You are on page 1of 15

Computers and Education: Artificial Intelligence 1 (2020) 100003

Contents lists available at ScienceDirect

Computers and Education: Artificial Intelligence


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

A fuzzy expert system-based adaptive learning approach to improving


students’ learning performances by considering affective and
cognitive factors
Gwo-Jen Hwang a, *, Han-Yu Sung b, Shao-Chen Chang c, Xing-Ci Huang a
a
Graduate Institute of Digital Learning and Education, National Taiwan University of Science and Technology, 43, Sec.4, Keelung Rd., Taipei, 106, Taiwan
b
Department of Allied Health Education and Digital Learning, National Taipei University of Nursing and Health Sciences, No.365, Ming-te Road, Peitou District, Taipei
City, Taiwan
c
Department of International Bachelor Program in Informatics and the Department of Information Communication, Yuan Ze University, No. 135, Yuandong Rd., Zhongli
Dist., Taoyuan City, 320, Taiwan

A R T I C L E I N F O A B S T R A C T

Keywords: Scholars have indicated the importance of providing guidance and support for individual learners. In the past
Artificial intelligence decades, most studies have developed adaptive learning systems to address this issue mainly based on students’
Adaptive learning cognitive status, such as their learning achievements. However, several educators have pointed out the need to
Fussy inference
take learners’ affective status into account. Therefore, this study proposed an expert system approach by taking
Expert system
Personalization
into account both the affective and cognitive status of individual learners. An adaptive learning system was
implemented based on the proposed approach. In addition, an experiment was conducted in a fifth-grade
Mathematics course to compare the learning performances and perceptions of the students who learned with
the adaptive learning system with affective and cognitive status analysis, a cognitive-status-based adaptive
learning system, and a conventional learning system. The ANCOVA results revealed that the adaptive learning
model with the affective and cognitive performance analysis mechanism outperformed the other two approaches
in terms of improving the students’ learning achievement (F ¼ 3.12, p < 0.05) and reducing their mathematical
anxiety (F ¼ 5.59, p < 0.01). In addition, it was found that the proposed approach helped the low achievers
successfully complete the learning tasks, while those learning with the conventional cognitive factor-based
approach were more likely to give up some learning tasks, and were more reliant on the detailed version of
the instructional materials.

1. Introduction that is, a model that takes students’ individual differences into account in
the design of the learning materials in order to obtain better learning
In the traditional one-on-many instructional approach, due to the achievement (Islam, Beer, & Slack, 2015).
limitation of such factors as teaching process, human power, and so on, Using artificial intelligence (AI) technologies to simulate teachers’
instructors can only provide all students with the same materials, and knowledge and experience to provide individual students with person-
they tend to employ an identical teaching approach and rate of progress. alized supports or guidance has been recognized as a potential solution
It is difficult to consider the learning needs of individual students (Pai, Kuo, Liao, & Liu, 2020; Xiao & Yi, 2020). Several researchers have
(Sampayo-Vargas, Cope, He, & Byrne, 2013). This means that students begun to develop such adaptive learning systems using AI technologies.
with better proficiency levels are limited by instructors’ teaching prog- For instance, Pai et al. (2020) proposed an intelligent tutoring system to
ress so they cannot make effective use of class time to learn. On the other help fifth-grade students learn the mathematical unit of multiplication
hand, students with insufficient proficiency levels may not be able to and division. They found that the intelligent tutoring system (ITS) not
understand the course content (Looi et al., 2009). To tackle this problem, only improved students’ mathematics learning performance, but also
researchers have proposed a student-centered adaptive learning model; increased their learning motivation. Hooshyar et al. (2016) applied an

* Corresponding author.
E-mail addresses: gjhwang.academic@gmail.com (G.-J. Hwang), hanyu.sung@gmail.com (H.-Y. Sung), shao.chen76@gmail.com (S.-C. Chang), ccha813@gmail.
com (X.-C. Huang).

https://doi.org/10.1016/j.caeai.2020.100003
Received 18 July 2020; Received in revised form 14 August 2020; Accepted 14 August 2020
2666-920X/© 2020 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
G.-J. Hwang et al. Computers and Education: Artificial Intelligence 1 (2020) 100003

online game-based formative assessment in a flowchart-based intelligent learners’ prior knowledge can increase learners’ learning performance
tutoring system. They reported that the 52 university students improved (Dochy, 1992; Moerkerke, 1996; Weibelzahl, 2001). Learning content
their learning interest, attitude and computer science learning achieve- that is either too difficult or too easy cannot facilitate learning motivation
ments after the learning activity. The results revealed that the experi- (Schunk, 1999); providing teaching models not based on learning needs
mental group exhibited better help-seeking strategies than the control may become the obstacle for students’ cognitive development (Looi et al.,
group. As can be seen, compared to traditional computer-assisted 2009). As a result, how to evaluate learners’ cognitive status and then
learning systems, adaptive learning systems can promote students’ design adaptive learning activities is an issue that is worth emphasizing.
learning performance (Truong, 2016). Chen, Liu, and Chang (2006) indicated that even though several re-
The cognitive domain includes one’s memory, thinking, recognition searchers have paid attention to developing learning systems appropriate
and application of things. It emphasizes that the main feature of for learners’ reading habits, they did not carefully consider the situation
knowledge is the acquisition and application of knowledge (Bloom, of the gap between students’ learning capabilities and the difficulty level
Engelhart, Furst, Hill, & Krathwohl, 1956). Researchers have also of the recommended courses, which resulted in cognitive load and sub-
revealed that adaptive learning should not just take cognitive status into sequently reduced learning performance. Therefore, Chen et al. (2006)
consideration (Nakic, Granic, & Glavinic, 2015); instead, learners’ designed a personalized learning system which was able to dynamically
emotional factors, such as the development and change in a person’s change learning paths and content, and developed and embedded into it
interest, attitude, value, appreciation, and adaptation, need to be taken numerous forms of learning content relevant to programming language.
into account (Krathwohl, Bloom, & Masia, 1964, p. 1). For example, Kim, Besides, they classified the learning content, and set up the coefficient
Lee, and Ryu (2013) showed that students’ characteristics and person- based on its difficulty level; they cross-analyzed students’ feedback on
ality traits are also crucial factors that influence personalized learning. the difficulty level after they had learned the content and the coefficient
Hence, in the development of adaptive learning, apart from students’ the system set up, and then arranged the following adequate learning
cognitive status, such factors as students’ learning attitude, learning in- content. The results showed that this adaptive learning system which
terest, and affection are issues worthy of attention. For instance, Ehrman, could adjust learning content based on students’ cognitive capabilities
Leaver, and Oxford (2003) indicated that learner difference includes the could significantly enhance students’ learning efficiency. Chen, Hsieh,
following three factors: learning strategy and affection (e.g., learning and Hsu (2007) specified that ignoring learners’ misconceptions would
anxiety and attitudes). In addition, Kobsa, Koenemann, and Pohl (2001) cause a decrease in learning achievement. The recent adaptive learning
further specified that in order to achieve true “adaptive” learning, system could arrange course content or ways of representation according
various and different aspects should be incorporated and considered. to learners’ characteristics, but it was difficult to assist learners in
To address this issue, this study aimed to develop an adaptive examining whether they were able to truly understand the learning
learning system by taking both the affective and cognitive factors into content embedded in the system on their own. In an effort to help
account. To explore the effectiveness of the proposed approach, an learners clarify their misunderstandings during their learning process,
experiment was conducted to evaluate the effects of the developed sys- Chen et al. (2007) developed systematic association rules for each
tem on students’ learning achievement, mathematical anxiety and learning concept, and presented personalized learning content and
cognitive load. learning paths based on learners’ different cognitive status after they
provided wrong answers, which then clarified their misconceptions and
2. Literature review increased their learning achievement. Tseng, Chu, Hwang, and Tsai
(2008) developed a learning system that provided personalized learning
2.1. Artificial intelligence in adaptive learning content based on cognitive status. The system set up the major learning
materials and achievement test for each unit; students had to read ma-
Artificial intelligence refers to a system that simulates humans to terials in the learning system and finish the achievement test. The
make judgments and decisions that exhibit the characteristics of human learning system would then provide learning materials at an adequate
intelligence (Akerkar, 2014). In the past decades, AI technologies have difficulty level for the next learning activity based on the student’s
been adopted by researchers to provide personalized learning guidance achievement test score in that unit as the reference for evaluating their
and supports for individual students in various courses, such as data cognitive status. This could effectively promote the learners’ learning
structure (Daghestani, Ibrahim, Al-Towirgi, & Salman, 2020), computer achievement. It was found that the experimental group using adaptive
programming (Palanca, Terrasa, Carrascosa, & Julian, 2019) and math- learning materials had better learning efficiency than the control group.
ematics (Pai et al., 2020). Monova-Zheleva (2005) further defined However, several educators have indicated that learning achievement
adaptive learning as a learning model that provides appropriate learning from a broad perspective should consider the three aspects of cognition,
content based on a user’s needs (Yau & Joy, 2010); these needs could be affection, and skills. Kim et al. (2013) specified that learners’ personal-
justified and evaluated by using the standards of learning technology ities and characteristics are crucial factors that influence personalized
such as learner profiles, competency definitions, sequencing rules, learning. Numerous researchers have further found that such individual
learning objects, and so on. Rosmalen et al. (2006) regarded adaptive differences as learners’ gender, prior knowledge and learning preferences
learning as a capability to obtain information through learners’ on-site might bring about different learning performances (Cassidy, 2012; Chen,
task performance or evaluation results in order to improve individual 2010; Chen & Macredie, 2010; Chen & Sun, 2012; Jonassen & Gra-
learners’ learning experience (Zhao & Okamoto, 2011). Yaghmaie and bowski, 2012). Kobsa et al. (2001) revealed that to achieve “adaptive”
Bahreininejad (2011) considered adaptive learning as a distinctive learning, the learner’s profile including his/her interests, capabilities as
learning process generated based on learners’ characteristics, interest, well as the learning portfolio and learning environment data such as the
and performance so as to accomplish such learning objectives as learning environment and adopted equipment should be considered to
enhancing learning performance, learning satisfaction, etc. Yang, accommodate the learner’s individual needs. In addition, researchers
Hwang, and Yang (2013) considered adaptive learning as the approach of have specified that in the field of teaching design, in order to develop an
revising learning content or ways of presentation in the learning process adaptive learning environment, individual differences between learners
based on learners’ capabilities, interest, conditions or personality traits. such as their prior knowledge, skills, abilities, and affective variables
Thus, adaptive learning includes the two major aspects of cognitive status should be taken into consideration, which could help them complete the
and affective status. learning tasks in a more effective way (Shute & Zapata-Rivera, 2012;
Considering learners’ individual cognitive status to provide corre- Vandewaetere, Desmet, & Clarebout, 2011). However, due to the diffi-
sponding learning content, learning paths or ways of presentation are an culty of evaluating overall mental capabilities, in most of the practical
essential mechanism; making use of learning content which matches applications, learning achievement only included comprehension ability

2
G.-J. Hwang et al. Computers and Education: Artificial Intelligence 1 (2020) 100003

of course content and memory as the evaluation standards in the cogni- (ACALS) is implemented using HTML5, CSS3, JavaScript and MySQL. It
tive aspect. Among the parameters of developing adaptive learning sys- consists of a learning module, a testing model, a learning status recording
tems, test scores are usually used to represent learning achievement module, and an expert system module, as shown in Fig. 1. The learning
(Hwang, Sung, Hung, & Huang, 2013). However, in order to achieve model enables students to learn by watching online videos and
truly adaptive learning, not only cognitive status but also other aspects completing learning sheets. The testing model is used to perform tests to
should be taken into account. evaluate individual students’ knowledge level. The learning status
recording module records the learning behaviors of individual students
2.2. Adaptive learning based on affection performance in the system logs for further analysis. The expert system module is used
to analyze students’ affective status and to determine the learning con-
Morshead (1965) defined the affective domain as the domain of tent for individual students based on their affective status and knowledge
exploring the development and change in a person’s interest, attitude, level.
value, appreciation, and adaptation. The educational objectives of the
affective domain can be divided into five phases: Receiving or Attending, 3.1. Expert system module
Responding, Valuing, Organization, and Characterization by value or
value complex. For each learning unit, there are three candidate versions of the
Receiving or Attending in the affective domain includes how to learning materials; that is, “standard version,” “detailed version” and
arouse, maintain, and guide students’ concentration, while Responding “advanced” version. Each version consists of the major learning mate-
shows whether students are able to actively participate in learning, and is rials, which present the standard learning content (see Fig. 2), and the
mostly relevant to learning interest (Al-Samarraie & Saeed, 2018). Apart “standard,” “detailed” or “advanced” instructions provided by the
from incorporating learning states into the learning process as the teacher.
adaptive factor for affective status, researchers have also integrated Fig. 3 shows the standard instructions. This version is the same as the
learner interest into developing an adaptive learning system. Ku and traditional instructional content the teacher delivers to students in the
Sullivan (2002) designed an adaptive learning system integrated with class. Therefore, it is designed for average students or for those students
learner interest. At first, the system investigated learners’ favorite food, whose learning status is unclear when starting a new unit.
sport types, and store types, and then integrated them into the mathe- Fig. 4 shows the detailed version, in which detailed explanations with
matics questions as the learning content. The results revealed that this reference to the prior knowledge required for learning the unit are pro-
could not only enhance students’ learning achievement, but also had a vided. This version is designed for those whose learning status is below
positive influence on their learning attitude. Moreover, numerous studies average.
have verified that paying attention to students’ affection in their learning Fig. 5 shows the advanced version, which provides extensive sup-
process could facilitate and improve their learning experiences (Johnson, plementary materials for those whose learning performances are above
Rickel, & Lester, 2000; Lehman, Matthews, D’Mello, & Person, 2008). the average.
Furthermore, when developing adaptive learning systems, students’ In the expert system, the students’ learning status, such as their an-
affection (e.g., happy, interested, depressed, bored, focused) should be swers to the exercises, is recorded as their cognitive status, and their
examined to provide adaptive feedback (Katsionis & Virvou, 2004) online learning behaviors are analyzed to determine their affectie status.
because these affective moods imply the possibility that students would Fuzzy inferencing is used to analyze the students’ cognitive and affective
begin to engage in activities which were not related to learning (Rodrigo status by using the fuzzy membership functions and inference rules
et al., 2007). In view of this, Hernandezet, Sucar, and Arroyo-Figueroa suggested by Hwang (1998). For example, the fuzzy membership func-
(2010) took advantage of a dynamic Bayesian network to develop an tion for analyzing students’ cognitive performance is shown in Fig. 6.
affective model in advance, and embedded it into an intelligent tutoring Moreover, the following fuzzy rules are used to determine the version
system to predict students’ affective reactions in an effort to improve of learning materials provided to individual students based on their
students’ learning achievement and learning motivation. recent learning status:
In brief, cognition can change affection, and vice versa. Learners’
individual differences not only exist in intelligence or learning achieve- (1) R1: If Current version is Advanced version and learning level is
ment, but may also be influenced by such personality traits as learning High
style, cognitive style, and so on. Thus, considering learners’ character-
istics in the learning context to develop an adaptive environment that Then go to Advanced version.
accommodates individual learning needs can not only have substantial
benefits for students’ learning achievement, but also has the potential to (2) R2: If Current version is Advanced version and learning level is
become the key to encouraging learners’ learning motivation and atti- Medium
tude. However, even though the field of education has long emphasized
that developing students’ interest, motivation, and attitude is an essential Then go to Advanced version.
objective, cognitive status has been taken as the foremost evaluation
indicator in schools at all levels, while the development of the affective (3) R3: If Current version is Advanced version and learning level is
field has received insufficient emphasis. Above all, adaptive learning Low
aims to attribute the learning agent to the learners themselves, and en-
ables every learner to be taught in accordance with their aptitude, per- Then go to Standard version.
sonal characteristics or learning status. As a result, the present study
aimed to develop an adaptive learning system integrated with individual (4) R4: If Current version is Advanced version and learning level is
learners’ affective and cognitive status, and to explore its effects on Poor
learners’ learning achievement, cognitive load and mathematical
anxiety. Then learn this lesson once again.

3. An expert system-based adaptive learning system with (5) R5: If Current version is Standard version and learning level is
affective and cognitive factors High

The affective and cognitive factors-based adaptive learning system Then go to Advanced version.

3
G.-J. Hwang et al. Computers and Education: Artificial Intelligence 1 (2020) 100003

Fig. 1. The structure of the adaptive learning system.

Fig. 2. An example of the major learning materials.

(6) R6: If Current version is Standard version and learning level is (8) R8: If Current version is Standard version and learning level is
Medium Poor

Then go to Standard version. Then learn this lesson once again.

(7) R7: If Current version is Standard version and learning level is Low (9) R9: If Current version is Detailed version and learning level is High

Then go to Detailed version. Then go to Standard version.

4
G.-J. Hwang et al. Computers and Education: Artificial Intelligence 1 (2020) 100003

Fig. 3. An example of the standard learning materials.

Fig. 4. An example of the detailed learning materials.

5
G.-J. Hwang et al. Computers and Education: Artificial Intelligence 1 (2020) 100003

Fig. 5. An example of the advanced learning materials.

Fig. 6. A graphic of a function on cognitive performance.

(10) R10: If Current version is Detailed version and learning level is (11) R11: If Current version is Detailed version and learning level is
Medium Low

Then go to Detailed version. Then go to Detailed version.

6
G.-J. Hwang et al. Computers and Education: Artificial Intelligence 1 (2020) 100003

(12) R12: If Current version is Detailed version and learning level is with the conventional learning system. The learning content in the sys-
Poor tem includes four topics, that is, parallelogram, triangle, trapezoid, and
compound graphics in the mathematics course. The learning processes of
Then learn this lesson once again. the three versions of the learning system are given as follows:
Based on the students’ cognitive performance, suitable learning ma-
terials are provided. (16) The adaptive learning process with the affective and cognitive
In terms of the affective performance analysis mechanism, students’ performance analysis mechanism (see Fig. 10): due to the cogni-
degrees of concentration, patience, and learning willingness are evalu- tive performance analysis mechanism, students had different
ated based on the system logs. versions of materials, which were determined after evaluating
their cognitive status by the fuzzy theory. After reading the
(13) “Concentration” refers to the degree to which students focus on teaching materials, there were 10 test items for practice. Based on
materials. During the process when students read through the this test result, the fuzzy theory was adopted to decide the ma-
materials, the system would randomly show a pop-up window. terials with an appropriate level of difficulty for the next unit. If
The study processed the time it took for students to respond to the the result was poor, the system would not show the next learning
window by using the fuzzy-theory engine; the quicker they unit; students must stay in that unit until they reached a certain
responded (see Fig. 7), the higher concentration level they had, proficiency level. At the same time, the system would investigate
and vice versa. When a student had low concentration, it would the students’ affective status (concentration, patience, and
pop up with short answer questions relevant to the learning ma- learning willingness) when they read the materials and did the
terials (see Fig. 8) to arouse the student’s concentration. exercise. Through the fuzzy theory, the system could judge
(14) “Patience” is an indicator to examine whether students were pa- whether adequate assistance was provided.
tient with the recent unit. When they entered into each unit about (17) The adaptive learning process with the cognitive performance
the area of a diagram, the system would start to calculate their analysis mechanism: due to the cognitive performance analysis
learning time. If a student’s learning time was higher than the mechanism, students had different versions of materials, which
average of all students, the system would show a tip or reminder to was determined after evaluating their cognitive status by the fuzzy
take a rest (see Fig. 9). theory. After reading the teaching materials, there were 10 test
(15) “Learning willingness” refers to students’ willingness to devote items for practice. Based on this test result, the fuzzy theory was
themselves to learning in the unit; it integrates the two variables adopted to decide the materials with appropriate difficulty level
of concentration and learning time to conduct the fuzzy theory. If for the next unit. If the result was poor, the system would not show
students had low learning willingness, the system would show a the next learning unit; students must stay in that unit until they
joke to encourage their learning willingness. reached a certain proficiency level.
(18) The learning process with the conventional learning system: stu-
dents used the learning materials in the standard version to learn
3.2. Adaptive learning process during the learning process.

Three versions of the learning system were implemented in this study: The interface of the adaptive learning system (see Fig. 11): students
(1) The adaptive learning process with the affective and cognitive per- learned in order and were required to finish the achievement test before
formance analysis mechanism; (2) The adaptive learning process with the entering the next unit. The check mark symbol represented those units
cognitive performance analysis mechanism; and (3) The learning process

Fig. 7. Pop-up window to examine concentration.

7
G.-J. Hwang et al. Computers and Education: Artificial Intelligence 1 (2020) 100003

Fig. 8. Pop-up window for short answer questions.

Fig. 9. Reminder to take a rest.

that had already been learned, while the play button was used to show 4. Experiment design
the unit that was currently being learned. When students acquired
learning materials of a graphic, they were asked to finish the learning 4.1. Participants
achievement test to strengthen their memory. The interface in the upper
part showed the progress bar, enabling the students to know how much A quasi-experimental design was adopted in the present study. The
they had finished. In addition, the options were all designed by the re- participants were six classes of fourth graders from an elementary school
searchers; they could provide corresponding evaluation and feedback on in southern Taiwan. Two classes with 53 students (26 male and 27 fe-
misconceptions based on the answers the students chose. male) were assigned to experimental group A (an adaptive learning

8
G.-J. Hwang et al. Computers and Education: Artificial Intelligence 1 (2020) 100003

Fig. 10. The affective and cognitive performance analysis mechanism.

Fig. 11. The interfaces of the adaptive learning system.

system with the affective and cognitive performance analysis mecha- system with the cognitive performance analysis mechanism; and the
nism), two classes with 53 students (26 male and 27 female) were control group used the conventional learning system with no adaptive
experimental group B (an adaptive learning system with the cognitive mechanism. In the learning activities, students from the three groups all
performance analysis mechanism), and the other two classes with 56 used tablet computers to enter the adaptive learning system. When
students (26 male and 30 female) were the control group (a conventional reading the learning materials, if there were calculation items, students
learning system). The average age of the participants was 10. All the could make use of scratch paper designed in the present study to help
participants had already learned the definition of basic graphics with memory and thinking. After they finished learning the materials of a
(including square, rectangle) and relevant concepts before the experi- unit, they began to do the exercise in the learning system. After the
ment, and they were taught by the same mathematics teacher. learning activities, all students were asked to finish the 40-min post-test
for the area unit, the mathematical anxiety post-questionnaire and the
4.2. Experimental procedure cognitive load questionnaire. The complete experimental process is
shown in Fig. 12.
Before the experiment, the three groups of students were all required
to take the 30-min pre-test for the area unit, and to complete the math- 4.3. Measuring tools
ematical anxiety pre-questionnaire. Afterwards, they started to engage in
the learning activities for 120 min. Experimental group A made use of the The learning achievement test in the present study was designed by
adaptive learning system with the affective and cognitive performance two experienced math teachers; the course content consisted of learning
analysis mechanism; experimental group B adopted the adaptive learning materials in regular elementary school mathematics courses. Before the

9
G.-J. Hwang et al. Computers and Education: Artificial Intelligence 1 (2020) 100003

Fig. 12. The experiment process.

experiment, a pre-test was administered to ensure that the participants assumption of homogeneity of regression coefficients for each group, was
had similar prior knowledge. The pre-test questions included five True/ satisfied for the remainder of the tests. The ANCOVA results revealed that
False items (10%), 12 Fill-in-the-Blanks (28%), two Try & Check (18%), after excluding the effects of the mathematics pre-test scores, there was a
and 12 Calculation items (42%); the perfect score was 100. There were significant difference between the two groups (F ¼ 3.12, p ¼ 0.04 <0
13 post-test questions, with six Fill-in-the-Blanks (30%) and seven .05), indicating that the students had significantly different learning
Calculation items (70%); the perfect score was 100. The reliability of the achievements as a result of using the different learning models. As shown
post-test was 0.74. in Table 1, the average achievement test score of experimental group A
The mathematical anxiety questionnaire developed by Fennema and (M ¼ 67.3) was significantly higher than that of experimental group B (M
Sherman (1976) was used to examine students’ mathematical anxiety. ¼ 58.17) and the control group (M ¼ 58.65). This indicated that the
The questionnaire consisted of four sub-scales, namely worry (10 items), students who used the adaptive learning system with the affective and
dislike (9 items), test anxiety (8 items), and perception of stress (5 items). cognitive performance analysis mechanism outperformed those who
The Cronbach’s α was 0.93. used the adaptive learning system with only the cognitive performance
The cognitive load questionnaire was modified by Hwang, Yang, and analysis mechanism and those who used the conventional learning sys-
Wang (2013) based on the measure developed by Paas, Tuovinen, Tab- tem. The findings showed that the adaptive learning system which took
bers, and Van Gerven (2003). The first five items were classified into the learners’ affective and cognitive status into account at the same time
mental load, while the last three items were for mental efforts. The was conducive to improving their learning achievement.
questionnaire with a total of eight items aimed to investigate students’
cognitive load while they were using the adaptive learning system to 5.1.1. Analysis of mathematical anxiety
learn. The questionnaire used a 5-point Likert scale (1 ¼ strongly disagree, One-way ANCOVA was used to uncover whether students in different
5 ¼ strongly agree). The overall Cronbach’s α of the cognitive load scale learning models had significant differences in the mathematical anxiety
was 0.92, while that of mental efforts and mental load was 0.86 and 0.85, post-questionnaire. The result (F ¼ 1.75; p ¼ 0.18 >0 .05) of the Levene’s
respectively. test for equality of variances showed that the assumption of the homo-
geneity of variances in the three groups was met. The test results (F ¼
5. Experimental results 0.22; p ¼ 0.79 >0 .05) of between-subjects effects, which indicated the
assumption of homogeneity of regression coefficients for each group, was
5.1. Analysis of mathematic learning achievement satisfactory for the remainder of the tests.
The ANCOVA results depicted that after excluding the effects of the
One-way ANCOVA was adopted to examine the three groups’ math- mathematical anxiety pre-questionnaire, the between-group test (F ¼
ematics learning achievement test scores. The result (F ¼ 0.41; p ¼ 0.66 5.59; p ¼ 0.004 <0 .01) reached a significant level, showing that the
>0 .05) of the Levene’s test for equality of variances indicated that the students had significant differences in their mathematical anxiety when
assumption of the homogeneity of variances in the groups was met. The using the different learning models. Table 2 shows that the average
test results (F ¼ 0.93; p ¼ 0.39 >0 .05) of the interaction effects between mathematical anxiety score of experimental group A was significantly
the independent variable and covariance, which indicated the lower than that of the control group, but experimental groups A and B did

Table 1
The ANCOVA results of students’ learning achievement.
Group N Mean SD Adjusted Mean SE F η2 Post Hoc (LSD)

(a)Experimental Group A 53 66.32 18.37 67.31 2.44 3.12* 0.038 (a)>(b)


(b)Experimental Group B 53 58.77 20.73 58.17 2.44 (a)>(c)
(c)Control Group 56 59.02 16.52 58.65 2.37

*p < 0.05.

10
G.-J. Hwang et al. Computers and Education: Artificial Intelligence 1 (2020) 100003

Table 2
The ANCOVA results of students’ mathematical anxiety.
Group N Mean SD Adjusted Mean SE F η2 Post Hoc (LSD)

(a)Experimental Group A 53 2.52 1.05 2.53 0.07 5.59** 0.066 (a)<(c)


(b)Experimental Group B 53 2.64 1.02 2.72 0.07
(c)Control Group 56 2.86 0.99 2.78 0.07

**p < 0.01.

not differ significantly in terms of their mathematical anxiety. This


Table 4
indicated that students using the adaptive learning system with the af-
The coding scheme for analyzing the learning records.
fective and cognitive performance analysis mechanism had significantly
lower levels of mathematical anxiety than those who used the conven- Versions of Learning units
materials
tional learning system. The results demonstrated that students’ mathe- Parallelogram Triangle Trapezoid Compound
matical anxiety could not be significantly decreased by using the (A) (B) (C) graphics (D)
adaptive learning system with only the cognitive performance analysis Detailed (1) A1 B1 C1 D1
mechanism or the conventional learning system. Besides, compared to Standard (2) A2 B2 C2 D2
the conventional learning system with no adaptive mechanism, the Advanced (3) A3 B3 C3 D3

adaptive learning system with the affective and cognitive performance


analysis mechanism could help reduce students’ mathematical anxiety. table of the high achievers in experimental group A and experimental
group B. According to the definition of Bakeman and Gottman (1997), we
5.1.2. Analysis of cognitive load converted the adjusted residual table into sequential behavior graphs.
One-way analysis of variance (ANOVA) was employed to explore the Figs. 13 and 14 show the learning paths of the high achievers in
effects of different learning models on students’ cognitive load. The ho- experimental group A, who learned with the adaptive learning system
mogeneity test showed no significant difference (Levene ¼ 0.17; p ¼ with the affective and cognitive performance analysis mechanism, and
0.84 > 0.05) among the three groups. The overall result of cognitive load those in experimental group B, who learned with the adaptive learning
is shown in Table 3. The average scores of experimental groups A and B system with the cognitive performance analysis mechanism. It was found
and the control group are 2.59, 2.70, and 2.72, respectively. There was that the learning paths of the students mainly consisted of the standard
no significant difference among the three groups in terms of their version (B2→C2→D2 in Fig. 13 and A2→B2→C2 in Fig. 14), and the
cognitive load (F ¼ 0.18; p ¼ 0.83 > 0.05). Besides, the average score of materials of the advanced version (C3→D3 in Fig. 13 and B3→C3→D3 in
cognitive load was between 2.5 and 2.7 with a moderate range, indi- Fig. 14). This indicates that the high achievers were capable of learning
cating that the difficulty level of the materials and the arrangement of the easier units (e.g., parallelogram and triangle) in the standard and
this system were all acceptable for the three groups of students. That is, advanced versions. However, when faced with the harder learning units
the arrangement of the learning activities was at a moderate difficulty (e.g., trapezoid and compound graphics), the two learning approaches
level and quantity, which did not cause the students cognitive load that brought about different influences. Unlike the learning paths of experi-
was either too high or too low. mental group B with only the cognitive performance analysis mechanism,
the support of the affective and cognitive performance analysis mecha-
5.1.3. Analysis of learning records nism enabled learners in experimental group A to maintain their cogni-
The learning records of the students in the two experimental groups tive status at a certain level even though they were faced with more
are further analyzed. The students are divided into high achievers and difficult learning units (C2→D2 and C3→D3 in Fig. 13). In contrast,
low achievers based on their post-test scores; that is, the students whose C1→D1 in Fig. 14 implied that experimental group B did not perform
post-test scores were in the top 33% are high achievers and those in the well without the affective status support. This indicated that the affective
last 33% are low achievers. The portfolio coding scheme is shown in status support could be conducive to improving students’ cognitive status
Table 4. It consists of the four units (i.e., parallelogram (A), triangle (B), when they came across difficult learning materials.
trapezoid (C), and compound graphics (D)) with three versions of In addition, we also generated sequential behavior graphs based on
learning materials (i.e., detailed (1), standard (2) and advanced (3)). the adjusted residual table of the low-achievement students (Table 7 and
To analyze the differences in the students’ learning behavior in the Table 8).
two experimental groups, behavioral sequence analysis was conducted. When comparing the learning paths of the low achievers, it was found
At first, we used the GSEQ 5.1 to generate the adjusted residual table that they generally learned with the detailed version (A1→B1→C1 in
(Quera, Bakeman, & Gnisci, 2007), and the z score was used to evaluate Fig. 15 and B1→C1→D1 in Fig. 16), implying that the low achievers in
the coding results of each group. When the z value is higher than 1.96, it experimental group B really needed detailed instructions from the
represents that the sequence behavior has statistical significance (Bake- teacher. Several learning paths (e.g., A1→B1→C1, B2→C2→D2 in Fig. 15
man & Gottman, 1997). Table 5 and Table 6 shows the adjusted residual and B1→C1→D1 in Fig. 16) further reveal that the cognitive status of the
low achievers was generally in the same level when they were learning
the more difficult units. Nevertheless, when comparing the learning
Table 3 paths of the two experimental groups, it was found that the affective
The ANOVA results of students’ cognitive load. performance analysis mechanism could improve the students’ cognitive
N M SD F status. For instance, C2→D3 in Fig. 15 specified that, by considering the
Cognitive load Experimental Group A 53 2.59 1.13 0.18 students’ affective status, it is possible to enhance low achievers’
Experimental Group B 53 2.70 1.06 cognitive performances. That is, it is important to encourage or remind
Control Group 56 2.72 1.15 the low achievers in time to help them overcome their learning
Mental efforts Experimental Group A 53 2.45 1.19 0.85
difficulties.
Experimental Group B 53 2.66 1.08
Control Group 56 2.73 1.20
Mental load Experimental Group A 53 2.83 1.18 0.20 6. Discussion and conclusions
Experimental Group B 53 2.76 1.15
Control Group 56 2.69 1.18
With regard to learning achievement, students using the adaptive

11
G.-J. Hwang et al. Computers and Education: Artificial Intelligence 1 (2020) 100003

Table 5
The Adjusted residual table of the high achievers in experimental group A.
Z A1 B1 C1 D1 A2 B2 C2 D2 A3 B3 C3 D3

A1 0.09 0 0.16 0 1 1.57 0.2 0.16 0 0.2 0.26 0.13


B1 0 0 0 0 0 0 0 0 0 0 0 0
C1 0.13 0 4.47* 0 1.41 0.65 0.29 0.22 0 0.29 0.37 0.18
D1 0 0 0 0 0 0 0 0 0 0 0 0
A2 0.78 0 2.25 0 3.59* 0.8 2.93 1.04 0 0.1 3.75 1.83
B2 0.56 0 1.78 0 2.04 0.34 3.03* 0.97 0 0.88 1.81 0.79
C2 0.16 0 0.27 0 0.57 1.12 0.35 3.57* 0 0.35 1.95 0.22
D2 0.13 0 0.22 0 1.41 0.65 0.29 4.47* 0 0.29 0.37 0.18
A3 0 0 0 0 0 0 0 0 0 0 0 0
B3 0.2 0 0.35 0 0.44 0.46 1.88 0.35 0 0.46 1.29 0.29
C3 0.18 0 0.32 0 2.02 1.3 0.41 0.32 0 0.41 3.66* 7.9*
D3 0.09 0 0.16 0 1.01 0.64 0.2 0.16 0 0.2 0.26 0.13

*p < 0.05.

Table 6
The Adjusted residual table of the high achievers in experimental group B.
Z A1 B1 C1 D1 A2 B2 C2 D2 A3 B3 C3 D3

A1 0.17 3.67* 0.2 0 1.31 0.85 0.56 0.4 0 0.12 0.12 0.12
B1 0.27 2.09* 2.9* 0 2.1 0.23 0.43 0.64 0 5.37* 0.19 0.19
C1 0.12 0.19 6.98* 0 0.93 0.57 0.4 0.28 0 0.08 0.08 0.08
D1 0 0 0 0 0 0 0 0 0 0 0 0
A2 1.23 0.12 2.03 0 5.65* 0.51 5.1 3.37 0 1.17 1.17 1.17
B2 0.73 1.17 0.53 0 2.13 0.69 4.02* 1.77 0 0.52 1.95 0.52
C2 0.48 0.76 0.59 0 2.67 1.05 2.37* 4.03* 0 0.34 0.34 2.99*
D2 0.36 0.58 0.45 0 2.16 0.95 0.79 5.68* 0 0.26 0.26 0.26
A3 0 0 0 0 0 0 0 0 0 0 0 0
B3 0 0 0 0 0 0 0 0 0 0 0 0
C3 0 0 0 0 0 0 0 0 0 0 0 0
D3 0 0 0 0 0 0 0 0 0 0 0 0

*p < 0.05.

Fig. 13. Comparison of learning paths with the affective and cognitive performance analysis mechanism: high-learning achievement (experimental group A).

learning system with the affective and cognitive performance analysis three groups of students, the difficulty level and the loading of the
mechanism outperformed those using the adaptive learning system with learning activities would not result in cognitive load that was either too
only the cognitive performance analysis mechanism and those using the high or too low.
conventional learning system. This revealed that the adaptive learning The results showed that experimental group A outperformed experi-
system with both the affective and cognitive performance analysis mental group B and the control group on mathematics learning
mechanism would be conducive to facilitating students’ learning achievement, which was in line with the points of several scholars who
achievement. As for mathematical anxiety, the adaptive learning system emphasized the importance of taking affective factors into account when
with only the cognitive performance analysis mechanism and the con- developing educational systems (Shephard, 2008; Snow & Farr, 1987;
ventional learning system could not significantly decrease students’ Tsai, Li, & Cheng, 2017; Tseng et al., 2008). For example, Snow and Farr
mathematical anxiety. In addition, compared to the conventional (1987) indicated that if a sound learning theory did not take holistic
learning system with no adaptive mechanism, the adaptive learning perspectives into account (e.g., not integrating the cognitive and affective
system with the affective and cognitive performance analysis mechanism aspects), this theory would be incomplete or impractical. Thus, learners’
could reduce students’ mathematical anxiety. In the aspect of cognitive affective status in their learning process is not only an issue worth
load, the difficulty level and arrangement of the materials in the system focusing on, but could also significantly influence education and learning
were all acceptable to the three groups of students. In other words, for the (Jones & Issroff, 2005).

12
G.-J. Hwang et al. Computers and Education: Artificial Intelligence 1 (2020) 100003

Fig. 14. Comparison of learning paths with the cognitive performance analysis mechanism: high-learning achievement (experimental group B).

Table 7
The Adjusted residual table of the low achievers in experimental group A.
Z A1 B1 C1 D1 A2 B2 C2 D2 A3 B3 C3 D3

A1 0 0 0 0 0 0 0 0 0 0 0 0
B1 0 0.08 0.11 0.08 1.76 0.68 0.42 0.47 0 0.21 0.17 0.13
C1 0 0.08 0.11 13.11* 0.57 0.68 0.42 0.47 0 0.21 0.17 0.13
D1 0 0.08 0.11 0.08 0.57 0.68 0.42 2.14* 0 0.21 0.17 0.13
A2 0 0.71 1 0.71 1.16 3.88* 1.64 3.49 0 1.38 1.6 1.23
B2 0 1.54 0.63 0.65 0.95 1.43 2.93* 3.56 0 0.06 0.51 1.13
C2 0 0.38 1.58 0.38 0.33 1.92 0.43 1.8 0 1.03 0.87 1.07
D2 0 0.43 0.61 0.43 1.27 3.38 1.22 7.16* 0 0.1 0.27 0.75
A3 0 0 0 0 0 0 0 0 0 0 0 0
B3 0 0.19 0.27 0.19 1.48 0.79 1.05 1.17 0 0.51 4.52* 0.33
C3 0 0.15 0.22 0.15 0.03 1.37 0.56 0.37 0 0.42 0.35 3.59*
D3 0 0.11 0.15 0.11 0.81 0.96 0.6 1.18 0 0.29 0.25 5.24*

*p < 0.05.

Table 8
The Adjusted residual table of the low achievers in experimental group B.
Z A1 B1 C1 D1 A2 B2 C2 D2 A3 B3 C3 D3

A1 0.08 0.12 0.12 0.15 0.58 1.27 0.4 0.39 0 0.15 0.08 0.08
B1 0.12 0.17 5.89* 0.21 0.83 0.34 0.57 0.56 0 0.21 0.12 0.12
C1 8.43* 0.17 0.17 4.76* 0.83 1.13 0.57 0.56 0 0.21 0.12 0.12
D1 0.15 0.21 0.21 3.82* 1.02 1.38 0.71 2.75* 0 0.26 0.15 0.15
A2 0.7 0.52 1 1.22 0.89 3.27* 1.83 2.75 0 0.02 0.7 0.7
B2 0.74 0.45 0.45 1.28 0.97 1 1.01 2.4 0 1.28 0.74 0.74
C2 0.4 0.57 0.57 0.98 0.57 2.82 0.84 3.79* 0 0.98 0.4 0.4
D2 0.34 0.49 0.49 0.6 0.49 2.11 1.5 3.22* 0 0.6 0.34 0.34
A3 0 0 0 0 0 0 0 0 0 0 0 0
B3 0.12 0.17 0.17 0.21 0.83 1.13 0.57 0.56 0 4.76* 8.43* 0.12
C3 0.08 0.12 0.12 0.15 0.58 0.79 0.4 0.39 0 0.15 0.08 11.96*
D3 0 0 0 0 0 0 0 0 0 0 0 0

*p < 0.05.

Mathematical anxiety is an important indicator of students’ poor implied that computer and information technology could decrease stu-
performance (Vukovic, Kieffer, Bailey, & Harari, 2013). In the current dents’ mathematical anxiety (Harris & Harris, 1987; Owolabi, Olanipe-
study, the mathematical anxiety of experimental group A was signifi- kun, & Iwerima, 2018), but also verified that mathematical anxiety
cantly lower than that of the control group. Previous research has spec- would have influences on mathematical learning and cognitive reflection
ified that the on-site learning atmosphere would affect students’ (Morsanyi, Busdraghi, & Primi, 2014). Consequently, paying attention to
formation of mathematical anxiety (Alexander & Martray, 1989); students’ affective change during their learning process could improve
moreover, the instructor’s attitude might become the reason for students’ their learning experiences (Johnson et al., 2000). A learning model
mathematical anxiety (Berebitsky, 1985; Vann, 1993). The learning integrating the cognitive and affective aspects is essential to mathematics
system developed in this study could dynamically investigate students’ learning (Wang et al., 2014). On the other hand, the results of the present
affective status and offer positive feedback and encouragement during study revealed that there was no significant difference in the cognitive
their learning process, which thus influenced the students’ mathematical load of the students in the three groups, indicating that the difficulty level
anxiety. Furthermore, students using the adaptive learning model with and arrangement of materials in the present study would not arouse a
the affective and cognitive performance analysis mechanism out- cognitive load that is too high or too low. Paas, Renkl, and Sweller (2003)
performed the control group on their learning achievement. This not only implied that when materials are carefully designed, learners’ cognitive

13
G.-J. Hwang et al. Computers and Education: Artificial Intelligence 1 (2020) 100003

Fig. 15. Comparison of learning paths with the affective and cognitive performance analysis mechanism: low-learning achievement (experimental group A).

Fig. 16. Comparison of learning paths with the cognitive performance analysis mechanism: low-learning achievement (experimental group B).

load can be well controlled, and therefore would be helpful for learning, and universities to investigate more regarding its effectiveness. In addi-
which corresponded to the findings of the present study. tion, more personal factors of students, such as their preferences and self-
According to the learning log analysis results, it was found that the efficacy, can be taken into account to develop more effective adaptive
high achievers in the cognitive and affective mode were able to complete learning environments.
the learning tasks with the standard version. However, those in the
cognitive-only mode might have failed to complete the tasks with the Statements on open data and ethics
standard version, and hence were guided to learn with the detailed
version when facing the more difficult content (e.g., trapezoid and The participants were protected by hiding their personal information
compound graphics). In terms of the low achievers, those learning with in this study. They were voluntary and they knew that they could with-
both the cognitive and affective mode and the cognitive-only mode draw from the experiment at any time. The data can be provided upon
generally needed to learn with the detailed version. However, it was requests by sending e-mails to the corresponding author.
found that some of the low achievers in the cognitive and affective mode
switched from the detailed to the standard version. This reveals that the Declaration of competing interest
cognitive and affective mode could keep the high achievers at the higher
knowledge level, while helping the low achievers make progress during The authors declare that they have no known competing financial
the learning process. interests or personal relationships that could have appeared to influence
The major contribution of the present study is to show that the con- the work reported in this paper.
ventional adaptive learning model with only the cognitive performance
analysis mechanism may not be good enough in terms of improving Acknowledgements
students’ learning performances as well as encouraging them to learn.
That is, it is important to take both the cognitive and affective factors into This study is supported in part by the Ministry of Science and Tech-
account when developing adaptive learning systems. nology of the Republic of China under contract numbers MOST-109-
Despite the valuable findings from the experimental results, there are 2511-H-011-002-MY3 and MOST-108-2511-H-011-005-MY3.
some limitations in the present study. First, the participants in this study
were fourth graders, and hence the findings might not be used to predict References
the learning outcomes and perceptions of students of different ages.
Second, the proposed approach was applied to a mathematics course; Akerkar, R. (2014). Introduction to artificial intelligence. New Delhi: Prentice-Hall.
therefore, to learn more about the effectiveness of the approach, further Al-Samarraie, H., & Saeed, N. (2018). A systematic review of cloud computing tools for
collaborative learning: Opportunities and challenges to the blended-learning
studies on other courses are required. environment. Computers & Education, 124, 77–91.
Therefore, for future studies, it is suggested that researchers can Alexander, L., & Martray, C. R. (1989). The development of an abbreviated version of the
consider applying the proposed approach to other courses in high schools mathematics anxiety rating scale. Measurement and Evaluation in Counseling and
Development, 22, 143–150.

14
G.-J. Hwang et al. Computers and Education: Artificial Intelligence 1 (2020) 100003

Bakeman, R., & Gottman, J. M. (1997). Observing interaction: An introduction to sequential Monova-Zheleva, M. (2005). Adaptive learning in Web-based educational environments.
analysis. Cambridge: Cambridge university press. Cybernetics and Information Technologies, 5(1), 44–55.
Berebitsky, R. D. (1985). An annotated bibliography of the literature dealing with mathematics Morsanyi, K., Busdraghi, C., & Primi, C. (2014). Mathematical anxiety is linked to reduced
anxiety. Retrieved ERIC Document Reproduction Service. No. ED 257684. cognitive reflection: A potential road from discomfort in the mathematics classroom
Bloom, B. S., Engelhart, M. D., Furst, E. J., Hill, W. H., & Krathwohl, D. R. (1956). to susceptibility to biases. Behavioral and Brain Functions, 10(1), 31.
Taxonomy of educational objectives, handbook I: The cognitive domain. New York: David Morshead, R. W. (1965). Taxonomy of educational objectives, handbook ii: Affective
McKay Co, Inc. domain. Studies in Philosophy and Education, 4(1), 164–170.
Cassidy, S. (2012). Exploring individual differences as determining factors in student Nakic, J., Granic, A., & Glavinic, V. (2015). Anatomy of student models in adaptive
academic achievement in higher education. Studies in Higher Education, 37(7), learning systems: A systematic literature review of individual differences from 2001
793–810. to 2013. Journal of Educational Computing Research, 51(4), 459–489.
Chen, L. H. (2010). Web-based learning programs: Use by learners with various cognitive Owolabi, J., Olanipekun, P., & Iwerima, J. (2018). Mathematics ability and anxiety,
styles. Computers & Education, 54(4), 1028–1035. computer and programming anxieties, age and gender as determinants of
Chen, C. M., Hsieh, Y. L., & Hsu, S. H. (2007). Mining learner profile utilizing association achievement in basic programming. GSTF Journal on Computing, 3(4), 109.
rule for web-based learning diagnosis. Expert Systems with Applications, 33, 6–22. Paas, F., Renkl, A., & Sweller, J. (2003). Cognitive load theory and instructional design:
Chen, C. M., Liu, C. Y., & Chang, M. H. (2006). Personalized curriculum sequencing Recent developments. Educational Psychologist, 38(1), 1–4.
utilizing modified item response theory for web-based instruction. Expert Systems with Paas, F., Tuovinen, J. E., Tabbers, H., & Van Gerven, P. W. (2003). Cognitive load
Applications, 30(2), 378–396. measurement as a means to advance cognitive load theory. Educational Psychologist,
Chen, S. Y., & Macredie, R. (2010). Web-based interaction: A review of three important 38(1), 63–71.
human factors. International Journal of Information Management, 30, 379–387. Pai, K. C., Kuo, B. C., Liao, C. H., & Liu, Y. M. (2020). An application of Chinese dialogue-
Chen, C. M., & Sun, Y. C. (2012). Assessing the effects of different multimedia materials based intelligent tutoring system in remedial instruction for mathematics learning.
on emotions and learning performance for visual and verbal style learners. Computers Educational Psychology, 1–16.
& Education, 59(4), 1273–1285. Palanca, J., Terrasa, A., Carrascosa, C., & Julian, V. (2019). Improving the programming
Daghestani, L. F., Ibrahim, L. F., Al-Towirgi, R. S., & Salman, H. A. (2020). Adapting skills of students in multiagent systems master courses. Computer Applications in
gamified learning systems using educational data mining techniques. Computer Engineering Education, 27(4), 836–845.
Applications in Engineering Education, 28(3), 568–589. Quera, V., Bakeman, R., & Gnisci, A. (2007). Observer agreement for event sequences:
Dochy, F. J. R. C. (1992). Assessment of prior knowledge as a determinant for future learning: Methods and software for sequence alignment and reliability estimates. Behavior
The use of knowledge state tests and knowledge profiles. Utrecht/London: Lemma B.V./ Research Methods, 39(1), 39–49.
Jessica Kingsley Publishers. Rodrigo, M., Baker, R., Maria, L., Sheryl, L., Alexis, M., & Sheila, P. (2007). Affect and
Ehrman, M. E., Leaver, B. L., & Oxford, R. L. (2003). A brief overview of individual usage choices in simulation problem solving environments. In Proceedings of the 13th
differences in second language learning. System, 31(3), 313–330. international conference on artificial intelligence in education (pp. 145–152). CA, USA:
Fennema, E., & Sherman, J. A. (1976). Fennema-Sherman mathematics attitudes scales: Marina Del Ray.
Instruments designed to measure attitudes toward the learning of mathematics by Rosmalen, P., Vogten, H., Van Es, R., Passier, H., Poelmans, P., & Koper, K. (2006).
females and males. Journal for Research in Mathematics Education, 7(5), 324–326. Authoring a full life cycle model in standards-based adaptive e-learning. Educational
Harris, A. L., & Harris, J. M. (1987). Reducing mathematics anxiety with computer Technology & Society, 9(1), 72–83.
assisted instruction. Mathematics and Computer Education, 21(1), 16–24. Sampayo-Vargas, S., Cope, C. J., He, Z., & Byrne, G. J. (2013). The effectiveness of
Hernandez, Y., Sucar, L. E., & Arroyo-Figueroa, G. (2010). November). Evaluating an adaptive difficulty adjustments on students’ motivation and learning in an
affective student model for intelligent learning environments. In Ibero-American educational computer game. Computers & Education, 69, 452–462.
conference on artificial intelligence (pp. 473–482). Berlin, Heidelberg: Springer. Schunk, D. H. (1999). Social-self interaction and achievement behavior. Educational
Hooshyar, D., Ahmad, R. B., Yousefi, M., Fathi, M., Horng, S. J., & Lim, H. (2016). Psychologist, 34(4), 219–227.
Applying an online game-based formative assessment in a flowchart-based intelligent Shephard, K. (2008). Higher education for sustainability: Seeking affective learning
tutoring system for improving problem-solving skills. Computers & Education, 94, outcomes. International Journal of Sustainability in Higher Education, 9(1), 87–98.
18–36. Shute, V. J., & Zapata-Rivera, D. (2012). Adaptive educational systems. Adaptive
Hwang, G. J. (1998). A tutoring strategy supporting system for distance learning on technologies for training and education, 7(27), 1–35.
computer networks. IEEE Transactions on Education, 41(4), 343–351. Snow, R., & Farr, M. (1987). Cognitive-conative-affective processes in aptitude, learning,
Hwang, G. J., Sung, H. Y., Hung, C. M., & Huang, I. (2013). A learning style perspective to and instruction: An introduction. Conative and affective process analysis, 3, l–10.
investigate the necessity of developing adaptive learning systems. Journal of Truong, H. M. (2016). Integrating learning styles and adaptive e-learning system: Current
Educational Technology & Society, 16(2), 188–197. developments, problems and opportunities. Computers in Human Behavior, 55,
Islam, N., Beer, M., & Slack, F. (2015). E-learning challenges faced by academics in higher 1185–1193.
education. Journal of Education and Training Studies, 3(5), 102–112. Tsai, C. Y., Li, Y. Y., & Cheng, Y. Y. (2017). The relationships among adult affective
Johnson, W. L., Rickel, J. W., & Lester, J. C. (2000). Animated pedagogical agents: Face- factors, engagement in science, and scientific competencies. Adult Education
to-face interaction in interactive learning environment. International Journal of Quarterly, 67(1), 30–47.
Artificial Intelligence in Education, 11, 47–78. Tseng, J. C., Chu, H. C., Hwang, G. J., & Tsai, C. C. (2008). Development of an adaptive
Jonassen, D. H., & Grabowski, B. L. (2012). Handbook of individual differences. Learning and learning system with two sources of personalization information. Computers &
Instruction. New York, NY: Routledge press. Education, 51(2), 776–786.
Jones, A., & Issroff, K. (2005). Learning technologies: Affective and social issues in Vandewaetere, M., Desmet, P., & Clarebout, G. (2011). The contribution of learner
computer-supported collaborative learning. Computers & Education, 44(4), 395–408. characteristics in the development of computer-based adaptive learning
Katsionis, G., & Virvou, M. (2004, October). A cognitive theory for affective user environments. Computers in Human Behavior, 27(1), 118–130.
modelling in a virtual reality educational game. In , Vol. 2. 2004 IEEE international Vann, B. D. N. (1993). Mathematics anxiety。parental anxiety as it relates to anxiety in
conference on systems, man and cybernetics (IEEE cat. No. 04CH37583) (pp. children. Dissertation Abstracts International, 53(8), 2724A–2725A.
1209–1213). IEEE. Vukovic, R. K., Kieffer, M. J., Bailey, S. P., & Harari, R. R. (2013). Mathematics anxiety in
Kim, J., Lee, A., & Ryu, H. (2013). Personality and its effects on learning performance: young children: Concurrent and longitudinal associations with mathematical
Design guidelines for an adaptive e-learning system based on a user model. performance. Contemporary Educational Psychology, 38(1), 1–10.
International Journal of Industrial Ergonomics, 43(5), 450–461. Wang, Z., Hart, S. A., Kovas, Y., Lukowski, S., Soden, B., Thompson, L. A., et al. (2014).
Kobsa, A., Koenemann, J., & Pohl, W. (2001). Personalised hypermedia presentation Who is afraid of math? Two sources of genetic variance for mathematical anxiety.
techniques for improving online customer relationships. The Knowledge Engineering Journal of Child Psychology and Psychiatry, 55(9), 1056–1064.
Review, 16(2), 111–155. Weibelzahl, S. (2001). July). Evaluation of adaptive systems. In International conference on
Krathwohl, D. R., Bloom, B. S., & Masia, B. B. (1964). Taxonomy of educational objectives, user modeling (pp. 292–294). Berlin, Heidelberg: Springer.
handbook ii: Affective domain. New York: David McKay Company. Inc, 0-679-30210-7, Xiao, M., & Yi, H. (2020). Building an efficient artificial intelligence model for
0-582-32385. personalized training in colleges and universities. Computer Applications in Engineering
Ku, H. Y., & Sullivan, H. J. (2002). Student performance and attitudes using personalized Education. https://doi.org/10.1002/cae.22235
mathematics instruction. Educational Technology Research & Development, 50(1), Yaghmaie, M., & Bahreininejad, A. (2011). A context-aware adaptive learning system
21–34. using agents. Expert Systems with Applications, 38(4), 3280–3286.
Lehman, B., Matthews, M., D’Mello, S., & Person, N. (2008). What are you feeling? Yang, T. C., Hwang, G. J., & Yang, S. J. H. (2013). Development of an adaptive learning
Investigating student affective states during expert human tutoring sessions. In system with multiple perspectives based on students’ learning styles and cognitive
Proceedings of the 9th international conference on intelligent tutoring systems (ITS 2008) styles. Journal of Educational Technology & Society, 16(4), 185–200.
(pp. 50–59). Berlin, Heidelberg: Springer. Yau, J. Y. K., & Joy, M. (2010). An adaptive context-aware mobile learning framework
Looi, C.-K., Wong, L.-H., So, H.-J., Seow, P., Toh, Y., Chen, W., et al. (2009). Anatomy of a based on the usability perspective. International Journal of Mobile Learning and
mobilized lesson: Learning my way. Computers & Education, 53, 1120–1132. Organisation, 4(4), 378–390.
Moerkerke, G. (1996). Assessment for flexible learning. Utrecht: Lemma. Zhao, X., & Okamoto, T. (2011). Adaptive multimedia content delivery for context aware
u-learning. International Journal of Mobile Learning and Organisation, 5(1), 46–63.

15

You might also like