USING CASUAL GAMES AND ONTOLOGIES TO TRAIN
UNSKILLED WORKERS
An Experience with the FoodWeb2.0 Platform
Nils Malzahn, Sabrina Ziebarth and H. Ulrich Hoppe
Department of Computer Science and Applied Cognitive Science, University Duisburg-Essen, 47048 Duisburg,Germany
Keywords: Web2.0, Collaborative Learning, Games with a Purpose (GWAP), Ontology Enhancement.
Abstract: Training and continuing education in and for the food industry face the challenge of reaching and
motivating workers without or with only a low level of formal qualification. In the FoodWeb2.0 project a
platform with specific support for training opportunities and selected Web2.0 enabled courses has been
established. This paper reports on first results of a Web2.0 course conducted on the platform and elaborates
on the usage of a casual game for training knowledge on food safety and hazardous material regulations.
Since the tentative results show that the students are motivated to influence the content of a course by
commenting provided material, we designed the casual game in such a way that it can be used in two
modes: as a learning game and as an interactive mechanism for ontology enrichment. Our evaluation at a
private professional training academy has shown promising results.
1 INTRODUCTION
The food industry in Germany is characterized by a
high amount of workers without or with only a low
level of formal qualification. While these workers
can be easily recruited and trained to perform simple
and often physically exhaustive tasks, there is a lack
of skilled workers who are able to use and control
the complex machines and processes of the food
production industry. Thus, human resource
managers try to develop some of the unskilled
workers to a higher qualification level to close this
gap. Typical obstacles are language problems (often
German is not the first language) and a lack of
motivation and confidence to learn because of
various reasons, e.g., education is not seen as an
asset or their current work is so exhausting that they
are not ready to learn. The project FoodWeb2.0
(funded by the German Ministry of Research and
Education) aims at training the employees of the
German food industry using two basic strategies:
motivating employees for vocational training and
performing education in collaborative, blended
learning scenarios using Web2.0 technologies. One
of the applications combining both strategies is the
use of a collaborative casual game called
“Matchballs”.
In the subsequent sections of this paper we
present the general approach of the FoodWeb2.0
project, the tools set up to achieve the project goals
as well as first results of on-going courses using this
platform from a qualitative study conducted with 40
students. The paper continues with a description of
the Matchballs game, its underlying architecture and
ontology enrichment mechanisms. Afterwards we
present the results of an evaluation of the game
conducted with 18 students of a vocational training
school specialized in sweets production. The paper
concludes with a discussion of the results and an
outlook on future work.
2 THE FOODWEB2.0 APPROACH
The FoodWeb2.0 project focuses on professional
training in the food industry. These trainings are
characterized by students of heterogeneous
background and knowledge. Furthermore, there are
only a few training facilities in Germany offering
specialized training for the food industry. At the
moment, these courses are conducted at the training
institutions. This implies that students from all over
Germany have to travel to the training facilities. To
reduce travel costs most of the courses are quite
531
Malzahn N., Ziebarth S. and Hoppe H..
USING CASUAL GAMES AND ONTOLOGIES TO TRAIN UNSKILLED WORKERS - An Experience with the FoodWeb2.0 Platform.
DOI: 10.5220/0003914605310537
In Proceedings of the 4th International Conference on Computer Supported Education (SGoCSL-2012), pages 531-537
ISBN: 978-989-8565-07-5
Copyright
c
2012 SCITEPRESS (Science and Technology Publications, Lda.)
intensive, i.e. much information is covered in a very
short time. The courses are often organized in phases
of training at a training facility and phases in which
the students have to apply their newly acquired
knowledge at their particular enterprise setting. In
the latter phase they are often not supported by
training facilities, since they are not in touch with
their trainers. This leads to a difficult learning and
transfer situation.
The FoodWeb2.0 project aims at improving this
situation by providing an online platform where
students and trainers can be in touch before, during
and after a face-to-face training block. We use a
Liferay portal as basis for our platform. It offers
basic tools like wikis, blogs, forums, document
libraries etc., and a sophisticated role/permission
system that allows to offer courses of several
training facilities on a single platform without
compromising their security and data privacy.
The most challenging part of further education
and learning at work is the application of “book
knowledge” to the specific work situation (cf.
Baldwin & Ford, 1988; Burke & Hutchins, 2007).
Web2.0 provides adequate ways of supporting this
kind of learning. Instead of just presenting
knowledge like in web based trainings or (video)
pod casts explaining the content with a most often
artificial case study (at best), the students are
instructed to share their experience regarding the
lessons learned with the other students by providing
blog entries, wiki based learning or in the case of the
more craft-oriented courses by recording videos of
their own performance.
In addition to providing learners with the
opportunity to communicate with their trainers
during phases of work at their specific enterprise the
platform is also used to provide preparatory courses.
Especially in the area of qualification courses from
unlearned workers to skilled workers, there is a
diversity of age and educational background
implying a wide variety of learning skills and
background knowledge. To harmonize the
knowledge levels with respect to specific courses,
the involved training facilities offer preparatory
courses, where the students may refresh their
knowledge on topics that are considered pre-
requisites for a particular course. Due to the Web2.0
spirit of the platform the students may always
discuss the subject matter in forums and present
their solutions in form of blog entries or videos,
which can be uploaded into the platform. Other
students are asked to peer-review these solutions.
We think that the Web2.0 spirit of the platform
helps to overcome the motivational issues of the
target group, because by providing user generated
content the students are allowed to shape the course
content by providing learning materials for other
students and getting feedback on their own real-
world solutions (for real-world problems) from their
colleagues. The basic pedagogical approach is
borrowed from collaborative learning designs like
jig-saw or gallery methods, but furthermore the
learners are always asked to comment on the
learning material provided by the trainers for further
improvement of the course content.
A first qualitative study during the testing phase
of a preparatory course (see Figure 1) with 40
students from a class of food technicians at a private
training facility was promising. The students were
excited by the possibility to give feedback on the
presented learning material and the comments of
their co-students. There was only one drawback. The
teachers had turned off the possibility to rate a
comment, because this did not seem necessary to
them. However, some of the students did not
comment on any resource, because they agreed to a
previously written comment. That explains why we
had only 28 comments from 40 students. The
follow-up group discussion confirmed this
hypothesis.
Figure 1: Glimpse of a preparatory course conducted on
the FoodWeb2.0 platform.
The students liked the possibility to repeat
lessons from school and prior courses by conducting
CSEDU2012-4thInternationalConferenceonComputerSupportedEducation
532
quizzes with immediate feedback. Whenever the
students were not able to solve an issue – regardless
if it was a technical issue or a task provided by the
trainer – they tried to solve it by requesting help
from peer students.
As the students liked the quizzes so much we
thought of using their motivation effect to pursue
three goals: 1) help the students to learn, 2) help the
trainers to detect misconceptions to improve their
teaching 3) provide a means to collect and formalize
knowledge about a specific domain. To achieve
these goals as sub goals of FoodWeb2.0 we
developed a framework for casual games.
3 MATCHBALLS
Matchballs is designed as a simple allocation game,
in which the player creates statements by linking
(“matching”) concepts displayed as balls (see Figure
2). A statement consists of two concepts linked by
one of four predetermined relation types. The game
can be played either as two player game or as single
player game with a bot. Each pair of players sees the
same game field (concepts) and the goal is to agree
with the teammate on as many relations as possible
in a given time. To agree on a relation both players
have to create it. If they agree on a relation, they
score points and get time bonuses. Players may see
the relations of their teammates, but not the relation
types. The connecting symbols representing the
relation change according to the state of agreement.
There are different symbols for proposals,
agreements and disagreements of the players.
Figure 2: Matchballs user interface.
As knowledge domain we use the domain of
food safety and hazardous material regulations,
which is an important topic of further education in
the German food industry. The considered concepts
are specific situations, actions, dangerous substances
and edibles, which can be linked by using the four
semantic relations “is similar to”, “is more general
than”, “results in” and “then you may not”. Such a
statement might be:
<Machine overheats> <results in> <fire danger>
To be flexible concerning the learning domain we
decided to use an ontology-based approach. The
ontology may be easily exchanged to adapt the game
to another domain. Furthermore, the ontology
principally offers the opportunity to encode specific
feedback for common misconceptions like it is often
done in intelligent tutoring systems. However,
ontologies are usually incomplete, since it is nearly
impossible to represent even a limited domain in
exhaustive detail. Accordingly, relations created by
the players that do not occur in the knowledge base
are not necessarily wrong, but possibly just missing,
especially if a significant amount of players creates
them. Thus, it is possible to use the so called
“wisdom of the crowd” of the game players to enrich
a pre-built ontology like it is done in games with a
purpose (gwap) for semantic applications (cf.
Siorpaes and Hepp, 2008).
The game has been designed as casual game: The
rules of the game are very easy to learn, the controls
are simple, single play sessions are short and
agreements are instantly rewarded. Casual games are
considered as “games for all”, which not only appeal
to gamers but to the mass audiences irrespective of
their age, gender or background (Kuittinen et al.,
2007). They are not very time consuming and can be
played occasionally. Thus, this game genre seems
appropriate to our divergent target group.
There are several different incentives for
different types of players: For competitive players
there are high scores and time bonuses, which are a
well-known and often used incentive since early
arcade games. Furthermore, players can collect
“achievements”, which are trophies for solving
certain predefined tasks (e.g. for team play with
another player or for scoring many points).
While playing the game the players have to
remember facts and rules in the context of food
safety and hazardous material regulations. Thus, it
can be used in a corresponding course for training
and recapitulation, not only in the class room, but
also online at home.
4 ARCHITECTURE
The central game server is based on a tuple space
middleware called SQLSpaces (Weinbrenner, et al.,
USINGCASUALGAMESANDONTOLOGIESTOTRAINUNSKILLEDWORKERS-AnExperiencewiththe
FoodWeb2.0Platform
533
2007). Software agents communicate by writing and
reading messages on and from the tuplespace. These
messages consist of tuples made of primitive data
types (integer, characters, booleans) and strings. A
single tuplespace server may contain several
tuplespaces used to divide the data stored in the
server into logic or semantic units.
The Matchballs architecture distinguishes four
different categories of tuplespaces: the Coordination
Space, the Game Spaces, the Intermediate Space,
and the Ontology Space (see Figure 3). The
Coordination Space is used to conduct the
matchmaking between two human players or to start
a single player game. The GameClients, which
reside in a Liferay portal, register at the
Coordination space to announce their availability for
a new game session and retrieve the information
about the Game Space they have to connect to. The
Game Client is implemented using HTML5 and
JavaScript to cover a wide amount of browsers and
operating systems.
Figure 3: Architecture of Matchballs.
Each game session has its own Game Space, to
which either two human players (multiplayer game)
or a human player and a GameBot (single player
game) are connected. The Game Space holds all
necessary information for a Matchballs game. That
means the Game Space consists of an excerpt of the
ontology space, the current timer as set by the Timer
agent, the current score, and the links made by the
players as well as its assessment by the Session
Manager. The GameBot has access to the whole
information stored in the ontology excerpt, i.e. it is
aware of the complete knowledge that is represented
in that ontology excerpt. Thus, all associations made
by the GameBot are correct assuming that the
ontology is adequately modeled. The excerpt from
the ontology is created by the Session Manager
agent. It takes care that there is always a minimum
of possible relations between balls in the beginning.
It also detects concordances of the two players with
respect to the links between balls made by each
player. If an agreement on an association is detected
by the Session Manager, i.e. both players’
GameClients wrote exactly the same relation tuple
into the Game Space. A game ends if the time is up.
Afterwards the links made by the players are
collected by the Collector and put into the
Intermediate Space for further inspection. At the
moment the Collector just counts the occurrence of
the specific relations made by the players and stores
or updates the amount in the Intermediate Space.
The Intermediate Space is used for analyses.
There is a threshold of at least five different players
linking two concepts with the same association type
(not present in the ontology yet) to propose this
association for inclusion into the ontology. Another
basic analysis looks for common misconceptions by
the players, i.e. the players frequently contradict an
association in the ontology by using another
association type than the one represented in the
ontology. Both analysis results are specifically
marked in the Ontology Space.
The Ontology Space holds a tuple
representation of the ontology. Every concept and
relation is represented by a tuple. The ontology
design is based on SKOS (Simple Knowledge
Organization System) (Miles and Bechofer, 2009).
We distinguish only four association types, easily
derived from SKOS-relations and we have few
concepts/classes and many individuals/instances,
which can be categorized in SKOS’s concept
schemes. The restrictions on the ontology are caused
by the spirit of the game. Since it shall be a casual
game for unskilled people a huge, sophisticated
system of associations and concepts/classes is
misleading. Furthermore, SKOS is suitable for
teachers to express their domain knowledge without
much help from ontology engineers. Last but not
least there is a plugin for Protégé for SKOS. Thus,
we can use a popular editor for ontology creation,
inspection and refinement.
The markers of the analyses agents are
translated into respective concept schemes:
misconception and new. Accordingly, the specific
individuals may easily be found by teachers and
ontology engineers. The ontology used in our
evaluation (see below) consists of 191 individuals
connected with 91 associations distributed on the
different association types.
5 EVALUATION
The Matchballs game has been evaluated with a
class of 18 students at the Academy of Sweets in
CSEDU2012-4thInternationalConferenceonComputerSupportedEducation
534
Solingen, Germany. Since there were only six
laptops available for conducting the experiment,
after a short plenary introduction of the game, the
class was divided into three groups of six students.
The students of each group had a timeslot of five
minutes for playing as many games as possible and
were encouraged to start with a single-player game
for learning the controls and then perform at least
one game with a human partner. After playing the
game they had to complete a questionnaire and at the
end there was a short plenary discussion.
The subjects consisted of five females and
thirteen males, sixteen were German native
speakers. They rated themselves to have high
knowledge in the fields of safety at work and danger
symbols (median of five on a scale from one to six)
and also some knowledge of hazardous materials
(median of four on a scale from one to six). Thus,
the participants are considered to have at least basic
knowledge of food safety and hazardous materials
and to represent a group recapitulating their
knowledge on these topics. Together the subjects
created 155 different relations, most were only
created by one user, but there were also relations
created by up to seven different participants.
Table 1: Comparison of minimal support for considered
relations, precision of relations and percentage of errors
based on wrong direction of the relation.
minimal support precision
errors based on wrong
direction
4 1.00 none
3 0.85 100,00 %
2 0.68 58,33 %
1 0.40 25,80 %
The higher the amount of users supporting a
relation, the higher is the probability that the relation
is correct (precision) (see Table 1). A frequent error
lies in choosing the wrong direction of the relation,
around 26% of the overall wrong relations are
correct except for the direction; if not all relations
but the ones with a minimal support higher than one
are considered, the percentage is even higher (see
Table 1). These errors can be considered as careless
mistakes instead of real misconceptions.
All frequently occurring, correct relations which
have a support of at least three are already in the
ontology. This is due to the initial composition of
the playing field containing two-thirds of concepts
being linked in the ontology and only one-third of
concepts being not linked with the other selected
concepts. But these relations only cover 17.7% of
the correct relations created by the users. Thus, the
relations with high support have a high precision,
but a low recall. 75.8% of the correct relations
already are in the ontology, 4.8% of the correct new
ones are supported by two and 19.4% by only one
user. The low support of the correct new concepts on
the one hand can also be explained by the initial
composition of the playing field and on the other
hand by the limited number of games played in the
experiment.
The “wisdom of the crowds” approach is often
criticized arguing that expert contributions would be
enough. In our case, the four “best” students (22.2%)
who created the biggest amount of correct new
relations could only provide 46.67% of the overall
number of correct new relations.
Table 2: Cluster centers.
Cluster 1
(14 students)
Cluster 2
(4 students)
Distinct Relations 14.07 5.25
Precision 0.49 0.51
Innovativeness 0.07 0.07
Sloppiness 0.14 0.00
The users show different profiles in creating
relations in terms of the number of different
relations created, their precision, the number of
correct relations missing in the ontology and the
number of errors based on direction issues. Each
user averagely created twelve relations, of which six
(0.51 %) were correct, one (8.3 %) was new and
correct and approximately two (15 %) were wrong
because of direction problems. There is no
correlation between these variables, but partitioning
the students into two clusters using the kMeans
algorithm (see Table 2), reveals one small group of
students (22.2 %), who create a number of distinct
relations far below average, are a bit more precise
and make no direction mistakes and a larger group,
who creates slightly more distinct relations than
average, are a bit less precise than the other group
and make much more direction errors.
On the whole the students had fun playing the
game and experienced it as motivating (each
variable had a median of four on a scale from one to
six). 77.8 % stated they would like to play the game
again and averagely stated they would play it once to
several times a month. The complexity is perceived
as medium (median of 3.5 on a scale from one to
six), which is an indicator that the game is neither
overstraining nor boring and hence appropriate.
USINGCASUALGAMESANDONTOLOGIESTOTRAINUNSKILLEDWORKERS-AnExperiencewiththe
FoodWeb2.0Platform
535
6 DISCUSSION
The evaluation results show that the Matchballs
game was perceived as a casual game. Based on the
data generated during the game sessions we were
able to identify 17 relevant associations or relations
previously not present in the ontology. These
relations were integrated by knowledge engineers. In
this sense, our learning game can also be seen as a
“game with a purpose” to enrich an initial ontology
is feasible.
Thus, the game may be used as an interactive
mechanism for closing gaps in the. At the moment,
the game can only be used for adding new relations
to the ontology. In the future, we plan to use the
game to acquire knowledge from new fields in class
sessions. The teacher may add an originally
disconnected set of new concepts to an existing
ontology. The students are then asked to play the
game by connecting these concepts with each other
as well as with the old ones, thus integrating them
into the existing ontology. This activity may be
viewed as a multi-player concept map creation game
in which the players create a shared concept map.
Concept maps have been successfully used as
learning tool for linking existing and new knowledge
as well as for evaluation and identifying valid and
invalid ideas of students (Novak & Canãs, 2006). If
the game is played in single player mode, the game
may still be used as an advanced vocabulary trainer.
Even when playing the game individually the
students still collaborate indirectly. Teachers can use
the game to extract information about typical
misconceptions of the group but also of individual
students.
In the context of the FoodWeb2.0 project there
have already been several requests by teachers and
students for transferring the game to further
knowledge domains. We will try to incorporate these
domains and enhance these ontologies with specific
feedback on the newly introduced relations. For the
multi-player scenario, feedback will be given about
the existence of these relations in the ontology. In
single-player scenarios the feedback will identify
possible misconceptions automatically based on
information on particular error types explicitly
represented in the ontology. Similar to intelligent
tutoring systems the semantic ontology structure will
be used for the generation of generic feedback.
Evaluation results concerning the FoodWeb2.0
platform in general (and not only the game lements)
are currently Janus-faced: On the one hand, the
perceived usefulness of the platform rated by the
students in general is very high. They like to be able
to look up subject matters on the internet, especially
if the trainer or peer students provide additional
information for further learning. On the other hand,
the trainers are either very enthusiastic or quite
reluctant to use the platform. Those of the trainers
that are enthusiastic often underestimate the time
needed for transforming their material and lesson
planning to an online supported course. The
reluctant ones overestimate the needed effort and
underestimate their students’ skill with Web2.0
tools. The trainers usually have a professional
background in one of the disciplines related to the
food industry (like veterinaries, food engineers or
biologists) but not in pedagogy. This may explain
why some of them are hesitant to employ
collaborative learning strategies in their courses and
consequently have issues with Web2.0 learning.
They use a “traditional” line of argumentation: loss
of control, quality control of the results, perceived
inefficiency of group work. Thus, they are surprised
by their students’ enthusiasm to work with the
platform and by the positive results that are
achieved. We plan to conduct an elaborate study
concerning the perceived usefulness of our platform
differentiated by roles and the changes to the course
content and pedagogical design that have happened
as a consequence of student feedback and the
projects’ “train-the-trainers” program.
ACKNOWLEDGEMENTS
We would like to thank Christopher Charles, Peter
Horster, Dominik Kloke, Carolin Pohl and Carsten
Wieringer, who implemented the Matchballs game
during a student project.
REFERENCES
Baldwin, T. T. And Ford, J. K., 1988. Transfer of training:
A review and directions for future research. In
Personnel Psychology, 41, 63–105.
Burke, L. A. andHutchins, H. M., 2007. Training transfer:
An integrated literature review. In Human Resource
Development Review, 6, 263-296.
Kuittinen, J., Kultima, A.; Niemelä, J., and Paavilainen, J.,
2007. Casual games discussion. In Proceedings of the
2007 conference on Future Play (pp. 105-112), ACM.
Liferay Inc., 2011. Liferay. http://www.liferay.com/en/.
Miles, A. and Bechofer, S., 2009. SKOS Simple
Knowledge Organization System Reference. Available
at http://www.w3.org/TR/2009/REC-skos-reference-
20090818/.
CSEDU2012-4thInternationalConferenceonComputerSupportedEducation
536
Novak, J. D. And Canãs, A. J., 2006. The theory
underlying concept maps and how to construct them.
Technical Report IHMC CmapTools 2006-01 Rev 01-
2008, Florida Institute for Human and Machine
Cognition.
Siorpaes, K. and Hepp, M., 2008. Games with a Purpose
for the Semantic Web. Journal IEEE Intelligent
Systems, 23(3), 50-60.
Weinbrenner, S., Giemza, A., Hoppe, H. U., 2007.
Engineering Heterogeneous Distributed Learning
Environments Using Tuple Spaces as an Architectural
Platform. In Procs of the ICALT 2007, 434-436.
USINGCASUALGAMESANDONTOLOGIESTOTRAINUNSKILLEDWORKERS-AnExperiencewiththe
FoodWeb2.0Platform
537