An effective analysis of online education model based on artificial intelligence

Abstract

With the development of artificial intelligence in education, online education has been recognized by the society as a new teaching method. It can make full use of the advantages of the network across regions, and make full use of the advantages of network technology to share the resources of colleges and universities, which is a promising educational method. In response to the demand of online education for learner information, this paper proposes the learner model Neighbor Mean Variation Multi-Objective Particle Swarm Optimization-Genetic Algorithm (NMVMOPSO-GA). This model includes the learner’s learning interest sub-model, the learner’s cognitive ability sub-model and the learner’s knowledge sub-model. The modelling techniques of the three sub-models are discussed separately, and their status and role in the online education system are analyzed. At the same time, for the knowledge model that reflects the learner’s learning progress and knowledge mastery, a learner knowledge sub-model constructed with Bayesian networks is proposed. The neighbor mean mutation operator is introduced to optimize the multi-objective particle swarm optimization algorithm and improve the convergence performance and stability of the multi-objective particle swarm optimization algorithm. We study the application of multi-objective particle swarm optimization algorithm in online course resource generation service. Through simulation experiments, it is verified that the multi-objective particle swarm optimization algorithm can improve the performance and stability of online course resource generation.

Keywords

Online education artificial intelligence intelligent learning particle swarm optimization algorithm

1 Introduction

Education informatization has entered a new stage of development and is changing from digital education to smart education supported by modern information technologies such as big data analysis and artificial intelligence [1 –3]. Due to the characteristics of personalized service, intelligent analysis, natural interaction, and ubiquitous access, smart education has become the trend of education informatization [4]. In recent years, online education platforms have gradually become mainstream, and their teaching concepts and practical effects have been recognized by many universities at home and abroad [5]. Some education platforms have opened their own excellent courses to enable global students to learn advanced ideas and concepts [6]. At present, this learning model has become mature. Compared with the traditional education model, the online education platform has the advantages of rich courses, free time, and easy to learn. Its concept satisfies the current education model of the information age, and its future development situation is in a favorable position [7].

Modern online education mostly adopts the “host college+learning center” model [8]. The learning center is a network that truly penetrates into the tentacles of students, and it is also the main undertaker of support services and supervision of student learning processes. Its functions cover many basic elements in the network education quality assurance system, such as resource delivery and daily teaching management, student learning support services, information transmission and feedback [9, 10]. Its important role in the management of online education is self-evident. Therefore, the development and application of the modern online education teaching management system will definitely bring qualitative changes to the standardized management of online education [11]. In recent years, foreign countries have achieved many successful experiences in sharing online education resources [12]. The British government has established the National Learning Grid, which is an education portal website in the United Kingdom [13]. It adopts modern communication technology and distributed network systems to enable all domestic schools and educational institutions to connect to the Internet to meet the learning needs of schools and families. With the strong support of the British government, the learning grid has become the largest education portal in Europe after several years of continuous improvement [14]. Users can obtain the content they need through the powerful integrated search function. At the same time, it also has the function of publishing courses, and allows teachers to establish new courses directly in the National Curriculum Center [15]. They can also formulate corresponding stages of learning plans and publish relevant resources [16]. The U.S. Educational Information Center is the most prestigious international database in the field of education and social sciences [17]. It was established under the auspices of the Educational Science Research Institute of the United States Department of Education. It aims to promote the development of American education. Family and other educational institutions provide comprehensive educational materials. Scholars believe that the online learning resource serialization service based on computational intelligence directly pays attention to the learner’s individual attribute features and online learning resource features, and provides learners with a learning resource set that matches the individual’s learning ability through the analysis of the differences between learner and learning resource features [18]. The learner attribute features include ability level, prior knowledge, expected goal, learning style, learning time, etc., and describe these feature attributes in a parametric way to complete the modeling of learner features. Through the parameterized representation of these features, the modeling process of online learning resources can be completed more accurately [19]. Scholars pointed out that the core of online learning resource serialization services based on computational intelligence is resource recommendation and path planning problem modeling [20]. The problem model is a formulaic expression that matches the characteristics of learners and online learning resources. At present, the online learning resource serialization service based on computational intelligence is mainly to construct the online learning resource service problem as a constraint satisfaction problem, and the specific matching process is constructed into multiple sub-objective optimization problems, which are solved using intelligent optimization algorithms. There are many related researches on improving the performance of online learning resource serialization services by improving intelligent optimization algorithms [21]. Research topics include online learning resource recommendation, online course content reorganization, and online learning path generation [22]. Different artificial intelligence algorithms were utilised for analyzing various models in the context of data mining techniques [23 –27].

This article discusses the learning process that embodies personalized concepts in the online learning system, and proposes the architecture of the learning system in this article for personalized activities. We construct a learner model containing information such as learner’s interest in learning, cognitive ability and knowledge control Neighbor Mean Variation Multi-Objective Particle Swarm Optimization-Genetic Algorithm (NMVMOPSO-GA). The modeling techniques of the learner’s cognitive ability sub-model and the learner’s knowledge sub-model were analyzed and studied. The neighbor mean mutation operator is introduced to improve the convergence performance and stability of the multi-objective particle swarm optimization algorithm. Combined with the speedless multi-objective particle swarm optimization algorithm, the NMVMOPSO algorithm is proposed. The NMVMOPSO algorithm is applied to solve the multi-objective optimization problem of online course resource generation, and an online course resource generation method NMVMOPSO-GA based on multi-objective particle swarm optimization algorithm is proposed. Experiments show that NMVMOPSO-GA helps to improve the performance and stability of online course resource generation.

The rest of this article is organized as follows. Section 2 discusses the online online learning process model and the functional structure of the online education system. Section 3 constructs the learner intelligence model in online education. Section 4 carried out model simulation experiments and effect analysis. Section 5 summarizes the full text.

2 Network online learning process model and functional structure of online education system

2.1 General online learning process model

The biggest disadvantage of the current network learning support system is that the three links of instructional design, learning support and system evaluation are isolated from each other. In most e-learning support systems, instructional design activities are completely offline activities. The specific teaching strategies and teaching activities are pre-designed and will not change due to changes in individual students and learning processes. Learning support activities are online activities, but are only responsible for presenting pre-stored web pages to learners. The content of the web pages does not change with individual learner differences. Some online learning systems with system evaluation function only complete the test activities, and the test results have nothing to do with instructional design and learning support activities. The general learning process model of the learning system is shown in Fig. 1.

Fig. 1

General learning process model of learning system.

In thepre-teaching stage, although the system provides a lot of course resources, it is impossible to choose the difficulty and goal of learning; in the middle stage of teaching, most e-learning support systems do not have diagnostic support. Although the learning materials provided by the system have been carefully designed by the subject teachers, their purpose is to meet the requirements of the public, so they are not targeted and cannot meet the individual requirements of learners. Since the current learning system basically views learners’ learning activities in an isolated and static view, little attention is paid to the navigation support that learners need in the network learning environment.

2.2 The model of personalized online learning process

In order to solve the shortcomings of the current online learning system in terms of personalization, this paper proposes a personalized interactive learning process between the online learning system and the learner from the perspective of development and connection by analyzing and comparing various existing personalized learning practice models,as shown in Fig. 2. Although this process is also composed of three stages before, during and after learning, each stage is more abundant than the above-mentioned general online learning process.

Fig. 2

Online education personalized learning process and system support model.

The mid-learning stage is the stage where the learner interacts with the network learning system with the strongest intensity. At this stage, the system needs to diagnose the learner’s knowledge status and personality characteristics according to the learner’s choice, and then determine the learning content, learning strategy and learning method according to the learner’s characteristics, and recommend appropriate courseware to students. Knowledge diagnosis in this system is to analyze the current ability of the learner from the learner model library. The system first finds each relevant knowledge point in turn from the domain knowledge, and then finds the learner’s understanding of the knowledge point from the learner’s current knowledge state. The level of understanding depends on the learner’s test feedback. If the current score of the knowledge point is good, it means that the learner has mastered the knowledge point and does not need to supplement the knowledge. If the current score is not good, it means that the learner does not understand this knowledge point thoroughly and needs to be supplemented. The learner completes the learning according to the form of learning activities provided by the system for him. The learning schedule organizes personalized learning materials based on the diagnosis results. Learning scheduling can be divided into two stages. The first stage is to determine the knowledge points of learners’ learning content. The learner finds the learning objects of all knowledge points that need to be learned from the knowledge model according to the diagnosis result; the second stage selects the appropriate media form, learning mode and organization form according to the learning preferences in the learner’s personality model.

2.3 Functional model of online learning system

The personalized knowledge representation dynamically generates learning materials suitable for the learner’s ability according to the learner’s personality characteristic model. It consists of three sub-modules: content generation, courseware recommendation and content presentation.

Learning is a process in which learners gradually approach their goals. Course goals usually consist of several sub-goals. The course generation module organizes the order in which the sub-goals are presented one after another based on the learner’s learning goals, knowledge level, and learning preferences—the learning path and the media type of the corresponding learning object. According to the feedback of the learner’s ability and the difficulty coefficient of the courseware, the courseware content suitable for the difficulty level is recommended to the learner. It recommends the learning materials organized by the content generation module to the learners according to the learners’ learning ability, for the learners to learn. The architecture diagram of the online learning analysis system is shown in Fig. 3.

Fig. 3

Architecture diagram of online learning analysis system.

In order to achieve online personalized learning, a complete learning system must first fully understand the personality characteristics of different learners. This requires building a model of learner personality as comprehensive as possible to analyze and discriminate the personality of different learners, so as to present different learning content to learners with different personalities. This requires that the learning system can present knowledge content according to specific needs, and the creation of knowledge models in related fields just provides strong support for accurately and flexibly generating and presenting learning content.

2.4 Online learning support subsystem functions

The student e-learning support subsystem builds a student space. A user who is a student can enter the student space. This subsystem provides students with the following functional modules to form a systematic network learning support environment.

The online learning function module is the core of online teaching, providing students with a personalized course list. Each student chooses a different course. The system displays a list of students according to the courses selected in the most recent semester, which is convenient for students to browse. At the same time, students can use keywords to search for related courses. The learning time and frequency of each knowledge point of the student will be recorded, which is convenient for the evaluation of the student and the related statistics of the knowledge point.

Personal space provides students with a private space to record their learning experience and other information. Personal space includes students’ own diaries and personal photo albums. Students exercise independent control over the openness of their personal space. If it is defined as a public state, anyone can browse and leave a message. If it is defined as a private state, only friends can browse and leave a message.

Students can send in-site messages to teachers, administrators or other students, and the system will automatically pop up a prompt for new messages when the recipient logs in. Messages in the station are one of the means of interaction. The structure diagram of online learning behavior evaluation framework is shown in Fig. 4.

Fig. 4

Structure diagram of online learning behavior evaluation framework.

3 Intelligent model of learners in online education

3.1 Online cognitive ability sub-model

The information needed to build the cognitive ability model is completed by the students entering and recording a series of question-making exercises when logging into the system. Since the cognitive model contains six types of learners’ goals to be learned, understood, applied, analyzed, synthesized and evaluated, that is, each knowledge point is marked with one or more cognitive skills. Therefore, you can carry out cognitive identification on the topics, that is, each topic tests the learner’s mastery of one or more cognitive abilities.

If the learner answers correctly to a certain cognitive ability tested by the test question, the score is 1, otherwise the score is -1. If the learner does not take the test or the test question does not involve a certain cognitive ability, the score is 0. According to the cognitive ability vector table, the learner’s cognitive ability of various knowledge items can be calculated.

From the vector table, the test question set corresponding to the knowledge item k can be found, so that the correct usage rate r(ai) of each cognitive ability ai under the knowledge item k can be calculated: $r (a_{i}) = {Na}_{i} (1) / [{Na}_{i} (1) + {Na}_{i} (- 1)]$ (1)

Among them, Nai (1) and Nai (–1) are the number of times that cognitive ability ai is correctly used and incorrectly used in the learner’s answer to the test question corresponding to knowledge item k.

Thus, the learner’s cognitive ability mastery vector under knowledge item k can be obtained: $R = [r (a_{0}), r (a_{1}), r (a_{2}), r (a_{3}), r (a_{4}), r (a_{5})]$ (2)

Thus, the learner’s comprehensive ability M for a certain knowledge point can be calculated, as shown in the following formula. $M = c_{i} • \sum_{i = 0}^{5} r (a_{i})$ (3)

ci is the weight of a certain cognitive ability under this knowledge point.

3.2 Online learning interest sub-model

In an education system that adopts standard standards for educational resources, the metadata specification information of each resource contains a description of all characteristic values of the resource, and each resource can be uniquely represented by its metadata information. Therefore, when collecting data on the learner’s interest information, it is not necessary to segment the entire learning resource and use the feature vector to represent the resource, but can directly use the value of the key information item in the metadata or its segmentation to represent the resource.

Similarly, the value of the feature vector in the learner’s learning interest model can also be limited by the range of the value space of the corresponding information item in the educational resource specification. When the learner’s interest is modeled by collecting the learner’s learned and tested educational resources, the educational resources can also be expressed using a vector space model. Each resource is represented as a feature vector: $R = [(c_{1}, t_{1}, w_{1}), (c_{2}, t_{2}, w_{2}), . . ., (c_{n}, t_{n}, w_{n})]$ (4)

ti is a characteristic item, which is the value of a key information item in the metadata information of the resource. wn is the weight of the feature item in R, and ci is the category of the information item to which the feature item belongs.

The user interest model in the personalized service is a model of the user’s interest obtained over a period of time through the user’s behavior and pages visited in the past. The scope of the user’s interest is often unlimited, and the pages visited by the user are also diverse. It is necessary to discover the feature values of each page through data mining and analysis for interest modeling. In the field of education, the interest of the learners concerned is mainly concentrated in the interest of learning in the field of education, specifically the interest of learners in the knowledge points of various disciplines. Types of educational materials are of interest, and the purpose of constructing an interest model is to combine students’ knowledge mastery to arrange students’ learning plans and training directions in a targeted manner.

Long-term interest is a relatively fixed interest preference of learners, which is relatively stable and is the main basis for judging the learner’s interest. Since the learner’s interest in the learning object will change and forget, it will gradually decrease with the passage of time, and at the same time, it will be stimulated by the new interest and increase again. Therefore, when calculating the long-term interest, the combined impact of time and current short-term interest needs to be considered. Only when the interest model obtained this time includes the short-term interest of this feature item, the long-term interest is updated. The calculation is as follows: $w_{i}^{1} = w_{i}^{s} + e^{- \frac{ln 2}{h * I} (d - d_{i})} • w_{i}^{1 - pre}$ (5)

Among them, $w_{i}^{1 - pre}$ is the user’s long-term interest weight before the update; $e^{- \frac{ln 2}{h * I} (d - d_{i})}$ is a forgetting factor, indicating that the interest decays according to the forgetting curve over time; di is the update time of the last long-term interest recorded in the model; h*I is Half-life of interest; $w_{i}^{s}$ is the short-term interest weight as an incentive for long-term interest.

3.3 Online knowledge sub-model

The learner knowledge model in this paper adopts an overlay model, that is, the knowledge set contained in the learner’s knowledge model is considered to be a subset of the domain knowledge set in the system at any time. Therefore, to model the learner’s knowledge, a domain knowledge model needs to be constructed by a domain expert first, and then a subset of knowledge is taken from the domain knowledge model according to the learner’s learning behavior to construct the learner’s knowledge model. The “cover model” mentioned here only means that the nodes and structure (node relationship) in the learner knowledge model come from the domain knowledge model, and the node parameters of the learner knowledge model need to be calculated by the learner’s learning behavior. It is concluded that the node parameters of the learner’s knowledge model also come from the domain knowledge model, and the node parameters of the domain knowledge model are only for reference.

The domain knowledge model constructs a Bayesian network model that can fully reflect the relationship between all knowledge items. The large scale of the domain knowledge model, the efficiency of Bayesian inference on it is very low, and it cannot meet the requirements of personalized learning timeliness. Therefore, it cannot be used directly as a learner’s knowledge model; but because of the comprehensiveness and versatility of the domain knowledge model, it can be used as a reference for constructing the learner’s knowledge model.

The learner knowledge model is structurally a subset of the domain knowledge model (the overlay model is used here). Therefore, the domain knowledge model can be derived from the domain knowledge model according to the current set of knowledge items learned by the learner and the learner’s staged learning goals. It extracts the required knowledge items to form a new Bayesian network as the learner’s knowledge model at the current stage. As the learner continues to learn more, he can continuously expand the structure of his knowledge model.

3.4 Online course resource generation method

The NMVMOPSO-GA method makes full use of the advantages of the multi-objective particle algorithm based on neighbor mean mutation. This method is based on the guides’ completion of the construction of the knowledge map, using the characteristics of the learner and the characteristics of the learning resources corresponding to the knowledge points contained on the knowledge map. First of all, we build a learner model, including the learning goals that the learner expects to achieve, the ability level of the learner, and the upper and lower limits of the learning time expected by the learner; second, we build a learning resource model, which consists of learning resources and learning resources contained in knowledge points. Finally, the neighbor mean mutation multi-objective particle algorithm is used to realize the online course resource generation service. The flow chart of NMVMOPSO-GA generation method is shown in Fig. 5.

Fig. 5

NMVMOPSO-GA method flow chart.

4 Model simulation experiment and effect analysis

4.1 Simulation experiment design

In order to verify the performance of the NMVMOPSO algorithm applied to multi-objective online course resource generation problem solving, five sets of experiments were set up respectively according to the learner and learning resource, and the performance of the NMVMOPSO-GA course resource generation method was observed as the learning resource changed.

In the Multi-Objective Particle Swarm Optimization (MOPSO) algorithm, the inertia weight is set to 0.4, and the inertia weight change rate is 0.96; in the Multi-Objective Ant Colony Optimization (MOACO) algorithm, the grid sorting method is used for external file maintenance, and the number of grid divisions is 8; the NMVMOPSO algorithm updates external files according to the dominant relationship.

1) Experimental environment

The algorithm simulation experiment environment is windows7 operating system, and the programming language environment is Matlab R2018a. The hardware environment is Intel Core processor i5-4570, the main frequency is 3.20 GHz, and the memory is 8GB.

2) Online course resource generation service experiment

Five sets of experiments were set up to verify the quality and stability of the online course resource generation method NMVMOPSO-GA implemented by the NMVMOPSO algorithm. Each set of experiments has five knowledge points on the knowledge map, and each knowledge point contains a different amount of learning resources.

Experiment 1, experiment 2, and experiment 3 ensure that the number of learners remains unchanged, and as the number of learning resources changes, we observe the course resource generation method NMVMOPSO-GA implemented by the NMVMOPSO algorithm; to ensure that the learning resources remain unchanged, as the number of learners changes, we observe the performance of the NMVMOPSO-GA generation method implemented by the NMVMOPSO algorithm. In the five sets of experiments, the learning resource is the number of learning resources corresponding to a knowledge point on the knowledge map. Therefore, in Experiments 1 to 5, the total number of learning resources is 600, 1200, 1500, 500, 600; the learner’s learning ability and the difficulty level of learning resources are randomly initialized to five levels; the random initialization corresponds to the knowledge points.

4.2 Service quality analysis

The data in Fig. 6 are obtained by running each method independently 50 times on the experimental platform. From the average data, it can be seen that the average value of the curriculum resource generation method implemented by the NMVMOPSO algorithm is the smallest among the curriculum resource generation methods implemented with the four core algorithms, indicating that the resulting online curriculum resources are more in line with the learner feature. It can be seen from the variance data that the method of generating the NMVMOPSO algorithm is that the obtained variance data is also the smallest of all the comparison methods, indicating that the online learning resources recommended by the NMVMOPSO algorithm for curriculum resource generation methods have better stability.

Fig. 6

Means and variance data of corresponding generation methods of each algorithm.

In order to more intuitively observe the quality of the online course resource generation service in five groups of experiments of different scales using the resource generation methods implemented by different algorithms, Fig. 7 shows the comparison of the single objective function values of the optimal individual solutions obtained by each method. In the figure, the optimal value curve of the resource generation method implemented by the NMVMOPSO algorithm is closest to the x-axis, which is lower than the other three groups of comparison methods. It can be clearly observed that the service quality of the resource generation method implemented by the NMVMOPSO algorithm is superior to the comparison method.

Fig. 7

Comparison of single objective function values of optimal individual solutions.

Figure 8 shows the service quality of the online course resource generation method implemented by the NMVMOPSO algorithm. From the convergence graph of online course resource generation, we can see that the course resource generation method implemented by the MOACO algorithm has good service quality on three objectives, and the overall service quality of the course resource generation method implemented by the Multi-Objective Artificial Fish Swarms Optimization (MOAFSO) algorithm obviously is at a disadvantage. The course resource generation method implemented by NMVMOPSO algorithm has higher service quality as a whole among other objective functions. Since the multi-objective optimization problem is looking for a set of compromise solutions, the course resource generation method implemented by the NMVMOPSO algorithm can obtain the smallest set of compromise solutions, which proves the service quality and stability of the resource generation method implemented by the NMVMOPSO algorithm. It can better meet the needs of solving online course resource generation problems from a multi-objective perspective. The experimental results also show that the resource generation method implemented by the NMVMOPSO algorithm exhibits better service quality and stability than the other three comparison algorithms. As the number of learning resources increases, the resource generation method implemented by the NMVMOPSO algorithm can still maintain high service quality and stability.

Fig. 8

Convergence graph generated by online course resources.

In order to compare whether the single objective function value of the optimal individual solution obtained by the multi-objective intelligent optimization algorithm has a statistically significant difference, we performed a paired t-test on the results of several generation methods, and the resulting probability value P is shown in Fig. 9. The online course resource service provided is more in line with the learner’s characteristic model, and the matching degree is higher. From the result of P value, t-test also confirms that the use of NMVMOPSO algorithm can better improve the matching of online learning resources and learners, and improve the service quality of online learning resources.

Fig. 9

The online course resource generation method implemented by each algorithm P value.

4.3 Analysis of online course resource generation time

When using different methods to generate online course resources, it is necessary to consider not only the quality of the resources generated by each method, but also the time taken by each method to generate course resources.

Figure 10 shows the comparison of the time taken by the resource generation methods under different algorithms in five sets of experiments. In the figure, you can visually observe the amount of time spent by each algorithm in generating online course resources. In Experiment 1 to Experiment 5, the resource generation method implemented by the NMVMOPSO algorithm proposed in this paper takes the least time. This shows that the resource generation method implemented by the NMVMOPSO algorithm for online course resource recommendation can improve the quality of course resource generation without costing a lot of time. It is a better online course resource generation algorithm.

Fig. 10

The average optimization time of five generations of resource generation methods under different algorithms.

Figure 11 shows the time course of the online course resource generation method NMVMOPSO-GA. As the number of learning resources increases, it will take more time to generate suitable learning resources for learners to provide better service quality.

Fig. 11

NMVMOPSO-GA experiment time curve.

4.4 Analysis of the success rate of each resource generation method

Figure 12 is a comparison of the number of times that each resource generation method under different algorithms achieves the specified accuracy in five sets of experiments. As can be seen from the figure, the number of times that the resource generation method using the NMVMOPSO algorithm achieves the specified accuracy in five sets of experiments.

Fig. 12

Comparison of the number of times the online course resource generation method reaches the specified accuracy under different algorithms.

Figure 13 shows the comparison of the success rates of the four methods in Experiment 1 to Experiment 5. It can be seen from the figure that the success rate of resource generation using the resource generation method implemented by the NMVMOPSO algorithm is much greater than the other three methods.

Fig. 13

Comparison of the success rate of resource generation methods under different algorithms.

5 Conclusion

This paper studies the development of online education and the environment and learning variables of distance education. Based on this, the learning process of online education is modeled and the functional structure is designed. The overall structure of the learner model NMVMOPSO-GA in the online education system is introduced. The neighbor mean mutation operator was introduced to improve the convergence performance and stability of the multi-objective particle swarm optimization algorithm, and the multi-objective particle swarm optimization algorithm NMVMOPSO based on the neighbor mean mutation was designed. The improved multi-objective particle swarm optimization algorithm NMVMOPSO is applied to the field of online course resource generation service problems, and an online course resource generation method NMVMOPSO-GA based on neighboring mean mutation multi-objective particle swarm optimization algorithm is proposed. Simulation experiments show that the online course resource generation method NMVMOPSO-GA designed with NMVMOPSO as the core can improve the service quality of online course resource generation. This article only studies the characteristics of learning resources from their original attributes, but the degree of intelligence of automatic semantic annotation of learning resources will also have an important impact on learning resource recommendation services. Therefore, using information technology to carry out data-driven personalized learning resource recommendation service is also a problem worthy of in-depth study.

Footnotes

Acknowledgments

This work was supported by the Topic: Research on the mode of teacher education based on the core literacy of rural teachers in Hebei (zd201803).

References

Abdulsahib

G.M.

and Khalaf

O.I.

, Comparison and Evaluation of Cloud Processing Models in Cloud-Based Networks [J], International Journal of Simulation–Systems, Science & Technology 19(5) (2018).

Arpaci

, What drives students’ online self-disclosure behaviour on social media? A hybrid SEM and artificial intelligence approach[J], International Journal of Mobile Communications 18(2) (2020), 229–241.

Cukurova

, Kent

and Luckin

, Artificial intelligence and multimodal data in the service of human decision-making: A case study in debate tutoring[J], British Journal of Educational Technology 50(6) (2019), 3032–3046.

Dos Santos

D.P.

, Giese

, Brodehl

, et al., Medical students’ attitude towards artificial intelligence: a multicentre survey[J], European radiology 29(4) (2019), 1640–1646.

Goodman

, Melkers

and Pallais

, Can online delivery increase access to education?[J], Journal of Labor Economics 37(1) (2019), 1–34.

Hrastinski

, Olofsson

A.D.

, Arkenback

, et al., Critical imaginaries and reflections on artificial intelligence and robots in postdigital K-12 education[J], Postdigital Science and Education 1(2) (2019), 427–445.

Ikedinachi

A.P.

, Misra

, Assibong

P.A.

, et al., Artificial intelligence, smart classrooms and online education in the 21st century: Implications for human development[J], Journal of Cases on Information Technology (JCIT) 21(3) (2019), 66–79.

Karsenti

, Artificial intelligence in education: the urgent need to prepare teachers for tomorrow’s schools[J], Formation et profession 27(1) (2019), 112–116.

Kataria

, Mishra

and Lalwani

, Role of Artificial Intelligence in Education[J], Studies in Indian Place Names 40(69) (2020), 193–201.

10.

Kavitha

and Lohani

, A critical study on the use of artificial intelligence, e-Learning technology and tools to enhance the learners experience[J], Cluster Computing 22(3) (2019), 6985–6989.

11.

, Artificial intelligence: a survey on evolution, models, applications and future trends[J], Journal of Management Analytics, 6(1) (2019), 1–29.

12.

Mao

, Ifenthaler

, Fujimoto

, et al., National policies and educational technology: a synopsis of trends and perspectives from five countries[J], TechTrends 63(3) (2019), 284–293.

13.

Masters

, Artificial intelligence in medical education[J], Medical Teacher 41(9) (2019), 976–980.

14.

Matheny

M.E.

, Whicher

and Israni

S.T.

, Artificial Intelligence in health Care: A report from the National Academy of Medicine[J], Jama 323(6) (2020), 509–510.

15.

Ocaña-Fernández

, Valenzuela-Fernández

L.A.

and Garro-Aburto

L.L.

, Artificial Intelligence and Its Implications in Higher Education[J], Journal of Educational Psychology-Propositos y Representaciones 7(2) (2019), 553–568.

16.

Prathik

, Uma

and Anuradha

, Particulate Matter on Human Health and their Feasibility Study Using Machine Learning Algorithms [J], Journal of Chemical and Pharmaceutical Research 8(9) (2016), 260–264.

17.

Wang

, Zhou

, Yang

, Sun

and Hsu

C.-H.

, Service Composition in Cyber-Physical-Social Systems, IEEE Transactions on Emerging Topics in Computing.

18.

Sharma

R.C.

, Kawachi

and Bozkurt

, The landscape of artificial intelligence in open, online and distance education: Promises and concerns[J], Asian Journal of Distance Education 14(2) (2019), 1–2.

19.

Shin

W.S.

and Shin

D.H.

, A study on the application of artificial intelligence in elementary science education[J], Journal of Korean Elementary Science Education 39(1) (2020), 117–132.

20.

Shyr

W.J.

, Yang

F.C.

, Liu

P.W.

, et al., Development of assessment indicators for measuring the student learning effects of artificial intelligence-based robot design[J], Computer Applications in Engineering Education 27(4) (2019), 863–868.

21.

Sithole

, Mupinga

D.M.

, Kibirige

J.S.

, et al., Expectations, challenges and suggestions for faculty teaching online courses in higher education[J], International Journal of Online Pedagogy and Course Design (IJOPCD) 9(1) (2019), 62–77.

22.

Sudin

M.N.

, Abdullah

S.S.

and Nasudin

M.F.

, Humanoid Localization on Robocup Field using Corner Intersection and Geometric Distance Estimation [J], International Journal of Interactive Multimedia and Artificial Intelligence, 2019, 50–6.

23.

Sun

, Tang

and Zuo

, Coronavirus pushes education online[J], Nature Materials 19(6) (2020), 687–687.

24.

Namasudra

, Devi

, Kadry

, Sundarasekar

and Shanthini

, Towards DNA based data security in the cloud computing environment [J], Computer Communications 151 (2020), 539–547.

25.

Williams

, Does competency-based education with blockchain signal a new mission for universities?[J], Journal of higher education policy and management 41(1) (2019), 104–117.

26.

, Wu

and Yang

, An online education data classification model based on Tr_MAdaBoost algorithm[J], Chinese Journal of Electronics 28(1) (2019), 21–28.

27.

Zanetti

, Iseppi

and Cassese

F.P.

, A “psychopathic” Artificial Intelligence: the possible risks of a deviating AI in Education[J], Research on Education and Media 11(1) (2019), 93–99.