Improving Play and Learning Style Adaptation in a Programming

Education Game

Renny S. N. Lindberg, Aziz Hasanov and Teemu H. Laine

Department of Software, Ajou University, Suwon, South Korea

Keywords:

Learning Styles, Play Styles, Educational Games, Adaptation, Programming Education.

Abstract:

The drive to teach programming to K-12 students has ampliﬁed in the past few years as several European

countries have added programming to their national curricula. Teaching programming is not simple as even

older students struggle with the topic. Educational games have been shown to increase motivation and learning

efﬁciency, and many games have been created to teach programming. Adaptation is a technique that could

improve these beneﬁts even further by personalizing the game to learners in a heterogeneous group. In this

study we presented Minerva, an adaptive programming education game designed for elementary school stu-

dents. The game uses Bartle’s Player Types and Honey and Mumford’s Learning Style Questionnaire to adapt

gameplay and learning content to match the player’s styles. We tested Minerva with 33 6th grade South Ko-

rean students using a post-test questionnaire, interviews, and a game log that was designed to keep track of

the students’ proﬁles and how Minerva adapted to them. Based on the results, we proposed how Minerva’s

adaptation system can be improved in the future. This paper can be of interest to anyone researching possible

uses of adaptivity in (programming) education games.

1 INTRODUCTION

Programmers are in great demand. A European Union

report stated that Europe will have up to 756 000

ICT job vacancies by 2020 (H

using et al., 2015).

South Korea is among the latest countries that have

decided to add programming education to their na-

tional curricula (Ministry of Education and Science,

2015). Several European countries have already up-

graded their K-12 curricula (Balanskat and Engel-

hardt, 2015), and non-proﬁt organizations, such as

Code.org, are pushing for similar reforms in the U.S.

Various tools, such as Scratch, are used to teach

programming to children. However, the need to teach

the core programming concepts still remains, and

with programming being a challenging topic to teach

to even older students (Gomes and Mendes, 2007;

Jenkins, 2002), this is an issue not to be taken lightly.

Educational games can yield positive results and are

one possible solution (Connolly et al., 2012). Major-

ity of programming education games tend to be one-

tracked puzzle games, which some players may ﬁnd

difﬁcult to approach if the game style does not suit

them (Charles et al., 2005; Magerko, 2008). Using

adaptation could alleviate these issues and improve

motivation and learning (Hwang et al., 2012).

We developed a programming education game,

Minerva, that adapts learning content and gameplay

to the learning and play styles of the player, respec-

tively. In this paper, we describe Minerva’s adapta-

tion model and show the results of its evaluation with

6th grade students in South Korea. Finally, we give

suggestions to improve adaptation in games such as

Minerva.

2 BACKGROUND

2.1 Learning and Play Styles

We have previously compared several learning style

models (Lindberg and Laine, 2016) and selected

Honey and Mumford’s LSQ and Bartle’s Player Types

for the basis of adaptation in Minerva because of their

simple terminology and wide use in related ﬁelds.

Honey and Mumford identify four learning styles

as follows: (i) Activist, who learns from experience

and prefers acting over thinking; (ii) Reﬂector, who

learns by observing others; (iii) Theorist, who learns

by analyzing and prefer to theorizing over doing; and

(iv) Pragmatist, who learn by applying theory in prac-

450

Lindberg R., Hasanov A. and Laine T.

Improving Play and Learning Style Adaptation in a Programming Education Game.

DOI: 10.5220/0006350304500457

In Proceedings of the 9th International Conference on Computer Supported Education (CSEDU 2017), pages 450-457

ISBN: 978-989-758-239-4

tice. Though the learner may utilize all four styles,

they naturally prefer one over the others, which be-

comes the primary learning style.

Bartle’s Player types was created as a taxonomy of

MUD (Multi-User Dungeon) players. Since then, also

single player gamers have been labeled directly with

Bartle’s Player Types or with other models that are

based on Bartle’s. In his model, Bartle divided play-

ers into four styles: (i) Killer, who prefers acting upon

other players using dominance and aggression; (ii)

Socializer, who prefers communication over game-

play; (iii) Explorer, who prefers discovery (e.g. hid-

den locations and bugs) driven by curiosity; and (iv)

Achiever, who prefers winning by points, achieve-

ments, and collectibles in a competitive manner.

2.2 Adaptive Learning Environments

Adaptivity in learning environments is the process of

transforming the learning environment (e.g., learn-

ing materials and activities, user interface) to match

the learner’s context, which can comprise different

dimensions, such as personal context (e.g. learn-

ing style, knowledge level, emotional state) (Akbari

and Taghiyareh, 2014), technical context (e.g. device

type), and spatio-temporal context (Hsu et al., 2016).

Recent surveys analyzed the integration of learn-

ing styles into learning environments (Akbulut and

Suzan Cardak, 2012; Truong, 2015). Diverse learning

style models have been utilized in the adaptation pro-

cesses (Akbari and Taghiyareh, 2014; Yaghmaie and

Bahreininejad, 2011; Latham et al., 2012), and Ruiz

et al. (Ruiz et al., 2008) proposed two ways in which

learning styles can be used in adaptation: a) adapt-

ing learning material type, and b) adapting learning

content structure. Some recent learning environments

do adaptation in ways that hardly ﬁt to the aforemen-

tioned classiﬁcation. For example, Akbari et al. (Ak-

bari and Taghiyareh, 2014) proposed a recommender

system that ﬁnds a helper (another learner) by com-

paring and matching learning styles of the help seeker

and the helper candidate. In other studies, the se-

quence of learning content has been adapted based on

learning styles (Yaghmaie and Bahreininejad, 2011;

Cabada et al., 2011; Latham et al., 2012). Vesin et

al., in their Java tutoring environment, implemented

adaptation of presentation method and navigation ac-

cording to the learning style (Vesin et al., 2012). For

example, if the learner with visual learning style can-

not earn required grade for a speciﬁc concept, the

learning environment changes his learning style to

verbal. Moreover, navigation is adapted according to

sequential and global learning styles.

Learning environments detect learning styles by

versatile methods. Most systems rely on question-

naires (Akbari and Taghiyareh, 2014; Vesin et al.,

2012; Cabada et al., 2011), but others methods, such

as discourse analysis (Latham et al., 2012), also exist.

2.3 Adaptive Educational Games

Adaptation is used in games to make them more in-

triguing for a heterogeneous group of players. Chen

discussed the need for adaptive game design and

how games can maintain the ﬂow for different play-

ers (Chen, 2007). Chen suggested giving control to

players through choices that affect the ﬂow of the

game, but warns against giving too much freedom

that could be overwhelming or annoying. To over-

come this conundrum, Chen proposed: “offer adap-

tive choices, allowing different users to enjoy the ﬂow

in their own way; and embed choices inside the core

activities to ensure the ﬂow is never interrupted”.

Educational games have a heterogeneous audience

with likely occurrence of different gaming prefer-

ences. Using adaptation has been perceived as a pos-

sible method for alleviating this issue. For example,

Magerko’s take on adaptation in educational games is

to use knowledge with pedagogical and entertainment

values: “Games for learning can incorporate model-

ing approaches for both pedagogical and entertain-

ment goals. A game can be tailored to both better en-

gage as well as educate the player by using both kinds

of knowledge.” (Magerko, 2008) Moreover, Magerko

listed common adaptation techniques for games: a)

Branch story into multiple paths; b) Track the player

type and change the game according to the player’s

preferences; and c) Use Non-Player Characters (NPC)

that adapt their behavior to different player types.

Several adaptive educational games have been

proposed. Kickmeier et al. (Kickmeier-Rust et al.,

2011) introduced an adaptive storytelling framework

in which the story is adapted according to the player’s

the needs and interests. Similarly, Peirce et al. (Peirce

et al., 2008) utilized adaptation in their physics educa-

tion game to give motivational and hinting support to

the player. Hwang et al. (Hwang et al., 2012) used

the Felder-Silverman learning style model to adapt

the appearance of their game.

3 MINERVA

3.1 The Game Concept

Minerva is a programming education game for ele-

mentary school students with no previous program-

ming experience. The player is given a clear goal and

Improving Play and Learning Style Adaptation in a Programming Education Game

451

purpose through a simple story in which the player is

remotely commanding a robot sent to a derelict space-

ship, Minerva, to repair it back to working order. The

game covers ﬁve core programming concepts: input,

output, math, repetition and decisions. Minerva does

not have separate puzzles for Math, Input and Out-

put, but they are inherently part of the gameplay (i.e.

giving input commands to the robot and observing its

actions as the output) and covered in the game story.

Minerva has several features that set it apart from

other programming education games. Firstly, many

programming games are simple puzzles without a sto-

ryline. Secondly, puzzle games’ levels are generally

not connected to each other except by the increase in

difﬁculty. In Minerva, the player is free to traverse

the ship. Thirdly, the player can use a chat commu-

nicate with other players in the same room. Fourthly,

some rooms can be solved in several different ways:

the player can shoot obstacles and aliens blocking the

way; talk to the aliens and convince them to leave

the room; or simply maneuver the robot past the ob-

stacles. Fifthly, Minerva has minigames for teaching

programming concepts. These repair missions, which

thematically connect to the story, are short puzzles

that focus on a single programming concept. The

game has two such puzzles on repetition and deci-

sions. Lastly, Minerva adapts content and gameplay

to the player’s learning style and the play style.

3.2 Adaptation

In this section, we explain how a questionnaire for

mapping the play and learning styles was created and

how the adaptation is done within Minerva.

3.2.1 Questionnaire

A questionnaire was created previously to identify

children’s learning and play styles (Lindberg and

Laine, 2016). In Minerva, the questionnaire is part of

player account creation and it has two parts. The ﬁrst

part covers play styles based on Bartle’s Player Types,

and the second half is for learning styles according to

Honey and Mumford’s Learning Style Questionnaire.

Each style has four Likert scale statements, resulting

in 32 statements. If the player scores evenly for mul-

tiple styles, an additional statement for each equally

rated style is shown, of which the player must choose

one to be the primary style.

3.2.2 Adaptation to Play Styles

Play style adaptation affects the number of aliens

and the active view shown to the player. For exam-

ple, Killers are provided with the maximum number

of aliens, whereas Sozializers will see the minimum

number of them. There are three different auxiliary

views in Minerva: score, chat and map. The active

view depends on the detected play style: Achievers

see the score view, map is for Explorers and chat for

Socializers. There is no active view for Killers when

the game starts. The player can change the active view

at any time. View contents are identical to all players,

with the exception of the map view: for Explorers the

map only shows the areas that the player has visited.

This technique is known as the fog of war.

Minerva adjusts play styles by tracking players’

behavior. For instance, if a Socializer behaves like a

Killer, or an Achiever shows tendencies towards So-

cializer, the style is updated accordingly. The parame-

ters used for tracking play behavior are aliens and de-

structible obstacles. Aliens are used to track Killers’

and Socializers’ behavior. Firstly, the player can have

a dialog with aliens, which is appealing to Socializers.

The dialog offers hostile options, which could appeal

to Killers. Players who decide to talk to aliens and

convince them to leave the room peacefully, will get

their Socializer score increased, whereas a player who

destroys aliens will have their Killer score strength-

ened. This way the game adapt itself to play styles be-

yond the questionnaire results. Explorer is an excep-

tion to this as an Explorer’s behavior is not tracked.

3.2.3 Adaptation to Learning Styles

Minerva’s learning style adaptation is based on

Magoulas et al.’s adaptive navigation in a web-based

learning system (Magoulas et al., 2003), which uses

three adaptation levels: (i) Remember, which is as-

sociated with recalling the theory; (ii) Use, which re-

lates to applying the learned theory; and (iii) Find,

which is the ability to propose and solve original

problems. In Remember, learning content is divided

into three parts: 1) Question, 2) Example, and 3) The-

ory. Learning content is shown in different order per

learning style. Fourth part, Activity, belongs to the

Use level.

We simpliﬁed and combined Remember and Use

levels to make Magoulas et al.’s framework ﬁt the

game environment and target group. The Find level

was omitted because the game is does not require

higher level pedagogical skills. Learning content in

Minerva is divided into four parts with different me-

dia modalities and simple terms: (i) WHY: text that

includes learning content explaining how the current

CSEDU 2017 - 9th International Conference on Computer Supported Education

452

topic may be used outside the learning activity (e.g.,

a puzzle). This is equivalent to Theory; (ii) WHAT:

images that show the learning activity to the player

and what the player is expected to do. This is equiv-

alent to Question; (iii) HOW: a video clip that shows

the player how a similar learning activity is solved,

equivalent to Example; and (iv) DO: This is the ac-

tual learning activity, equivalent to Activity.

Table 1 presents Minerva’s adaptive learning con-

tent ordering model. This ordering adheres to

Magoulas et al.’s level Remember, with the exception

of Activist for which we placed DO after WHAT. This

was done to ensure that the puzzles’ goals are clear to

the player before solving them. Unlike play styles,

learning styles are not updated during gameplay.

Table 1: Adaptive ordering of learning content.

Activist WHAT DO HOW WHY

Reﬂector WHY HOW WHAT DO

Theorist WHAT WHY HOW DO

Pragmatist HOW WHY WHAT DO

4 EVALUATION

4.1 Research Design

Minerva was evaluated at a South Korean elementary

school with 32 grade 6 students (F: 15, M: 17). A

teacher and 12 students were also interviewed. Here

we focus on evaluating Minerva’s adaptation model.

In addition to utilizing the play and learning style

questionnaire, we created a post-test questionnaire

with Likert scale statements and open answer ques-

tions. The post-test questionnaire was ﬁlled right af-

ter gameplay. Semi-structured interviews for students

and their teacher were also prepared.

Minerva collects essential game data such as: play

and learning style proﬁles, room entering and exit

times, obstacles destroyed, aliens removed by talk-

ing or destroying, puzzle entering and exit times, total

moves performed, and game start/end times. Addi-

tionally, all chat messages are stored to a database.

4.2 Results

The results in this section are based on Minerva’s log

system with some results from the questionnaire and

interviews when they relate to adaptation.

4.2.1 Play Styles

Minerva has a simple internal point system that is

meant for adaptation handling. Shooting aliens, talk-

Table 2: Initial and ﬁnal play styles scoring by Minerva.

Achiever Killer Explorer Socializer

µ 4.64 3.15 3.03 4.97

Start σ 3.34 3.64 3.31 3.75

∑

153 104 100 164

µ 4.70 4.85 3.03 6.42

End σ 3.36 4.57 3.31 3.14

∑

155 160 100 212

ing to them, and destroying obstacles give points for

Killers, Socializers and Achievers, respectively. As

we mentioned before, Minerva currently does not

track adaptation towards Explorer.

Figure 1 shows play styles at the start and end

of the game. In the beginning, Socializer was the

most common primary play style and Achiever was

the second most popular. Secondary play style re-

sults were more even, with only minor leaning toward

Achiever. The least favorite style was Killer. It should

be noted that questionnaire results could all be given

negative results, in which case the least negative is

marked as the primary play style. Additionally, some

players do not have tertiary or quaternary play styles;

instead, they could have several equally strong play

styles grouped as the secondary style. For example,

a female player’s results were as follows: Achiever:

4p, Killer: -2p, Explorer: 0p, Socializer -8p. This

makes her primary play style Achiever, however, de-

spite been given 0p to it, the player’s secondary group

was Explorer.

In total, eight players’ styles changed during

gameplay. The changes were either to Socializer

(three) or to Killer (two), and three players’ primary

play styles were identiﬁed as combinations: Killer-

Socializer, Killer-Achiever and Achiever-Socializer.

Style changes are quantiﬁed in Table 2 with mean

(µ), standard deviation (σ) and sum(

∑

) of points per

style. The largest increase in points was for Killer,

with a total of 56p being added to players’ proﬁles by

the end of the game. Killers were followed closely

by Socializers with 48p. These are fairly substantial

point hikes compared to Achievers’ 2p.

Tendencies towards Socializer were observed dur-

ing the gameplay, with players talking together and

helping each other to solve rooms and puzzles. Fur-

thermore, several players chatted actively through the

provided chat system, though majority of the con-

versation were mainly emoticons and jokes. Some

played did ask for help over the chat system, but they

did not receive any help through the chat. Addition-

ally, one player identiﬁed as Socializer made a request

of just chatting, instead of playing the game:

Male player (Chat): “Hey guys, let’s chat

rather than beat this game”

Improving Play and Learning Style Adaptation in a Programming Education Game

453

Figure 1: Play styles of the players: A) at the start of the game B) at the end of the game.

A majority of chatting occurred in the ﬁrst and

ﬁnal rooms. There were also several messages left

in other rooms, but these were largely from a male

player who spammed seemingly random letters and

occasional emoticons. At the start of the game, this

player was identiﬁed as an Explorer, with only one

point difference to Socializer and Achiever as sec-

ondary styles. Generally, female players did not use

the chat as much as males for chatting; either they

left one or two comments, or simply asked for an ad-

vice. Out of 15 females who played the game, only

ﬁve wrote something on the chat, whereas 14 out of

17 male players partook the chat. Female players also

wrote fewer messages, though this is slightly skewed

due the substantial spamming done by a handful male

players, some of whom found it amusing:

Male player (Chat): “Spamming this text is

so fun”

One female player, who had asked for help in the

chat, suggested an improvement to the game:

Female player (Questionnaire): “It would be

nice if this game provided an easier tutorial”

At the beginning of the game, she was identiﬁed

as an Achiever with one point difference to Killer and

Socializer. By the end of the game, Socializer had

been set as her primary style and secondary style was

a combination of Killer and Achiever. These styles

are also supported by the player’s comment:

Female player (Questionnaire): “I had fun

when I talked with aliens and killed aliens”

Though only commented by a few students, the

chat system garnered only positive feedback:

Female player (Interview): “The chatting

system is so fresh, like an online game”

Male player (Questionnaire): “I liked the

chat”

Interacting with aliens was also positively re-

ceived by players, even if they only would talk to one

of the aliens and then proceed to destroy others. For

instance, the following comment was left by a player

who talked to an alien, destroyed the others, and had

been identiﬁed as a Killer from the start:

Female player (Questionnaire): “I like this

game when I talked with aliens, because their

appearance was so funny”

4.2.2 Learning Styles

Primary learning styles are identiﬁed from question-

naire results and cannot change during gameplay. Fig-

ure 2 shows that Activist was the most popular learn-

ing style with 14 designees. The results have been

calculated in a manner similar to play styles, meaning

that actual points do vary more, as we see in Table 3.

Activist and Reﬂector are closer to each other with a

difference of 19 points, which is substantial compared

to Theorist and Pragmatist.

Figure 2: Learning styles.

We observed that many players ignored tutorials

completely and just clicked to get to the game as

quickly as possible. This behavior supports the results

above suggesting that a majority of players leaned to-

wards the Activist style. Some negative feedback was

CSEDU 2017 - 9th International Conference on Computer Supported Education

454

Table 3: Learning styles scoring.

Activist Reﬂector Theorist Pragmatist

µ 2.82 2.42 1.45 1.82

σ 3.73 3.64 3.03 3.54

∑

93 80 48 60

towards the tutorials and learning content, for instance

one female player complained about lack of instruc-

tions for controlling the robot:

Female player (Questionnaire): “It was dif-

ﬁcult to understand the game tutorial [...] It

would be better if the tutorial has more expla-

nation about robot control, obstacles etc”

This was despite Minerva showing a set of images

and a video describing how the robot controls work.

Other students complained about instructions in gen-

eral, and lack of information about learning objectives

was pointed out by the teacher:

Female player (Questionnaire): “The in-

structions were hard to understand”

Male player (Questionnaire): “It would be

nice if the instructions explained in more de-

tail and precisely”

Female teacher (Interview): “The game was

excellent but to use it in a real programming

class it can be improved a bit. [...] the learn-

ing objective was not so clear. Core contents

and learning objective should be apparent.”

5 DISCUSSION

In the following sections, we discuss how Minerva

could be updated based on our evaluation results to

handle the play and learning style adaptation even

better. These ideas are applicable to other adaptive

games based on the same models.

5.1 Improving Play Style Adaptation

The initial play styles mirrored the players’ behavior

in most cases, as only eight players’ play styles were

changed during gameplay. Additionally, a large num-

ber of players had a close number of points in more

than one style. This was also indicated by comments

from the questionnaire where a few players stated

both enjoying talking and shooting aliens.

Minerva does not consider the relevance of sec-

ondary styles. For instance, if Achiever and Explorer

are both valued much higher than Killer and Social-

izer, it would be safe to assume that the player has

tendencies for both styles. Consequently, the game

should adapt its behavior to support both styles.

Minerva’s adaptation system is currently rather

simple, and it is unsure whether it sufﬁces in a longer

test scenario. Some of the originally planned adap-

tation functionalities were not fully realized simply

due to lack of time. For instance, Achievers are only

mildly supported, with only one room ﬁlled with bar-

rels that they could destroy to collect points. Some

simple modiﬁcations in future iterations will be di-

rected towards the Achiever class, such as random

destructible obstacles and dropping collectibles that

will give extra points to the player. Essentially, these

drops can happen with any play style, but if the player

keeps destroying barrels, the chance of gaining col-

lectibles would increase. Additionally, for more com-

petitive Achievers it would be appealing if each room

and puzzle would display the highest score that any

player has cleared it with, and scoreboard would dis-

play the total high scores. Lastly, by adding col-

lectible achievements, for instance clearing a room in

a speciﬁc way, would be motivating for Achievers.

When adapting a game for Explorers, extra game

content, such as hidden or visible objects and rooms,

must be generated. Another possible way to support

Explorers is to leave harmless bugs in the game that a

player can use to their beneﬁt. Killers and Socializers

are the easiest styles to handle, which could explain

why they were prominent player types at the end of

the game. In a future version, both will be supported

with diverse dialog paths for Socializer and stronger

feedback in form of audio and graphics for Killers

when they destroy aliens and obstacles.

Table 4 summarizes our play style adaptation im-

provement ideas. Additionally, the game world could

also change according to secondary styles. For exam-

ple, Socializers would normally have the minimum

number of NPCs and the default number of destruc-

tible obstacles. A Socializer with Achiever as sec-

ondary style would have the usual number of NPCs,

with a higher number of destructible obstacles and

elevated chance for collectibles. A Socializer hav-

ing Killer and Achiever as secondary styles would,

in turn, have the maximum number of NPCs.

5.2 Improving Learning Style

Adaptation

Although players complained about difﬁcult puzzles

and tutorial contents, it does not mean that the current

method of providing educational materials is fully at

fault. The players were understandably excited for

having a gap from their usual activities. We ob-

served their hurry to get to the game as several players

skipped most tutorials in the start of the game. Further

testing is required to see how well the learning content

Improving Play and Learning Style Adaptation in a Programming Education Game

455

Table 4: Play style adaptation improvements. Default value is between minimum and maximum numbers a room can handle.

Play Style User Interface Game World Destructible Obstacles NPCs

Neutral Unaltered Unaltered Default number; Default chance

for collectibles

Default number

Killer Unaltered Unaltered Default number; Increased

audio-visual effects; Default

chance for collectibles

Maximum number; In-

creased audio-visual ef-

fects

Achiever High-score shown;

Achievements collected;

Achievements available

Room high-

score shown;

Maximum number; Increased

chance for collectibles

Default number

Explorer Deliberate bugs that give

advantages

Hidden rooms

and objects;

Extra rooms

Default number; Default chance

for collectibles

Default number

Socializer Can initiate chat with any

online player

Unaltered Minimum number; Default

chance for collectibles

Minimum number;

More complex dialog

options

and the game are received in long-term. Nonetheless,

it is clear that there is room for improvement.

Because Minerva is an educational game, learn-

ing content provision must be done with care. Learn-

ing content should be spread so that it will not distract

gameplay. To achieve this, the usage of learning styles

is important, as well as the correct design of puzzles,

especially when the topics start to become more com-

plex. A puzzle itself should be presented without ad-

ditional text convey the topic at hand. In repetition

and decision puzzles, this was achieved with relative

ease, but with more complex topics maintaining the

same level of clarity can be challenging.

An unobtrusive method of learning material de-

livery is voice-over. This would particularly support

players who prefer listening over reading. Moreover,

Minerva already uses images in tutorial content, but

these were based on screenshots of the game. More

carefully crafted and visually pleasing tutorial images

could yield increased attentiveness from the players.

Minerva changes the content presentation order

depending on learning style. This can be improved by

considering more the needs of each style. Activists

could directly start the puzzle without any tutorials

shown, with an option to open the tutorial later. For

Reﬂectors, a larger change is recommended: record-

ing other players solving a similar puzzle and allow-

ing Reﬂectors to view a recording. For Theorists

and Pragmatists, the focus should be on how to make

learning materials more appealing. This could be

done for example by splitting materials into practical

and theoretical parts, or highlighting relevant parts.

For Activists and Reﬂectors, both materials could be

made available and Minerva could also track which

type of material the player prefers to see.

Table 5 lists the aforementioned improvement

ideas. Learning content is marked as: theoretical,

Table 5: Learning style adaptation improvements.

Learning

Style

User Interface Learning Con-

tent

Neutral View tutorials in any

order

Theoretical and

Practical content

highlighted

Activist On-demand tutorials Theoretical and

Practical content

highlighted

Reﬂector View a recording of

another player solv-

ing the same puzzle

Full content

Theorist Show tutorial before

puzzle

Theoretical con-

tent highlighted

Pragmatist Show tutorial before

puzzle

Practical content

highlighted

practical or the combination of the two. Theoretical

includes the theoretical aspects of the topic, and prac-

tical explains the theoretical aspects brieﬂy followed

by real-life examples. This type of division works bet-

ter with more complex topics; simpler concepts can

be presented with the same content for all styles.

5.3 Limitations

This study has several limitations that should be con-

sidered in future research: (i) the evaluation did not

measure the inﬂuence of adaptation to learning and

motivation; (ii) Minerva does not adjust all styles dy-

namically (i.e. Explorer, all learning styles), which

may cause imbalance and mismatch; (iii) currently

unknown inﬂuence of game design to learning/play

styles could help make design decisions that support

adaptation; and (iv) existing criticism against learning

and play styles should be accounted for.

CSEDU 2017 - 9th International Conference on Computer Supported Education

456

6 CONCLUSIONS

In this paper we introduced Minerva, an adaptive pro-

gramming education game that uses Bartle’s Player

Types and Honey and Mumford’s LSQ to adapt both

gameplay and learning content. We also showed the

results of adaptation evaluation obtained from test-

ing Minerva at a South Korean elementary school,

and proposed improvements for the adaptation model

based on evaluation results. There is room for im-

provement especially on how the learning content is

shown to players and on the adaptation of play and

learning styles.

ACKNOWLEDGMENTS

This work was supported by the Korean National Re-

search Foundation (NRF-2015R1C1A1A02036469).

REFERENCES

Akbari, F. and Taghiyareh, F. (2014). E-SoRS: A personal-

ized and social recommender service for E-learning

environments. In International Conference on e-

Learning and e-Teaching, pages 1–12.

Akbulut, Y. and Suzan Cardak, C. (2012). Adaptive edu-

cational hypermedia accommodating learning styles:

A content analysis of publications from 2000 to 2011.

Computers & Education, 58(2):835–842.

Balanskat, A. and Engelhardt, K. (2015). Computing our

future : Computer programming and coding Priorities,

school curricula and initiatives across Europe. Tech-

nical report, Brussels.

Cabada, R. Z., Barr

on Estrada, M. L., and Reyes Garc

ıa,

C. A. (2011). EDUCA: A web 2.0 authoring tool for

developing adaptive and intelligent tutoring systems

using a Kohonen network. Expert Systems with Appli-

cations, 38(8):9522–9529.

Charles, D., Kerr, A., and McNeill, M. (2005). Player-

centred game design: Player modelling and adaptive

digital games. In Proceedings of the digital games re-

search conference, volume 285, pages 285–298.

Chen, J. (2007). Flow in games (and everything else). Com-

munications of the ACM, 50(4):31.

Connolly, T. M., Boyle, E. A., Macarthur, E., Hainey, T.,

and Boyle, J. M. (2012). Computers & Education A

systematic literature review of empirical evidence on

computer games and serious games. Computers & Ed-

ucation, 59(2):661–686.

Gomes, A. and Mendes, A. J. (2007). Learning to program -

difﬁculties and solutions. In International Conference

on Engineering Education.

Hsu, T., Chiou, C., Tseng, J. C. R., and Hwang, G. (2016).

Development and Evaluation of an Active Learning

Support System for Context-Aware Ubiquitous Learn-

ing. Learning Technologies, IEEE Transactions on

Learning Technologies, 9(1):37–45.

using, T., Korte, W. B., and Dashja, E. (2015). e-Skills in

Europe: Trends and Forecasts for the European ICT

Professional and Digital Leadership Labour Markets

(2015-2020). Technical Report November, empir-

ica Gesellschaft f

ur Kommunikationsund Technolo-

gieforschung mbH, Bonn.

Hwang, G.-J., Sung, H.-Y., Hung, C.-M., Huang, I., and

Tsai, C.-C. (2012). Development of a personalized ed-

ucational computer game based on students’ learning

styles. Educational Technology Research and Devel-

opment, 60(4):623–638.

Jenkins, T. (2002). On the difﬁculty of learning to program.

In Annual Conference of the LTSN Centre for Infor-

mation and Computer Sciences, pages 53–58.

Kickmeier-Rust, M. D., Augustin, T., and Albert, D. (2011).

Personalized storytelling for educational computer

games. In Proceedings of the International Confer-

ence on Serious Games Development and Applica-

tions, pages 13–22.

Latham, A., Crockett, K., McLean, D., and Edmonds, B.

(2012). A conversational intelligent tutoring system

to automatically predict learning styles. Computers &

Education, 59(1):95–109.

Lindberg, R. S. N. and Laine, T. H. (2016). Detecting Play

and Learning Styles for Adaptive Educational Games.

In International Conference on Computer Supported

Education, volume 1, pages 181–189.

Magerko, B. (2008). Adaptation in Digital Games. Com-

puter, 41(6):87–89.

Magoulas, G., Papanikolaou, K., and Grigoriadou, M.

(2003). Adaptive web-based learning: accommo-

dating individual differences through system’s adap-

tation. British Journal of Educational Technology ,

34(4):511–527.

Ministry of Education and Science (2015). Sw . Technical

report.

Peirce, N., Conlan, O., and Wade, V. (2008). Adaptive ed-

ucational games: Providing non-invasive personalised

learning experiences. In International Conference on

Digital Game and Intelligent Toy Enhanced Learning,

pages 28–35.

Ruiz, M. d. P. P., D

ıaz, M. J. F., Soler, F. O., and P

erez, J.

R. P. (2008). Adaptation in current e-learning systems.

Computer Standards and Interfaces, 30(1-2):62–70.

Truong, H. M. (2015). Integrating learning styles and adap-

tive e-learning system: Current developments, prob-

lems and opportunities. Computers in Human Behav-

ior, 55:1185–1193.

Vesin, B., Ivanovi

c, M., Kla

snja-Mili

cevi

c, A., and Budi-

mac, Z. (2012). Protus 2.0: Ontology-based seman-

tic recommendation in programming tutoring system.

Expert Systems with Applications, 39(15):12229–

12246.

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

aware adaptive learning system using agents. Expert

Systems With Applications, 38(4):3280–3286.

Improving Play and Learning Style Adaptation in a Programming Education Game

457