Journal of Computer Assisted Learning

2.1 Mobile learning in context‐aware environment

Brown, Collins, and Duguid (1989) emphasized the importance and necessity of solving problems in authentic learning activities. Learning in authentic context or environment has been considered as context‐aware ubiquitous learning (Looi et al., 2011). As a cognitive tool, mobile application can help learners absorb knowledge from complex situation (Shadiev, Hwang, Huang, & Liu, 2015). Mobile learning, which can link authentic context and digital learning materials, continues to be concerned by the researchers. Researchers implemented educational experiments to explore the support role of mobile applications in context‐aware ubiquitous learning for the learners, such as in science courses, social science courses, and language courses (Hung, Lin, & Hwang, 2010).

Some studies show that mobile applications that provide guidance and clues for learners to observe the learning objects can help learners get better results in context‐aware environment. Chen et al. (2003) designed a mobile learning system for students to acquire outdoor bird knowledge by searching information through the system, and the study shows that the experimental group performed better than the control group using a learning sheet with a guide book. Hwang and Chang (2011) developed a mobile system for students to gain historical and cultural knowledge by leading students to observe the learning object via asking questions; the study shows that the experimental group performs better than control group with conventional tour‐based learning approach and there was no significant difference between the two groups in cognitive load. Also, Sung, Hwang, and Chang (2016) explored the positive effectiveness of using a mobile application to learn in local temple.

By contrast, some studies show that mobile applications may not always lead to positive results in context‐aware environment. Researchers developed a virtual butterfly ecological system for students to gain science knowledge in campus and found that there was no significant difference between the experimental group using the system and the control group using traditional observation method in post‐test (Tarng, Ou, Yu, Liou, & Liou, 2015). Also, Chu (2014) found that there were negative effects on students' learning achievement and cognitive load mainly because of overloading of working memory; that is, students have to interact with both the digital materials and environment. In a mobile learning activity, the interaction between real world and mobile devices could influence cognitive load of students (Chu, 2014). Therefore, it is important for researchers to take interaction complexity of mobile learning into consideration and rethink the learning process guided by mobile applications with cognitive load, so as to find a more appropriate learning approach and interactive process for students to learn in context‐aware environment.

2.2 Cognitive load in mobile learning

Cognitive load theory was initially developed by Australian cognitive psychologist John Sweller in the late 1980s. The theory emphasizes that working memory capacity has limitations when dealing with novel information (Van Merrienboer & Sweller, 2005). Cognitive load is the total amount of mental activity applied to individual cognitive system within a given time, and there are three types of cognitive load (Sweller, Van Merrienboer, & Paas, 1998; Van Merrienboer & Sweller, 2005), which are intrinsic, extraneous, and germane load. Intrinsic load is associated with the inherent nature of the learning materials, learners' expertise, and an interaction among them. Extraneous load is caused by the format and the manner that information is presented as well as the working memory required to perform learning activities. Germane load is associated with the efforts of learners, which are used to process and comprehend the learning material and can be induced by appropriate instructional design. The sum of these three kinds of load is the total cognitive load (Sweller et al., 1998). The composition of cognitive load is complex and implicit, so researchers have not yet reached a unified understanding of the concept and dimension division. Paas and Van Merrienboer (1994) divided cognitive load into assessment factors and causal factors, and the assessment factors include mental load (ML), mental effort (ME), and behaviour performance. ML, which is associated the interaction between task and subject characteristics, is based on students' knowledge of the task and the number and degree of information interaction (Paas & Van Merrienboer, 1994). ME refers to cognitive capacity required to complete the learning task and can be imposed by improper instructional design; that is, it reflects cognitive load related to the way of structuring and presenting the learning content or the strategy used to guide students, and it can affect learning performance. Meanwhile, it is considered by Deleeuw and Mayer (2008) that performance is an objective and indirect indicator of cognitive load measure; usually, better learning results indicate higher germane load.

To evaluate the effectiveness of mobile learning, many studies have included cognitive load as an investigation dimension. Based on the cognitive load theory of Paas and Van Merrienboer (1994) and Sweller et al. (1998), some researchers converted the Paas cognitive load rating scale into two dimensions: ML and ME (Chu, 2014; Hwang, Yang, & Wang, 2013). The majority of these studies show that mobile learning can help students in experimental group perform better than students in control group (e.g., Wu et al., 2012). Nevertheless, when it comes to cognitive load, conclusions are various. Some show that both the ML and ME in experimental group was lower than those in control group (Lin & Lin, 2016; Shadiev et al., 2015); some show that both the ML and ME in experimental group was higher than those in control group (Sung, Hwang, Liu, & Chiu, 2014), whereas others find that there was no significant difference between two groups in both ML and ME (Chang, Shih, & Chang, 2017). Additionally, some studies show that students in experimental class did not perform better than students in control class as their ML was higher (Chu, 2014), as they require interacting not only with digital learning materials. In general, the influence of mobile application and its interaction complexity on learner's cognitive load has not been consistently reached.

Meanwhile, perception of one's own ability will affect the degree of ME by affecting the firmness of performance goal (Salomon, 1984). Low SE is likely to cause anxiety and increases the processing of unrelated information. Clark (1999) proposed an inverted u‐shaped relationship of SE and ME: individuals with low SE will try to escape the task as they are expected to fail; that is, when faced with a task beyond the capacity of working memory, students will automatically adjust their performance goals, thereby reducing the commitment of ME; overconfident individuals may encounter over confidence default as they tend to use inherent strategy, whereas individuals with medium SE believe that if they do not work hard they will not get good results, so they will invest the most ME into learning task. But, Yoshida (2002) held that there is a positive linear relationship between SE and ME. Therefore, to find how interaction complexity in mobile application influences learning performance and cognitive load, it is necessary to explore the relationship among cognitive load and SE simultaneously.

2.3 Purposes of this study

In order to explore the relationship among learning performance, interaction complexity, and cognitive load, this study proposed two different mobile applications for plant learning. One named Plant‐E is based on expert system and needs students to interact with relevant questions, whereas another named Plant‐G just provides information pages of target plants. The interaction between students and mobile application of Plant‐E is richer than Plant‐G. The main purpose of this study is to explore the influence of mobile learning with interaction complexity on learning effectiveness and cognitive load. The following points of interests were researched:

1. Explore the relationship of interaction complexity in mobile learning with learning performance and cognitive load.
2. Explore the relationship of cognitive load and SE in mobile learning.
3. Analyse students' perception of proposed systems in plant classification scenarios, and the impact on learning effectiveness of proposed learning approaches.

3.1 Rule‐based mobile learning expert system

Previous studies have reported that mobile systems can support students to learn plants knowledge in context‐aware environment (e.g., Zacharia et al., 2016). Huang, Lin, and Cheng (2010) developed a mobile plant learning system, which can provide relevant questions for students to interact with and information by observing features of leaves. Chu, Hwang, Tsai, and Tseng (2010) proposed a two‐tier test approach for students to answer questions about the characteristics of the plant, which can help students gain a better understanding by answering reasons for the choices of former questions. Chen, Hwang, and Tsai (2014) designed a mobile application for campus plant learning, by which learners can scan a QR code to get prompts increased gradually to observe plant characteristics and answer related questions. Most of learning processes of mobile learning in previous studies include multiple‐choice questions with feedback or prompts. These studies have proved that appropriate prompts and interaction in mobile application can produce better learning performance and better attitudes towards the proposed approaches than traditional learning approaches.

In this study, a rule‐based expert system is proposed to support plants learning in context‐aware environment for secondary school students, expert system means a system that has a domain expert knowledge experience, and rule‐based expert system, also called the generative rules system, is to use a series of rules to express expert knowledge (Clancey, 1983). Compared with novices, experts pay more attention to the characteristics of situations or problems; that is, there is a great difference between experts and novices in their attention, which suggests that one may not simply learn from other people's experience but also need to learn to experience it (Sawyer, 2006, pp.30–31). This study sought to combine the plant‐related knowledge in secondary school curricula with the characteristics of mobile learning and design a plant factual information database. Facts stored in the database include objects, attributes, and values; objects are selected plants, and value of a certain attribute that used to describe characteristics of plants is assigned as “yes” or “no.” The rule in the database is a series of if‐then statements, which means attributes are associated with each other.

Students can use the rule‐based expert system in mobile learning to observe characteristics of target plants and make decisions by answering “yes or no” questions, which will improve learners' participation in mobile learning and frequency of observing plant characteristics. The rule‐based expert system provides questions for students to gain information and practice relevant knowledge, which is similar to the systems developed in previous studies (Hwang & Chang, 2011; Sung et al., 2016). It is important to note that the innovation of the plant learning system this study lies in the database is not constituted with questions in tests but a decision tree made up with “yes or no” attributes. The “yes or no” questions related with each other can help students have high level cognitive processing to understand how the classification of plants actually works and can provide them an opportunity to know how botanist work with plant classification. Rule‐based expert system for plant learning will not provide feedback for wrong answers but gives student opportunities for reselection, which means when students are answering current questions, they can return to previous questions and make decisions again or review their previous choices.

3.2 Learning contents and materials

To be in line with the curriculum standard of secondary school in China, the learning content needs to meet the requirements of teaching material and syllabus. Plant knowledge in Chinese textbooks (PEP edition) is in seventh grade, named green plants in the biosphere. As seed plants are the most common plants on campus, this learning contents focus on the classification of seed plants. Plant database in this study was composed of only 10 plants for not making students cognitively overloaded: cedar, Ginkgo biloba, palm, Phyllostachys propinqua, Pittosporum tobira, Crape myrtle, Camphor tree, Chinese holly, Koelreuteria paniculata, and Dandelion. The learning objectives of the mobile learning activity designed by researchers and teachers in this study are to identify common gymnosperms and angiosperms in seed plants, classify plants according to certain characteristics, and summarize main characteristics of seed plants and their relationship with human life.

The reasoning mechanism of rule‐based expert system enables learners to infer plant information through characteristics of plant and can improve availability and accuracy of knowledge. The factual information database in this study is in the form of a Binary Tree (see Figure 1). To begin with, students find a target plant on campus and stand in front of it. Next, students visit the root node and answer the first question by interacting with the mobile application; if students choose “yes”, then visit left subtree, otherwise, visit right subtree; each node has one “yes” or “no” question for students to answer. Until there is no child node, the mobile system will present an information page of the related plant.

Take Phyllostachys propinqua as an example to illustrate the learning process of rule‐based expert system (see Figure 2). Before starting to observe the plant, students enter the system from start interface and get learning objectives. Then, students observe plants via questions and prompts of plant characteristics presented by the system, which will prompt relevant information when characteristic is difficult to observe, and students make “yes” or “no” decisions. When the choice is “yes”, the system will continue to present a question about other characteristic; otherwise, the system will present another question of the same characteristic. For instance, when students are observing Phyllostachys propinqua, they need to answer the question “Are the leaves palmate?” numbered 4, and they should choose “no,” so the system will present another question “Are the leaves sword‐shaped?” numbered 4′.

In this study, the mobile application named Plant‐E is based on expert system, which needs students to interact with relevant questions (see Figure 3); another named Plant‐G only provides plants information pages, which enables students to switch between information pages of different plants at any time.

4.1 Hypothesis

The theoretical foundation of this hypothesis is based on the cognitive load theory and related conceptual models about relationship of ME, learning performance, and SE mentioned earlier, such as the schematic representation of construct cognitive load (Paas & Van Merrienboer, 1994) and u‐shaped relationship of SE and ME (Clark, 1999) mentioned earlier. Accordingly, the research framework of this study is shown in Figure 4.

In this research framework, we assumed that interaction complexity would have an impact on ML and ME, as well as results of students in posttest and delayed test; ME and SE might interact with each other and would have influence on the results in posttest and delayed test. In the light of purposes and research framework in this study, hypotheses are as follows:

4.2 Participants

There are 137 participants in total, who are included in three classes of seventh graders of a secondary school located in Hunan Province, south China. All participating students of three classes taught by the same biology teacher are aged 12–14 years. Most of them (about 74%) have more than 5 years' experience in using mobile phonestechnology and computers; others have at least 3 years' experience. Before the experiment was carried out, the school had a midterm examination, and there was no significant difference among the three classes in results of biology test (Sig. = 0.914 > 0.05), which means students of the three classes have the same level of competence in the biology discipline. One class (Class A) was randomly assigned to be the experimental Class, and other two (Class B and C) were the control group. Forty‐six students (25 boys, 21 girls) in Class A were asked to use Plant‐E in campus with target plants; 47 students (24 boys, 23 girls) in Class B were assigned to use Plant‐G in the same campus; 44 students (22 boys, 22 girls) in Class C used Plant‐G in traditional classroom. According to cognitive load theory (Paas & Van Merrienboer, 1994), the number and degree of information interaction of Plant‐E are more than Plant‐G, so the interaction complexity of Class A is higher than that of Class B. Also, we supposed that the number and degree of information interaction in traditional classrooms are less than those in campus in where students need to observe plants; interaction complexity of Class B is higher than that of Class C. Thus, the three learning conditions constitute three different levels interaction complexity (Class A > Class B > Class C).

4.3 Procedure

The learning procedure is shown in Figure 5. At the beginning of the quasi‐experiment, all students took the pretest. Before the learning activity, the teacher conducted prior knowledge teaching about plants. During the learning activity, students in Class A learned by answering related questions provided by Plant‐E in campus where the target plants in learning content were planted; those in Class B learned with Plant‐G in the same context‐aware environment, whereas Class C only used Plant‐G in a classroom to learn the same learning content. The mobile learning situation of three classes is shown in Figure 6. After completing the learning tasks, students were asked to take posttest and questionnaire survey. Finally, 1 week later, students took the delayed test.

This study asked students to carry their own mobile devices for learning activities. We have tested that all the mobile phones of students are able to run the software smoothly before the experiment. Besides, it proved that screen size of mobile devices has no significant impact on the learning effect (Molnar, 2016), so this study does not consider the impact of screen size. During the learning activity and the data analysis process, all of the students' personal information, including their names, were hidden and replaced by their school number as identification.

4.4 Instruments

The instruments applied in this study included a pretest, a posttest, a delayed test, and a questionnaire to measure students' cognitive load, SE, and perception of proposed learning method and systems.

5.1 Learning performance

5.1.2 Delayed test analyses

The delayed test aimed to evaluate students' retention of plant classification knowledge. The mean and standard deviations values of the delayed test scores were 69.630 and 16.420 for Class A, 59.255 and 19.940 for Class B, and 50.205 and 13.866 for Class C. One‐way ANOVA test shows that there was significant difference between three groups on delayed test (F[2,134] = 14.738, p = 0.000 < 0.05).

It has been tested that the delayed test score in three groups was not homoscedasticity, so Dunnett's t test was used in post hoc test. It turned out that Class A performed significant better than Class B (mean difference = 10.375, p = 0.022 < 0.05), Class B performed significant better than Class C (mean difference = 9.051, p = 0.040 < 0.05), and Class A performed significant better than Class C (mean difference = 19.426, p = 0.000 < 0.05). In this study, we have found that the higher interaction complexity is, the better the retention of memory students will have; what is more learning in campus benefits more than in a traditional classroom, as Class B (using Plant‐G in campus) performed better than Class C (using Plant‐G in classroom), and learning with Plant‐E is more effective than using Plant‐G as Class A performed better than Class B. Moreover, as shown in Figure 7, the result change of Class B and Class C from post‐test to delayed test was slight, whereas delayed test result of Class A was significant higher than its posttest result.

5.2 Cognitive load

5.2.2 Cognitive load with different SE

It is found that there was positive correlation between ME and SE (correlation coefficient = 0.184, p = 0.032 < 0.05), which has also been pointed out by Salomon (1984). In order to explore the relationship among ML, SE, and learning performance, this study divided students in the experiment into three groups according to their self‐reported SE, high SE group, medium SE group, and low SE group. One‐way ANOVA test shows that there were significant difference between three groups on ML (F[2,134] = 4.033, p = 0.020 < 0.05). Tukey's HSD method was used in post hoc test, and it is found that ME devoted in learning activities of medium SE group was notable higher than that of low SE group (mean difference = 0.950, p = 0.015 < 0.05). In particular, medium SE group invested the most ME (5.550), followed by high SE group (5.157), whereas low SE group invested the least (4.600).

In the meantime, one‐way ANOVA tests show that there were significant difference on posttest (F[2,134] = 5.858, p = 0.004 < 0.05) among different SE groups, so was that on delayed test (F[2,134] = 3.234, p = 0.042 < 0.05). In post hoc test, medium SE group gained the highest scores in both posttest and delayed test (see Table 3) and performed significant better than low SE group.

Table 3. Descriptive analysis of the posttest and delayed test results in different self‐efficacy groups

	Posttest		Delayed test		n
	M	SD	M	SD
L	51.532	14.043	54.383	16.214	47
M	62.523	19.144	63.432	17.734	44
H	61.152	17.027	61.957	20.769	46

It can be found that each SE group had a similar rule in ME and learning performance, that is, the medium SE group input the most ME to fulfil the task; hence, students in medium group gain the best grades, rather than the high group (see Figure 8).

5.3 Perceptions and attitudes

The mean and standard deviations values of students' perceptions in the three groups were shown in Table 4. One‐way ANOVA tests revealed that there was no significant difference on PU and PEOU, and students' average perception of using mobile applications were at upper middle level.

Students' views on pros and cons of proposed mobile applications and learning methods were also collected through the questionnaire. It is discovered that there was no obvious difference among three groups on the pros and cons, which coincided with the perceptions of students. In terms of its advantages, students pointed out the novel and interesting activities, convenience of knowledge acquisition, improvement of interest in learning, and more focused, and more relaxed, and so on. Because Plant‐E were used in Class A, some students reported the advantages of “providing information through topics” and “having clearer learning objectives and knowledge points” according to the learning interaction of Plant‐E. In terms of shortcomings, students generally pointed out the disadvantages of “mobile devices can damage vision” and “students who are not self‐sufficient can easily become distracted.” In addition, due to the complexity difference in the interactive learning processes, there are differences in time and information perception, such as students in Class A thought learning time was tense, whereas students in Class B considered that the information is not enough, and students in Class C held that learning content was too simple.