Professional Documents
Culture Documents
Entertainment Computing: Harits Ar Rosyid, Matt Palmerlee, Ke Chen
Entertainment Computing: Harits Ar Rosyid, Matt Palmerlee, Ke Chen
Entertainment Computing
journal homepage: www.elsevier.com/locate/entcom
A R T I C L E I N F O A B S T R A C T
Keywords: The ultimate goals of serious educational games (SEG) are to facilitate learning and maximizing enjoyment
Serious educational game during playing SEGs. In SEG development, there are normally two spaces to be taken into account: knowledge
Serious game development space regarding learning materials and content space regarding games to be used to convey learning materials.
Learning material deployment How to deploy the learning materials seamlessly and effectively into game content becomes one of the most
Game content generation
challenging problems in SEG development. Unlike previous work where experts in education have to be involved
Chem Dungeon
heavily, we proposed a novel approach that works toward minimizing the efforts of education experts in
User survey
mapping learning materials to content space. For a proof-of-concept, we apply the proposed approach in de-
veloping a SEG game, named Chem Dungeon, as a case study in order to demonstrate the effectiveness of our
proposed approach. This SEG game has been tested with a number of users, and the user survey suggests our
method works reasonably well.
⁎
Corresponding author.
E-mail addresses: [email protected] (H.A. Rosyid), [email protected] (M. Palmerlee), [email protected] (K. Chen).
https://doi.org/10.1016/j.entcom.2018.01.001
Received 15 March 2017; Received in revised form 31 December 2017; Accepted 1 January 2018
Available online 05 January 2018
1875-9521/ © 2018 Elsevier B.V. All rights reserved.
H.A. Rosyid et al. Entertainment Computing 26 (2018) 1–9
development [9,10] which diverts the purpose of an existing commer- limitation. The mapping task becomes difficult if one of the content
cial game for knowledge learning. This approach exploits the properties space is large and complex. Hence, we do not think this approach is
of existing commercial games which fit a learning process, e.g., a player extensible in SEG development.
has to learn game rules, objectives and strategy to succeed a game. Such Unlike the above approaches, our proposed SEG framework would
a typical learning process is also applicable in traditional education exploit the instructional resources and makes use of appropriate PCG
systems. As an alternative game development methodology, Procedural techniques towards minimizing the cost. Thus, our proposed framework
Content Generation (PCG) technique can generate game content auto- is expected to connect knowledge and game content spaces seamlessly
matically via algorithms using a random or pseudo-random process that in SEG development.
produces an unpredictable range of possible gameplays, for instance,
[11]. This will significantly lower the cost of game development. 3. Methodology
Moreover, the latest PCG work [11] suggests that a proper use of the
categorized game content may facilitate eliciting positive gameplay In this section, we propose an alternative framework for SEG de-
experience. Motivated by the previous works, our framework would velopment especially for addressing the mapping issue pertaining to
suggest making use of PCG and existing entertainment games in SEG two spaces. To accommodate that, the framework exploits learning
development (see Section 3.2 for details). In particular, we believe that resources and making use of the latest PCG techniques.
embedding annotated knowledge chunks into categorised game con- The advantage of structuring serious game content in two spaces of
tent/elements makes the mapping easier to accomplish. learning materials and game content provides a higher degree of con-
We summarise the main contributions of the work presented in this trol for the game generation. In the existing SEG approaches, however,
paper as follows: (a) we propose an alternative framework for effective education experts have to be the prominent force in the process of
and efficient SEG development; (b) under our proposed framework, we deploying learning materials into a SEG. Thus, an expert is expected to
develop a proof-of-concept SEG, Chem Dungeon, to demonstrate the deeply understand characteristics of learning materials and game con-
usefulness of our proposed framework; and (c) we test this SEG with tent according to their expertise in order to link the two spaces.
human players via user survey and statistical analysis. However, it becomes infeasible and unscalable in the presence of
The rest of the paper is organized as follows. Section 2 reviews the complex learning or game content space. Hence, game developers are
related works. Section 3 presents our SEG framework and Section 4 demanded to utilize the natural and inherent game elements to deal
describes our proof-of-concept SEG, Chem Dungeon. Section 5 and 6 with the knowledge deployment issue. This is feasible since sophisti-
reports user test analysis results and discussion, respectively. Finally, cated education resources are accessible easily and the PCG techniques
the last section concludes the research. allow for flexibly controlling game elements to embed knowledge
chunks. Thus, we believe that making use of learning resources and
2. Related work making use of the latest PCG techniques could slash the expense of SEG
development. Furthermore, given the semantic descriptions of those
In this section, we outline connections and main differences to re- content spaces, the developer can formulate different aspects between
levant SEG development approaches. them, which sparks a proper deployment.
As argued by Damir et al. [5] based on their interviews with edu- To address the issues mentioned above, we propose an alternative
cation experts, game developers and players who involve themselves in framework for SEG development as illustrated in Fig. 1. First, learning
SEG, it is crucial to have a seamless connection between knowledge and materials and game elements are in separate spaces. In one hand, an-
game content spaces in SEG development. Moreover, they further em- notation takes place to describe education materials naturally from the
phasize that two spaces must be controllable [5] to allow for gaining meta-data retrievable from reliable resources. Then, we need to estab-
the controllability in tailoring game elements that are likely affecting lish the strategy for presenting them to players, based on their retrieved
different kinds of the player’s experience such as learning, enjoyment, properties or using the corresponding learning resources. On the other
motivation, engagement and so on. In addition, it is suggested by hand, categorisation of game content space consists of a couple of steps.
Hussaan et al. [6] that there are three components in SEG. Apart from It starts with a difficulty categorisation which groups game content
learning and game resources, domain concept should be introduced to according to the level of challenge. Subsequently, within each of the
specify the relationships between learning materials. Specifically, it pre-defined content categories, e.g., difficulty levels, and given a
facilitates using learning resources to formulate strategies in carrying
out learning based on game resources. Nevertheless, this approach [6]
emphasizes that all of those components have to be taken care by
education experts via interactions with students or game players.
Gamification [12] is a typical SEG development approach that ex-
plicitly takes knowledge and game content spaces into account in de-
velopment. The basic idea underlying gamification is directly embed-
ding game elements (e.g., avatar, badges, levels and scores) into the
learning process. Doing so make students more actively engaged in the
learning process when they are situated in a game-like presentation of
the learning materials. In [12], education experts and game developers
handcraft the combination of the two spaces, which is laborious and
time-consuming. Similarly, Belloti et al. proposed an approach for
adaptive SG via building up the proper connection between knowledge
and game content spaces [13]. Their approach breaks down a serious
game into subsequent tasks by considering diversified connections be-
tween learning materials and game elements. Then, adaptation is car-
ried out by offering a proper task sequence to an individual player to
maximize their positive learning and positive affective experience [13].
However, the game design (in particular, the mapping between two
spaces) relies heavily on education experts. Hence, the development
Fig. 1. An alternative SEG development framework.
cost is often very high. Technically, such an approach is also subject to
2
H.A. Rosyid et al. Entertainment Computing 26 (2018) 1–9
number of education materials, clustering analysis is applied to group experience as well. In such a way, the player can sufficiently prepare
similar game content. Hence, the aspects underlying the descriptive him/herself to challenge the enemies.
learning materials and game elements can guide a developer to use Our method applies PCG in that it provides details of the game
their logic in formulating the mapping between learning materials and content in the parameters configurations. Given the large space for the
game content. The outcome is a SEG content library comprised of generated content, manually identifying the category for the content
playable games for learning. space becomes impractical. Therefore, we adopt a set of steps applied in
entertainment game’s PCG [11], including: difficulty categorisation
3.1. Knowledge space and similarity categorisation. “Difficulty” categorisation provides
games for players with different abilities for playing the game [17].
Knowledge space of a SEG refers to all the relevant materials Meanwhile, “similarity” categorisation benefits the space of the game
consisting of items to be learned by a player. Our framework considers content which provides abundant choices of games which support re-
the subjects in the low-level learning category. Commonly, learners petitive sessions of learning.
acquire this category of knowledge through recalling or repetitions. For Robert and Chen suggest that generating categorised game content
instance, alphabet learning (i.e. visual appearances, pronunciations and for players can facilitate positive affective experience [11] via a proof-
constructions) by repeatedly looking at, listening to and trying to write of-concept first-person shooter game. Prior, developers annotated some
them. Another example from a more advanced domain is vocabulary game examples and let the categorisation model learned from it. For
which maps words from another language to their meanings in the specifying difficulty levels, a developer can also adopt a rule-based
student’s mother tongue. These learning materials should be de- approach by taking into account a small number of game controlling
scriptive; thus, they enable (semi) automatic serious game develop- parameters. At this point, developers have to decide the threshold va-
ment. lues of those parameters to split content space into proper regions of
Bellotti [14] demonstrates an annotation technique for serious different difficulty levels. Consequently, content categorisation natu-
games’ tasks. However, the author employs experts to annotate sub- rally takes place with the specified difficulty levels. Nevertheless,
jective attributes. Again, there is no assurance whether this approach compared to the aim in [11], a different purpose of clustering analysis
can handle the growing size of learning materials. Especially for serious occurs here. Given the value of k as a total number of chunks of
games, where a large number of learning materials have to be recalled, knowledge, the analysis identifies k groups of similar game content for
such as biology terms, geographical items or languages. each education material. As such should prevent boredom developing
On the other hand, we argue that the ideal properties for learning when multiple repetitions of a game session required.
materials originate from descriptions provided by a reliable education
resource, teaching handout for instance, and the representation of the 3.3. Mapping between knowledge and game content spaces
knowledge (e.g., text, image, audio or video). Hence, we operate the
annotation based on the objective descriptions of the learning mate- Mapping between knowledge and game content spaces is the es-
rials with reduced involvement from experts. Accordingly, the devel- sential step that deploys each learning material into game content
oper selects the relevant properties/attributes. Given the available based on their underlying characteristics. Often, serious game designers
documented resources, an information retrieval technique – beyond our view knowledge units as the learning tasks in an educational game. In
scope – automates the annotation process. Then, another program our framework, we follow the same perspective to allow straightfor-
measures attributes of the knowledge representation as part of the an- ward mapping. Thus, the player has to address the learning task in a
notation process. For instance, the number of words of the text-based specific game mechanic, such as collection, match-making, destruction
learning material or the length of an educational video. Consequently, or text narration. In the developer’s window, the selection of one or
the education content space contains comprehensive detail to initiate more game mechanics from the existing game content is practical to
the strategy for delivering the learning materials. If no relationship handle. This will become the “container” of the learning mission. For
exists between education materials, an automated method (e.g., instance, the original Pac Man game requires the avatar to collect all
sorting) establishes the strategy based on the attribute values. the dots in the game. Meanwhile, the education version replaces these
Otherwise, a syllabus or a teaching handout provides the strategy ex- dots with answers to a specific learning-task mission, such as math
plicitly; thus, players will recall the knowledge accordingly. subjects in Number Muncher educational game [18].
One must ensure that this assignment promotes learning to the
3.2. Game content space subject of interest. Commonly, the game mechanics that directly lead to
the game’s goal are the candidates. We expect them to imply learning
Game content space of an educational game refers to all the tasks as the prime mission of the game. Thus, players will respect their
playable games generated by an entertainment game engine to facilitate learning experience to spotlight during game sessions. However, we do
the learning defined via knowledge space for a player. Game content is not rule out other existing game mechanics to become the container of
known as elements and objects of a game that the player interacts the learning task as long as they can promote knowledge acquisition
during a game session. For instance, the type of enemy controls the way and possess non-contrasting perspective with the learning goal. So, the
an enemy behaves in the game environment. In that, a non-player developer must have an adequate knowledge of the game mechanics
character (NPC) with artificial intelligence can search and traverse the and s/he must be able to identify their importance in the game session.
nearest route to fight the avatar. Another example is the type of weapon Referring to the underlying game mechanic(s), the mapping be-
in a First Person Shooter (FPS) game that determines the destructive tween learning units with game content may employ an arbitrary or
power to a target. Commonly, a level designer manually designs a sampling-based mapping. However, it may promote ineffective learning
limited number of game levels, each of them has a unique configuration for different players. In fact, learning in an educational game involves
of these game content. However, this approach is impractical when the various factors [19]. For instance, arbitrary mapping potentially assigns
game consists of abundant elements and parameter values. Fortunately, an uncomplicated recall materials with “difficult” games. Hence, a
we can randomly or pseudo-randomly generating these elements via novice player is struggling to play such games trying to overcome the
Procedural Content Generation (PCG). It generates various game con- challenges. This situation could hinder a player’s aim to recall the
tent based on practices or methods (e.g., [11,15,16]) that ensure a game learning material. In other words, using arbitrary or sample-based
content is playable for the players. For instance, in FPS, NPCs are better mapping can produce imbalanced outcomes for the players.
to be spawned further from the player’s starting point. Distributing Therefore, the mapping should follow specific conditions that pro-
them in various places of the game set will create a balanced gaming duce acceptable deployment of learning materials; thus, it
3
H.A. Rosyid et al. Entertainment Computing 26 (2018) 1–9
their lives. Each round contains a purchasing mode, one turn for the
player to defend and another for attacking the NPC. The purchasing
mode allows each player to buy atoms from the periodic table. An atom
has a price specified by the atomic number (e.g., Helium [He] with
atomic number 2 costs two Atom Bucks). On the first turn, one player
attacks with a single atom. The opposing player (defender) only see the
valence electron of the attacking atom. Thus, it earns a chance to ap-
point a number of atoms for defence. If the attacking atom creates a
chemical bond with one or more of the defender’s chosen atoms (a
successful defence), the defending player receives rewards composed of
a number of Atom Bucks and Energy Units. Otherwise, if there is no
known possible compound between the attacking element and any of
the defending elements, the attack is successful and the defending
player receives a penalty for those unbonded defending elements. In
fact, such a rule should discourage players from just defending with
every element they own each time. Meanwhile, regaining the unused
defending elements costs a decrease in energy. However, if the player
has insufficient energy, their health decreases in proportion to the
deficit. Once each turn ends, players earn a number of Atom Bucks to
Fig. 2. The Chem Dungeon layout.
allow them to spend on additional elements.
4
H.A. Rosyid et al. Entertainment Computing 26 (2018) 1–9
Victory. Otherwise, losing all lives or running out of time issues a Defeat. Table 1
There are some helpful hints for players to play the game. Each Difficulty categorisation rule set.
game aims to form one compound (repeatedly). Novice players can
Difficulty Enemy-type Total-enemy Total-bullets Maze
adopt a trial-and-error strategy and are fully aware not to lose all their
lives. Therefore, the player ought to actively read the text message at Easy 0 <4 Any
the top-centre position of the game which holds the latest result for the Medium 0 >3
1 <3
compound-forming attempt. Meanwhile, if only one life remains, a
Hard 1 <2
player can regain full lives by collecting a potion. Or, similarly, by
accumulating experience (XP) bars through accurate shots and hit weak
enemies. Once the XP reaches a full bar, one additional life replaces it. strategy in our case associates with structuring education materials in a
However, such an endeavour should consider the remaining bullets/ specific order based on the priority of attributes for sorting. Therefore,
ammunition and the 90-s time limit. These restrictions impede players based on recall priority, compounds are ordered based on total-types-of-
abusing such tactical practices merely for entertainment while dis- atom, total-atom, atom-1-number and atom-2-number, total-character-
regarding the goal of playing the game: memorising compounds’ atoms. symbol-1 and total-character-symbol-2, respectively. As a result, the ea-
All these game mechanics existed in the original rogue-like game. siest compound to remember is H2 (composed of two Hydrogen atoms)
Only the chosen game mechanics were converted to represent the and the hardest to recall is CaB6 (formed from one Calcium atom with
question (i.e. avatar’s atomic shield) and choices of the learning task six Boron atoms). Hence, the sorted compounds are then represented by
(i.e. coin collection). Additional elements were also added (i.e. the CompoundID attribute which has numeric values from 1 to 100.
learning-related information) to reinforce the knowledge acquisition
task.
The next subsection outlines the construction of Chem Dungeon via 4.3.2. Game content space: rogue-like maze
our framework. The game content space was constructed from an existing rogue-like
and maze game to confirm that it segregates from the learning mate-
4.3. Chem Dungeon: Game development rials. Henceforth, the categorization and mapping processes become
revealing for our demonstration. As an overview, generating game
Inspired by Chem Fight, we apply our proposed approach in Section elements using parameter values applies here which consist of maze-id
3 to develop a new SEG: Chem Dungeon. As a matter of fact, we use the (categorical), enemy-type (0: random-move enemy, and 1: smart
library of education materials and the basic rule (pairing atoms to enemy), total-enemy (1–5), total-bullets (1–5). By default, the game
create a compound) as the core of Chem Dungeon. Moreover, an ex- content space counts 48,600 different parameter configurations.
isting rogue-like game 4 is employed to represent the game content. In difficulty categorisation, three levels of challenges separate the
Given both spaces are available, the following subsections describe the game content. To our best knowledge, the parameters enemy-type and
process details. total-enemy distinguish the difficulty quite noticeably within the rule set
in Table 1. As a result, 22,365 of game content is categorised as Easy,
15,660 is Medium-level game content and 10,575 of content has a
4.3.1. Learning Materials: Chemical compounds
Hard difficulty level. Fig. 3 illustrates three different levels of difficulty.
The educational game has a purpose in promoting the memorization
The image on the left is identified as an Easy game. Due to this fact,
of chemical compounds for players. For this case study, there are 100
there is merely a single obstacle from one enemy which moves ran-
compounds composed of at least two atoms. The textual representation
domly, but the avatar can wander around the maze freely without very
informs a compound’s symbol, name and the atoms. For instance, 2
much concern being hit by the sole enemy. The image in the middle and
Hydrogen and 1 Oxygen construct an H2O which is known as the water
on the right are Medium and Hard difficulty levels, respectively. In a
compound. The single atom appears as a game object with a text-based
medium-difficulty game, the avatar can still move freely although there
atomic symbol, e.g., O, Ca, Cl. Meanwhile, if the compound comprises
are four enemies moving randomly. Thus, the game provides additional
of numerous atoms of the same type, it appears as a concatenation of
challenges for the player to avoid collision with these enemies. Mean-
strings between the total atom and symbol, such as 2O, 2H, 2Cl, 6B.
while, the game content with five Smart enemies demands a high level
According to Fig. 1, there are two general steps to proceed. First,
of tactical practice in decision-making because these enemies are cap-
given the periodic table data, attributes appointment operates ac-
able of traversing the shortest path to the avatar.
cording to the forming atoms and compound representation. Attributes
Our goal in the similarity categorisation is to provide a selection of
of the forming atoms (atom-1 and atom-2) include atom-1-number
similar game content for each learning material. To accommodate that,
(discrete),atom-2-number (discrete), total-types-of-atom (discrete) and
a clustering analysis builds (k = 100 ) clusters of similar game content
total-atom (discrete). Attri-butes associated with compound and atom
inside a difficulty group. Given that maze-id parameter does not de-
representations include: total-character-symbol-1 (discrete) and total-
scribe a maze explicitly, five numeric parameters provide details of the
character-symbol-2 (discrete). Subsequently, a computer program re-
corresponding maze. The details are measures of maze’s total-path, total-
trieves necessary data from the periodic table and measures the total
corners, total-intersections, total-deadend and complexity. Aside from this,
characters for the involving atoms, then, annotates the attributes au-
we are aware of some issues: (1) the large size of game content space,
tomatically. For instance CO2 contains one Carbon and two Oxygen
and (2) the dynamic size of the content space due to the previously
atoms. The annotated values of this compound are atom-1-number = 6
played games. Accordingly, our choice falls to Balanced Iterative
(C), atom-2-number = 8 (O), total-types-of-atom CO2 is 2 (1 C + 1 O), the
Reducing and Clustering using Hierarchies (BIRCH) which is fast and
total-atom is 3 (1 C + 2 O), total-character-symbol-1 and total-character-
flexible even with very large samples (details available in [20]).
symbol-2 are both 1.
For our case, configuring BIRCH with k = 100 and setting the
Second, with the fact that no correlations exist between compounds,
branching factor B = 2 constructs a binary tree of game content space.
the strategy of remembering them takes into account the properties. In
Subsequently, the BIRCH operates to search for an optimum threshold
fact, recalling them should be driven by the complexity of each com-
value T which identifies 100 clusters with the highest silhouette score
pound. In other words, the more complex the representation of a
as an evaluation measure. The result of BIRCH on game content space
compound, the more difficult it is to memorise. Accordingly, the
under normalised values attests to Low, Medium and High difficulty
groups using a threshold Tl = Tm = Th = 0.02 to reach the highest sil-
4
Available from http://www.kiwijs.org/. houette of 0.23, 0.2 and 0.23, respectively.
5
H.A. Rosyid et al. Entertainment Computing 26 (2018) 1–9
Overall, 300 clusters are identified and equally divided into three However, additional criteria ensure an appropriate mapping based on
difficulty levels which are ready for deployment with the educational our notion of possible learning conditions between simple versus com-
materials. plex compounds. We assume that the likelihood of failures to recall
complex compounds may be higher than the easier ones. Thus, a higher
4.3.3. Mapping: a rule-based approach number of games to support such repeated attempts may transpire for
Before carrying out the mapping, we manually identified, analysed learning complex compounds. As a consequence, a slight difference in
and assigned candidates of game mechanics that fit our SEG construc- the game elements for recurrence of memorization may accustom the
tion. We took such steps by assuming a learning task as a multiple- player to those games without the fear of boredom growing. Therefore,
choice question. Consider two groups of atoms (e.g., X and Y) that form the player may have a wider space for focusing on the learning goal.
a chemical compound. The question is one of the atom group (X) which Given these expectations, the cluster details resemble those aforemen-
appears on the avatar’s atomic shield. And group Y is among the tioned conditions including the quantity of game content (represents
choices. We spotted some game mechanics that may represent the the number of repetitions) and the sum of standard deviations of game
question-choices pair. Then, we sort them based on the interest level in content features (represents the variety of games) under non-normal-
the existing game content. ized parameter values. The following pseudo-code shows the deploy-
ment rules in practice.
1. Battle an enemy.
2. Coins collection. 1. ASSIGN education materials with string CompoundID Ej , where
3. Bullets collection for adding ammunition. j: 0,1,…,(n−1) (n is the learning materials size).
4. Potions collection to refresh the lives. 2. Within each cluster ( j: 0,1,…,(n−1) ) of each difficulty level
(i: 0,1,…(m−1) total difficulty levels): COUNT total games (N ij ), SUM
Based on our knowledge, fighting enemies to combine their atoms has the standard deviations of parameters (Sij ) and ASSIGN the game
a conflict of purposes between the game mission and the learning content with string ID Gij .
process. In one hand, the fighting scenario performs destructive actions 3. SORT clusters within each difficulty level based on the value of N
to be undertaken. In contrast, the avatar collects the correct atom(s) (ascending) and S (descending), respectively.
that construct a chemical compound. To some extent, the first game 4. Create PAIRS of [Ej,Gij], where j: 0…(n−1) and i: 0…(m−1) , enabling
mechanic fails to meet our goal. Meanwhile, the remaining three game an education material gets a cluster of game content from each
mechanics have purposes that potentially promote rote learning. difficulty level.
However, the collecting-bullet interaction is considered less important
because it is the prerequisite for shooting obstacles. Similarly, the po- Mapping priority starts with the number of games in a cluster and is
tion collection is not convincing as well. Because there is an alternative followed by the standard deviation of the cluster. Fig. 4 depicts result
to resuscitating lives by accumulating XP level to the maximum value. details regarding the number of games for each compound and Fig. 5
This game attribute is achievable via shooting obstacles, killing enemies shows deployment details with respect to variations of the game con-
or collecting coins. Alternatives to reach one goal are often introduced tent. SQL-based tables store the mapping result and details of both
in various games to engage strategic actions. Thus, we prefer to keep content spaces.
potion-collecting as is. But other researchers or developers may alter its
functionality as a hint towards the correct coin-atom to be formulated
with the avatar’s atom. In our case, the fittest game mechanic to en-
dorse rote learning is the coins collection. In the SEG, we transform coins
as atom objects that interact with the avatar’s atomic shield. Moreover,
the abundance of atom objects can serve the repetitions required for
reinforcing the rote learning. Meanwhile, the enemies and incorrect
atom objects are the obstacles of the SEG.
Previous steps categorised the game content and learning materials
for the mapping process. The game content clusters carry details in-
cluding total game content, the linear-sum of each parameter, the sum-
of-squares of each parameter and the centroid of each cluster. These
statistics can serve as the game content’s description.
Our deployment strategy operates a rule-based method. Given the
specifications found in learning material and content space, the fol-
lowing crucial rule applies: a compound deploys into three unique
content clusters, all from different difficulty levels. Indeed, this rule
Fig. 4. Mapping result in terms of number of games in a cluster.
ascertains no duplicates of game content for multiple compounds.
6
H.A. Rosyid et al. Entertainment Computing 26 (2018) 1–9
Finally, the game engine generates the newly selected game composed of
Fig. 5. Mapping result in terms of the sum of standard deviation of game content para-
meters in a cluster.
parameters incorporating the CompoundID (based on the value of E) and
the value of V which associates the game content features.
5. User test
7
H.A. Rosyid et al. Entertainment Computing 26 (2018) 1–9
Table 2 Table 4
Z-test on proportion of gained enjoyment. Survey results: learning outcome vs. affective experience.
Indicators Ha: π ≠ 0.5 Ha: π > 0.5 Ha: π < 0.5 Not fun 42 65
p-value 0.00000 0.00000 1 Fun 48 154
99% Conf. intervals 0.59–0.74 0.6–1.0 N/A
H0 status Rejected Rejected Rejection failed
the game sessions. It is evident from Table 4 that our SEG allows more
players to gain positive learning outcome and Fun together as there are
We observed the game log and found that there are slight differ- 154 out of 309 falling into this category. This clearly demonstrates that
ences between various gaming activities that separate the reported Fun the use of separate content spaces and a proper mapping proposed in
and Not Fun. We suspect that players interpret differently to such a our framework may lead to a SEG that fits all the characteristics de-
subjective experience. One player feels ’entertained’ if the game content scribed by Abt in 1970s [27]. He mentions that in a serious game,
fits his skill. Meanwhile, another player experiences an enjoyment “learning” may be primary but other experiences involved should not
when the game content is more difficult to conquer. This is a factor be overlooked. Furthermore, serious games involve learning and en-
among many others that different players could have various percep- tertainment dimension as a unity during game sessions [28–31]. Recent
tual/cognitive experience in response to the same stimuli. Moreover, research by Pavlas found that enjoyment arising from the playing ac-
the affective experience may change overtime or known as concept drift tivities may affect the learning of a player in a serious game [19]. While
[11]. Providing thorough questionnaires that accommodate various the learning in serious games is a primary objective that any players
aspects of enjoyment [25] can produce a consistent report. have to achieve, our work emphasizes the importance of enjoyment
Regarding the learning performance of the players, each question (entertainment). Overall, the experimental results reported above in-
item in an exam represented a learning material. Thus, pre and post- dicate that, to a great extent, our game content and rules may elicit
game exams produced binary values indicating prior knowledge and positive affective experience and many players gain such enjoyment
recalling results, respectively. The difference in scores between pre and when they engage in learning via game playing.
post-game exams produced three types of learning performances: un-
changed, improvement and decay. However, we only use the un-
changed and improved learning performances here. Because the nega- 6. Discussion
tive score (decay) likely originated from arbitrary answers or random
guess [26]. Therefore, we divided 309 reports involving not known It is worth stating that our current case study based on Chem Fight is
prior knowledge into 219 game sessions of “improvement” and 90 subject to limitation. Chem Fight is designed purely for SEG in which
sessions of “unchanged”. For this case, the same z-tests operate using the game mechanics strongly correlate the properties of atoms and
the same values for the null hypothesis and alternative hypotheses compounds. Nevertheless, the Chem Dungeon introduces a PCG-based
while π indicates the proportion of improved Learning. Table 3 sum- SEG which currently aiming at rote-learning. It presents repetition
marises three z-tests results. Given the 0.01 significance level, two z- strategy that can be applied directly to the game content. However,
tests reject the null hypothesis while 99% confident the proportion of there is an opportunity to apply our scheme for a more complex type of
Improved Learning reports (0.694) is greater than 0.5 proportion. knowledge, such that found in the understanding category. Our frame-
Furthermore, we investigate the recorded gaming activities corre- work also accepts some learning types under this category that includes
sponding to learning and not learning outcome and find gaming activities grouping, identification, recognition, selection or translation. Based on
seem correlated to “learning” outcome. In general, a game session the fact that they allow learning strategy similar with recalling, i.e.
where players recalled most of the education materials has more repetition. Meanwhile, learning methods (in the understanding cate-
gaming activities than a game session where players only recalled few gory) such as describe, discuss, explain and report are not suitable for
or no the education materials. In fact, the total time spent in reading the our development framework. They require learning strategies beyond
successfully collected compound corresponding to learning actions take repetition. In that, the game content should provide a proper platform
around 15 s on average. In contrast, the not learning actions always take to facilitate such a high order of learning. For instance, a narrative-
less than three seconds. Overall, the total actions in learning game based game and multi-player games are the potentials because a player
sessions over the not learning in-game activities have been doubled can convey or present his/her knowledge to other players. Additionally,
approximately. This is due to the fact that the goal of such an educa- the scenario generation (such as via narratives [32,33]) or meta-cog-
tional game is designed to collect as many correct atoms as possible (i.e. nitive learning support in the game [34] maybe helpful embedding a
reflects a “learning”). complex knowledge. Hence, the higher the complexity of learning
The statistical evidence in Table 2 confirms that the Chem Dungeon materials to be embedded into a game content space, the smaller ap-
game considered as successful from the players’ perspective regarding plicability of our framework to build the SEG.
their learning and enjoyment, which is consistent with Pavlas’ testi- Currently, our framework fits suitably with the existing action game
mony [19]. genre. Other game genres like action-adventure, adventure, logic (e.g.,
On the other hand, we also look into the relationship between puzzles or mazes) and trivia are also suitable because their gameplay
learning outcome and affective experience reported by the survey and game mechanics allow direct mapping with the rote-learning sub-
participants. Table 4 summarizes such information collected from all jects. It requires relatively the same portion of modifications to the
target game content. For instance, a platformer game such as Mario
Bros. can be modified to ask the player to collect atom items instead of
Table 3 power items. Then, Mario/Luigi must enter a warp-pipe to form che-
Z-test on proportion of improved learning.
mical compounds accordingly. We can also modify a puzzle game,
Z-test, H0 against: Tetris for instance, by transforming its puzzle objects into atom objects
(boxes) and whenever the adjacent atom boxes form a chemical com-
Indicators Ha: π ≠ 0.5 Ha: π > 0.5 Ha: π < 0.5 pound, a corresponding information pops up. Basically, our framework
p-value 0.00000 0.00000 1
accepts a wide range of game genres, especially, the games with simple
99% Conf. intervals 0.65–0.73 0.66–1.0 N/A
H0 status Rejected Rejected Rejection failed and easy-to-play mechanics. A game genre like Real Time Strategy
(RTS) may be too complex to be applied in our framework. In that, the
8
H.A. Rosyid et al. Entertainment Computing 26 (2018) 1–9
developer is taking a critical role in altering the current game me- References
chanics, rules or scenarios to allow mapping process.
Meanwhile, our method is focusing on single-player games. Indeed, [1] D. Michael, S. Chen, Serious Games: Games that Educate, Train, and Inform,
Thomson Course Technology, 2006.
the produced SEG was distributed to be played in a single-player mode
[2] J.P. Gee, Good video games and good learning, Phi Kappa Phi Forum, vol. 85, The
only. Typical players, especially Achievers and Killers types [35], Honor Society of Phi Kappa Phi, 2005, p. 33.
should prefer learning-playing in this way. We have not tested our [3] S. de Freitas, S. Jarvis, A framework for developing serious games to meet learner
needs, in: Interservice/Industry Training, Simulation and Education Conference.
framework in a multiplayer game setting. If we deploy the produced [4] R. Ibrahim, A. Jaafar, Educational games (eg) design framework: Combination of
SEG as a multiplayer game, we have to provide an additional feature in game design, pedagogy and content modeling, International Conference on
the SEG to give an impression of a multi-player game. The simplest to Electrical Engineering and Informatics, 2009. ICEEI’09, vol. 1, IEEE, 2009, pp.
293–298.
add is the player ranking based on their overall learning tasks [5] D. Ismailović, J. Haladjian, B. Köhler, D. Pagano, B. Brügge, Adaptive serious game
achievements, scores or other measurement methods. Alternatively, we development, 2012 2nd International Workshop on Games and Software
Engineering (GAS), IEEE, 2012, pp. 23–26.
can also apply a multiplayer game content into our framework. The [6] A.M. Hussaan, K. Sehaba, A. Mille, Tailoring serious games with adaptive peda-
produced serious game will have to undergo through a higher degree of gogical scenarios: a serious game for persons with cognitive disabilities, in: ICALT,
modifications from the source game content. This is driven by the 2011, pp. 486–490.
[7] D. Djaouti, J. Alvarez, J.-P. Jessel, Classifying serious games: The G/P/S model, in:
added dimensionality of interactions involved in the game session, such P. Felicia (Ed.), Handbook of Research on Improving Learning and Motivation
as communication and competition between players. Notably, social- through Educational Games, IGI Global, 2011, pp. 118–136 (Ch. 6).
type and exploration-type players are the hardcore-fan of this typical [8] J. Alvarez, D. Djaouti, An introduction to serious game definitions and concepts,
Serious Games & Simulation for Risks Management (2011) 11.
serious game where they exploit those multiplayer-game features to [9] J.P. Gee, What video games have to teach us about learning and literacy, Comp.
engage themselves [35]. Thus, the developer has to do some creativity Entertain. (CIE) 1 (1) (2003) 20.
[10] D.W. Shaffer, How Computer Games Help Children Learn, Macmillan, 2006.
to ensure the optimal utilisation of multi-player facilities for players’ [11] J. Roberts, K. Chen, Learning-based procedural content generation, IEEE Trans.
learning. For instance, one can modify the scenario that encourages Comput. Intell. AI Game 7 (1) (2015) 88–101.
players using the chat room to discuss the solution of a learning task. [12] K.M. Kapp, The Gamification of Learning and Instruction: Game-Based Methods and
Strategies for Training and Education, John Wiley & Sons, 2012.
Sung et al. give the insight to apply collaborative learning in a multi- [13] F. Bellotti, R. Berta, A.D. Gloria, L. Primavera, Adaptive experience engine for
player SEG through what so-called knowledge engineering process that serious games, IEEE Trans. Comput. Intell. AI Game 1 (4) (2009) 264–280.
[14] F. Bellotti, R. Berta, A. De Gloria, L. Primavera, A task annotation model for
promotes a higher level of cognition [36].
sandbox serious games, IEEE Symposium on Computational Intelligence and
Games, 2009. CIG 2009, IEE, 2009, pp. 233–240.
[15] J. Togelius, G.N. Yannakakis, K.O. Stanley, C. Browne, Search-based procedural
7. Conclusion content generation: a taxonomy and survey, IEEE Trans. Comput. Intell. AI Game 3
(3) (2011) 172–186.
[16] G.N. Yannakakis, J. Togelius, Experience-driven procedural content generation,
We have successfully developed Chem Dungeon using a strategy IEEE Trans. Affect. Comput. 2 (3) (2011) 147–161.
that combines education materials and an entertainment game. [17] M. Csikszentmihalyi, Flow: The Psychology of Optimal Experience, Harper
Retrieving inherent details of the learning materials demonstrates ad- Perennial, USA, 1990.
[18] M.E.C. Consortium, Number Munchers Game < https://archive.org/details/
vantages in two regards. First, the learning materials has a natural Number_Munchers_v1.2_1986_MECC > (cited 3 November 2017).
description held by the attributes enabling a developer to organize [19] D. Pavlas, A Model of Flow and Play in Game-Based Learning: The Impact of Game
Characteristics, Player Traits, and Player States, Ph.D. thesis, University of Central
them semantically. In the second, computer programs can auto-
Florida, 2010.
matically annotate those attributes with little interference from experts [20] T. Zhang, R. Ramakrishnan, M. Livny, Birch: an efficient data clustering method for
and with a concern merely for the learning materials’ size. On the other very large databases, ACM Sigmod Rec. vol. 25, ACM, 1996, pp. 103–114.
[21] C. Pedersen, J. Togelius, G.N. Yannakakis, Modeling player experience for content
hand, the procedurally generated game elements in our approach un- creation, IEEE Trans. Comput. Intell. AI Games 2 (1) (2010) 54–67.
lock another route towards rapid development of SEG in which cate- [22] F. Jäkel, F.A. Wichmann, Spatial four-alternative forced-choice method is the pre-
gorization becomes automated using a combination of rule-based ap- ferred psychophysical method for naïve observers, J. Vis. 6 (11) (2006) 13.
[23] M. Papastergiou, Digital game-based learning in high school computer science
proach and machine learning. Hence, those detailed descriptions education: Impact on educational effectiveness and student motivation, Comp.
underlying both content spaces facilitate a developer in establishing the Educ. 52 (1) (2009) 1–12.
[24] M. Manske, C. Conati, Modelling learning in an educational game, Proceedings of
mapping rules based on his/her knowledge. Besides, the two-space the 2005 Conference on Artificial Intelligence in Education: Supporting Learning
structure could be a baseline for further research in a procedural serious Through Intelligent and Socially Informed Technology, IOS Press, Amsterdam, The
game generation wherein attributions are concerning learning mate- Netherlands, 2005, pp. 411–418.
[25] F.-L. Fu, R.-C. Su, S.-C. Yu, Egameflow: a scale to measure learners enjoyment of e-
rials, game elements, and game rules. Moreover, the new educational learning games, Comp. Educ. 52 (1) (2009) 101–112.
game we have developed using our method has shown reasonable re- [26] L. Bao, Dynamics of Student Modeling: A Theory, Algorithms, and Application to
sults in supporting players’ learning and entertaining them. Quantum Mechanics, Ph.D. thesis, University of Maryland at College Park, 1999.
[27] C. Abt, Serious Games, Viking Press, 1970.
In our ongoing research, we are further developing the game to [28] L. Michaud, J. Alvarez, Serious Games: Advergaming, Edugaming, Training and
enable it to predict players’ experiences via gaming data, which would More, IDATE Consulting & Research.
[29] K.P. Jantke, Toward a taxonomy of game based learning, 2010 IEEE International
lead to a corresponding adaptation method for personalized learning in Conference on Progress in Informatics and Computing (PIC), vol. 2, IEEE, 2010, pp.
the SEG. In the future, we are also going to investigate our proposed 858–862.
approach to new SEG development by combining PCG-based or multi- [30] M. Zyda, From visual simulation to virtual reality to games, Computer 38 (9) (2005)
25–32.
player entertainment game platforms and different learning materials. [31] U. Ritterfeld, M. Cody, P. Vorderer, Serious Games: Mechanisms and Effects,
Routledge, 2009.
[32] J. Rowe, B. Mott, S. McQuiggan, J. Robison, S. Lee, J. Lester, Crystal island: a
Acknowledgement narrative-centered learning environment for eighth grade microbiology, in:
Workshop on Intelligent Educational Games at the 14th International Conference on
Artificial Intelligence in Education, Brighton, UK, 2009, pp. 11–20.
The authors would like to thank the action editor and anonymous [33] N. Peirce, O. Conlan, V. Wade, Adaptive educational games: Providing non-invasive
reviewers for their valuable comments that improve the presentation of personalised learning experiences, in: 2008 Second IEEE International Conference
on Digital Games and Intelligent Toys Based Education, 2008, pp. 28–35,
this manuscript. doi:http://dx.doi.org/10.1109/DIGITEL.2008.30.
[34] B. Kim, H. Park, Y. Baek, Not just fun, but serious strategies: using meta-cognitive
strategies in game-based learning, Comp. Educ. 52 (4) (2009) 800–810.
Appendix A. Supplementary material [35] R. Bartle, Hearts, clubs, diamonds, spades: players who suit muds, J. MUD Res. 1 (1)
(1996) 19.
[36] H.-Y. Sung, G.-J. Hwang, A collaborative game-based learning approach to im-
Supplementary data associated with this article can be found, in the proving students’ learning performance in science courses, Comp. Educ. 63 (2013)
online version, at http://dx.doi.org/10.1016/j.entcom.2018.01.001. 43–51.