An Adaptive System Architecture Model for the Study of Logic and

Programming with Learning Paths

Adson M. da S. Esteves

, Aluizio Haendchen Filho

2,3

, André L. A. Raabe

and Rudimar L. S. Dazzi

Laboratory of Technologic Innovation in Education, University of the Itajaí Valley (UNIVALI),

Rua Uruguay, 458, Itajaí, Brazil

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

Univertity Center of Brusque (UNIFEBE), Brusque, Brazil

Keywords: e-Learning, Intelligent Tutoring Systems, Adaptive System, Constructivism, Constructionism.

Abstract: The number of dropouts and evasion rates in computing courses are among the highest in Brazilian

universities. To reduce this rate, eLearning technologies are being used to compose solutions. Because of such

reality, this work aims at showing an adaptive system architecture with learning paths that best fit the student’s

profiles and interests. In order to take student’s profiles and interests into account, two theories will be used:

constructivist and constructionist. The fundamentals of these theories were analysed to formulate a teaching

structural model for the system. The literature was researched to find adaptive systems with theories similar

to constructivism and constructionism. Then it was designed a collaborative agent system based on intelligent

software agent techniques to help the student on its paths and content choices. In this system, the student’s

difficulties, characteristics, and knowledge obtained from other users can be reused. An environment with a

content hierarchy which allows more attractive learning path construction options may ease the learning, make

the study more interesting and help reduce evasion rates in computing courses.

1 INTRODUCTION

Dropout is a negative phenomenon present in

Brazilian higher education. Related to the negative

phenomenon, it is possible to explicit: the students

themselves, the teaching institutions, and the market.

For students who drop out of college, it reduces their

chances of personal and professional growth. For the

institutions, they fail to fulfil their social function of

educating, on the one hand, and fostering the labour

market, on the other. Related to the market, it is

noteworthy that it is increasingly interested in new

professionals related to the areas of computing and

information technology (IT), as shown by various job

search sites (Guia da Carreira, 2018; MichaelPage,

2019; Pattabiraman, 2019; CareerCast, 2019; Trade

Schools 2019). Young people seeking professional

growth in this market choose computer-related

courses to meet this demand.

Even though there is a good number of students

entering these courses, the dropouts in Brazil are quite

high. Dropout rates can range from 22% to 32% and

is the second highest rate among courses at Brazilian

universities (Filho et al, 2007; Lobo, 2017). Among

the reasons that lead to this, it is the difficulty of the

initial subjects that are the pillars of the course:

Algorithms and Programming. These subjects are

considered difficult by many beginners because of the

need for the required abstraction and logical-

mathematical skills they do not have in their daily life

(Raabe and da Silva, 2005).

Seeking to better understand this problem, Giraffa

and Mora (2016) conducted a survey with students

who stopped attending computer science courses at

PUCRS during the period of 2012-2013. In this

research, it was found that the main reasons that

contributed for the dropouts were: (i) lack of study

time, as many have working hours; (ii) difficulties in

understanding issues in the classroom and in

activities; and (iii) teachers not so well qualified to

serve several students. Such difficulties point not only

to the problem of lack of prior skills, but to the

didactic organization of teachers and the formulation

of activities, as well as students’ individual problems.

Teachers specific to each individual student

would be ideal for solving problems, but the cost

makes this practice unfeasible (Weragama, 2013). In

addition, the difficulty of classroom issues may be

Esteves, A., Filho, A., Raabe, A. and Dazzi, R.

An Adaptive System Architecture Model for the Study of Logic and Programming with Learning Paths.

DOI: 10.5220/0009412406790690

In Proceedings of the 22nd International Conference on Enterprise Information Systems (ICEIS 2020) - Volume 1, pages 679-690

ISBN: 978-989-758-423-7

679

related to the different types of students entering

computer courses. Heterogeneous profiles such as

gender, age, education level, way of learning, and

problem-solving skills make it difficult to create

unique content that addresses them all (Oliveira et al,

2015).

In order to reduce the problem, two points are

highlighted: changing teaching methodologies and E-

learning.

E-learning is a web-based learning ecosystem

integrating several stakeholders with technology and

processes. It provides a flexible and personalized way

to learn, allowing learning on demand and reducing

its cost (Cidral, 2018).

The application of E-learning is possible with the

Intelligent Tutoring Systems (ITS). Nowadays, also

called Adaptive Systems, they incorporate Artificial

Intelligence (AI) techniques to develop tutors who

know what, for whom and how they teach (Nwana,

1990), that is, they consider the peculiarities of the

student.

On ITS development, many authors have

successfully used Multi-Agent Systems (MAS)

proposed (Giraffa, 1999; Yaghmaie and

Bahreininejad, 2011; Dolenc and Aberšek, 2015;

Hooshyar et al, 2015; Harley et al, 2015; Vaidya and

Sajja, 2016). Using MAS can help with complex tasks

such as monitoring student activity, capturing

information about their dynamic contexts,

recommending content based on their profiles, and

more (Frade, 2015).

Another way to solve problems may be by

changing teaching methodologies. Some suggest a

change in the form of general education to facilitate

learning, such as constructivism (Piaget, 1967) and

constructionism (Papert, 1980). In the constructivist

approach, during the learning process, the student

may be able to decide how to learn and act proactively

in knowledge building (Bada 2015). In addition,

constructionism points out that building a product

related to students’ interests helps learning to occur in

a more efficient way.

One way to enable students to conduct their

learning in computing is to provide choices on

learning paths. Paths are different types of skills or

knowledge that follow a user-definable sequence.

Thus, led by the adaptive system, the students can

define their own learning path, facilitating the

acquisition of the content.

Learning environments can be tools for applying

constructionism (Baranauskas, 1999). They can be

applied with microworld building, hypermedia text,

and programming environments. Programming

environments that help the user and facilitate the

learning of logic present constructivist

characteristics.

This paper presents an adaptive system

architecture model with learning paths for

programming teaching. For the development of

learning paths, the IDE Portugol Studio (Noschang,

2014) is used. It was developed with the purpose of

facilitating the learning of programming logic in

Brazil and has architectural features that facilitates to

create adaptable learning paths for the students.

The adaptive system is designed to provide the

following features: (i) learning paths with adaptive

options, led by an intelligent tutor so that students can

choose the knowledge of interest; (ii) adapted

resolution tips and support materials during the

exercise, if the student has difficulties; and (iii)

providing exercises adapted to the student's personal

interests. The proposed solution utilizes MAS

technology, and the MIDAS platform provides

infrastructure services such as communication,

management, and interaction between agents.

2 BACKGROUND

The following subsections describe the main

foundations used for solution development.

2.1 e-Learning and Adaptive Systems

Considering the rapid growth of Technology and

Population, it seems inevitable that E-learning is

going to be the main agent for education. It involves

innovative pedagogy, advanced teaching strategies

and learning methodologies and approaches tightly

connected with flexibility, accessibility, openness,

and communication through modern services.

Personalized intelligent agents and recommender

systems have been widely accepted as solutions

towards overcoming information retrieval challenges

by learners arising from information overload (Tarus,

2017).

That’s why Intelligent Tutoring Systems are great

systems to integrate into E-learning environments

(Phobun, 2010). For years, there have been several

proposals on how ITS can be modelled. The

following works have similar themes and ideas that

helped in the development of the architecture

proposed in this work.

Cabada et al (2017) developed Java Sensei, an

adaptive learning environment based on student

preferences. The system recommends exercises based

on other students’ grades who have previously used

the system. It uses sentiment analysis captured

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

680

through cameras to provide personal instruction to

users. It also allows the creation of exercises without

the need to program. The system architecture consists

of the following layers: (i) presentation, which is the

interface that communicates with the user; (ii) tutor,

comprising a domain module, a tutor module, a

student module and an exercise recommendation

system; (iii) intelligence, which infers the

characteristics of the students using the system by

analysing their emotions; (iv) web content, which

contains key files for the logic of previous layers; and

(v) data, which contains the student and domain saved

data.

De Meo et al (2007) used the MAS technique to

develop the X-Learn. The virtual learning system is

structured in XML. The system consists of 3 main

agents: (i) User Device Agent (UDA); (ii) Skills

Management Agent (SMA); and (iii) Learning

Program Agent (LPA). Knowledge is stored in a

learning object repository. By activating the UDA

associated with it, the student has access to a list of

skills they can learn. This list is made in collaboration

with the SMA, which analyses the skills he has

already learned, and the skills he can learn. After the

user selects what they want to learn, SMA brings up

a list of subjects needed to learn that specific skill. It

then cooperates with LPA to define the best learning

program based on knowledge and user profile

information.

Panagiotis et al (2016) developed the APLe

(Agents for Personalized Learning). They based their

system in 2 ontologies: (i) Learner Model and (ii)

Learning Object and Outcomes. The first models the

student’s characteristics like learning and social style,

use of technology, literacy, experience, time of study

and reasons for education. The second stands for the

knowledge domain. It used Learning Objects, which

are units of educational digital content and Learning

Outcomes which are what the student is expected to

know. The tutor Agents designated to each student

uses a LSM (Learning Space Management) and a

LTC (Learning Tactic Control). The LSM is the

knowledge to be learned; it used learning objects to

create a graph structure. The Learning Objects and

Outcomes are placed in this graph with the actual

states of the learner. They tell if the student learned

something depending on the states they are. The LTC

is a reactive component that selects the tactic to apply

to the LSM. When the learner asks for a

recommendation, the system triggers the LTC to use

a formula to balance the LSM nodes and determine

which nodes of content should be recommended.

All these works use Multi-Agent Systems. While

they are not obligatory to make them function, MAS

allows for more scalability, error tolerance,

robustness and security. These works help

formulating some structures from the proposed

system as it will be seen in sections 3 and 4.

2.2 Constructivism and

Constructionism in Learning Paths

The constructivism, proposed by Piaget (1980), is

based on the principle that learning is a process of

knowledge construction in which the student is an

active part of the process. Children not only record

what they are taught, but also formulate their own

ideas with prior knowledge. Previously acquired

knowledge is not necessarily wrong, but with

experience, by validating this knowledge, it is

possible to correct and define new solutions.

By giving the learner a leading role in learning,

during the process of knowledge building

constructivism allows one to confront the results of

experiments and re-evaluate their ideas about how the

world works (Philips, 1995).

However, Papert (2008) studies Piaget’s research

and criticizes some aspects. He points out how

concrete learning, described by Piaget, is a subjective

term that leads to misinterpretation. Papert proposes

constructionism as a form of “reconstructed” learning

of Piaget’s constructivism.

For Papert, learning is not just about

experimentation, building objects in the world plays

a more important role in knowledge building. For

him, the construction is not only theoretical, but also

practical of a real product. Moreover, the context of

this construction being relevant to the student allows

the construction of the mental model itself to be

facilitated by association with a subject that already

knows.

To provide the student with the possibility of

construction and protagonism, the chosen form was

the use of learning paths. These paths are like a

sequence of nodes, where each node is related to a

skill or subject learned. These subjects and skills have

various contents within them that serve as a learning

possibility for the student. The student who wants to

learn some skill can look at the contents and look for

the one of their choice. A student looking at various

content in sequence will build a content path,

ultimately called the learning path. Details of the path

structure can be seen in section 4.1.

Raabe et al (2016) points out how contexts

applied in exercise statements can influence the

resolution of activities, as they show a real application

of this knowledge. This, however, cannot be

generalized since students have their own context. If

An Adaptive System Architecture Model for the Study of Logic and Programming with Learning Paths

681

an exercise that is not in a context is applied, that

application may not be relevant to learning. Thus, in

addition to the paths, each tree node contains a series

of content with different contexts to learn. The tutor

suggests to the student the one with the best context

for him to learn, depending on his interests.

2.3 Portugol Studio

Many authors propose tools that facilitate and remove

barriers in early learning. There are several

environments in the literature that help beginners in

programming (Cooper, 2000; Ng, 2005; Resnick,

2009; Wolber, 2011; Paiva, 2016; Romagosa 2019)

and Portugol Studio (PS) (Noschang 2014) is an IDE

focused on Brazilian learners.

PS is a beginner-oriented programming IDE. The

language syntax is defined in Portuguese and has a

simplified interface for easy learning. Even as a

beginner IDE, it has several libraries that allow more

complex programs to be developed.

The IDE has a language like C and PHP. It has

syntax documentation tabs, and library

documentation in the help tab. It has several examples

of programs already made, from simpler programs

with “Hello World” writing to complete games using

the existing graphics mode. It also has didactic error

messages as standard which can help smart tutors.

These messages tell the user the location of the error,

why the error occurred, and examples of

troubleshooting.

Some works that enable more effective intelligent

tutoring have already been developed for this tool

(Pelz, 2011; Hodecker, 2014). They deal with the

development of an automatic exercise corrector as a

plugin for PS.

The PS platform is open source and is already

used in more than 7 universities. It has over 200,000

downloads and keeps up to date. Besides being able

to change IDE source code, it is also possible to

develop plugins, create buttons with functions in the

code editor, add libraries, among others.

3 PROPOSED SOLUTION

The system was designed with features of an adaptive

STI. It offers content on various topics with different

themes. This enables the student to choose the content

they want to learn and set up their own learning path.

Meanwhile the system helps the student to choose the

content that fits his profile. In addition, the system

provides solving tips and support materials when

performing exercises tailored to their profile if the

students experience difficulties. The following

subsections present the learning path structure, the

generic architecture, and the proposed role model for

the agents.

3.1 Path Model

The path model developed for the system uses a level

and topic structure. In the system domain there are

several different topics of programming learning,

however some need previously taught topics to be

learned. Because of that, the framework uses levels to

define the order in which some topics need to be

learned. This structure can be seen in Figure 1.

Figure 1: Structure of knowledge domain levels.

Existing topics at the same level can be learned in

any order. Conversely, topics that are at different

levels need to be learned according to the order of the

levels at which they are located, starting from the

lowest level to the highest level. The student using the

system will need to complete all topics at the same

level to advance to the next one. However, this does

not mean that he will need to complete all existing

content on that topic.

This is possible because each topic has several

contents that validate the topic as complete if used by

the student. These contents may vary between: (i)

exercises and (ii) support materials. The contents

have attributes that represent their characteristics that

differentiate them from each other. Attributes range

from: (i) Difficulty, (ii) Complexity, (iii) Content

Type, (vi) Taxonomy, and (v) Tags. This can be seen

in Figure 2.

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

682

Figure 2: Topic structure, and knowledge domain exercise.

The tags in each content are related to the themes

that the content addresses outside the programming.

These themes are related to different preferences that

students may have when using the system. For

example, a student may prefer songs, so music-tagged

content will be recommended to them.

When selecting content, students will have their

option recorded in their account and linked to their

previous content. This connection in sequence of

contents is called the learning path. This can be seen

in Figure 3.

Figure 3: Generic example of a learning path.

The topics in a path do not necessarily have to be

in order of levels. Topics that have already been

completed by the student can always be revisited, so

lower level exercises after higher levels can occur on

the paths. This can happen in cases where the student

realizes or is driven by the system that needs to

reinforce content they should have already learned.

The students, in these cases, will be directed to a

different exercise of the same topic and placed on

their path.

The track is recorded in the system, allowing it to

be reused to help new students select their content.

Similarities between student characteristics may

identify paths that best suit them because they have

already been used and successful by previous

students.

3.2 Generic Architecture

The generic architecture is an abstract representation

of the system, as shown in Figure 2. In Localhost, it

is the PS system with its functions (AutoCorrect and

Error Messages). The proposed solution is

represented in two modules: the Web Platform and

the Tutor Plugin.

Figure 4: Proposed system generic architecture.

Domain and student models are contained in the

web platform. They keep the content and student data

respectively. Structured content data for learning

paths as well as students’ doubts registered in the

system are stored in the domain model. In the student

model, the consolidated profile data of all students in

the system is stored.

Students using the system must create an account

so that their profile and usage information is saved

and help the tutor develop their teaching method. For

the adaptive system to learn over time, information

from all students using the system must be

consolidated into a database.

In the web platform, the following five agents act:

(i) Pedagogical Tutor, who collaborates with

Classifier and Recommender agents to provide

students with exercises recommended by their

preferences and similarities with other students;

(ii) Interface, which interacts with the student by

recording their questions and answering them, and by

creating an interface for the system administrator and

the student;

(iii) Tracker, which is responsible for capturing

student interactions with the system and storing them

in the database;

An Adaptive System Architecture Model for the Study of Logic and Programming with Learning Paths

683

(vi) Classifier, which is responsible for clustering

students according to their interactions and

preferences;

(v) Recommender, to recommend to the tutor

module the best student exercise based on other

students. With these agents working together, we seek

to meet the needs of students in programming

learning.

The web platform uses the MIDAS platform

(Haendchen Filho, 2017) responsible for the

execution of the multi-agent system, the management

of communication between agents and the

mechanisms of access to the database. Information on

the operation of this system is described in section

2.1.

3.3 Tutor Module Specifications

The specifications of the agents that make up the

Tutor Module were defined using the role model. This

model is used in several approaches (Gonçalves 2009;

Haendchen Filho, 2019) to provide a summary of

agent activities. According to theory, a role can be

described by two basic attributes: (i) Responsibilities:

the role of obligations that indicate functionality; and

(ii) Permissions: the rights associated with the role,

indicating the features that the agent can use. Table 1

presents the role model of the agents.

Table 1: Agent Responsibility Table.

The Tracking Agent is responsible for capturing

student interactions with Portugol Studio. It is located

on the web platform and receives user interaction data

through the plugin interface. This agent is therefore

responsible for recording in the student module the

following student information: (i) clicks on the

interface; (ii) length of stay in the system; (iii)

exercise response time; (iv) solved exercises; (v)

chosen paths.

The Interface Agent is responsible for formatting

the interface for the system administrator. It has

access to domain and student module information to

allow the administrator to see this information on

their screen. It also receives questions from users

through an existing plugin interface. When a user asks

a question, the agent searches the domain module for

answers to similar questions and show the student. If

they do not have similar questions already logged in

the system, it will be logged until an administrator

answers it. Therefore, their responsibilities are: (i) to

record student’s questions; (ii) notify the

administrator about unresolved student’s questions;

(iii) send the students who asked the questions the

respective answers of the administrators; (iv) answer

student’s questions without the need for an

administrator if a similar question has already been

asked by another student and answered.

The Tutor Agent is the one who chooses the best

teaching strategy for the student. It has access to

student module and domain module information. This

agent collaborates with the Classifier agents to cluster

students and discover similarities, and the Advisory

agent to suggest student-tailored content. This

enables new students to learn from the tutor’s

experiences with previous students. In addition, to

better identify these groups, it identifies errors in

exercises by enumerating by types of errors, and

students' personal preferences and content selection.

Finally, it analyses the reading time and resolution of

each content, identifying which students spent more

and less time. Thus, this agent’s responsibilities are:

(i) to suggest learning paths based on student

preferences and profile (collaboration with Classifier

and Recommender Agent); (ii) suggest real-time

exercise solving tips with the aid of the automatic

correction module to help you learn programming

content more easily; (iv) identify errors in exercises

(collaboration with Tracking Agent); (v) identify

student preferences according to chosen exercises

(collaboration with Tracking Agent); (vi) identify the

average exercise resolution time (collaboration with

Tracking Agent).

The Classifier Agent uses Clustering techniques

to group students with similar characteristics, and

Machine Learning techniques to identify students

with potential dropout characteristics. The Classifier

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

684

Figure 5: Content request process in BPMN model.

collaborates with the Pedagogical Agent. Once the

student is classified into a group, you can check

which exercises best fit them and their preferences.

The profile data that allow classification are: (i) age,

(ii) gender, (iii) educational level, (iv) personal

preferences, (v) chosen paths, (vi) resolved exercises,

(vii) average time exercise resolution and (viii) most

frequent types of errors. This information is stored in

the user accounts created by each student using the

system. The first four attributes are obtained at the

time of user’s account creation, and the others are

captured while using the system.

Referring Agents are responsible for

recommending content to a student. It receives from

the Pedagogical Agent the student group to which the

student belongs and uses this group to find the content

that had the best results in terms of time and user

errors. When the analysis and selection is completed,

it sends to the Pedagogical Agent the content that best

fits your profile and level of knowledge.

This communication between agents can be

represented in a BPMN model (Küster, 2012). The

BPMN (Business Process Modelling Notation) is a

leading process modelling language that was

designed to be readable by all its business users

(OMG, 2006). Figure 3 presents an example of

communication between system agents using the

BPMN model.

The initial process with the user requesting the

pedagogical agent and then requesting the interest

group that student is to the classifying agent. This

agent will identify the group that the student

requesting belongs to and will return to the

pedagogical one. Then the pedagogical agent will

send the recommending agent the group of students

requesting the exercise recommendations for him.

The recommender will compare content paths among

students in the group and indicate the most common

content among them for the student who requested it.

This list returns to the pedagogical agent who

responds to the student with the list of contents

recommended to him, along with the exercises of the

level the student is at.

An Adaptive System Architecture Model for the Study of Logic and Programming with Learning Paths

685

4 IMPLEMENTATIONS

An implementation of the system is being developed

for this work. The system is already structured on the

MIDAS platform with all agents and their services.

Agents on the platform interface show only

communication services, which allow collaboration

between agents. Responsibilities that do not require

communication with other agents such as user

tracking through the interface or calculating user time

when using some content are intrinsic to the platform

interface as can be seen in Figure 6.

Figure 6: Agent structure modelled on MIDAS platform.

The life cycle of each agent differs according to

its objectives and services. The tracking agent must

capture and organize the user's information while

using the system. The interface agent must

communicate with the user taking questions and with

the administrator presenting information. The tutor

agent must select the teaching strategies for the user,

using the sociability between the agents of the system.

The classification agent must group similar students

to facilitate the identification of characteristics.

Finally, the recommendation agent must select the

most adapted content according to the group of

student characteristics.

The system in the Portugol interface will

communicate with agents on the web platform. Figure

7 shows the system content selection image. It is

under development and it will be changed, but it can

already be seen inside the Portugol Studio system.

Figure 7: Interface prototype working on PS.

The above presented interface is in Portuguese

idiom. It contains, on the top panel, the contents

recommended by the system for the user. In the panel

below, the contents that can be selected by the user

are They can be filtered by: (i) type, (ii) level, (iii)

difficulty and (iv) themes.

4.1 Methodological Procedures

The project was planned in several phases described

below:

(i) Systematic literature review to find related work

and platforms to assist the implementation

(ii) Definition of the learning trails structure

(iii) Elaboration of the system architecture

(iv) Specification of the role model for system

agents

(v) Specification of agent services workflow

(vi) Adaptation of the Portugol Studio tool to

receive the system as its plugin

(vii) Population of the database

(viii) Implementation of the agents

(ix) Development of the Graphical User Interface

(x) Experiments with students

(xi) Analysis of the experiments

Phases (i) to (iv) are concluded. Phases (vii), (viii)

and (ix) are in progress. A partial implementation of

the Agents and the User Interface was illustrated in

figures 6 and 7 and the population of the database is

currently being held. Phases (x) and (xi) will be held

in the near future.

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

686

5 DISCUSSION

In trying to solve the problem of dropout, intelligent

tutoring systems focused on programming teaching

were researched and applied in Brazilian universities,

but they use tools with the problems already reported

in research with the PS. In recent updates, PS allows

the development of plugins in the tool, so the

possibility of an adaptive system being integrated into

the tool has become palpable.

Thus, researching the works with Adaptive

Systems and teaching methodologies, it was possible

to develop the architecture presented in this work.

The system described is based on two points: (i)

Learning Paths and (ii) Multi-Agent System

Architecture.

The learning path structure was developed by

trying to use the constructivist and constructionist

concepts developed by Piaget and Papert

respectively. However, assisting the student in what

activities he should do and allowing him a freedom to

choose what he wants to do are almost opposite

concepts. To solve this point, the literature sought the

uses of learning paths.

In the literature, the concepts of learning paths are

already used by some authors such as De Meo (2007)

and Panagiotis (2016). With De Meo, paths are used

to define the way a student must take to complete a

project. The options to choose from in his work are

only from existing projects. The paths and topics the

student must follow to achieve their project goal are

predefined by the system. On the other hand,

Panagiotis uses learning paths as suggestions for the

student. The student can select a learning object

within that track without necessarily having to follow

it as recommended.

Works such as De Meo’s only let a momentary

protagonism to the student. The student, even doing a

project he has chosen, must still follow predetermined

instructions. Panagiotis' work already gives the

student greater freedom, the intelligent tutor being

just someone who suggests what he should learn.

Both works, however, do not allow the student to set

up their own track.

The proposed track structure for the system uses

student choices to assemble the paths. Each student

using the system will receive suggestions for the

following paths, but he or she can assemble their own

by selecting the order of content they want to watch

or play. Paths are made of content that is like

Panagiotis learning objects and uses tags that define

the characteristics of each content. They are an

expansion of the work of Santos el al (2013). Not only

are exercise topics, taxonomy and complexity

highlighted, but also the topics they cover in

explaining the topics. This allows the student to learn

according to the subject of interest.

Recommend content according to the student’s

topic of interest and interactions with other users is

part of the smart tutor’s responsibilities. Modeling the

smart tutor is an important part of the development

process and one of the contributions of this work.

Many works on intelligent and adaptive tutors are

developed with multi-agent system technologies as

already mentioned in section 1. As these systems are

constantly updated, they need to be modular to reduce

errors and facilitate maintenance. Multi-agent

systems are great for that.

In this paper, we use the concept of service

oriented intelligent agents as proposed through

MIDAS. Analyzing some works in the literature it

was possible to see which agents are well used and

which are adaptable to this work.

The use of recommending agents is an existing

practice among intelligent tutors, as in the study by

Cabada et al (2017). He proposes an architecture

composed by a tutor and an exercise recommender. In

the proposed model, the tutor is used solely to assist

the user in activities with suggestions, while the

recommender only provides exercises based on the

user’s characteristics.

This separation is applied in the proposed

architecture. The recommending agent has only the

function of recommending the exercises based on the

characteristics of the students who have already used

the system. However, the discovery of these similar

characteristics is the responsibility of another agent,

the classifying agent.

Both agents collaborate with the main agent, the

pedagogical tutor. It coordinates agent collaboration

and uses existing exercise correction modules in

Portugol Studio to deliver exercise resolution tips.

This separation of intelligence from

recommendation and activity aid is like that of

Cabada’s work. Meanwhile two other agents were

added. The tracking agent, whose primary function is

to track all user interactions and store them, and the

interface agent that is responsible for communicating

interfaces with the administrator.

Both improve system maintenance. The tracker

further modularizes tasks, and the interface agent

makes it easy for the administrator to change system

settings without having to make changes to the source

code. In addition, the interface agent allows the

administrator to answer student questions and save

them so that they can answer similar questions from

other students.

An Adaptive System Architecture Model for the Study of Logic and Programming with Learning Paths

687

These system features and Portugol Studio

functionality can therefore help students improve

their programming learning with the support of the

intelligent tutor.

The tutor architecture presented here follows

current technological and methodological standards.

The main difference from similar work is the learning

paths structure. It allows different and personal

learning paths to be suggested for new students

according to previous students with similar

characteristics.

This might reduce the slope of the students

learning curve since successful cases will be used to

help them. The Portugol Studio (PS) tool also helps

in the same direction since it was developed to aid

novice programming students in universities, and it is

widely used in Brazil.

The plugins the PS have are easily available on its

interface. This will allow the proposed Adaptive

System to be found by many Computer Science

students in Brazil and also independent learners and

will certainly foster the development of new research

projects.

6 CONCLUSION AND FUTURE

WORKS

There is a recurring need in the Computer Science

area for tools to assist the learning process, especially

programming logic. Students often do not have time

to study or cannot adapt to classroom teaching.

Adaptive systems emerge as one of the alternatives to

this reality.

Alternatively, constructivist and constructivism

methodologies can also help students learn new

concepts in programming using IDEs. Systems with

adaptive features and assisted learning techniques

like Portugol Studio can be a way to facilitate learning

for beginning programmers.

The main contribution of this work is the proposal

of an architectural model with low-level diagrams

representing the functionalities of the agents for an

adaptive system. The purpose is to facilitate the logic

learning and programming with learning paths. This

can generate gains in preparing a generation in a high

demand market for this knowledge type.

The next steps in implementing the system are as

follows: (i) implementation of the services, (ii)

population of the content tags and (iii)

implementation of the communication interface

between Portugol and the Agents on the web.

The implementation of the services will allow

communication between agents on the MIDAS

platform. Each agent service implemented must

follow the definitions shown in subsection 3.3.

Content tags are one of the bases of the system's

intelligence. They will allow the recommending agent

to identify topics of interest to students in each

content. This will permit him to learn to program

within his context of interest, following

constructionist theory.

Finally, Portugol's communication interface with

web agents will start the connections that will allow

agents to receive student information and collaborate

to send students learning suggestions.

As the system is finished, the future works in the

project are: (i) experiments with students and (ii)

analysis of the experiments, following the procedures

described on subsection 4.1.

The experiments will be held made with classes of

computer science real students of the authors from the

local university. The students will use the system as a

support to discover their difficulties in programming.

The main goal is to assess whether the system

gives them coherent personalized suggestions

according to their difficulties and preferences. The

student’s acceptance on the recommendations are

also going to be measured.

Since the Portugol Studio tool is widely adopted

in Brazilian universities, further work also can be

done to collected data to find patterns about general

difficulties of Brazilian programming learners and

which learning paths can help the most.

These researches might help find better ways to

help students learn programming and reduce the

dropouts on universities caused by the difficulties

they have in the initial courses.

REFERENCES

Bada, S. O., & Olusegun, S. (2015). Constructivism

learning theory: A paradigm for teaching and learning.

Journal of Research & Method in Education, 5(6), 66-

70.

Baranauskas, M. C. C., Vieira, H., Martins, R. M. C., &

D’ABREU, J. V. (1999). Uma taxonomia para

ambientes de aprendizado baseados no computador. O

computador na sociedade do conhecimento, 45.

Cabada, R. Z., Estrada, M. L. B., Hernández, F. G.,

Bustillos, R. O., & Reyes-García, C. A. (2018). An

affective and Web 3.0-based learning environment for

a programming language. Telematics and Informatics,

35(3), 611–628. doi:10.1016/j.tele.2017.03.005

CareerCast (2019) The Toughest Jobs to Fill in 2019,

Available at: https://www.careercast.com/jobs-

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

688

rated/2019-most-difficult-jobs-to-fill (Accessed: 30th

December 2019).

Carmen-Leocadia Pesantez Pozo (2017) 'Smart Education

and Smart e-Learning', Conference proceedings of

»eLearning and Software for Education« (eLSE),

III(01), pp. 89-95.

Cidral, W. A., Oliveira, T., Di Felice, M., & Aparicio, M.

(2018). E-learning success determinants: Brazilian

empirical study. Computers & Education, 122, 273–

290. doi:10.1016/j.compedu.2017.12.001

Cooper, S., Dann, W., & Pausch, R. (2000). Alice: a 3-D

tool for introductory programming concepts. Journal of

Computing Sciences in Colleges, 15(5), 107-116.

De Meo, P., Garro, A., Terracina, G., & Ursino, D. (2007).

Personalizing learning programs with X-Learn, an

XML-based, “user-device” adaptive multi-agent

system. Information Sciences, 177(8), 1729–1770.

doi:10.1016/j.ins.2006.10.005

Dolenc, K., & Aberšek, B. (2015) "TECH8 intelligent and

adaptive e-learning system: Integration into technology

and science classrooms in lower secondary schools"

Computers and Education, 82, 354–365, doi:10.1016/

j.compedu. 2014.12.010

Elham Mousavinasab, Nahid Zarifsanaiey, Sharareh R.

Niakan Kalhori, Mahnaz Rakhshan, Leila Keikha &

Marjan Ghazi Saeedi (2018) Intelligent tutoring

systems: a systematic review of characteristics,

applications, and evaluation methods, Interactive

Learning Environments, doi: 10.1080/10494820.2018.

1558257

Frade, R. V. C. (2015) "UNIVIRTUAL – Ambiente Virtual

3D Multiagente com Recomendação Personalizada de

Objetos de Aprendizagem", Dissertação (Mestrado em

Ciência da Computação) - Universidade Estadual do

Rio Grande do Norte, Universidade Federal Rural do

Semi-Árido, Mossoró.

Giraffa, L. M. (1999). Uma arquitetura de tutor utilizando

estados mentais. 1999 (Doctoral dissertation, Tese

(Doutorado em Ciências da Computação) – Instituto de

Informática, UFRGS, Porto Alegre).

Giraffa, L. M., & da costa Mora, M. (2013). Evasão na

disciplina de algoritmo e programação: um estudo a

partir dos fatores intervenientes na perspectiva do

estudante. In Congresos CLABES.

Gonçalves, E. J. T. (2009). Modelagem de arquiteturas

internas de agentes de software utilizando a linguagem

MAS-ML 2.0 (Doctoral dissertation, Dissertação de

Mestrado. Universidade Estadual do Ceará. Centro de

Ciência e Tecnologia. Fortaleza).

Guia da Carreira (2018) Saiba quais as 8 profissões que

mais crescem no Brasil, Available at:

https://www.guiadacarreira.com.br/profissao/profissoe

s-que-mais-crescem/ (Accessed: 30th December 2019).

Haendchen Filho, A. (2005). Um Framework do tipo

Middleware para Sistemas Multi-Agentes na Internet

(Doctoral dissertation, PUC-Rio).

Filho A., Thalheimer J., Dazzi R., Santos V. and Koehntopp

P. (2019). Improving Decision-making in Virtual

Learning Environments using a Tracing Tutor Agent. In

Proceedings of the 21st International Conference on

Enterprise Information Systems - Volume 1: ICEIS,

ISBN 978-989-758-372-8, pages 600-607. DOI:

10.5220/0007744006000607

Harley, J. M., Bouchet, F., Hussain, M. S., Azevedo, R., &

Calvo, R. (2015) "A multi-componential analysis of

emotions during complex learning with an intelligent

multi-agent system". Computers in Human Behavior,

48, 615–625, doi:10.1016/j. chb.2015.02.013

Hodecker, Andrei. Aprimoramento e avaliação do corretor

de questões do Portugol Studio. 2014. TCC (graduação

em Ciência da Computação) - Universidade do Vale do

Itajaí, Itajaí, 2014.

Hooshyar, D., Ahmad, R. B., Yousefi, M., Yusop, F. D., &

Horng, S.-J. (2015) "A flowchart-based intelligent

tutoring system for improving problem-solving skills of

novice programmers", Journal of Computer Assisted

Learning, 31(4), 345–361.

Küster, T., Lützenberger, M., Heßler, A., & Hirsch, B.

(2012). Integrating process modelling into multi-agent

system engineering. Multiagent and Grid Systems, 8(1),

105–124. doi:10.3233/mgs-2012-0182

Lobo, R. (2017) 'A Evasão No Ensino Superior Brasileiro –

Novos Dados', Estadão, 07 October.

MichaelPage (2019) The World’s Most In Demand

Professions, Available at: https://www.michaelpage.co.

uk/minisite/most-in-demand-professions/ (Accessed:

30th December 2019).

Ng, S. C., Choy, S. O., Kwan, R., & Chan, S. F. (2005). A

Web-Based Environment to Improve Teaching and

Learning of Computer Programming in Distance

Education. Lecture Notes in Computer Science, 279–

290. doi:10.1007/11528043_28

Noschang, L. F., Pelz, F., & Raabe, A. (2014). Portugol

studio: Uma ide para iniciantes em programaçao. Anais

do CSBC/WEI, 535-545.

Nwana, H. S. (1990). Intelligent tutoring systems: an

overview. Artificial Intelligence Review, 4(4), 251-277.

Oliveira, C. M., Pimentel, A., & Krynski, E. M. (2015,

October). Estudo sobre o sequenciamento inteligente e

adaptativo de enunciados em programaçao de

computadores. In Anais dos Workshops do Congresso

Brasileiro de Informática na Educação (Vol. 4, No. 1,

p. 1320).

OMG Business Process Modeling Notation. "Version 1.0."

OMG Final Adopted Specification. OMG (2006).

Paiva, J. C., Leal, J. P., & Queirós, R. A. (2016). Enki.

Proceedings of the 2016 ACM Conference on

Innovation and Technology in Computer Science

Education - ITiCSE ’16. doi:10.1145/2899415.289

9441

Papert, Seymour. "A máquina das crianças." Porto Alegre:

Artmed (1994).

Panagiotis, S., Ioannis, P., Christos, G., & Achilles, K.

(2016). APLe: Agents for Personalized Learning in

Distance Learning. Computer Supported Education,

37–56. doi:10.1007/978-3-319-29585-5_3

Pattabiraman K. (2019) LinkedIn’s Most Promising Jobs of

2019, Available at: https://blog.linkedin.com/2019/

january/10/linkedins-most-promising-jobs-of-2019

(Accessed: 30th December 2019).

An Adaptive System Architecture Model for the Study of Logic and Programming with Learning Paths

689

Pelz, Fillipi Domingos. Correção automática de algoritmos

no ensino introdutório de programação. 2011. TCC

(graduação em Ciência da Computação) - Universidade

do Vale do Itajaí, Itajaí, 2011.

Philips, D. C. (1995). The Good, the Bad, and the Ugly. The

many Faces of Constructivism. I.

Phobun, P., & Vicheanpanya, J. (2010). Adaptive

intelligent tutoring systems for e-learning systems.

Procedia - Social and Behavioral Sciences, 2(2), 4064–

4069. doi:10.1016/j.sbspro.2010.03.641

Piaget, J. (1980). The psychogenesis of knowledge and its

epistemological significance. In M. Piatelli-Palmarini

(Ed.), Language and learning (pp. 23-34). Cambridge,

MA: Harvard University Press.

Raabe, A., Zanini, A. S., Santana, A. L. M., & Vieira, M. F.

V. (2016). Influência dos enunciados na resolução de

problemas de programação introdutória. Revista

Brasileira de Informática na Educação, 24(1), 66.

Raabe, A. L. A., & Silva, J. D. (2005). Um ambiente para

atendimento as dificuldades de aprendizagem de

algoritmos. In XIII Workshop de Educação em

Computação (WEI’2005). São Leopoldo, RS, Brasil

(Vol. 3, p. 5). sn.

Resnick, M., Maloney, J., Monroy-Hernández, A., Rusk,

N., Eastmond, E., Brennan, K., ... & Kafai, Y. B.

(2009). Scratch: Programming for all. Commun. Acm,

52(11), 60-67.

Romagosa i Carrasquer B. (2019) The Snap! Programming

System. In: Tatnall A. (eds) Encyclopedia of Education

and Information Technologies. Springer, Cham.

Santos, A., Gomes, A., & Mendes, A. (2013). A taxonomy

of exercises to support individual learning paths in

initial programming learning. 2013 IEEE Frontiers in

Education Conference (FIE). doi:10.1109/fie.2013.

6684794

Silva Filho, R. L. L., Motejunas, P. R., Hipólito, O., &

Lobo, M. B. C. M. (2007). A evasão no ensino superior

brasileiro. Cadernos de pesquisa, 37(132), 641-659.

Tarus, J. K., Niu, Z., & Mustafa, G. (2017). Knowledge-

based recommendation: a review of ontology-based

recommender systems for e-learning. Artificial

Intelligence Review, 50(1), 21–48. doi:10.1007/

s10462-017-9539-5

Trade Schools (2019) 31 High-Demand Jobs in 2019 for

Almost Every Type of Person, Available at:

https://www.trade-schools.net/articles/high-demand-

jobs.asp (Accessed: 30th December 2019).

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

for collaborative learning environment in an

educational habitat. 2016 International Conference on

ICT in Business Industry & Government (ICTBIG).

doi:10.1109/ictbig.2016.7892644

Weragama, D. S. (2013). Intelligent tutoring system for

learning PHP (Doctoral dissertation, Queensland

University of Technology).

Wolber, D. (2011). App inventor and real-world

motivation. Proceedings of the 42nd ACM Technical

Symposium on Computer Science Education -

SIGCSE’11. doi:10.1145/1953163.1953329

Yaghmaie, M., & Bahreininejad, A. (2011). A context-

aware adaptive learning system using agents. Expert

Systems with Applications, 38(4), 3280-3286.

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

690