for notifying the agents about the

changes in the environment. When a MIDAS

Container is running in stand-alone mode, it has a

local Blackboard agent that deals with intra-

container communication, and when a Container

connects to a MIDAS Server, each local Blackboard

of each MIDAS Container synchronizes with the

MIDAS Server Blackboard in order to provide

transparent communication within the whole

ecosystem.

2.2 Data WebHouse

The Web allows recording practically all

behavioural actions of the user in a single click

(Kimball and Merz, 2000). It means that one can

capture not only the page accessed but also

navigability information. The recording of all

interactions made by anyone via an application or

web site, is called a clickstream. Activities carried

out by the user such as click capturing, form filling,

and others, create conditions for analysis, profile

identifications, preferences and trends of each

particular user.

Figure 2 shows a very simple example of a

dimensional model for a Data Webhouse (DWH) for

a VLE. As for Data Warehouse, a DWH is based on

an architecture called Dimensional Model.

Dimensional modelling is a discipline that seeks to

model data for the purposes of understandability and

performance.

Figure 2: Simple dimensional DWH example for a VLE.

All dimensional models rely on the concept of

measured facts. The Facts table, represented by the

entity Clickstreams, stores users clicks on the VLE.

The dimensions relate to the entities that serve as

perspectives of analysis in any subject of the model.

In the example, Student, Discipline, Time and Tutor

are the dimensions connected to the fact table.

Dimensions are rich in descriptions. For example,

the Student dimension stores all the student profile

data.

Besides manipulating information and

discovering knowledge, a VLE needs to be prepared

to react immediately to students’ actions in the

environment, eliminating the time between the

Collaborative Agents in Adaptative VLEs: Towards an Interface Agent for Interactivity and Decision-making Improvement

693

occurrence of an event and the execution of an

action (Sassi, 2010). This is called Zero Latency

Enterprise (ZLE). The idea in a ZLE strategy is to

use DWH integrated with other Business

Intelligence tools to deliver real-time, zero-latency

information for much faster decision making.

2.3 Case-Based Reasoning

Case-Based Reasoning (CBR) is an Artificial

Intelligence technique for problem solving and

knowledge acquisition based on the principle that

“similar problems have similar solutions” (Aamodt

and Plaza, 1994). According to Vitorino (2009), the

use of CBR methodology and its application in VLE

is based on a broad cognitive theory that involves

the process of remembering, as a problem-solving

phenomenon, and the process of reusing past

episodes to solve new problems, that corresponds to

a frequent and powerful way of human reasoning.

In CBR approach, knowledge maintenance is

simplified by the ability to learn new information in

the form of cases. Other advantage is the fast

response time and the ability to work in domains that

are not completely known. These features enable its

application in many types of tasks such as diagnostic

systems, help desk systems, evaluation systems,

decision support systems, and project systems

(Kolodner and Leake, 1996).

The basic elements of a CBR system are: (i)

knowledge representation, carried out by means of

concrete experiences; (ii) similarity measure, which

looks for similar situations for the current problem

in a knowledge base; (iii) adaptation, where past

situations not identical to the current problem can be

adapted to find a suitable solution for the new one;

and (iv) learning, which occurs every time a case is

resolved and a new experience is retained and

integrated into the knowledge base.

A conceptual model for the cycle CBR (Figure 3)

was proposed by Aamodt and Plaza (1994). It

encompasses a continuous cycle of reasoning,

consisting of four main tasks: (i) recovering the most

similar case(s) from the case base, in which the goal

is to find a case or a small set of cases in the base

that contains a problem description near to the

current problem or situation; (ii) reusing this case(s)

to solve the problem; (iii) review the proposed

solution in order to transfer it to the present

situation; if necessary, the recovered solution can be

adapted to fully meet the requirements of the present

situation; and (iv) retaining the experience

represented by the current case (or parts of that

experience) for future reuse.

Figure 3: Cycle of Case-Based Reasoning (Aamodt and

Plaza,1994).

2.4 Felder-Silverman Model

Felder and Silverman (1988) developed a theory that

states there is difference in the way students learn:

seeing or hearing; reflecting and acting; reasoning

logically and intuitively; memorizing, visualizing,

drawing analogies and building mathematical

models; steadily or not.

The authors mapped learning styles and created a

questionnaire entitled Index of Learning Styles (ILS)

base on them. The ILS is an instrument that

evaluates seven identified dimensions (Table 1).

Based on this assessment, according to the answers

provided, an index is calculated that establishes the

predominant dimension present in each profile. The

index ranges from 1 to 11 and represent the intensity

of the categories.

Table 1 presents the proposed dimensions:

perception (sensory or intuitive), input (visual or

verbal), organization (inductive or deductive),

processing (active or reflective) and understanding

(sequential or global).

This model has been widely used to classify

profiles (Freitas et al., 2006; Aguiar et al., 2014;

Trevelin et al., 2013). Some of the features identified

by ILS are:

 Active Students. Tend to understand and retain

information if they can turn that knowledge into

action.

 Reflective Students. Prefer to think about

information before acting and tend to enjoy

working alone.

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

694

Table 1: Learning dimensions (Felder and Silverman,

1988).

 Intuitive Students. They like innovations, but

not repetitions; they may be better at

understanding new concepts and tend to be more

innovative and work faster than sensory students.

 Sensory Students. They like to learn facts and

solve problems by established hands-on methods

and don’t like surprises and complications.

 Visual Students. They easily remember what

they see such as movies, photos, diagrams and

demonstrations.

 Verbal Students. Acquire and assimilate

knowledge based on written and spoken

explanations.

 Sequential Students. Tend to gain knowledge in

linear, logically interconnected steps, and follow

step-by-step ways to find solutions.

3 RELATED WORKS

Simbine et al. (2018) proposed a model for student

interaction visualization on the basis of their

learning trajectory, along with an interactive

visualization system of learning trajectories in VLE.

Thus, it generated a model of data collection,

visualization and analysis in the form of graphs,

according to the characteristics of student

interaction. By analysing these data, it was possible

to verify the order of access of students’ interaction

with existing content, which can be used to improve

the organization of educational content in the VLE.

Referring to the issue of interaction with students

in VLE, Maciel et al. (2014) propose a virtual

assistant integrated with the Moodle environment in

order to offer daily support to the academic activities

of distance learning students. This wizard exposes

the content orally through a visual avatar, making it

more interesting for students. Besides, it allows the

human tutor to contact the student through this

avatar, sending messages in BackOffice.

Regarding the adaptability of a VLE, Vaidya and

Sajja (2017) proposed an agent-based system for

collaborative learning environment in an educational

habitat. The approach provides an agent that not

only offers the student learning facilities, but also

calibrates content and learning outcomes. Dorça

(2012) presents a probabilistic approach using

reinforcement learning, in which a dynamic,

interactive and gradually updated student model is

implemented through a stochastic process. Model

updating occurs based on information about student

performance within the learning environment. This

approach adopts the ILS.

Zapparolli et al. (2017) develop a tool called

FAG that uses Business Intelligence and Learning

Analytics techniques to assist in knowledge

management and decision support in a VLE. The

tool provides analytical and consolidated reports

with cross-sectional and systemic views, considering

all virtual classrooms and contexts of a specific

teacher. It enables corrective actions to be taken,

ensuring quality work and preventing dropout risk.

In the context of knowledge management,

Heinzen (2002) presented a tool to assist the

teaching of programming logic. Departing from the

problem statement, the system retrieves solutions to

similar problems previously solved. The work

emphasizes problem solving based on analogy,

being focused on code. Nascimento et al. (2016)

apply CBR to suggest a pedagogical action for a

student-learning problem. This system also uses

learning objects to support pedagogical actions

aimed at facilitating the understanding of complex

concepts of the Introduction to Programming

discipline.

Although the many efforts to improve

interaction, interactivity, and decision-making

processes in VLE (CBR, ILS questionnaire, and BI

techniques, and so on), none of them provide an

integrated platform with the most relevant

functionalities, unlikely the proposal presented in

next section. It is also important to note that neither

approach adopts a proactive procedure, all are just

responsive.

4 IAG DESCRIPTION

This section introduces the main methodological

procedures for the proposed approach.

4.1 Defining the Generic Architecture

The generic architecture of IAg is shown in Figure 4

including the following components: (i) a

microservice-oriented platform for MAS

development and management (Haendchen Filho et

Collaborative Agents in Adaptative VLEs: Towards an Interface Agent for Interactivity and Decision-making Improvement

695

al., 2019), as presented in Section 2.1; (ii) a DBW

structure for data representation; (iii) an organization

of collaborative agents, instantiated on the platform;

and (iv) the Moodle virtual environment, used for

the case study. As described in Section 2, the

platform facilitates the development of agents,

providing communication, management, and

database access.

Figure 4: Generic Architecture.

The VLE MAS instantiated in the platform is

composed of the following collaborative agents:

(i) Tracing Agent (TAg), responsible for storing and

managing the data structure; (ii) Interface Agent

(IAg), that perform interaction with human actors;

(iii) Knowledge Agent (KAg), which manages AI

techniques to perform predictions and prescriptions;

(iv) Pedagogical Agent (PAg), which performs

content management, learning objects and trails; and

(v) Student, Tutor and Professor, which represent

virtual instances of these human actors.

The databases are represented by the Academic

Information System (AIS) and the DWH. The AIS

contains academic data, such as student profile and

history, data from tutors, teachers,

discipline/courses, and so on. The DWH is a

dimensional model composed by a central table of

facts, connected with the dimensions.

Listeners are placed in relevant spots in the

interface, waiting for the clicks triggering the script

to store the information. Locations do not

necessarily have to be on the links since data can be

stored with simple interactions. Clickstreams should

be sent to the webhouse data for storage.

All actions the user take can disclose knowledge

about the use of the system. DWH is widely used to

process analyses, obtaining information from two

main sources: (i) communication protocol data,

stored in the web services logs; and (ii) behaviour

seized with site scripts after establishing a session.

User behaviour on pages is a critical part because it

is not so simple to change a site to capture the

information.

4.2 Data Gathering

In order to promote collaboration, the TAg interacts

with the Interface Agent for populating the DWH

with the necessary data on student interaction in

VLE.

Figure 5 shows the data structure stored in the

dimensional model from which IAg can obtain the

data with the collaboration of TAg. A central Fact

table is connected with the dimensions relevant to

the context of the VLE.IAg have access to data to

turn it into real-time information that may be used to

interact with students, tutors, and teachers. Besides,

it will use BI tools to generate OLAP dashboards,

assisting teachers, and tutors in the decision-making

process.

Figure 5: VLE Webhouse Model.

The summary descriptions of dimensions are:

 Calendar Date. Attributes may include days of

the week, seasons, and holidays, among others.

 Time_of_day. Time slots during the day,

including hours, minutes, and time slots like

lunchtime, class time, and so forth.

 Academic Date. Associated with different

structures that differ on the number of modules

(semesters, four-month periods and trimesters).

 Page. The page source (e.g., static, dynamic),

function (content, exercise, video, forum) and so

forth.

 Session. Session is the collection of actions taken

by a visitor to a site while it navigates without

leaving this site.

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

696

 Causal. Describes the conditions of the current

progress of the subject, such as beginning of the

subject, period of tests, etc.

 Student. Information about the student profile.

 Academic Records. Provides information on the

student trajectory.

 Discipline. The attributes can include class

hours, credits, opening and closing dates, and so

forth.

 Teacher. Information about teacher profile.

 Tutor. Information about the tutor including the

degree of training, number of tutored disciplines,

etc.

 Domain Model. Describes the course schedule

of the learning path in the VLE.

 Referrer. Brings information about the URL

from where the user came from.

4.3 Functionalities Specification

The role model has been used (Gonçalves, 2009,

Haendchen Filho, 2017) to provide a summary of

software agents. A role can be described by two

basic attributes: (i) responsibilities are obligations

and indicate functionality, and (ii) permissions are

the rights associated with the role and indicate the

resources that the agents can use. Interaction and

interactivity are key concepts in the IAg role model.

Interactivity can be defined as the

communication human-machine and refers to a

mediated environment in where participants can

communicate synchronous or asynchronously and

participate in reciprocal message exchanges

(Kiousis, 2002). Interaction occurs among same

nature entities: human-human or machine-machine.

Table 2 presents the IAg role model.

The IAg responsibilities are:

 Login Procedures. IAg has two responsibilities

in the login procedure. The first is to apply the

ILS questionnaire when the student’s first login

occurs. In this procedure, he works in

collaboration with the PAg. According to the

information collected in the questionnaire, the

student is inserted in one of the profiles provided

for in the Felder-Silverman model. The second

responsibility is to create an interactive message

each time the student logs in, containing a set of

information about their last access.

 Knowledge Representation. In representing

knowledge, IAg’s responsibilities are to record

new questions and allow both the student, the

tutor and/or the teacher to consult on cases that

have already been resolved. In this case, IAg’s

main responsibilities are to collect cases and

Table 2: Partial IAg Role Model.

questions and consult via KAg the most similar

cases that can be reused for a specific problem.

 Interactivity with the Human Tutor. The IAg

is also responsible for enabling the tutor to

interact with students by sending notifications

with relevant information. These include: (i)

notify busiest time in class, allowing the tutor

and teacher to interact in real time with students

through chats, forums and any other activity that

may motivate interaction with groups; (ii)

generate notification of absence list, low/high

frequency in the discipline; and (iii) inform about

high or underperformers.

 Interactivity with the Student. IAg’s

interaction with students occurs by sending

interactive message at login, invitations to chat

and forums with other students and tutors. It also

maintains a proactive stance, informing the

student of their latest actions in the environment.

In addition, it sends welcome messages to new

students, messages for low-frequency students

offering help.

 Interactivity among Tutors, Teachers and

VLE Managers. The data structure stored in a

dimensional model enables IAg to use OLAP

tools to provide important information to VLE

tutors, teachers, and managers. for decision-

making. The tool provides analytical and

consolidated dashboards with cross-sectional and

Collaborative Agents in Adaptative VLEs: Towards an Interface Agent for Interactivity and Decision-making Improvement

697

systemic views, considering all dimensions and

contexts of a specific classes and students.

 Interaction. In this group, microservices are

placed. They involve interaction among IAg and

other collaborative agents. The microservice

Send student profile identification to PAg means

that IAg collects information from the ILS

questionnaire, scores the answers, and identifies

the student profile in one of 7 possible Felder-

Silverman Model categories. With this

information, PAg can define which learning

paths best fit this profile. The microservice

Receive TAg notification from missing student

who signed in enables preventive measures to be

taken when a student who has spent time without

logging in to VLE logs in. In this case, it is

important to find out why he is absent and if he

needs help with any difficulties. Agent

interactions can occur synchronously or

asynchronously. In asynchronous mode,

Blackboard is used as a mediator, as will be

shown in the following section.

4.4 IAg’s Responsibilities Specification

In order to specify IAg’s responsibilities, the

HEFLO (https://app.heflo.com/) tool was used. It

has strong adherence to the Business Process Model

and Notation (BPMN) for process diagram

(orchestration). BPMN diagramming is intuitive and

allows the representation of complex process details

as a standard language.

Figure 6 presents a workflow of collaboration

and interaction among agents and other actors

through task and service modelling. Links and

messages describe how they are related and how

they interact.

According to the service workflow, the

procedure starts when the student logs in to the

Moodle environment. After logon, the system begins

to collect log data, which is stored in a database

(DB). At this time TAg, which is responsible for

coordinating and maintaining the DWB, performs

the task of identifying the student profile.

If it identifies linked profile, TAg runs a new

service for retrieving information from its last login

and sends this data to IAg. The IAg upon receiving

this information initiates an interactivity service by

creating a welcome message with information from

the student's last login and where they left off.

Figure 6: Workflow adaptation - login procedure.

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

698

However, if the student does not have a profile

linked to his/her account, TAg writes on Blackboard

(BB) that it is necessary to apply the questionnaire

for this student. Upon receiving the message, BB

generates a notification for IAg to apply the

questionnaire. So, the IAg begins the service of

applying ILS questionnaire to rate the student

profile. When it finishes, IAg posts a message on the

BB, notifying TAg for storing profile data. After

reading the message on BB and storing the profile

data, TAg writes on the BB a message for the PAg

informing that there is a new student with a defined

profile.

Thus, PAg will be able to offer this new student

the layout and the most appropriate track for his/her

learning. The IAg now performs an interactive

welcoming message for the student and forwards

his/her to its adapted environment. environment.

5 IMPLEMENTATION

As mentioned in previous sections, the VLE system

was instantiated and implemented on the MIDAS

platform. On the platform, the S-Manager

infrastructure component provides a GUI wizard to

assist in global management tasks, as shown in

Figure 7.

In the panel on the right side of the figure, all

agents that make up the VLE can be seen. The panel

shows a navigable resource hierarchy organized by

agents. When an agent is selected, details are

displayed in the Details panel on the left side of the

window. At the bottom of the window, the Server

Log panel shows details of all transactions being

executed. In the Containers panel on the top left, you

can see the two containers registered on the platform

and used in this domain: the instantiated VLE and

the Academic Information System, which works

integrated with the VLE, from which the data for the

DW is extracted.

The details of the Interface Agent in the right-

hand panel show the services it provides. The + sign

in front of the services indicates that these services

are broken down into microservices at a lower level

of detail, as previously shown (Figure 6).

For the implementation of graphical BI

interfaces, the API of the Power BI tool (Microsoft,

2019) was used. Once the data is loaded in the tool,

it is possible to transform them by means of the

Query Editor option. This function offers several

data preparation functions such as dividing and

grouping columns, creating calculated columns,

applying filters and even building relationships

Figure 7: GUI to view instantiated agents and its services.

between tables.

The home screen allows the creation of

visualizations in an intuitive way with drag-drop

functions. The visualization of the dashboard or

ready-made report with drill-up and drill-down

operations, as well as the application of filters for

data analysis and knowledge acquisition was carried

out successfully. Many ways for visualizing are

available in the side menu and the columns can be

selected, as shown in Figure 8.

The Power BI JavaScript API provides

bidirectional communication between Power BI

reports and the application. The JavaScript API

enables to more easily embed reports into

applications and to programmatically interact with

those reports so that the applications and the reports

are more integrated.

The software also has the attractive feature that

the reports created in the tool can be accessed from

mobile devices through Power BI Mobil, being

made available free of charge for operating systems,

Android, IOS and Windows Mobile.

Knowledge representation was implemented with

the development of a question base, previously fed,

that allows the student to carry out research. The

prototype was developed using CBR techniques and

rules, applying similarity based recover procedure

for answers retrieval. Figure 9 shows the results of a

Collaborative Agents in Adaptative VLEs: Towards an Interface Agent for Interactivity and Decision-making Improvement

699

Figure 8: Power BI filters and the partial view of a query.

search, presenting several questions organized by

degree of similarity to the searched subject.

For developing this part of the solution, the

NetBeans IDE 8.1 development environment was

used. Bootstrap was chosen as a framework for

developing the web interface, and MySQLi was used

as the database management system.

The results achieved so far have been shown to

be adequate to what was proposed in this work. In

addition to its proactive nature, the solution offers

yet another tool for student learning, providing a

knowledge base that will assist the course with

several classes of students. The recovery, adaptation

and learning of registered cases are still under

development.

6 DISCUSSION

The work presented in this paper is primarily based

on suggestions from Open University (UK),

Brazilian Association of Distance Learning

Maintainers (ABMES, 2019), Brazilian Association

of Distance Learning (ABED, 2018), and the

Figure 9: CBR

Similarity

Search Result.

previously cited authors, beyond other researchers

(Choi et al., 2018; García-Álvarez et al., 2018) that

unanimously recommend the development of

proactive solutions for VLEs. The literature shows

that the vast majority of existing solutions do not

meet this requirement, or only partially.

As previously mentioned, the following works

focuses in the same problem: (i) Zapparolli et al.

(2017), which provide analytical and consolidated

BI reports; (ii) Dorça (2012), that uses a dynamic,

interactive, and gradually updated student model

based on ILS profiles; (iii) Maciel et al. (2014), that

use an avatar for facilitating interaction; (iv) Vaidya

and Sajja (2017), that provides an agent that not only

offers student learning facilities, but also calibrates

content; (v) Simbine et al. (2018), that focus on the

analysis and adaptation of learning trails; (vi)

Nascimento et al. (2016), that apply CBR to suggest

a pedagogical action for a student-learning problem.

The solution presented in this paper uses all these

techniques in one approach, offering a complete and

comprehensive proposal. In addition, it offers an

environment composed of a set of collaborative

agents with the proposal to create a virtual

environment for proactive learning, in contrast to all

approaches that act in a reactive way. As it is still a

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

700

work in progress, it was not possible to show in this

article all the potentialities that are in the

development phase, mainly the proactive procedures

of the solution. But the preliminary results achieved

were beyond our initial expectations.

7 CONCLUSIONS AND FUTURE

WORKS

Preliminary studies show that current VLEs are

mostly reactive. This characteristic is a source of

demotivation and feeling of abandonment for

students, leading to high dropout rates and low

graduation rates. More adaptable environments to

student profile, with large-scale interactivity, and

proactivity, can promote the expected benefits of

VLEs. It is well-known that when technology

expectations are unrealistically high and

subsequently not met in practice, the result can be

dissonance and dissatisfaction among stakeholders,

especially students (García-Álvarez et al., 2018).

The main contribution of this work is a solution

based on the needs of the educational market, aiming

to guarantee the student expectation meeting and

generation of adequate levels of motivation and

satisfaction. For this, proactive characteristics such

as adaptation, interactivity, and interaction were

included to obtain a strong sense of satisfaction,

possibly reducing dropout rates and increasing

graduation rates.

For implementing the solution, a platform

developed in the Applied Intelligence Laboratory at

University of Vale do Itajaí was used. The platform

has been already applied to successfully implement

other collaborative agents of the system (Haendchen

Filho et al., 2019).

As future works it is ongoing the implementation

of proactive procedures and adaptative interfaces.

Proactive procedures of social skill and autonomous

behavior are being developed in the agent's

workflow. Autonomy refers to the agent property of

running without interacting with humans, and social

ability indicates that they are able to interact by

sending and receiving messages and not by explicit

task invocation. For implementing interactivity, two

approaches are being applied: (i) the specification of

pre-defined rules (eg, welcoming students who are

absent for x days), and (ii) the use of a knowledge

base acquired by means of a machine learning

process. In certain circumstances, the agent must

have autonomy to communicate with the human

actors Student or Tutor. Interactivity based on

machine learning must consider that the knowledge

acquired by the KAg can be used collaboratively by

the IAg to assist the student in a proactive way.

Adaptative interfaces aim at providing an

interface design with learning objects appropriate to

the student's profile defined by the ILS

questionnaire. A user model will be created in order

to represent the way the developer will build the

system based on computational thinking. That is,

according to a logical sequence, observing the

requirements, tasks, and user experiences and

capabilities. The VLE graphical interface refers to

the environment in which the user effectively

interacts to accomplish a task. This user model will

be managed by IAg, according to the student's

profile, the learning trails, and the content objects,

handled by the Pedagogical Agent.

REFERENCES

Aamodt, A. and Plaza, E. (1994). Case-based reasoning:

Foundational issues, methodological variations, and

system approaches. AI communications, 7(1), 39-59.

ABED - Associação Brasileira do Ensino a Distância

(2018). Brazilian e-Learning Census 2017.

http://www.abed.org.br/site/pt/midiateca/censo_ead/15

54/2018/10/censoeadbr_-_2017/2018. In Portuguese.

ABMES - Associação Brasileira de Mantenedoras do

Ensino Superior, 2019. Brazilian Higher Education

Census http://www.abmes.org.br/documentos/detalhe/

720/censo-da-educacao-superior-2018-principais-

resultados. In Portuguese.

Aguiar, J. J. B., Fechine, J. M., and Costa, E. B. (2014).

Cognitive Styles and Learning Styles in Informatics in

Education: a systematic mapping focused on SBIE,

WIE and RBIE. Brazilian Symposium on Computers

in Education. In Portuguese.

Alencar, M. and Netto, J. M. (2013). Facilitating e-

learning tutoring using SISACAD. 19º Congresso

Internacional ABED de Educação a Distância,

Salvador/BA, Brazil.

Burkhardt, M. (2018). Towards a Pragmatic Approach to

Engineer Cognitive, Agent-Oriented, Distributed

Applications. IEEE 3rd International Workshops on

Foundations and Applications of Self Systems.

Choi, S., Lam, S., Li, K., and Wong, B. (2018). Learning

Analytics at Low Cost: At-risk Student Prediction with

Clicker Data and Systematic Proactive Interventions.

Journal of Educational Technology & Society,

21(2):273-290.

Collier, R. W., Russel, S., and Lillis. D. (2015). Reflecting

an agent programming with AgentSpeak. LNCS, 9387.

Dorça, F. A., 2012. A stochastic approach based on

reinforcement learning for automatic and dynamic

modeling of student learning styles in adaptive and

intelligent systems for distance education. PhD Thesis.

Collaborative Agents in Adaptative VLEs: Towards an Interface Agent for Interactivity and Decision-making Improvement

701

Universidade Federal de Uberlândia, Brazil. In

Portuguese.

Dragoni, N., Giallorenzo, S., Lafuente, A. L., Mazzara,

M., Montesi, F., Mustafin, R., and Safina, L. (2017).

Microservices: Yesterday, Today, and Tomorrow. In:

Mazzara, M., Meyer, B. (eds) Present and Ulterior

Software Engineering. Springer.

Felder, R. M. and Silverman, L. K. (1988). Learning and

teaching styles in engineering education. Engineering

Education, 78 (7), 674-681.

Freitas, A. A., Dornellas, D. V., and Belhot, R. V. (2006).

Production engineering student professional

requirements: a view through learning styles. Revista

GEPROS, 1. In Portuguese.

García-Álvarez M.T., Novo-Corti I., and Varela-

Candamio, L. (2018). The effects of social networks

on the assessment of virtual learning environments: A

study for social sciences degrees. Telematics and

Informatics, 35(4), 1005-1017.

Gonçalves E. J. T. (2009). Modeling Software Agent

Internal Architectures Using the MAS-ML Language.

MSc Thesis, Universidade Estadual do Ceará, Brazil.

In Portuguese.

Haendchen Filho, A. (2017). Multiagent Systems on the

Internet: A service-oriented architecture for the

development of intelligent software agents on the

Internet. Novas Edições Acadêmicas, 2017.

Haendchen Filho, A., Thalheimer, J. M., Dazzi, R. L. S.,

Santos, V., and Koehntopp, P. I. (2019). Improving

Decision-making in Virtual Learning Environments

using a Tracing Tutor Agent. 21nd ICEIS.

Heinzen, L. A. (2002). An Application of Case Based

Reasoning to support programming logic teaching,

Regional University of Blumenau. In Portuguese.

Higashino, M., T., Kawato, T., and Kawamura. T. (2018).

A Design with Mobile Agent Architecture for

Refactoring a Monolithic Service into Microservices.

Journal of Computers, 13(10).

Homayouni H., S. and Ghosh, I., Ray (2018). Data

Warehouse Testing. Advances in Computers.

Kimball, R. and Merz, R. (2000). The Data Webhouse

toolkit. John Wiley & Sons.

Kiousis, S. (2002). Interactivity: a concept explication.

New Media & Society. 4.

Kolodner, J. L. and Leake, D. (1996). A tutorial

introduction to case-based reasoning. In: Leake, D.

(ed.). Case-Based: Experiences, Lessons and Future

Directions. MIT Press, p.31-65.

Maciel, A. M. A., Rodrigues, R. L., and Carvalho, E. C. B.

(2015). Development of an integrated Moodle virtual

assistant to support online learning. XXV Simpósio

Brasileiro de Informática na Educação (SBIE 2014).

In Portuguese.

Microsoft, (2019). Introduction to Power BI Desktop.

https://docs.microsoft.com/pt-br/power-bi/desktop-

getting-started. In Portuguese.

Nascimento, P. B., Oliveira, E. H. T. de, Silva, F. S. da,

Ramos, I. M. M. and Ramos, D. B. (2016). Case-

Based Reasoning in Virtual Learning Environments: A

Systematic Literature Review. In: Sánchez, J. (ed).

Nuevas Ideas en Informática Educativa, 12, 551-556.

In Portuguese.

Sassi, R. J. (2010). Data Webhouse and operational

Business Intelligence: revisiting technology and

analyzing Data Warehouse Trends. XXX Encontro

Nacional de Engenharia de Produção. São Carlos, SP,

Brasil. In Portuguese.

Simbine, F. B., Lima, J. V., Torre, M. R., and Chiguvo, S.

S. (2018). Analysis of Learning Paths in Virtual

Learning Environments through Information

Visualization. InfoDesign - Revista Brasileira de

Design da Informação, 15(2), 183-196. In Portuguese.

Trevelin, A. T. C., Pereira, M. A. A., and Oliveira Neto, J.

D. (2013). The use of “inverted classroom” in

technology courses: a comparison between the

traditional model and the flipped classroom inverted

model adapted to learning styles. Revista de estilos de

aprendizaje, 6(12). In Portuguese.

Vaidya, N. M. and Sajja, P. S. (2016). Agent based system

for collaborative learning environment in an

educational habitat. International Conference on ICT

in Business, Industry, and Government, p. 1–5.

Vitorino, T. A. S. (2009). Case-Based Reasoning:

Concepts and Applications. MSc Thesis. UFMG,

Brazil. In Portuguese.

Woodley, A. and Simpson, O. (2014). The elephant in the

room. In: Anderson, T. and Zawacki-Richter, O.

(Eds.). Online distance education: Towards a research

agenda, 459-485.

Zapparolli, L., Stiubiener, I., Pimentel, E., and Braga, J. C.

(2017). Applying Business Intelligence and Learning

Analytics technics in Distance Learning

Environments. XXVIII Simpósio Brasileiro de

Informática na Educação (SBIE 2017).

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

702