Expert Systems with Applications 39 (2012) 8030–8043

Contents lists available at SciVerse ScienceDirect

Expert Systems with Applications

journal homepage: www.elsevier.com/locate/eswa

The design and implementation of a competency-based intelligent mobile

learning system
Chien-Chang Hsu ⇑, Chih-Chiang Ho
Department of Computer Science and Information Engineering, Fu-Jen Catholic University, 510, Chung Cheng Rd., Hsinchuang, New Taipei City 24205, Taiwan

a r t i c l e i n f o a b s t r a c t

Keywords: Mobile learning provides a ubiquitous learning context for the learners to select appropriate learning
Competency-based learning paths and learning objects. Adaptive learning methods and correct learning path planning can help to
Mobile learning achieve the goal of learning anytime and anywhere. Moreover, the display ability of mobile learning
Learning path selection devices has become a key factor affecting the interest and acquisition time of learners. Achieving the
desired functionality is currently an important topic in the field of mobile learning. This paper uses
competency-based learning as the basis to evaluate the knowledge deficiency that the learner must
overcome. We then use carrier selection, fuzzy interpolation computation, and ant-genetic algorithm
techniques to select the appropriate learning paths and objects. Finally, we use NFC’s point-to-point
technology to transfer the learning content in the learning device to a larger screen with NFC capability
in the user’s environment to display the same content, thus providing a complete learning system.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction developed an M-learning system using a formative assessment-

based mobile learning mechanism for local cultural learning with
With the popularization of M-learning (mobile learning) devices wireless networks and PDAs. Moreover, some researchers have also
and the recent advances in wireless networking, learners have uses radio frequency identification (RFID) to identify the learning
been able to take advantage of adaptive learning methods to select objects for conducting outdoor or location-aware ubiquitous learn-
suitable learning objects and learning paths that satisfy their learn- ing (Chu, Hwang, Tsai, & Tseng, 2010; Liu, Tan, & Chu, 2009; Shih,
ing needs to achieve the goal of learning anytime and anywhere. Chu, Hwang, & Kinshuk, 2010). For example, Rogers et al. (2005)
M-learning takes place in an environment where learners learn used RFID, PDA, mobile computers, and substance detectors to cre-
by using mobile devices and a wireless network. This type of learn- ate an outdoor learning system. This system allows students to
ing features a personalized learning environment and the ability to integrate what they explore and learn outdoors into an indoor
learn anytime and anywhere (Chang, Chen, & Hsu, 2010; Chen, learning system. In this regard, major topics in M-learning include
Chang, Lin, & Yu, 2009; Chiou, Tseng, Hwang, & Heller, 2010; curriculum planning, creating teaching materials, personalized
Chu, Hwang, Huang, & Wu, 2008; Huang, Kuo, Lin, & Cheng, learning, and the interface display of learning devices.
2008; Huang, Lin, & Cheng, 2010; Huseyin, Nadire, & Erinc, 2009; Curriculum planning, which refers to the establishment of
Kinshuk, Sutinen, & Goh, 2003; Motiwalla, 2007; Quinn, 2001; learning paths, is critically important to M-learning. An optimal
Sharples, 2000; Triantafillou, Georgiadou, & Economides, 2008). curriculum plan keeps learners on the proper track throughout
Adaptive learning, teaching materials, and learning time prediction the learning process. The importance of the curriculum path is par-
are also important factors for learners participating in mobile ticularly true in personalized adaptive learning. Most current per-
learning (Hsu & Tang, 2006). For instance, Heath et al. (2005) sonalized learning systems consider course scheduling and the
developed a learning system that allowed students to connect their preferences, interests, and browsing behavior of the learner in
personal digital assistant (PDA) to a student response system (SRS) the systems. However, these systems overlook the competencies
so they could interact with their teachers. Teachers could use the of the learner. They do not consider whether the difficulty of the
system to monitor students’ progress in a timely manner. Sampson courses that are provided is suitable for every learner. Generally
(2006) created the SMILE system, with which learners or teachers speaking, unsuitable courses are those that are beyond the learn-
use materials on a learning platform through PDAs, mobile phones, ers’ comprehension, which may cause learning disorientation,
or laptops for educational purposes. Hwang and Chang (2011) resulting in low learning efficiency (Carro, Pulido, & Rodriguez,
1999; Chen & Chung, 2008; Huang, Huang, & Chen, 2007; Hwang,
⇑ Corresponding author. Tel./fax: +886 2 29053876. Kuo, Yin, & Chuang, 2010; Marquez, Ortega, Abril, & Velasco, 2008;
E-mail address: cch@csie.fju.edu.tw (C.-C. Hsu). Wang, Wang, & Huang, 2008; Yang & Wu, 2009). Chen, Liu, and

0957-4174/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2012.01.130
 C.-C. Hsu, C.-C. Ho / Expert Systems with Applications 39 (2012) 8030–8043 8031

Chang (2006) proposed an agent-based personalized internet

teaching system. This system uses four agents and applies item re- Intelligent Personalized
sponse theory for personalized course scheduling. It considers the learning learner
interface
difficulty and consistency of the courses as well as the learners’
competencies throughout the learning process. During the learning Learner
process, the learning paths are modified in accordance with the
learners’ responses. Acampora, Gaeta, Loia, Ritrovato, and Salerno Fig. 2. System architecture.
(2008) proposed a self-adaptive E-learning system based on an
ontological description of a learning environment and a memetic
algorithm. The memetic algorithm adds a local search algorithm peer-to-peer
after the optimal solution for each generation is generated. Hong, transmission
Chen, Chang, and Chen (2007) proposed a personalized learning Migration of
content
system based on a genetic algorithm in an attempt to build suitable
learning paths for users. This system builds personalized learning
paths based on the number of wrong answers given by learners NFC reader of the
Display device including
in a pre-test. Scores on the pre-test are conceptual scores that indi- M-learning device
NFC reader
cate the difficulty levels of various courses. The scores serve as the
basis for establishing learning paths. A personalized E-learning sys- Fig. 3. Learning contents transformation.
tem based on a genetic algorithm is better than free browsing
learning patterns because it provides learners with better quality
learning with easier learning paths for learners. In addition to hybrid ant-genetic algorithms can speed up the establishment of
applying a genetic algorithm for building learning paths, Huang learning paths.
et al. (2007) introduced case-based reasoning to facilitate the oper- However, curriculum planning is closely related to teaching
ation of this personalized E-learning system. If the learners fail to materials (Kontopoulos, Vrakas, Kokkoras, Bassiliades, & Vlahavas,
acquire mastery of the course, the system uses a genetic algorithm 2008). Certain characteristics of competency-based education
to recommend a personalized curriculum to assist the learners in (CBE) make it applicable to the curriculum planning of M-learning.
comprehending the course. Finally, after several courses have been CBE focuses on job skill training and core ability cultivation (Hafeez
completed, the case-based reasoning system can provide a conclu- & Essmail, 2007; McClelland, 1973; Noe, 2005). It is suitable for
sive evaluation for each learner. Some researchers also tried to im- personalized adaptive learning, which allows learners to look for
prove the performance of the genetic algorithm by integrating appropriate learning content based on self-competency tests. CBE
different algorithms. For example, Yang and Wu (2009) proposed includes five fundamental elements: demonstration of competen-
an ant colony system based on attributes and ant colony optimiza- cies, criteria for competency assessments, evaluations of students’
tion in an attempt to select suitable learning objects more effi- competencies, learning processes, and education programs. CBE
ciently. This system uses rule induction to improve the quality of features personal, self, systematic, active, and mastery learning.
pheromones and helps learners to find suitable learning objects CBE helps learners to achieve a pre-determined competency stan-
more easily. In short, genetic algorithms and ant colony algorithms dard and master what they are learning. Thus, it is particularly
share some similar search methods. Both are types of heuristic applicable for the learning of skill subjects. CBE, which focuses
algorithms. The integration of ant colony and genetic algorithms on learning outcomes, is a self-oriented learning method. Prior to
has been previously applied to solve various problems (Altiparmak each learning unit, instructions are always provided that describe
& Karaoglan, 2007; Dorigo & Stutzle, 2004; Lee, Lee, & Su, 2002; all the knowledge and skills required for learning this particular
Stutzle & Hoos, 2000; Xu, Lim, Ong, & Tang, 2006). These algo- unit. Learners can determine, based on their current abilities,
rithms can find a feasible solution and avoid early convergence whether this unit is suitable for them. During the learning process,
by combining the advantages of both algorithms. They can also learners can examine whether they are actually acquiring knowl-
avoid falling into locally optimal solutions. The development of edge of the unit with the help of learning evaluations. Fig. 1 illus-
trates a complete framework of CBE teaching materials. The
framework covers occupations, tasks, learning units, post-learning
evaluations, learning goals, and learning evaluations. Using the
Occupations learning goals as learning objects and the post-learning evalua-
tions as pre-test questions, learners can select suitable learning ob-
jects for personalized learning.
Tasks Tasks Tasks Many researchers have applied different methods to achieve the
goal of personalized adaptive learning (Chen & Duh, 2008; Hwang,
Chen, Lai, & Lin, 2009; Liu et al., 2009). Chen et al. (2006) had ex-
perts design a course difficulty index and used learner feedback
Learning units

……
Learning units

……
Learning units

……
for adaptive learning. Hong et al. (2007) also had experts design
Post-learning Post-learning Post-learning a difficulty index, and used pre-tests to select suitable courses for
evaluations evaluations evaluations
adaptive learning. Yang and Wu (2009) had experts determine var-
ious attributes of learners and learning objects and used these
attributes for adaptive learning.
The learning interface display is also a critical problem in per-
Learning goals Learning goals Learning goals sonalized adaptive learning. Because learners may come from var-
ious age groups, there are many innate limitations to mobile
Learning Learning Learning learning devices, such as screen size limitations, simplistic input
evaluations evaluations evaluations interfaces (number pads and handwriting recognition), and limited
internet access. Most current systems are designed exclusively for
Fig. 1. Framework of CBE teaching materials. desktop computers, making it inconvenient to learn through
8032 C.-C. Hsu, C.-C. Ho / Expert Systems with Applications 39 (2012) 8030–8043

Occupations
Microcomputer control

tasks
Data book Welding Distribution ……………
study training technology Practice visit

semiconductor devices
Understanding of

tools and solder

Understanding of welding
…… …… …… ……
Learning
units

evaluations
Post-learning

evaluations
Post-learning

evaluations
Post-learning

evaluations
Post-learning
Students can identify the Students can state the Students can state the
<Learning goals>
Learning object

symbols and texts of names and specifications names, specifications

control wires without of magnetic switches …………… and usage of wires and
any references without any references. connectors without any
references.

Learning Learning Learning

evaluations evaluations evaluations

Fig. 4. Example of a course tree diagram.

Table 1
Examples of pre-test evaluation items.

Items Questions (TRUE or FALSE question)

Evaluation item The measured output is called the step response, which uses the step variation to stimulate the system
1
Evaluation item Personal computer software can be divided into two categories: operating systems and application software
2
Evaluation item The most commonly used file types in Visual Basic are files, module files, and project files
3
Evaluation item A set of input/output wires used for data transfer between peripheral devices and a computer is called the input/output bus (I/O Bus)
4
Evaluation item A way to programmatically simulate equipment in the LabView environment for other relevant applications is called a virtual instrument
5
Evaluation item RS-232 and RS-485 are used as the standard serial interface for data transmission
6
Evaluation item The SCADA system is composed of a long-distance network, data communications equipment, host devices, and remote control devices
7
Evaluation item The goal of text and equipment symbols is to improve the ability to draw the control chart, enhance chart reading and interpretation, and increase
8 the understanding of the control system
Evaluation item ISaGRAF is a synchronous system (an input variable is only one state in each cycle). The input variables will not be updated in the same cycle during
9 FOR recursive execution time
Evaluation item The sequential function chart (SFC) program is represented by a series of steps and a transfer condition diagram. The steps and transfer conditions
10 are connected by the directional lines
Evaluation item PLC is the most commonly used logic processing unit in the automatic control system. It can easily collect the status of sensors, switches, and buttons
11 by software program. After some logical operations, it can generate the corresponding output actions
Evaluation item The OPEN PLC standard defines five languages, namely, ladder diagram, function block diagram, structured text, sequential function chart, and
12 instruction set

mobile device displays. However, with the integration of Wi-Fi and handheld devices. Hua and Lu (2006) collected and arranged sev-
mobile devices and the plan to include long-term evolution (LTE) eral mobile devices around the learner to display learning content.
technology in the future, the current problems with mobile inter- Due to the limited screen size of mobile devices, learning content
net access are being mitigated. Wi-Fi/LTE allows learners to gain must be divided into several blocks and transferred to adjacent de-
access to online learning sites and obtain learning materials wher- vices to be displayed. This approach can solve the problem of small
ever they are. Nevertheless, the limitation of screen size remains screen size. Yang and Chen (2006) adjusted the size of the learning
an unsolved problem for mobile devices. Some related research content based on the screen size of a particular mobile device. In
has sought to mitigate this problem (Chen, Chang, & Wang, summary, most of the abovementioned methods divide a large
2008). Churchill and Hedberg (2008) provided eight recommenda- page into several smaller blocks to fit the screen sizes of the vari-
tions for designing appropriate learning objects in small screen ous mobile devices. However, this method is still inconvenient for
 C.-C. Hsu, C.-C. Ho / Expert Systems with Applications 39 (2012) 8030–8043 8033

Table 2
Example of learning objects. (a)
Learning Learning goal
object
Object 1 You can construct a virtual instrument (Virtual Instrument, VI)
application program in the LabVIEW environment with the
help of learning object explanations and references
Object 2 You can use VB to design a ‘‘phone book’’ program with the 0 Intensity of relationship
0.5 1
following menu: insert, delete, and query functions
Object 3 You can learn how to use LabVIEW debugging features to
debug all your established VI with the help of learning object (b) (c)
explanations and references
Object 4 You can state the meaning of digital input/output and give a
few examples of digital input/output applications without any
reference books and materials
Object 5 You can use the computer interface to adjust the parameters of
PID control by giving a real servomotor. You can also find the
influence of different parameters on the servomotor by
0.5 1 0.5 1
observing the relationship between parameter adjustment and
the servomotor’s operation
Object 6 Students can identify the symbols and texts of control wires (d) (e)
without any references

users. This study has discovered that NFC technology can serve as a
good solution for mobile device screen size. NFC technology en-
ables convenient short range communication between mobile 0.5 1 0.5 1
learning devices or with other electronic equipment. By placing
two NFC devices together, the two devices can be made to ex- Fig. 5. Partition of trapezoidal fuzzy sets.
change information. Such an intuitive method of operation is very
suitable for learners of all ages.
This paper proposes a competency-based adaptive M-learning the convenience of learners. Such an interface is particularly suit-
system that selects suitable learning objects based on pre-tests able for elderly learners.
and career planning to achieve the goal of adaptive learning. The The structure of this paper is as follows: Section 2 introduces
system uses an ant-genetic algorithm to establish learning paths the structure of this system, Section 3 presents the system imple-
through the accumulation of pheromone. It also considers the con- mentation, and Section 4 discusses and concludes the paper.
nections between learning objects and the learners’ past perfor-
mance to build learning paths that best fit the learners. The 2. System architecture
benefit of using an ant-genetic algorithm is that it speeds up the
establishment of learning paths, especially when there are a large Fig. 2 shows the architecture of the competency-based intelli-
number of genes. Finally, the system applies NFC technology to gent M-learning system. The system contains two modules: the
move learning content to mobile devices with bigger screens for intelligent learning interface and the personalized learner. The

Table 3
Fuzzy relationships between career plans and learning objects.

Occupations Learning object

Learning object 1 Learning object 2 Learning object 3 Learning object 4 Learning object 5 Learning object 6
Automatic control planning [0.75,0.8, 0.95, 1] [0.35, 0.45, 0.5, 0.6] [0.75, 0.8, 0.85, 1] [0.45, 0.65, 0.75, 0.85] [0.75, 0.8, 0.85, 1] [0.2, 0.35, 0.4, 0.45]
Control programming [0.75, 0.8, 0.95, 1] [0.75, 0.85, 0.9, 1] [0.75, 0.8, 0.85, 1] [0.3, 0.5, 0.6, 0.7] [0.75, 0.8, 0.85, 1] [0, 0.15, 0.2, 0.25]
Maintenance engineer [0.2, 0.35, 0.4, .045] [0, 0.1, 0.15, 0.25] [0.4, 0.45, 0.5, 0.65] [0.6, 0.8, 0.9, 1] [0.55, 0.6, 0.65, 0.8] [0.75, 0.9, 0.95, 1]

Table 4
Fuzzy relationships between pre-test items and learning objects.

Evaluation item Learning object

Learning object 1 Learning object 2 Learning object 3 Learning object 4 Learning object 5 Learning object 6
Evaluation item 1 [0.55, 0.7, 0.75, 0.8] [0, 0.1, 0.15, 0.25] [0.4, 0.45, 0.5, 0.65] [0, 0.2, 0.3, 0.4] [0.75, 0.8, 0.85, 1] [0, 0.15, 0.2, 0.25]
Evaluation item 2 [0.75, 0.9, 0.95, 1] [0.75, 0.85, 0.9, 1] [0.75, 0.8, 0.85, 1] [0.3, 0.5, 0.6, 0.7] [0.55, 0.6, 0.65, 0.8] [0.2, 0.35, 0.4, 0.45]
Evaluation item 3 [0.75, 0.9, 0.95, 1] [0.75, 0.85, 0.9, 1] [0.75, 0.8, 0.85, 1] [0.3, 0.5, 0.6, 0.7] [0, 4, 0.45, 0.5, 0.65] [0, 0.15, 0.2, 0.25]
Evaluation item 4 [0.55, 0.7, 0.75, 0.8] [0, 0.1, 0.15, 0.25] [0.55, 0.6, 0.65, 0.8] [0.6, 0.8, 0.9, 1] [0.75, 0.8, 0.85, 1] [0, 0.15, 0.2, 0.25]
Evaluation item 5 [0.75, 0.9, 0.95, 1] [0.35, 0.45, 0.5, 0.6] [0.75, 0.8, 0.85, 1] [0.3, 0.5, 0.6, 0.7] [0.2, 0.25, 0.3, 0.45] [0, 0.15, 0.2, 0.25]
Evaluation item 6 [0.2, 0.35, 0.4, 0.45] [0, 0.1, 0.15, 0.25] [0.4, 0.45, 0.5, 0.65] [0.6, 0.8, 0.9, 1] [0.75, 0.8, 0.85, 1] [0.35, 0.5, 0.55, 0.6]
Evaluation item 7 [0.35, 0.5, 0.55, 0.6] [0.2, 0.3, 0.35, 0.45] [0.4, 0.45, 0.5, 0.65] [0, 0.2, 0.3, 0.4] [0.55, 0.6, 0.65, 0.8] [0, 0.15, 0.2, 0.25]
Evaluation item 8 [0, 0.15, 0.2, 0.25] [0, 0.1, 0.15, 0.25] [0, 0.05, 0.1, 0.25] [0.3, 0.5, 0.6, 0.7] [0.2, 0.25, 0.3, 0.45] [0.55, 0.7, 0.75, 8]
Evaluation item 9 [0.2, 0.35, 0.4, 0.45] [0.2, 0.3, 0.35, 0.45] [0, 0.05, 0.1, 0.25] [0, 0.2, 0.3, 0.4] [0, 4, 0.45, 0.5, 0.65] [0, 0.15, 0.2, 0.25]
Evaluation item 10 [0, 0.15, 0.2, 0.25] [0, 0.1, 0.15, 0.25] [0.2, 0.25, 0.3, 0.45] [0, 0.2, 0.3, 0.4] [0, 4, 0.45, 0.5, 0.65] [0.2, 0.35, 0.4, 0.45]
Evaluation item 11 [0.55, 0.7, 0.75, 0.8] [0.2, 0.3, 0.35, 0.45] [0.4, 0.45, 0.5, 0.65] [0.6, 0.8, 0.9, 1] [0.55, 0.6, 0.65, 0.8] [0.2, 0.35, 0.4, 0.45]
Evaluation item 12 [0, 0.15, 0.2, 0.25] [0.35, 0.45, 0.5, 0.6] [0.4, 0.45, 0.5, 0.65] [0.3, 0.5, 0.6, 0.7] [0.2, 0.25, 0.3, 0.45] [0, 0.15, 0.2, 0.25]
8034 C.-C. Hsu, C.-C. Ho / Expert Systems with Applications 39 (2012) 8030–8043

Non-specialized courses

2 1 1 1.5 1 1.5 1 1 2
Simple Moderate Difficult

Specialized courses

1.5 0.5 0.5 1 0.5 1 0.5 0.5 1.5

Simple Moderate Difficult

Fig. 6. Shapes of trapezoidal fuzzy sets.

Difficulty level: Moderate

Courses type: Non-specialized
very weak weak moderate strong very strong
1

0
Intensity of
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 relationship

Difficulty level: Difficult

Courses type: Non-specialized
very weak weak moderate strong very strong
1

Intensity of
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 relationship

Difficulty level: Simple

Courses type: Non-specialized
very weak weak moderate strong very strong
1

Intensity of
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 relationship

Fig. 7. Trapezoidal fuzzy sets of non-specialized courses.

intelligent learning interface analyzes learners’ competencies, se- 2.1. Intelligent learning interface
lects learning objects, and transfers information on the learning
content. The personalized learner uses an ant-genetic algorithm The intelligent learning interface is responsible for transferring
to establish learning paths that are suitable for the learners’ com- NFC teaching material and analyzing the learner’s individual
petencies in an attempt to facilitate learning efficiency. competencies. The NFC teaching material transfer step applies
 C.-C. Hsu, C.-C. Ho / Expert Systems with Applications 39 (2012) 8030–8043 8035

Difficulty level: Moderate

Courses type: Specialized
very weak weak moderate strong very strong
1

0
Intensity of
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 relationship

Difficulty level: Difficult

Courses type: Specialized
very weak weak moderate strong very strong
1

0
Intensity of
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 relationship

Difficulty level: Simple

Courses type: Specialized
very weak weak moderate strong very strong
1

Intensity of
0
0.2 0.5 0.8 1
relationship
0.1 0.3 0.4 0.6 0.7 0.9

Fig. 8. Trapezoidal fuzzy sets of specialized courses.

peer-to-peer transmission to transfer learning content from the

A T B
NFC reader of an M-learning device to the NFC reader of a display
1
device (Fig. 3). After receiving the learning content from the NFC
reader of the M-learning device, the NFC reader of the display de-
vice displays the content on a larger screen, making reading of the
GA GT GB learning object easier.
The intelligent learning interface carries out personalized com-
petency analyses with the help of a pre-test evaluation in an at-
tempt to discover the competencies that learners lack for their
0 respective career plans. It applies fuzzy interpolation to select suit-
a1 a2 a3 a4 b1 b2 b 3 b4 able learning objects. By exploiting the features of competency-
based teaching curricula, the intelligent learning interface converts
Fig. 9. Centers of gravity of fuzzy sets.
questions from the post-learning evaluation into true or false ques-
tions for the pre-test evaluation to analyze an individual’s learning
index. An example of the tree diagram of competency-based
courses controlled by a microcomputer, an example of the pre-test
A C B evaluation, and an example of learning objects are shown in Fig. 4,
1 Tables 1 and 2, respectively.
Suitable learning objects (flearning_objects) are selected based on
the competencies that learners lack for their career plans:

GC
flearning objects ¼ fcareer  fpretest ; ð1Þ

where fcareer, fpretest, and  denote the fuzzy values of the learner ca-
reer analysis, the knowledge and competencies that the learner
0
c1 c2 c3 c4 lacks, and the fuzzy interpolation, respectively. The fuzzy values
of the learner career analysis reflect a learner’s career plans and
Fig. 10. Calculation of fuzzy interpolation. directions of their careers:
8036 C.-C. Hsu, C.-C. Ho / Expert Systems with Applications 39 (2012) 8030–8043

LO18 LO0C LO13 LO0B LO0F

LO02 LO01 LO17 LO03

LO06 LO16

LO0D LO15 LO0E

Start End
LO05 LO08

LO10 LO14

LO11 LO12 LO09

LO04 LO07 LO0A

Fig. 11. Example of a learning concept graph.

fcareer ¼ Pkm¼1 fg ðC m Þ; ð2Þ

LO22 LO21
where k, fg, and C denote the number of occupations selected by the
learner, the fuzzy values of the must-have learning objects corre-
sponding to the learner’s career plans, and the occupations in the
learner’s career plans, respectively. LO1 LO2 LO3 LO4
The knowledge and competencies that the learner lacks indicate
what the learner does not have as a learning object. It measures the
questions that are answered wrongly and the relationships among
learning objects: LO20
 
fpretest ¼ Pni¼1 ft P wrong
i ; ð3Þ
Fig. 13. Optional learning objects.
where n, ft, and Pwrong denote the number of questions answered
wrongly in the pre-test, the fuzzy values of the learning objects cor-
responding to the questions of the pre-test, and the questions an- actual career plans, and the pre-test questions. Fig. 5 shows the par-
swered wrongly in the pre-test, respectively. Tables 3 and 4 show titions of the trapezoidal sets. The X-axis indicates the relationship
the fuzzy relationship between the learner’s career plans and the between the learning objects and the learners’ career plans or the
learning objects, and the fuzzy relationship between the pre-test intensity of the pre-test items and the learning objects. The values
items and the learning objects, respectively. The fuzzy values in are between 0 and 1. The higher the values are, the stronger the
the tables use trapezoidal fuzzy sets. The values are determined intensity of the relationship. For instance, the relationship between
based on the relationships between the learning objects and the a program design course and a career as a programmer is strong. On
learners’ career plans, the learning objects and the pre-test items, the other hand, the relationship between a program design course
the difficulty of learning objects, and the complexity of the learning and a career as a maintenance engineer is weak (Fig. 5(a)). The lar-
objects. The same learning objects can have different fuzzy values ger the area of the right triangle of the trapezoidal fuzzy set is, the
for different career items or pre-test questions. The fuzzy values more difficult the learning objects are (Fig. 5(b)). The larger the area
are usually determined by experts in the design process of the var- of the left triangle of the trapezoidal fuzzy set is, the easier the
ious teaching materials. The fuzzy values of this system are deter- learning objects are (Fig. 5(c)). The area of the middle square of
mined based on the teaching materials selected, the learners’ the trapezoidal fuzzy set is used for fine tuning the difficulty of
the learning objects. If the overall area of the trapezoidal fuzzy set
is large, it means that the learning objects are less specialized gen-
Learning object

Learning object

eral courses (Fig. 5(d)). If the overall area of the trapezoidal fuzzy
set is small, it means that the learning objects are more specialized
courses (Fig. 5(e)).
Based on the above partitions, this paper categorizes courses
into three levels: simple, moderate, and difficult. By distinguishing
23 4D 1F 0D 67 3C 6A 29 specialized courses from non-specialized courses, we determine
the shape of the trapezoidal fuzzy set (Fig. 6). Finally, we categorize
Fig. 12. Gene encoding. the intensity of the relationships in Fig. 5(a) into five levels: very
 C.-C. Hsu, C.-C. Ho / Expert Systems with Applications 39 (2012) 8030–8043 8037

LO2A is the prerequisite of LO23

5A 53 4A 43 3A 33 2A 23 LO3A is the prerequisite of LO33
LO4A is the prerequisite of LO43
LO5A is the prerequisite of LO53
Chromosome sequencing
direction of learning paths
Take LO 33 as an example:

9A 3A 93 ........ 43 33 2A 23 5A 3A 53 4A 43 33 23

(a) (b)
5A 53 3A 4A 43 33 2A 23 5A 53 4A 43 3A 33 2A 23

(c) (d)
?

5A 53 4A 43 33 2A 23 3A

(e)
Fig. 14. Learning benefits.

Pheromone Values A Indicates higher pheromone values

Indicates lower pheromone values

LO1 LO 2

48 3A 5A 53 4A 33 2A 23 48 3A 5A 53 4A 33 2A 23
Pheromone Values B
P arent A P ar en t A
Fig. 15. Accumulation of pheromone values.

5A 53 4A 43 3A 33 9A 28 5A 53 4A 43 3A 33 9A 28

Parent B Parent B
weak, weak, moderate, strong, and very strong. A total of 30 stan-
dard trapezoidal fuzzy sets are established for various fuzzy values.
Figs. 7 and 8 show the standard trapezoidal fuzzy sets of non-spe-
cialized and specialized courses. 48 3A 48 3A 33
The fuzzy interpolation method uses fuzzy interpolation with a
Offspring Offsprin
center of gravity to calculate the relationship between two fuzzy
sets (Huang & Shen, 2006). For instance, A and B represent two Fig. 16. Crossover operation.
known trapezoidal fuzzy sets, and T is the trapezoidal fuzzy set
formed by the two far-left points and the two far-right points of
A and B together. The normal trapezoid based on the center of grav-
ity is used to calculate the values of the X-coordinates of the cen- c1 ¼ ð1  kÞ  a1 þ k  b1 ; ð7Þ
ters of gravity of all fuzzy sets. GA, GB, and GT denote the centers c2 ¼ ð1  kÞ  a2 þ k  b2 ; ð8Þ
of gravity of A, B and T, respectively (Fig. 9): c3 ¼ ð1  kÞ  a3 þ k  b3 ; ð9Þ
a2 þ a24  a21  a22 þ a3 a4  a1 a2 c4 ¼ ð1  kÞ  a4 þ k  b4 ; ð10Þ
GA ¼ 3 ; ð4Þ
3ða3 þ a4  a1  a2 Þ dðGA ; GT Þ
k¼ ; ð11Þ
dðGA ; GB Þ
2 2 2 2
b3 þ b4  þ b3 b4  b1 b2
b1 b2
GB ¼ ; ð5Þ where k, d(GA, GT), and d(GA, GB) denote the fuzzy interpolative ratio
3ðb3 þ b4  b1  b2 Þ
based on the center of gravity, the distance between GA, GT, and the
8 distance between GA, GB, respectively (Huang & Shen, 2006):
> T 1 ¼ minimumða1 ; b1 Þ;
>
>
T 23 þ T 24  T 21  T 22 þ T 3 T 4  T 1 T 2 < T 2 ¼ minimumða2 ; b2 Þ;
GT ¼ ; GC ¼ ½c1 ; c2 ; c3 ; c4 center of gravity
3ðT 3 þ T 4  T 1  T 2 Þ >
> T 3 ¼ maximumða3 ; b3 Þ;
>
:
T 4 ¼ maximumða4 ; b4 Þ: c23 þ c24  c21  c22 þ c3 c4  c1 c2
¼ : ð12Þ
3ðc3 þ c4  c1  c2 Þ
ð6Þ
Eqs. (7)–(9) generate the trapezoidal fuzzy set of the trapezoidal
fuzzy interpolation. The resulting trapezoidal fuzzy set is
C = [c1, c2, c3, c4]. The fuzzy set C is then de-fuzzed with a normal
trapezoid based on the center of gravity to generate GC and its
5A 53 4A 43 3A 33 2A 23 5A 53 2A 43 3A 33 4A 23
X-coordinate value (Fig. 10). Within the fuzzy range of 0.0–1.0,
the system selects only those learning objects whose values are
greater than or equal to 0.5: Fig. 17. Mutation operation.
8038 C.-C. Hsu, C.-C. Ho / Expert Systems with Applications 39 (2012) 8030–8043

Learning paths generate

Learning object
Display device with
database
NFC reader

NFC teaching
material interface

Server

WLAN Personal
knowledge
Analyzing personalized
fil
competencies
Learner M-learning device
with NFC reader

Fig. 18. System environment.

Display Device
M-learning with NFC Reader
device
NFC embedded NFC embedded

(a) NFC reader (b) M-learning and display devices with NFC
reader

Fig. 19. System devices.

First, hexadecimal encoding is used for gene encoding to form

chromosomes. Every hexadecimal value represents one learning
object, and every learning object represents a gene in a chromo-
some (Fig. 12). By LPij, the learning path from LOi to LOj is denoted.
The gene sequencing in a chromosome is exactly the same as the
sequencing of the learning path. For instance (LO23, LO4D, LO1F) is
the sequencing of the learning path LP 23;4D ) LP 4D;1F . After a suit-
able learning path is established for learners, other learning objects
that are not taken as necessary are regarded as optional learning
objects. These objects branch off from the selected learning path
(Fig. 13), as optional learning objects (indicated with dashed lines
in the figure), so that learners can decide whether or not they wish
to learn them.
Fig. 20. System display. The fitness functions use maximum learning benefits (LB) to
search for suitable learning paths:
!!
2.2. Personalized learner
X
n X
m 1 SPass
LOKþ1 SPass
LOk
LB ¼ maximum bðLOi Þ þ  ; SPass
LO – 0;
i¼1 k¼1
100 100
The personalized learner is responsible for establishing learning ð13Þ
paths in a learning concept graph. The learning concept graph con-
sists of five parts (S, F, C, E, P), the start node, final node, learning where n, b, LO, m, and SPass
represent the total number of learning
LO
concept, essential condition, and pass core. Fig. 11 shows an exam- objects, the benefits of each learning object, the learning objects,
ple of the learning concept graph. The must-have conditions of the total number of learning objects on which the learners have past
LO0A in the graph are LO04 and LO07, meaning that prior to learning performance, and the learners’ past performance on those learning
LO0A, learners must acquire LO04 and LO07. objects, respectively. The benefits of learning objects (b) are deter-
The learning path scheduling uses an ant-genetic algorithm to mined by the distance between the learning objects and the prereq-
construct suitable learning paths. By integrating some of the char- uisites in the genetic chromosomes (Fig. 14). The fitness functions
acteristics of ant colony algorithms into a genetic algorithm, the consider two factors. The first is the shortest learning path between
calculation of pheromone can accelerate the establishment of each learning object and its prerequisites. The other is the learners’
learning paths. The ant-genetic algorithm includes gene encoding, past performance in the objects they have learnt. The fitness func-
fitness functions, crossover and mutation, and stop conditions. tions also place objects not learnt or objects learnt with a poorer
 C.-C. Hsu, C.-C. Ho / Expert Systems with Applications 39 (2012) 8030–8043 8039

Table 5
Fuzzy relationships between career plans and learning objects.

Occupations Learning object (16)

01 02 03 04 05 06
Automatic control planning [0.75, 0.8, 0.95, 1] [0.35, 0.45, 0.5, 0.6] [0.75, 0.8, 0.85, 1] [0.45, 0.65, 0.75, 0.85] [0.75, 0.8, 0.85, 1] [0.2, 0.35, 0.4, 0.45]
Control programming [0.75, 0.8, 0.95, 1] [0.75, 0.85, 0.9, 1] [0.75, 0.8, 0.85, 1] [0.3, 0.5, 0.6, 0.7] [0.75, 0.8, 0.85, 1] [0, 0.15, 0.2, 0.25]

Learning object (16)

07 08 09 0A 0B 0C
Automatic control planning [0.6, 0.8, 0.9, 1] [0.55, 0.6, 0.65, 0.8] [0.45, 0.6, 0.7, 0.85] [0.6, 0.8, 0.9, 1] [0.55, 0.6, 0.65, 0.8] [0.45, 0.65, 0.75, 0.85]
Control programming [0.4, 0.65, 0.75, 0.85] [0.4, 0.45, 0.5, 0.65] [0.6, 0.75, 0.85, 1] [0.3, 0.5, 0.6, 0.7] [0.75, 0.8, 0.85, 1] [0.6, 0.8, 0.9, 1]

Learning object (16)

0D 0E 0F 10 11 12
Automatic control planning [0.3, 0.5, 0.6, 0.7] [0.75, 0.85, 0.9, 1] [0.6, 0.75, 0.85, 1] [0.75, 0.85, 0.9, 1] [0.45, 0.65, 0.75, 0.85] [0.45, 0.6, 0.7, 0.85]
Control programming [0, 0.2, 0.3, 0.4] [0.75, 0.85, 0.9, 1] [0.6, 0.75, 0.85, 1] [0.75, 0.85, 0.9, 1] [0.6, 0.8, 0.9, 1] [0.6, 0.75, 0.85, 1]

Learning object (16)

13 14 15 16 17 18
Automatic control planning [0.6, 0.7, 0.8, 1] [0.55, 0.6, 0.65, 0.8] [0.55, 0.65, 0.7, 0.8] [0.55, 0.6, 0.65, 0.8] [0.75, 0.85, 0.9, 1] [0.75, 0.8, 0.85, 1]
Control programming [0.6, 0.7, 0.8, 1] [0.75, 0.8, 0.85, 1] [0.2, 0.3, 0.35, 0.45] [0.2, 0.25, 0.3, 0.45] [0.75, 0.85, 0.9, 1] [0.75, 0.8, 0.85, 1]

performance at the front of the learning path as often as possible, so generation is considered to determine which of the two paternal
that learners can relearn these objects: generations has a learning path that starts with this particular ob-

8
> þ0:1; prerequisite learning objects come before more than 5 learning objects ðFig: 14ðaÞÞ;
>
>
>
> þ0:3; prerequisite learning objects come before 3  4 learning objects ðFig: 14ðbÞÞ;
>
>
>
< þ0:5; prerequisite learning objects come before 1  2 learning objects ðFig: 14ðcÞÞ;
b¼ ð14Þ
>
> þ1; prerequisite learning objects come before and close to learning objects ðFig: 14ðdÞÞ;
>
>
>
> 0; no prerequisite for learning objects;
>
>
:
1; prerequisite learning objects do not come before learning objects ðFig: 14ðeÞÞ:

The crossover procedure uses pheromone calculation of the ant ject. The learning paths of the paternal generations are compared,
algorithm to decide the learning path direction in which the most and the one with the higher pheromone value is selected as the
pheromone accumulates. Fig. 15 demonstrates accumulated phero- learning path for the offspring generation (Fig. 16(b)). If any learn-
mone values in a learning path. The pheromone values in the learn- ing object of the paternal generations has already appeared in the
ing path between two learning objects vary depending on the former part of the path of the offspring generation, another object
direction. LP1,2 and LP2,1 represent learning paths in opposite direc- that has not yet appeared in the path of the offspring generation is
tions between the learning objects 1 and 2. In addition to the tradi- selected randomly, so that the algorithm can carry on. This process
tional single point crossover, the ant-genetic algorithm also uses a repeats until the chromosomes of the offspring generation are
whole piece of chromosome for crossover. The amount of phero- filled with genes. The crossover rate of the traditional single point
mone accumulated in a learning path determines each section of method is 30%, and the crossover rate of the ant-genetic algorithm
the path of an offspring generation (LPoffspring): alone is 70%. Mutation is carried out by switching the positions of
   two randomly selected genes, and the probability of mutation is 1%
LP offspring ¼ LP maximum sparentA
i;j ; sparentB
i;j ; ð15Þ (Fig. 17).
When calculating pheromone levels, the ant-genetic algorithm
where LP; sparentA , and sparentB denote the learning paths, the phero- uses the max–min ant system to limit pheromone values to the
i;j i;j
mone values in the learning path from i to j of paternal generation range between a given maximum smax and a minimum smin value.
A, and the pheromone values in the learning path i to j of paternal During the calculation, if a pheromone value is greater than the
generation B, respectively. maximum value, the maximum value is selected; if the pheromone
The ant-genetic algorithm uses maximum pheromone values to value is smaller than the minimum value, the minimum value is
generate a generation of offspring. After the crossover of two pater- selected.
nal generations, only one offspring generation with a higher pher- 1
omone value will be generated. If the pheromone values of the two smax ¼ ; q ¼ 0:1; ð16Þ
LBinit
best  q
paternal generations are the same, a learning path is selected ran-
smax
domly for the offspring generation. When crossover begins, the smin ¼ ; ð17Þ
n
first learning path section with the maximum pheromone value
of the two paternal generations is selected as the first learning path where LBinit
best , q, and n denote the values retrieved by the fitness
section of the offspring path (Fig. 16(a)). Next, the learning object functions of the optimal solutions of the initial generation in the
in the next part of the first learning path section of the offspring ant-genetic algorithm, the pheromone decay value, and the total
8040 C.-C. Hsu, C.-C. Ho / Expert Systems with Applications 39 (2012) 8030–8043

Table 6
Fuzzy relationships between pre-test evaluation items and learning objects.

Evaluation item Learning object (16)

01 02 03 04 05 06
Evaluation item 1 [0.55, 0.7, 0.75, 0.8] [0, 0.1, 0.15, 0.25] [0.4, 0.45, 0.5, 0.65] [0, 0.2, 0.3, 0.4] [0.75, 0.8, 0.85, 1] [0, 0.15, 0.2, 0.25]
Evaluation item 2 [0.75, 0.9, 0.95, 1] [0.75, 0.85, 0.9, 1] [0.75, 0.8, 0.85, 1] [0.3, 0.5, 0.6, 0.7] [0.55, 0.6, 0.65, 0.8] [0.2, 0.35, 0.4, 0.45]
Evaluation item 3 [0.75, 0.9, 0.95, 1] [0.75, 0.85, 0.9, 1] [0.75, 0.8, 0.85, 1] [0.3, 0.5, 0.6, 0.7] [0, 4, 0.45, 0.5, 0.65] [0, 0.15, 0.2, 0.25]
Evaluation item 4 [0.55, 0.7, 0.75, 0.8] [0, 0.1, 0.15, 0.25] [0.55, 0.6, 0.65, 0.8] [0.6, 0.8, 0.9, 1] [0.75, 0.8, 0.85, 1] [0, 0.15, 0.2, 0.25]
Evaluation item 6 [0.2, 0.35, 0.4, 0.45] [0, 0.1, 0.15, 0.25] [0.4, 0.45, 0.5, 0.65] [0.6, 0.8, 0.9, 1] [0.75, 0.8, 0.85, 1] [0.35, 0.5, 0.55, 0.6]
Evaluation item 7 [0.35, 0.5, 0.55, 0.6] [0.2, 0.3, 0.35, 0.45] [0.4, 0.45, 0.5, 0.65] [0, 0.2, 0.3, 0.4] [0.55, 0.6, 0.65, 0.8] [0, 0.15, 0.2, 0.25]
Evaluation item 8 [0, 0.15, 0.2, 0.25] [0, 0.1, 0.15, 0.25] [0, 0.05, 0.1, 0.25] [0.3, 0.5, 0.6, 0.7] [0.2, 0.25, 0.3, 0.45] [0.55, 0.7, 0.75, 8]
Evaluation item [0, 0.15, 0.2, 0.25] [0, 0.1, 0.15, 0.25] [0.2, 0.25, 0.3, 0.45] [0, 0.2, 0.3, 0.4] [0, 4, 0.45, 0.5, 0.65] [0.2, 0.35, 0.4, 0.45]
Evaluation item [0.55, 0.7, 0.75, 0.8] [0.2, 0.3, 0.35, 0.45] [0.4, 0.45, 0.5, 0.65] [0.6, 0.8, 0.9, 1] [0.55, 0.6, 0.65, 0.8] [0.2, 0.35, 0.4, 0.45]
Evaluation item [0, 0.15, 0.2, 0.25] [0.35, 0.45, 0.5, 0.6] [0.4, 0.45, 0.5, 0.65] [0.3, 0.5, 0.6, 0.7] [0.2, 0.25, 0.3, 0.45] [0, 0.15, 0.2, 0.25]

Learning object (16)

07 08 09 0A 0B 0C
Evaluation item 1 [0.45, 0.65, 0.75, 0.85] [0.75, 0.8, 0.85, 1] [0, 0.15, 0.25, 0.4] [0.3, 0.5, 0.6, 0.7] [0.4, 0.45, 0.5, 0.65] [0.15, 0.35, 0.45, 0.55]
Evaluation item 2 [0, 0.2, 0.3, 0.4] [0, 0.05, 0.1, 0.25] [0.25, 0.35, 0.4, 0.45] [0.15, 0.35, 0.45, 0.55] [0.75, 0.8, 0.85, 1] [0.6, 0.8, 0.9, 1]
Evaluation item 3 [0, 0.2, 0.3, 0.4] [0.2, 0.25, 0.3, 0.45] [0.3, 0.45, 0.55, 0.7] [0, 0.2, 0.3, 0.4] [0.75, 0.8, 0.85, 1] [0.6, 0.8, 0.9, 1]
Evaluation item 4 [0.6, 0.8, 0.9, 1] [0.55, 0.6, 0.65, 0.8] [0.15, 0.3, 0.4, 0.55] [0.3, 0.5, 0.6, 0.7] [0.4, 0.45, 0.5, 0.65] [0.15, 0.35, 0.45, 0.55]
Evaluation item 6 [0.6, 0.8, 0.9, 1] [0.75, 0.8, 0.85, 1] [0.15, 0.3, 0.4, 0.55] [0.45, 0.65, 0.75, 0.85] [0.2, 0.25, 0.3, 0.45] [0, 0.2, 0.3, 0.4]
Evaluation item 7 [0.3, 0.5, 0.6, 0.7] [0.4, 0.45, 0.5, 0.65] [0.3, 0.45, 0.55, 0.7] [0.15, 0.35, 0.45, 0.55] [0.4, 0.45, 0.5, 0.65] [0.15, 0.35, 0.45, 0.55]
Evaluation item 8 [0, 0.2, 0.3, 0.4] [0.75, 0.8, 0.85, 1] [0, 0.15, 0.25, 0.4] [0, 0.2, 0.3, 0.4] [0, 0.05, 0.1, 0.25] [0, 0.2, 0.3, 0.4]
Evaluation item [0, 0.2, 0.3, 0.4] [0.4, 0.45, 0.5, 0.65] [0.6, 0.75, 0.85, 1] [0.15, 0.35, 0.45, 0.55] [0.55, 0.6, 0.65, 0.8] [0.45, 0.65, 0.75, 0.85]
Evaluation item [0.3, 0.5, 0.6, 0.7] [0.4, 0.45, 0.5, 0.65] [0.6, 0.75, 0.85, 1] [0.3, 0.5, 0.6, 0.7] [0.75, 0.8, 0.85, 1] [0.6, 0.8, 0.9, 1]
Evaluation item [0.15, 0.35, 0.45, 0.55] [0.2, 0.25, 0.3, 0.45] [0.45, 0.6, 0.7, 0.85] [0.3, 0.5, 0.6, 0.7] [0.75, 0.8, 0.85, 1] [0.45, 0.65, 0.75, 0.85]

Learning object (16)

0D 0E 0F 10 11 12
Evaluation item 1 [0.15, 0.35, 0.45, 0.55] [0.75, 0.85, 0.9, 1] [0.45, 0.6, 0.7, 0.85] [0.55, 0.65, 0.7, 0.8] [0.15, 0.35, 0.45, 0.55] [0.15, 0.3, 0.4, 0.55]
Evaluation item 2 [0, 0.2, 0.3, 0.4] [0.35, 0.45, 0.5, 0.6] [0.6, 0.75, 0.85, 1] [0.3, 0.45, 0.5, 0.6] [0.3, 0.5, 0.6, 0.7] [0.3, 0.45, 0.55, 0.7]
Evaluation item 3 [0, 0.2, 0.3, 0.4] [0.2, 0.3, 0.35, 0.45] [0, 0.15, 0.25, 0.4] [0, 0.1, 0.15, 0.25] [0.15, 0.35, 0.45, 0.55] [0.45, 0.6, 0.7, 0.85]
Evaluation item 4 [0, 0.2, 0.3, 0.4] [0.2, 0.3, 0.35, 0.45] [0, 0.15, 0.25, 0.4] [0.3, 0.45, 0.5, 0.6] [0, 0.2, 0.3, 0.4] [0.15, 0.3, 0.4, 0.55]
Evaluation item 6 [0.45, 0.65, 0.75, 0.85] [0.2, 0.3, 0.35, 0.45] [0.15, 0.3, 0.4, 0.55] [0.3, 0.45, 0.5, 0.6] [0, 0.2, 0.3, 0.4] [0.15, 0.3, 0.4, 0.55]
Evaluation item 7 [0.6, 0.8, 0.9, 1] [0.75, 0.85, 0.9, 1] [0.45, 0.6, 0.7, 0.85] [0.55, 0.65, 0.7, 0.8] [0.15, 0.35, 0.45, 0.55] [0, 0.15, 0.25, 0.4]
Evaluation item 8 [0, 0.2, 0.3, 0.4] [0, 0.1, 0.15, 0.25] [0, 0.15, 0.25, 0.4] [0.2, 0.3, 0.35, 0.45] [0, 0.2, 0.3, 0.4] [0, 0.15, 0.25, 0.4]
Evaluation item [0.15, 0.35, 0.45, 0.55] [0.35, 0.45, 0.5, 0.6] [0.6, 0.75, 0.85, 1] [0.75, 0.85, 0.9, 1] [0.45, 0.65, 0.75, 0.85] [0.45, 0.6, 0.7, 0.85]
Evaluation item [0.15, 0.35, 0.45, 0.55] [0.35, 0.45, 0.5, 0.6] [0.15, 0.3, 0.4, 0.55] [0.75, 0.85, 0.9, 1] [0.6, 0.8, 0.9, 1] [0.6, 0.75, 0.85, 1]
Evaluation item [0, 0.2, 0.3, 0.4] [0.2, 0.3, 0.35, 0.45] [0, 0.15, 0.25, 0.4] [0.55, 0.65, 0.7, 0.8] [0.45, 0.65, 0.75, 0.85] [0.6, 0.75, 0.85, 1]

Learning object (16)

13 14 15 16 17 18
Evaluation item 1 [0.15, 0.25, 0.35, 0.55] [0.4, 0.45, 0.5, 0.65] [0.35, 0.45, 0.5, 0.6] [0.4, 0.45, 0.5, 0.65] [0.35, 0.45, 0. 5, 0.6] [0.2, 0.25, 0.3, 0.45]
Evaluation item 2 [0.6, 0.7, 0.8, 1] [0.55, 0.6, 0.65, 0.8] [0.2, 0.3, 0.35, 0.45] [0, 0.05, 0.1, 0.25] [0.55, 0.65, 0.7, 0.8] [0.75, 0.8, 0.85, 1]
Evaluation item 3 [0.6, 0.7, 0.8, 1] [0.75, 0.8, 0.85, 1] [0, 0.1, 0.15, 0.2] [0, 0.05, 0.1, 0.25] [0.55, 0.65, 0.7, 0.8] [0.75, 0.8, 0.85, 1]
Evaluation item 4 [0.3, 0.4, 0.5, 0.7] [0.2, 0.25, 0.3, 0.45] [0.55, 0.65, 0.7, 0.8] [0.2, 0.25, 0.3, 0.45] [0.35, 0.45, 0.5, 0.6] [0.4, 0.45, 0.5, 0.65]
Evaluation item 6 [0.15, 0.25, 0.35, 0.55] [0.2, 0.25, 0.3, 0.45] [0.75, 0.85, 0.9, 1] [0.75, 0.8, 0.85, 1] [0.2, 0.3, 0.35, 0.45] [0.2, 0.25, 0.3, 0.45]
Evaluation item 7 [0.3, 0.4, 0.5, 0.7] [0.4, 0.45, 0.5, 0.65] [0.75, 0.85, 0.9, 1] [0.2, 0.25, 0.3, 0.45] [0.35, 0.45, 0.5, 0.6] [0.4, 0.45, 0.5, 0.65]
Evaluation item 8 [0, 0.1, 0.2, 0.4] [0.2, 0.25, 0.3, 0.45] [0, 0.1, 0.15, 0.2] [0.75, 0.8, 0.85, 1] [0, 0.1, 0.15, 0.25] [0.2, 0.25, 0.3, 0.45]
Evaluation item [0.3, 0.4, 0.5, 0.7] [0.55, 0.6, 0.65, 0.8] [0.2, 0.3, 0.35, 0.45] [0, 0.05, 0.1, 0.25] [0.55, 0.65, 0.7, 0.8] [0.75, 0.8, 0.85, 1]
Evaluation item [0.3, 0.4, 0.5, 0.7] [0.75, 0.8, 0.85, 1] [0.35, 0.45, 0.5, 0.6] [0.55, 0.6, 0.65, 8] [0.75, 0.85, 0.9, 1] [0.75, 0.8, 0.85, 1]
Evaluation item [0.15, 0.25, 0.35, 0.55] [0.75, 0.8, 0.85, 1] [0.35, 0.45, 0.5, 0.6] [0, 0.05, 0.1, 0.25] [0.75, 0.85, 0.9, 1] [0.5, 0.6, 0.65, 0.8]

(
number of learning objects, respectively. A pheromone decay value 1
LBbest
; if the path from LOi to LOj is the optimal path;
is usually within the range 0 < q 6 1, and the greater the value is, Dsi;j ¼
0; otherwise;
the faster the pheromone decays. To prevent the pheromone from
decaying too fast and to generate a more optimal solution, this sys- ð19Þ
tem selects 0.1 as the decay value.
where si;j ; Dsi;j , and LBbest denote the pheromone values in the learn-
Initially, the ant-genetic algorithm sets pheromone values in all
ing path LOi to LOj, the pheromone values to be increased, and the
learning paths at the maximum value smax. After the optimal solu-
learning benefits of the optimal solution of this generation,
tion of each generation is generated, the learning path of the solu-
respectively.
tion updates the pheromone value. If the learning path does not
appear in the optimal solution at this point, the pheromone value
needs to be decreased. However, if the learning path appears in the 3. System implementation
optimal solution at this point, the pheromone value needs to be
increased. Fig. 18 shows the environment of the competency-based intel-
ligent M-learning system, which includes an embedded NFC
si;j ¼ ð1  qÞsi;j þ Dsi;j ; M-learning device, an embedded NFC display, and an intelligent
 C.-C. Hsu, C.-C. Ho / Expert Systems with Applications 39 (2012) 8030–8043 8041

Table 7
Calculation results of fuzzy relationships among learning objects.

Result Learning object (16)

01 02 03 04 05 06
Trapezoidal fuzzy set [0.62, 0.69, 0.82, 0.87] [0.52, 0.62, 0.67, .077] [0.64, 0.69, 0.74, 0.89] [0.35, 0.55, 0.65, 0.75] [0.63, 0.68, 0.73, 0.88] [0.09, 0.24, 0.29, 0.34]
Center of gravity 0.75 0.65 0.75 0.57 0.74 0.23

Learning object (16)

07 08 09 0A 0B 0C
Trapezoidal fuzzy set [0.39, 0.59, 0.69, 0.79] [0.4, 0.45, 0.5, 0.65] [0.52, 0.67, 0.75, 0.92] [0.37, 0.57, 0.67, 0.77] [0.66, 0.71, 0.76, 0.91] [0.53, 0.73, 0.83, 0.93]
Center of gravity 0.61 0.51 0.72 0.59 0.77 0.75

Learning object (16)

0D 0E 0F 10 11 12
Trapezoidal fuzzy set [0.1, 0.3, 0.4, 0.5] [0.62, 0.72, 0.77, 0.87] [0.49, 0.64, 0.74, 0.89] [0.71, 0.82, 0.87, 0.97] [0.53, 0.73, 0.83, 0.93] [0.54, 0.69, 0.74, 0.94]
Center of gravity 0.32 0.75 0.69 0.84 0.75 0.73

Learning object (16)

13 14 15 16 17 18
Trapezoidal fuzzy set [0.5, 0.6, 0.7, 0.9] [0.66, 0.71, 0.76, 0.91] [0.32, 0.42, 0.47, 0.57] [0.27, 0.32, 0.37, 0.52] [0.73, 0.83, 0.88, 0.98] [0.68, 0.75, 0.8, 0.95]
Center of gravity 0.68 0.77 0.45 0.38 0.86 0.8

The bold values are computed using Eq. (12).

Table 8
3.1. Intelligent learner interface
Relationships between learning objects and prerequisite objects.

Prerequisite object Prerequisite object When logging into the M-learning interface of the M-learning
LO01 LO02   LO0C LO18   device, learners enter user IDs and passwords for identification
LO02    LO0E LO12   (Fig. 20(a)). Learners then take a pre-test for competency assess-
LO03 LO17   LO0F LO0B  
ment and select items for career planning. The example here de-
LO04    LO10 LO04 LO13 LO17
LO05 LO02   LO11    scribes a situation in which a learner answers 10 questions
LO07 LO04   LO12 LO02 LO11  wrongly in the pre-test shown in Table 1 (items 1, 2, 3, 4, 6, 7, 8,
LO08 LO05 LO07 LO13 LO13 LO0C   10, 11, and 12). Fig. 20(b) demonstrates a career-selection inter-
LO09 LO07 LO12  LO14 LO10   face, and Table 5 shows the fuzzy relationship between the se-
LO0A LO07   LO17 LO01  
LO0B LO09 LO13  LO18   
lected career plan and the learning objects. Table 6 indicates the
fuzzy relationship between the learning objects and the pre-test,
excluding the correctly answered questions. With fuzzy interpola-
tion and the two relationship tables, the system can figure out the
Table 9 learner’s competency index for each learning object in relation to a
Learner’s past performance. learner’s career plan. Table 7 shows the products of multiplication
Past performance Past performance
of each learning object in the pretest and career plans in Tables 5
and 6. Learning objects 06, 0D, 15, and 16 are unnecessary learning
LO01 65 LO0C 
LO02  LO0E 
objects, meaning there are a total of 20 learning objects that are
LO03  LO0F 70 must-have courses for the learner.
LO04  LO10 
LO05  LO11 
3.2. Personalized learner
LO07 85 LO12 
LO08  LO13 75
LO09 80 LO14  Table 8 shows the relationship between the 20 must-have
LO0A  LO17  learning objects and their prerequisite objects. If a given course
LO0B  LO18 
has two prerequisite objects, the learner must finish both prereq-
uisites prior to beginning the must-have course. Table 9 indicates
the learner’s past performance in the 20 learning objects.
learner. The embedded NFC display is composed of an embedded Fig. 21 demonstrates the crossover of two sets of randomly se-
computer and an USB-based NFC reader. The NFC M-learning de- lected chromosomes based on the optimal solution of the initial
vice also has an embedded computer and an USB-based NFC reader generation. The gray shade indicates the paths with the higher
(Fig. 19(a)), and it connects to a back-end server through Wi-Fi. The pheromone values. Fig. 22 shows the optimal learning path with
two devices communicate through their built-in NFC (Fig. 19(b)). the maximum learning benefit of 15.90. Fig. 23 illustrates the
The back-end server is responsible for receiving and transmitting chromosomes of the optimal solution of the initial generation.
information to the M-learning device, such as pre-test information The maximum learning benefit is 1.15, which is used to calculate
and the content of learning objects. Personalized competency anal- smax and smin. The optimal solution of the initial generation is taken
yses and learning paths are also completed by the back-end server. as an example for pheromone updates in all learning paths.
This system was developed in Windows XP with Visual Studio Fig. 24(a) and (b) show the pheromone before and after the
2003, C++/MFC-related mobile phone software programs. updates, respectively.
8042 C.-C. Hsu, C.-C. Ho / Expert Systems with Applications 39 (2012) 8030–8043

08 0B 14 07 11 02 03 12 0E 10 05 17 09 13 0A 18 01 0F 04 0C

04 11 12 14 13 07 01 0A 05 17 09 0E 18 0C 03 08 0F 10 0B 02

04 11 12 0E 18 01 0A 05 17 09 13 07 0C 03 08 0F 10 0B 14 02

Fig. 21. Chromosome crossover.

02 05 11 12 0E 01 17 03 04 10 14 07 0A 18 0C 13 08 09 0B 0F

Fig. 22. Optimal learning path.

04 07 02 13 12 05 17 0E 18 09 0B 14 01 0A 0F 0C 03 08 10 11

Fig. 23. Optimal chromosomes of initial generations.

8.70 8.70 7.83 7.83

LO07 LO07
8.70 8.70
8.70 8.70
8.70 8.70 7.83 7.83
8.70 8.70 8.70 7.83 7.83 7.83
8.70 7.83
LO04 LO02 LO04 LO02
8.70 8.70 8.70 7.83 7.83 7.83
8.70 7.83
8.70 8.70 7.83 7.83
8.70 8.70 7.83 7.83
L LO01

8.70 8.70 7.83 7.83

(a) (b)
Fig. 24. Pheromone update of learning paths.

4. Discussion and conclusion to constructing learning paths based on genetic algorithm, Huang
et al. (2007) introduced case-based reasoning for conclusive tests
Many studies have proposed various methods for establishing and assessments in an attempt to adjust the content of the learning
learning paths. Chen et al. (2006) proposed an agent-based person- paths. The advantages include its consideration of the course diffi-
alized internet teaching system. Its advantages include its ability to culty and consistency and the learners’ competencies, producing a
consider the difficulty and consistency of courses and the learners’ higher quality of learning paths; the disadvantages include con-
competencies and to adjust the learning path in a timely manner. cerns about the accuracy of standardized assessments. Yang and
The disadvantages include having too few options for feedback as Wu (2009) proposed an ant colony system based on attributes
well as expressed concerns about the accuracy of the course diffi- and ant colony optimization. The advantages are higher quality
culty index. Acampora et al. (2008) proposed a self-adaptive E- learning tailored to the attributes of learners and learning objects,
learning system based on ontological descriptions of a learning and consideration for learners’ learning styles and course difficulty.
environment and a memetic algorithm. Its advantages include low- This study proposes a competency-based intelligent M-learning
er calculation complexity and faster speed in generating a sub- system. The advantages of this system are the selection of suitable
optimal solution. The disadvantages are the lack of consideration learning objects based on pre-tests and career plans and its consid-
of course difficulty and learner competencies. Hong et al. (2007) re- eration for the connections among learning objects and the learn-
ferred to the learners’ pre-test outcomes and established learning ers’ past performance.
paths using a genetic algorithm. The advantages include its ability This study proposes a competency-based intelligent M-learning
to consider the course difficulty and the learners’ competencies to system, which applies peer-to-peer transmission of NFC technol-
produce higher quality learning paths based on pre-test outcomes. ogy to the migration of learning content from a smaller mobile de-
The disadvantages are lower calculation speed, concerns for the vice to a bigger screen for the convenience of learners. This system
accuracy of pre-tests, and the course difficulty index. In addition analyzes learners’ competencies through a pre-test. It generates a
 C.-C. Hsu, C.-C. Ho / Expert Systems with Applications 39 (2012) 8030–8043 8043

competency index based on the wrong answers given by learners Hong, C. M., Chen, C. M., Chang, M. H., & Chen, S. C. (2007). Intelligent web-based
tutoring system with personalized learning path guidance. In Paper presented in
in the pre-test. In addition, it considers the connection between
IEEE international conference on advanced learning technologies (pp. 512–516).
learners’ career plans and learning objects, and it uses fuzzy inter- Hsu, C. C., & Tang, Y. W. (2006). An intelligent mobile learning system for on-the-job
polation to determine the connection between the competency in- training of luxury brand firms. In Australian conference on artificial intelligence
dex and learning objects in an attempt to select suitable objects. (Vol. 4304, pp. 749–759).
Hua, Z., & Lu, H. (2006). Web browsing on small-screen devices: A multi-client
The ant-genetic algorithm includes the defining characteristic of collaborative approach. IEEE Pervasive Computing, 5(2), 78–84.
ant colony algorithms, which is the calculation of pheromone. Huang, M. J., Huang, H. S., & Chen, M. Y. (2007). Constructing a personalized E-
The pheromone calculation determines the amount of pheromone learning system based on genetic algorithm and case-based reasoning
approach. Expert Systems with Applications, 33(3), 551–564.
accumulating in every learning path of the optimal solution of Huang, Y. M., Kuo, Y. H., Lin, Y. T., & Cheng, S. C. (2008). Toward interactive mobile
every generation. During this process, the pheromone value of a synchronous learning environment with context-awareness service. Computers
good learning path increases, while the value of a bad learning path & Education, 51(3), 1005–1226.
Huang, Y. M., Lin, Y. T., & Cheng, S. C. (2010). Effectiveness of a mobile plant learning
decreases. The calculation process is thus accelerated. system in a science curriculum in Taiwanese elementary education. Computers
This learning system has the following features. First, it applies & Education, 54(1), 47–58.
peer-to-peer transmission of NFC technology to the migration of Huang, Z., & Shen, Q. (2006). Fuzzy interpolative reasoning via scale and move
transformations. IEEE Transactions on Fuzzy Systems, 14(2), 340–359.
learning content to a bigger screen to address the problem of the Huseyin, U., Nadire, C., & Erinc, E. (2009). Using mobile learning to increase
small screen sizes of mobile devices, effectively displaying the con- environmental awareness. Computers & Education, 52(2), 381–389.
tent on a suitable interface, particularly, for elderly learners. Sec- Hwang, G. J., & Chang, H. F. (2011). A formative assessment-based mobile learning
approach to improving the learning attitudes and achievements of students.
ond, this system combines learners’ career plans and current
Computers & Education, 56(4), 1023–1031.
competencies and then uses fuzzy interpolation to determine the Hwang, G. H., Chen, B., Lai, S. W., & Lin, C. H. (2009). Construction, application, and
connections between learning objects, their competencies, and effect analysis of the adaptive guide system for a museum. Digital Learning
their career plans. In this manner, it can select learning objects that Technology, 1(4), 307–325.
Hwang, G. J., Kuo, F. R., Yin, P. Y., & Chuang, K. H. (2010). A heuristic algorithm for
meet the learners’ needs. Third, when there are a large number of planning personalized learning paths for context-aware ubiquitous learning.
genes in the chromosomes, the hybrid algorithm based on genetic Computers & Education, 54(2), 404–415.
and ant colony algorithms can overcome the problem of the slow Kinshuk, S. J., Sutinen, E., & Goh, T. (2003). Mobile technologies in support of
distance learning. Asian Journal of Distance Education, 1(1), 60–68.
calculation speed of traditional genetic algorithms. Finally, by con- Kontopoulos, E., Vrakas, D., Kokkoras, F., Bassiliades, N., & Vlahavas, I. (2008). An
sidering the connections among learning objects and learners’ past ontology-based planning system for E-course generation. Expert Systems with
performance, this system can use the ant-genetic algorithm to Applications, 35(1–2), 398–406.
Lee, Z. J., Lee, C. Y., & Su, S. F. (2002). An immunity based ant colony optimization
establish learning paths efficiently. algorithm for solving weapon-target assignment problem. Applied Soft
Computing, 2(1), 39–47.
References Liu, T.-Y., Tan, T.-H., & Chu, Y.-L. (2009). Outdoor natural science learning with an
RFID-supported immersive ubiquitous learning environment. Educational
Technology & Society, 12(4), 161–175.
Acampora, G., Gaeta, M., Loia, V., Ritrovato, P., & Salerno, S. (2008). Optimizing
Marquez, J. M., Ortega, J. A., Abril, L. G., & Velasco, F. (2008). Creating adaptive
learning path selection through memetic algorithms. In Proceedings of the IEEE
learning paths using ant colony optimization and Bayesian networks. In
international joint conference on neural networks (pp. 3869–3875).
Proceedings of the IEEE international joint conference on neural networks (Vol. 1,
Altiparmak, F., & Karaoglan, I. (2007). A genetic ant colony optimization approach
pp. 3834–3839).
for concave cost transportation problems. In Proceedings of the IEEE congress
McClelland, D. C. (1973). Testing for competence rather than for intelligence.
evolutionary computation (CEC-2007) (pp. 1685–1692).
American Psychologist, 28(1), 1–24.
Carro, R. M., Pulido, E., & Rodriguez, P. (1999). Dynamic generation of adaptive
Motiwalla, L. F. (2007). Mobile learning: A framework and evaluation. Computers &
Internet-based courses. Journal of Network and Computer Applications, 22(4),
Education, 49(3), 581–596.
249–257.
Noe, R. A. (2005). Employee training and development (3rd ed.). Singapore: McGraw-
Chang, C. S., Chen, T. S., & Hsu, W. H. (2010). The study on integrating WebQuest
Hill.
with mobile learning for environmental education. Computers & Education.
Quinn, C. (2001). Get ready for M-learning. Training and Development, 20(2),
doi:10.1016/j.compedu.2010.12.005.
20–21.
Chen, T. S., Chang, C. S., Lin, J. S., & Yu, H. L. (2009). Context-aware writing in
Rogers, Y., Price, S., Randell, C., Fraser, D. S., Weal, M., & Fitzpatrick, G. (2005). Ubi-
ubiquitous learning environments. Research and Practice in Technology Enhanced
learning integrates indoor and outdoor experiences. Communications of the ACM,
Learning, 4(1), 61–82.
48(1), 55–59.
Chen, G. D., Chang, C. K., & Wang, C. Y. (2008). Ubiquitous learning website: Scaffold
Sampson, D. (2006). Exploiting mobile and wireless technologies in vocational
learners by mobile devices with information-aware techniques. Computers &
training. In Proceedings of the IEEE international workshop on wireless, mobile and
Education, 50(1), 77–90.
ubiquitous technology in education (pp. 63–65).
Chen, C. M., & Chung, C. J. (2008). Personalized mobile English vocabulary learning
Sharples, M. (2000). The design of personal mobile technologies for lifelong
system based on item response theory and learning memory cycle. Computers &
learning. Computers & Education, 34(3–4), 177–193.
Education, 51(2), 624–645.
Shih, J. L., Chu, H. C., Hwang, G. J., & Kinshuk, S. J. (2010). An investigation of
Chen, C. M., & Duh, L. J. (2008). Personalized web-based tutoring system based on
attitudes of students and teachers about participating in a context-aware
fuzzy item response theory. Expert Systems with Applications, 34(4), 2298–2315.
ubiquitous learning activity. British Journal of Educational Technology.
Chen, C. M., Liu, C. Y., & Chang, M. H. (2006). Personalized curriculum sequencing
doi:10.1111/j.1467-8535.2009.01020.x.
utilizing modified item response theory for web-based instruction. Expert
Stutzle, T., & Hoos, H. H. (2000). MAX–MIN ant system. Future Generation Computer
Systems with Applications, 30(2), 378–396.
Systems, 16(8), 889–914.
Chiou, C. K., Tseng, J. C. R., Hwang, G. J., & Heller, S. (2010). An adaptive navigation
Triantafillou, E., Georgiadou, E., & Economides, A. A. (2008). The design and
support system for conducting context-aware ubiquitous learning in museums.
evaluation of a computerized adaptive test on mobile devices. Computers &
Computers & Education, 55(2), 834–845.
Education, 50(4), 1319–1330.
Chu, H. C., Hwang, G. J., Huang, S. X., & Wu, T. T. (2008). A knowledge engineering
Wang, T. I., Wang, K. T., & Huang, Y. M. (2008). Using a style-based ant colony
approach to developing e-libraries for mobile learning. Electronic Library, 26(3),
system for adaptive learning. Expert Systems with Applications, 34(4),
303–317.
2449–2464.
Chu, H. C., Hwang, G. J., Tsai, C. C., & Tseng, J. C. R. (2010). A two-tier test approach to
Xu, Y. L., Lim, M. H., Ong, Y. S., & Tang, J. (2006). A GA-ACO-local search hybrid
developing location-aware mobile learning systems for natural science courses.
algorithm for solving quadratic assignment problem. In Proceedings of the
Computers & Education, 55(4), 1618–1627.
international conference on genetic and evolutionary computation (pp. 599–606).
Churchill, D., & Hedberg, J. (2008). Learning object design considerations for small-
Yang, J. H., & Chen, Y. L. (2006). Universal access and content adaptation in mobile
screen handheld devices. Computers & Education, 50(3), 881–893.
learning. In International conference on advanced learning technologies (pp. 1172–
Dorigo, M., & Stutzle, T. (2004). Ant colony optimization. NY: MIT Press.
1173).
Hafeez, K., & Essmail, E. A. (2007). Evaluating organization core competences and
Yang, Y. J., & Wu, C. (2009). An attribute-based ant colony system for adaptive
associated personal competencies using analytical hierarchy process.
learning object recommendation. Expert Systems with Applications, 36(2),
Management Research News, 30(8), 530–547.
3034–3047.
Heath, B. P., Herman, R. L., Lugo, G. G., Reeves, J. H., Vetter, R. J., & Ward, C. R. (2005).
Project numina: Enhancing student learning with handheld computers.
Computer, 38(6), 46–53.