di culty adjustment using player’s performance and ......in games from di erent genres, from...

The Journal of the Society for Art and Science, Vol. 18, No. 5, pp. 143 – 155 (2019)

Difficulty Adjustment Using Player’s Performance andElectroencephalographic Data

Henry Fernandez1) Koji Mikami2) Kunio Kondo3)

Graduate School of Media Science, Tokyo University of Technology

1)[email protected], 2)[email protected], 3)[email protected]

Abstract

For high skilled players, an easy game might become boring and for low skilled players, a difficult game might

become frustrating. The purpose of this research was to create new and better ways to offer players with different

skills, an appropriate experience. We focused on adapting the difficulty levels of a simple 2D platform game,

designing and building levels automatically. The proposed method consists of Dynamic Difficulty Adjustment

(DDA) and Rhythm-Group Theory (a procedural content generation method), combined with levels of attention

obtained from Electroencephalographic (EEG) data. Experiments were designed in the way that players had to

clear five different levels that were created automatically using the player’s performance and EEG data obtained

from a biosensor while playing. Results showed that the method successfully adapts the level difficulty according

to the player’s status. In addition, the designed method calculates difficulty using values computed in real time

to decide the shape and structure of the levels. The method designed in this research can be implemented not

only in platformers but also in other genres that involve elements of rhythm, additionally, it could be used

by game developers as a tool for playtesting in order to improve the game design, receive quantitative and

numerical feedback from players and create an overall better experience for their players.

– 143 –


1 Introduction

In the field of video games different types of play-ers exist, some players are more skilled than oth-ers. Although the case is particular for compet-itive games, we can see how in games where theplayers’ skill levels are not suitable for the diffi-culty of a game, for both types of players (highand low skilled players) the result might be anunsuitable experience [1].

Usually these problems exist as a consequenceof an unbalanced game design and can be solvedchanging the game rules or difficulty to make itappropriately challenging for players [2].

Dynamic Difficulty Adjustment (DDA) is onesuitable solution for this particular problem whendesigned well and it has proved to be successfulin the past [3]. However, the experience of play-ers can be ruined by a poor implementation whenthey realize that difficulty is being adjusted de-liberately [4]. In order to avoid this issue, wefocused on finding a complementary componentthat could support traditional difficulty adjust-ment, adding variety and better results.

The level of challenge in video games is one ofthe most relevant aspects that affect the playerexperience, however it’s not the only one [5]. Im-mersion plays a very important role determininghow the player experience is shaped in general [6]and the levels of attention of players while play-ing a game influence how immersion fluctuatesthrough time [7].

The word immersion in the video games field isused in a symbolic way for explaining the expe-rience of feeling surrounded by a different realityas if players were submerged in water, it’s a wayto refer to the sensation of a deep feeling, hav-ing all our senses focused on a specific reality, thevirtual reality [8].

Player experience is defined as the relationshipbetween the player and the game, the influencethat causes the game on the player while play-ing and the reactions triggered by that interac-tion [9]. The proper balance between frustration,challenge, and immersion, transforms into a goodplayer experience [10].

Prompted by the relationship between immer-sion and player experience, we decided to include

an Electroencephalography (EEG) component tomeasure attention values. This adds variety tothe adjustment, making it less predictable forplayers.

For the automatic creation of levels we usedRhythm-Group theory [11], a successful Proce-dural Content Generation method to constructlevels with a sense of rhythm for the player.

This article constitutes the result of an ex-tended and improved explanation of a work pre-sented at the NICOGRAPH International 2017conference in June 2017 [12].

2 Related Work

Dynamic Difficulty Adjustment improves theplayer experience in different ways [13, 14] andeven in its most basic or elemental form, whendone in the appropriate way, it could successfullyadapt the game, making the levels of challengemore suitable for players.

This method is useful and can be implementedin games from different genres, from racing games[1], to platform games [15], from multiplayer dig-ital games [16] to board games [17]; DynamicDifficulty Adjustment is a method that can beadapted to each developer’s or designer’s necessi-ties when using the proper approach.

For the particular case of this research, one ofthe most relevant studies is the work of MartinJennings-Teats et. al [15] which propose an ap-proach similar to ours; using DDA and Procedu-ral Content Generation, they created a personal-ized and structural experience for players. Ouroriginal contribution is the inclusion of the EEGcomponent to the previous approach.

In recent years there have been several of stud-ies related to BCI and games, mobile games [18]showing how useful these devices could be work-ing together with mobile devices; PC-based FPSgames [19] focused on the player experience andinteraction; and even for conducting research re-lated to how players learn to play [20], etc. Thesedevices, specifically EEG-based devices, have alsoproved to be useful for making adaptive games[21], which is particularly our field of study.

– 144 –


Figure 1: General design of our approach

EEG biosensors, are suitable to be used forcognitive games [22]; in multiplayer cooperativegames [23], for evaluating how the cognitive ac-tivity changes while playing with teammates; alsoin serious games, in order to promote rehabilita-tion with patients that suffer from motor deficitsdue to a stroke, these researchers developed a newgame that aims to help them with the process ofrehabilitation [24].

Neurosky Mindwave Mobile device [25], whichhas shown positive results in reliability and com-mercial contexts, was used to capture the atten-tion data from players [26]. In addition, a re-search particularly focused on video games hasshown that the device accurately reads values ofattention from players [27].

Previous researchers used a similar approachcombining DDA and BCI to improve performancewhile doing a specific task [28]. Our researchdiffers from theirs in different aspects: we useEEG and they use fNIRS; our field of test wasgames, theirs was general tasks’ performance; re-sults show that performance was improved by de-tecting boredom, in our case, we didn’t only im-prove performance but also adapted the difficultydepending on the players’ skills to achieve a moresuitable experience.

Procedural Content Generation (PCG) meth-ods are a good alternative to automatically mod-ify the level design according to how players playin a game [29].

In previous research the effectiveness of PCGwas demonstrated using games [30]. GeorgiosN. Yannakakis and Julian Togelius introduced aframework for procedural content generation ap-plied with computational models of user expe-

rience, they created a method for developers totrigger specific experiences depending on the userdecisions or status inside the game.

3 Method

Our method consists of the combination of threedifferent elements: Dynamic Difficulty Adjust-ment, Rhythm-Group Theory and Brain Com-puter Interface. A general flow of our approachcan be seen in figure 1.

3.1 Dynamic Difficulty Adjustment

Dynamic Difficulty Adjustment (DDA), is theprocess of changing game elements automaticallyin real-time, based on the player’s performance,in order to adapt the game to each player andavoid frustration or boredom [31].

We calculate the difficulty using the numbers ofthreats present in a level, specifically the numberand width of gaps, number of enemies in eachplatform and the type of these enemies (beatableor unbeatable).

Difficulty is adapted according to performanceand attention levels calculated by the EEG devicewhile playing.

3.2 Rhythm-Group Theory

A Rhythm-based method for 2D platform gamesis a type of technique for automatic level creationin which rhythm is what the player feels with hishands while playing [11]. This method is a toolfor the developer to create levels with a sense ofrhythm, levels that are playable and have a natu-ral feeling for players when it comes to level shapeand experience. We are using only a part of themethod created by Gillian Smith et al., here wedescribe the parts of the method used for ourresearch, to read the details about the originalmethod, see [11].

3.3 Brain Computer Interface

Brain Computer Interface or BCI [32] systems arebased on obtaining electroencephalogram (EEG)data, extracting relevant information to translate

– 145 –


it into commands to be read by software or par-ticular applications [33].

For input extraction we are using NeuroskyMindwave Mobile, a bioensor that digitizes braindata into concise inputs for developers to be ableto interact with it in real time, using a set of pre-designed algorithms API to monitor brain activ-ity.

For this research we use the eSense value Atten-tion, which is a value between 0 and 100 (being 0the least focused and 100 the most focused) thatdescribes the levels of attention of the user in realtime.

4 Implementation

The game is a side-scrolling 2D platformer inwhich the player has to reach a goal placed onthe right-most part of the level, very similar toSuper Mario Bros [34]. The player is required toovercome simple challenges: gaps between plat-forms, beatable enemies and unbeatable enemies,both types of enemies static. In the beginning ofeach level players get two bars of health, once theplayer loses both, has start from the beginning ofthe level, in addition, the game time is shown ontop-center of the screen.

4.1 EEG Data

According to the documentation [25], values from80 to 100 are considered elevated ; values from 60to 80 are considered slightly elevated ; values from40 to 60 is neutral ; values from 20 to 40 reducedand values from 1 to 20 strongly lowered.

By default, the device outputs data once a sec-ond, it means, we get as many values as the timethe player plays a level. We calculate the averageof all values obtained in a level, the calculation isshown in equation (1).

If a player is concentrated in a task, would per-form better than if concentration levels were poor[35]. Considering this, when attention values arehigh, it means the player is concentrated in thegame, it also means should perform better so wedecided to add a higher level of challenge whenthe player is focused, in contrast, if the player isnot focused, we reduce the level of difficulty.

A =1

n

n∑i=0

ai (1)

Arithmetic Mean (A): ai is the attention cal-culated by the device per second and n is thenumber of seconds that the player takes to finisha level. The interval time for this equation de-pends on how long each player takes to completea level.

4.2 Player’s Performance

To calculate the player’s performance we consid-ered two parameters: number of deaths or hitsby an enemy and gameplay time (time from startto end). Low values for number of hits and playtime result in a high performance, on the oppo-site case, high values for these parameters resultin low performance.

P =1

1 + h + dX1 +

g

eX2 (2)

Performance (P ): d is the number of deaths;h is the number of times hit by an enemy; g isthe gameplay time; e is the expected completiontime and X1, X2 are weights that represent theinfluence of each term in the final calculation.

The term 11+h+d is calculated by cross-

multiplication, the number of times hit or deathsis inversely proportional to how well players areplaying, we sum 1 to avoid division by zero.The term g

e is calculated by the same cross-multiplication principle, when players take moretime to complete a level, performance decreases.

4.3 Dynamic Difficulty Adjustment

We used the attention value calculated in equa-tion (1) and the player’s performance value cal-culated in equation (2) and combined them inequation (3) to get a general a global value thatinvolves both parameters. These two values cancomplement each other and affect the final calcu-lation. Weight for this equation were both set to0.5, same amount of influence for both parame-ters, attention and performance.

G = PW1 + AW2 (3)

– 146 –


Combination of performance and attention(G): P is the performance calculated in equa-tion (2); A is the attention average calculated inequation (1) and W1,W2 are weights that repre-sent the influence of each term on the final calcu-lation.

At =1

nt

nt∑i=0

ati (4)

Arithmetic Mean (At) for specific elements oftype t: ati is the attention obtained from the de-vice when the player interacts with elements oftype t ; nt is the number of times the playerinteracts with elements of type t and t = [lowenemy, medium enemy, high enemy, gap]. Theinterval time for this equation depends on howmany times the player interacts with elements oftype t.

Pt =1

1 + (ht + dt)(5)

Performance per type (Pt): ht is the number oftimes the player has been hit by enemies of typet; dt is the number of times a player has died dueto enemies of type t; t = [low enemy, mediumenemy, high enemy, gap].

The term 11+(ht+dt)

is calculated by cross-multiplication, the number of times hit or deathsis inversely proportional to how well players areplaying, we sum 1 to avoid division by zero.

Equation (6) shows how to calculate the globalvalue to decide how many elements of each kindare included in the level. Equation (5) shows theperformance calculation for a particular type ofelement and the calculation for attention values ofa particular type of element is the same as equa-tion (1) but instead of using all values, we cal-culated how attention values behaved when theplayer interacted with elements of that kind.

For example, low jump is one of the elementtypes; to decide how many elements (percentageof occurrence) we assigned to low jump actions,when the player dies due to an element of typelow jump, we reduce the number of low jump typeelements in the next level, it means, we are trying

to add elements that increase the player’s perfor-mance. In the case of attention, if the calculatedattention was registered high for low jump typeelements, we increase the number of this type ofelements to increase get better results in the nextlevel.

There is a compensation between both values,performance and attention, that work togetherto calculate the final percentage of occurrence ofeach element, it really adds variation to the game-play and the experience in general.

Gt = PtZ1 + AtZ2 (6)

Combination of performance and attention forelements of type t (Gt): Pt is the performancecalculated in equation (5); At is the attention av-erage calculated in equation (4) and Z1, Z2 areweights that represent the influence of each termon the final calculation.

4.4 Rhythm-Group System

A set of parameters are important to take intoconsideration to construct a rhythm, these are:rhythm type, rhythm density, action types, num-ber of actions. For our research we used the fol-lowing values:

• Action Type: we chose the simplest ones runand jump. For the action jump, there arethree different types: low, medium and high

• Rhythm Type: we are using a regular type ofrhythm which means that actions are evenlydistributed in the rhythm, other types of ac-tions are random or swing.

• Rhythm Density: number of actions in arhythm, we choose this depending on howthe player results are, the better the result,the higher the density. The minimum valueis 5 actions and maximum value is 50 actions.

• Rhythm Length: this is how long the game-play time should be according to the levellength (horizontally), this is decided depend-ing on the player results, the better the re-sults, the longer the level. The minimumvalue is 5 seconds, the maximum value is 30seconds.

– 147 –


Figure 2 shows an example of rhythms that oursystem can create.

Figure 2: Rhythm Representation

To calculate the number of elements for eachrhythm’s value, we defined a difficulty functionthat is shown in equation (7).

D1(p0, p1) =

{D0 + SG if p1 ≥ p0,

D0 − SG if p1 < p0(7)

Difficulty (D1(p0, p1)): p0 is the performancecalculated in the previous level with equation (5);p1 is the performance calculated in the currentlevel with equation (5); D0 is the difficulty of theprevious level; S is a constant value to make thechange between level and level smoother; G iscalculated with equation (3).

The difficulty for the current level is calcu-lated using the difficulty of the previous level plusa variation of the global value. This variationcan take positive or negative values depending onwhether we make the level more difficult or easier.The constant S represents a STEP value definedto make smooth changes of difficulty between lev-els so players do not feel an abrupt change.

For our implementation, the value S was setto 0.125 (calculated empirically after testing withother values). The value v is 1 if the differencebetween the performance for the current level andthe previous level is positive and -1 if the differ-ence is negative.

E1 = E0 + D1M (8)

Rhythm density (E1): E0 is the rhythm den-sity calculated in the previous level; D1 is the

difficulty calculated with equation (7); M is themaximum value for rhythm density.

L1 = L0 + D1N (9)

Rhythm length (L1): L0 is the rhythm lengthcalculated in the previous level; D1 is the diffi-culty calculated with equation (7); N is the max-imum value for rhythm length.

The rhythm density and rhythm length are cal-culated using equation (8) and (9) respectively.Density and Length are directly proportional todifficulty.

We simplified the elements that could be builtby our system to the simplest elements in plat-formers, there are no special items or moving ene-mies, only the basic features to show the rhythm-group method working with the rest of our sys-tem.

We selected three types of challenges for eachjump type, it means, we have three types for lowjump, three for medium jump and three typesfor high jump, in total nine different elements.The first element is a gap, a separation betweenplatforms, if the player falls through a gap, dies;the second element is a spike, the player dies whentouching a spike; finally an enemy that can bebeaten by the player jumping on its head. Figure3 shows the geometry elements that the currentgeometry system can create.

Figure 3: Rhythm System Elements

Figure 4 shows a piece of the level result fora low performance player, it’s an easy level withnot so many enemies or gaps; on the other hand,we can see the result for a challenging level in fig-ure 5 which is different from the low performanceresult, with more enemies, gaps and challenges.

– 148 –


Figure 4: Low performance player: Few challenges, easier to complete.

Figure 5: High performance player: More challenges, more difficult to complete.

5 Experiments & Results

Players were asked to complete five levels createdautomatically by our method and they wore thebiosensor while playing to record their brain ac-tivity and adapt the difficulty of each level. Fig-ure 6 shows the process.

5.1 Players and Environment

25 people between 21 and 30 years old, 7 (28%)women and 18 (72%) men completed the experi-ment. Before starting, players were asked: ”Haveyou played Super Mario Bros before?” (A) and”Have you completed Super Mario Bros?” (B).

For question A, only 3 people (12%) answeredno, the rest (22 people, 88%) answered yes; forquestion B, 9 people (36%) said yes, the rest (16people, 64%) answered no.

The game was played on a 15inch widescreenmonitor of a Dell XPS LX502 laptop, at approx-

imately 60cm from the player, using an Xbox360gamepad , with the left thumbstick to move andA button to jump. Sound was played using thelaptop’s speakers at a volume of 25%.

Figure 6: Experiments Process

5.2 General Features

For the first level that players played, we set thedifficulty to 50%. Depending on the results of the

– 149 –


first level the second level was created to adapt itto the player’s results. Same process is repeateduntil players completed five levels.

5.3 Limitations

We faced some issues with the biosensor whenperforming the experiments. Sometimes the con-nection between the device and the computer(bluetooth) was not strong enough to establisha successful connection.

Besides connection issues, we found that afterplayers wear the device for more than 10 min.,they didn’t feel comfortable, which lead us to de-cided to keep the experiments short.

For some players the device just didn’t fit well,regardless its adjustable functionality, sometimesis not easy to fit comfortably in all types of play-ers.

The result of those players that experiencedproblems with the connection while playing, nui-sance or any kind of bad experience that couldaffect the results were excluded from the experi-ment.

In addition to Neurosky related limitations,we consider that this method would be suitablefor games that involve jumping as their mainmechanic and games that involve some kind ofrhythm in their core mechanics. Runners, 2D or3D, sidescroller platformers and rhythm gamesare among the types of games we consider thiswould be a good method for.

5.4 Analysis

The overall results for all players across time areshown in figure 7 (A). The graph shows the av-erage results for all 25 players, attention, perfor-mance, global value (attention & performance)and difficulty.

5.4.1 General Results

Comparing the behavior of the global value(green curve) and the difficulty (purple curve)from level to level, we can see how in the endof the experiment both curves get close to eachother, an increasing difficulty higher than the

global value. This means that the method ismatching the difficulty to the player’s skills.

In addition there is a balance between atten-tion (blue curve) and performance (red curve),the method combines both values and makes surethat both of them contribute with the final cal-culation. For example we can see the result fromlevel 2 to level 3, performance decreases and theattention value increases, the result (green curve)is a stable value, which turns into an increase ofdifficulty, keeping the pace and attention for play-ers.

Difficulty increases, comparing level 1 and level5, showing that the level of challenge is changing,from 0.5 to 0.66, an overall increase of the dif-ficulty until gets closer to the global value, weestimate that difficulty in upcoming levels woulddecrease a little and then raise along with theglobal value. We have to test with more levels toconfirm this.

With the challenge increasing, we can also seehow the global value changes, in the end of theexperiment it ends up slightly higher than thebeginning, meaning that players perform betterwith a higher difficulty.

If we performed experiments including morelevels, we would expect that curves vary together,specially the global value and difficulty, the ex-pectation is to keep increasing smoothly and bothof them close to each other, demonstrating theadaptation process. We still have to make moreexperiments to confirm this.

5.4.2 Player Groups

In addition to the general results of our exper-iments, we separated and classified players’ re-sults by experience. The main reason to do thisis that these players have features in common andit makes it easier and more meaningful when an-alyzing the results.

Figure 8 shows the results for all grouped fea-tures. We know that the number of players fromgraph B is low however, we can see how the be-havior for both types of groups (A and B) in theend is similar, curves get closer in the end, whichmeans that for both types of players the algo-rithm is adapting the difficulty.

– 150 –


Figure 7: Experiments results: Including EEG (A), Excluding EEG (B). Each graph shows: Attention (blue

line); Performance (red line); Attention and Performance (green line); Difficulty (purple line)

One of the interesting things about this par-ticular group is we can see how the difficultybecomes lower to match players’ results and inthe end, curves get closer, reducing the gap whileplaying.

Comparing graphs C and D we can deduce thatplayers from group C are more experienced thanplayers from group D. In fact, by the overall per-formance of each group (73% for group C and66% for group D), we can assume this. We cansee that for players with different experience, thealgorithm is, step by step, adapting to changeand offer a suitable experience. Curves for bothgroups end up with a similar shape.

The difficulty for almost all groups adapts ata constant pace. Values of attention, difficultyand performance start at a particular point foreach graph and in the end of the experiment theyincrease, it means players are getting better at thegame and also challenge is increasing accordingly.

In general, for different types of players, withdifferent characteristics and different experiencewe can see how the method is adapting, perfor-mance and attention are adjusting values of dif-ficulty, the green and purple curve end close toeach other.

We expect that if we perform experiments withmore players and levels, the difficulty curve andglobals (att. perf.) curve will keep moving atthe same pace, ideally increasing with the player’sabilities.

5.4.3 Players’ Feedback

After playing each level of the game, we inter-viewed few players and collected feedback of theirexperience while playing. This section shows ashort sample of those interviews.

All players said after that the game was funto play however, they also mentioned that somelevels (the easiest ones) are too simple and theywould enjoy levels with more elements and chal-lenges.

As a recurring comment from high skilled play-ers, they all agree that difficult levels are more in-teresting than easy levels, in contrast, low skilledplayers felt better with easier levels but they alsosaid that a higher level of challenge would be in-teresting.

When we designed experiments for this re-search, we implemented a game with the mostbasic elements from a platformer, to adjust thegame and make it fit for the Rhythm-group the-ory parameters, in addition, graphics and in-game feedback are also very basic. Based on thefeedback from players and the results of the ex-periment we should increase the level of challengeof the game, adding variety and improving theimplementation of it.

5.4.4 EEG Influence

In order to validate the results obtained from ourapproach of combining player’s performance andEEG to adapt the difficulty and to be able toprove the value and novelty of including this new

– 151 –


Figure 8: Results for Groups of Players. Each graph shows: Attention (blue line); Performance (red line);

Attention and Performance (green line); Difficulty (purple line)

component, we conducted new experiments with-out the EEG data.

Using the same experiment layout shown in fig-ure 6 and with the same approach described infigure 1, we removed the biosensor data from thecalculation and gathered results from 29 players.Results from this experiment can be seen in figure7 (B).

Comparing the results for the approach withand without the EEG component, we can observesignificant differences.

As we explained before, the adaptation appearsfor the result with EEG at the end of the experi-ment, where difficulty and performance/attentionvalues are getting close to each other, in con-trast, we can see that for the experiment with-out EEG, these values are more separated, thisdemonstrates that the adaptation with EEG isbetter, in less time, was able to adjust the diffi-culty accordingly.

Results for the experiment without EEG showhow the adaptation occurs abruptly, between lev-els we can see that changes are not smooth. One

could argue that increasing or decreasing a con-stant in the calculation to soften these changes,the abrupt change problem would be solved, how-ever, this could make the game less interesting bylacking variety (levels would be too similar for toolong).

We can see a behavior for the experimentswithout EEG, show a pattern where perfor-mances decreases and increases in each level, thisis due to the difficulty being too high or too lowin the calculation; on the other hand we can seethe results for the approach with EEG, changesoccur smoothly and slower than without EEG.

6 Conclusions

We created a new method that is capable ofadapting the level of challenge for players depend-ing on their performance and degree of attention(using EEG data).

The adaptation matches different types ofplayer’s skills and status, not only experiencedplayers but also inexperienced players, people

– 152 –


with different characteristics. We also validatedour results by comparing the results with an ap-proach that didn’t include the EEG component,demonstrating that the inclusion of brain wavesrelated data can lead to better results.

Due to the biosensor issues and limitations,also the necessity of having the device for theprocess, we do not envision this method for com-mercial use yet, however, with a better perfor-mance and more comfortable device, this methodcould be implemented in other genres that involvejumping as a core mechanic or elements in whichthe rhythm group theory can help to add a senseof harmony to the game. Among the types ofgames we consider this method could be suitablefor are: endless runners, 2D or 3D side-scrollingplatformer or rhythm games

For game developers, this would be a goodplaytesting tool, this method would enable themto gather helpful information to create better lev-els. For instance, when designing a platformer,developers could create a base design and iterateon it dynamically using our method, evaluate andimprove the design.

As future work we plan to improve the designedmethod, testing with different EEG biosensors,planning and performing more experiments, test-ing with more players and modifying the initialvalues to compare which values adapt the best tothis approach. We also would like to include otherfactors on the difficulty calculation, so far we haveonly tested with number of elements in the level,we should also consider position and relationshipbetween elements, which describes another levelof difficulty.

In addition, we consider it would be meaning-ful to include new elements, other than difficultyto evaluate the player’s experience. The harmonythat gameplay, graphics and sounds constitute alltogether to shape experiences for players influ-ence how these players perceive the game in gen-eral. As future ideas, involving graphics, soundsand new elements from gameplay would increasethe value of this approach and its results.

For this research we didn’t consider a specificstudy on psychophysical elements and their in-fluence on the attention of players, for our nextsteps with this research we would like to address

this area too.

References

[1] Jared E. Cechanowicz, Carl Gutwin, andScott Bateman. Improving player balanc-ing in racing games. In Proceedings of CHIPLAY ’14, pages 47–56. ACM New York,NY, USA c©2014, October 2014.

[2] Eva Kraaijenbrink, Frank Gils, Quan Cheng,Robert Herk, and Elise Hoven. Balancingskills to optimize fun in interactive boardgames. In Proceedings of INTERACT ’09,pages 301–313. Springer-Verlag Berlin, Hei-delberg c©2009, August 2009.

[3] Hunick Robin. The case for dynamic dif-ficulty adjustment in games. In Proceedingsof ACE ’05, pages 429–433. ACM New York,NY, USA c©2005, June 2005.

[4] Andrew Rollings and Ernest. Adams. An-drew Rollings and Ernest Adams on GameDesign. New Riders, 201 West 103rd Street,Indianapolis, Indiana 46290 An Imprint ofPearson Education, 2003.

[5] Nacke Lennart and Lindley Craig A. Flowand immersion in first-person shooters: mea-suring the player’s gameplay experience. InProceedings of Future Play ’08, pages 81–88.ACM New York, NY, USA c©2008, Novem-ber 2008.

[6] Cox Anna L., Cairns Paul, Shah Pari, andCarroll Michael. Not doing but thinking:The role of challenge in the gaming expe-rience. In Proceedings of CHI ’12, pages 5–10. ACM New York, NY, USA c©2012, May2012.

[7] Nordin A. Imran, Ali Jaron, AnimashaunAishat, Asch Josh, Adams Josh, and CairnsPaul. Attention, time perception and immer-sion in games. In Proceedings of CHI EA ’13,pages 1089–1094. ACM New York, NY, USAc©2013, April 2013.

– 153 –


[8] Janet Murray. Hamlet on the Holodeck: TheFuture of Narrative in Cyberspace. Cam-bridge, MA: The MIT Press, 1997.

[9] Lennart E. Nacke, Anders Drachen, KaiKuikkaniemi, Joerg Niesenhaus, Hannu J.Korhonen, Wouter M. van den Hoogen,Karolien Poels, Wijnand A. IJsselsteijn, andYvonne A. W. de Kort. Playability andplayer experience research. In Proceedingsof DiGRA ’09, 2009.

[10] Christopher Perdersen, Julian Togelius, andGeorgios N. Yannakakis. Modeling player ex-perience for content creation. ComputationalIntelligence and AI in Games, IEEE Trans-actions, 2(1):54–67, 2010.

[11] Gillian Smith, Mike Treanor, Jim White-head, and Michael Mateas. Rhythm-basedlevel generation for 2d platformers. In Pro-ceedings of FDG ’09, pages 175–182. ACMNew York, NY, USA c©2009, April 2009.

[12] Henry D. Fernandez B., Koji Mikami, andKunio Kondo. Adaptable game experiencebased on player’s performance and eeg. InProceedings of NICOGRAPH International2017. IEEE, June 2017.

[13] Denisova Alena and Cairns Paul. Adaptationin digital games: The effect of challenge ad-justment on player performance and experi-ence. In Proceedings of CHI PLAY ’15, pages97–101. ACM New York, NY, USA c©2015,October 2015.

[14] Yun Chang, Trevino Philip, HoltkampWilliam, and Deng Zhigang. Pads: enhanc-ing gaming experience using profile-basedadaptive difficulty system. In Proceedings ofSandbox ’10, pages 31–36. ACM New York,NY, USA c©2010, July 2010.

[15] Martin Jennings-Teats, Gillian Smith, andNoah Wardrip-Fruin. Polymorph: Dynamicdifficulty adjustment through level genera-tion. In Proceedings of PCGames ’10. ACMNew York, NY, USA c©2010, June 2010.

[16] Alexander Baldwin, Daniel Johnson, andPeta A. Wyeth. The effect of multiplayerdynamic difficulty adjustment on the playerexperience of video games. In Proceedingsof CHI EA ’14, pages 1489–1494. ACM NewYork, NY, USA c©2014, April 2014.

[17] Gomez-Hicks Guillermo and KauchakDavid. Dynamic game difficulty balancingfor backgammon. In Proceedings of ACM-SE’11, pages 295–299. ACM New York, NY,USA c©2011, March 2011.

[18] Paul Coulton, Carlos Garcia Wylie, andWill Bamford. Brain interaction for mo-bile games. In Proceedings of MindTrek ’11,pages 37–44. ACM New York, NY, USAc©2011, September 2011.

[19] Kenneth Chan, Koji Mikami, and KunioKondo. Measuring interest in linear singleplayer fps games. In Proceedings of SA ’10.ACM New York, NY, USA c©2010, Decem-ber 2010.

[20] Rina R. Wehbe, Dennis L. Kappen, DavidRojas, Matthias Klauser, Bill Kapralos, andLennart E. Nacke. Eeg-based assessment ofvideo and in-game learning. In Proceedingsof CHI EA ’13, pages 667–672. ACM NewYork, NY, USA c©2013, April 2013.

[21] Fabien Lotte. Brain-computer interfaces for3d games: hype or hope? In Proceedings ofFDG ’11, pages 325–327. ACM New York,NY, USA c©2011, June 2011.

[22] Kyrgyzov Olexiy and Souloumiac Antoine.Adaptive eeg artifact rejection for cognitivegames. In Proceedings of ICMI ’12, pages567–570. ACM New York, NY, USA c©2010,October 2012.

[23] Johnson Daniel, Wyeth Peta, Clark Madi-son, and Watling Christopher. Cooperativegame play with avatars and agents: Differ-ences in brain activity and the experience ofplay. In Proceedings of CHI ’15, pages 3721–3730. ACM New York, NY, USA c©2015,April 2015.

– 154 –


[24] Hou Xiyuan and Sourina Olga. Emotion-enabled haptic-based serious game for poststroke rehabilitation. In Proceedings ofVRST ’13, pages 31–34. ACM New York,NY, USA c©2013, October 2013.

[25] Inc. NeuroSky. Neurosky.com, (2015).biosensors and algorithms — ecg — eeg— neurosky, 2015. [Online; accessed 26-January-2016].

[26] Grierson Mick and Kiefer Chris. Better braininterfacing for the masses: progress in event-related potential detection using commercialbrain computer interfaces. In Proceedings ofCHI EA ’11, pages 1681–1686. ACM NewYork, NY, USA c©2011, May 2011.

[27] Genaro Rebolledo-Mendez, Ian Dunwell,Erika A. Martinez-Miron, Maria Do-lores Vargas-Cerdan, Sara Freitas, FotisLiarokapis, and Alma R. Gacria-Gaona.Assessing neurosky’s usability to detectattention levels in an assessment exercise.In Proceedings of HCI Part I ’09, pages149–158. Springer-Verlag Berlin, Heidelbergc©2009, July 2009.

[28] Afergan Daniel, Peck Evan M., SoloveyErin T., Jenkins Andrew, Hincks Samuel W.,Brown Eli T., Chang Remco, and JacobRobert J.K. Dynamic difficulty using brainmetrics of workload. In Proceedings of CHI’14, pages 3797–3806. ACM New York, NY,USA c©2014, April 2014.

[29] Noor Shaker, Julian Togelius, and Mark J.Nelson. Procedural Content Generation inGames. Springer International Publishing,Switzerland, 2016.

[30] Julian Togelius and Georgios N. Yannakakis.Experience-driven procedural content gener-ation. IEEE Transactions on Affective Com-puting, 2(3):147–161, 2011.

[31] Wikipedia. Dynamic game difficulty bal-ancing, 2012. [Online; accessed 18-January-2016].

[32] Blanchard Gilles and Blankertz Benjamin.Bci competition 2003-data set iia: spa-tial patterns of self-controlled brain rhythmmodulations. Biomedical Engineering, IEEETransactions, 51(6):1062–1066, 2004.

[33] P. J. McCullagh, M. P. Ware, and G. Light-body. Brain computer interfaces for inclu-sion. In Proceedings of AH ’10. ACM NewYork, NY, USA c©2013, April 2010.

[34] Nintendo. Super Mario Bros, 1985.

[35] Adrian Moran and Shane M. Murphy. TheOxford Handbook of Sport and PerformancePsychology: Concentration: Attention andPerformance. Oxford University Press, 198Madison Avenue, New York, NY 10016,2012.

– 155 –

di culty adjustment using player’s performance and ......in games from di erent genres, from...

Documents