Mapping Knowledge Domain Analysis in Deep Learning Research of Global Education

Abstract

With the rapid development of the global digital knowledge economy, educational activities are facing more challenges. Sustainable development education aims to cultivate students’ thinking ability to better integrate with the contemporary world view, so classroom practice should involve innovative teaching and learning. The goal of sustainable development education is to cultivate talents with high-level thinking and sustainable development abilities. The concept of deep learning emphasizes mobilizing students’ internal motivation, focusing on problem-solving ability, improving students’ critical thinking level, and developing students’ lifelong learning ability. The concept of deep learning has evolved with the times. The introduction of the concept of deep learning in teaching can enhance students’ understanding of the nature of knowledge, cultivate students’ high-level thinking, and enable students to achieve better learning results. Integrating the concept of deep learning into teaching has extremely important significance and value for sustainable development education. It has become a hot topic in the world to comprehensively analyze the research status of deep learning and explore how deep learning can help education achieve sustainable development. In this study, CiteSpace (6.1.R2) visualization analysis software was used to visualize and quantitatively analyze the literature on deep learning in the Social Science Citation Index (SSCI). The visualized analysis is conducted on the annual publication amount, authors, institutions, countries, keywords, and high-frequency cited words of deep learning, to obtain the basic information, development status, hot spots, and evolution trends of deep learning research. The results show that the annual publication volume of deep learning is on the rise; deep learning research has entered a rapid growth stage since 2007; the United States has published the most papers and is the center of the global deep learning research collaboration network; the countries involved in the study were often interconnected, but the institutions and authors were relatively dispersed; research in the field of deep learning mainly focuses on concept exploration, influencing factors, implementation strategies and effectiveness of deep learning; learning method, learning strategy, curriculum design, interactive learning environment are the high-frequency keywords of deep learning research. It can be seen that deep learning research has the characteristics of transnationality, multidisciplinary nature and multi-perspective. In addition, this paper systematically analyzes the latest progress in global deep learning research and objectively predicts that using intelligent technology to design appropriate teaching and learning scenarios and evaluation methods may become the future development trend of deep learning. The research results of this paper will help readers to have a comprehensive understanding of deep learning research, provide deeper and more targeted resources for integrating deep learning concepts into teaching, and promote better sustainable development of education.

1. Introduction

In today’s world, with the rapid progress of science and technology, the rapid development of information technology, and the fierce competition for talent training, traditional teaching methods are no longer applicable. Sustainable development education is more respected in the information age. The development of the era needs talents such as creativity, initiative awareness, self-learning ability and lifelong learning. The outbreak of COVID-19 in 2019 has brought new challenges to education. The intervention of information technology has changed the teaching methods of teachers and the learning methods of students. At the same time, the teaching effect faces some new difficulties. The use of intelligent technology easily leads to students only paying attention to the acquisition of knowledge in learning, ignoring the in-depth development of knowledge, and students unable to grasp the essence of knowledge. To solve this practical problem, we need to pay attention to the nature of learning. Learning is integrated with other developmental processes and must take into account the whole child (emotion, identity, cognition) [1]. Deep learning means that learners learn new knowledge actively and critically on the basis of understanding, realize the transfer and application of knowledge, and develop the ability to solve problems and higher-order thinking skills [2]. Different from shallow learning, deep learning not only involves memory, but pays more attention to the understanding and application of knowledge, focuses on the development of students, and emphasizes students’ learning ability and problem-solving ability. In terms of learning objectives, deep learning requires the formation of critical thinking, the improvement of problem-solving ability, and the realization of independent learning. In terms of learning results, it points to meaningful learning and learning transfer. In terms of learning methods, it pursues understanding and construction. In terms of the learning process, it shows initiative and high commitment [3]. Deep learning focuses on students’ “learning”. The essence of deep learning provides a new theoretical perspective and method for breaking through the dilemma of ineffective teaching. Deep learning can solve the problem of shallow and fragmented knowledge learning, promote students to grasp knowledge as a whole, and promote knowledge transfer and construction [4]. Deep learning is of great value to improve students’ ability to discover, analyze and solve problems and is an important way to promote the development of students’ innovative thinking [5].

Talents are the driving force of sustained development education, and how to cultivate talents has always been a key concern in the field of sustained development education, teaching is not only a question of “how to teach”, but also “how to learn”; the focus of education is not teaching, but students’ learning ability [6]. With the development of information technology, the shortcomings of traditional teaching began to appear [7], and after continuous exploration and reflection, people began to realize the importance of meaningful learning activities [8]. “Deep Learning” has opened up new ideas for traditional teaching and has received the attention of many researchers. Decades of empirical studies have shown that deep learning can improve students’ academic performance, improve teaching effectiveness [9,10], and that students who use deep learning methods perform better than those who use surface learning methods in terms of remembering, integrating, and transferring information [11,12,13]. Over the past 40 years or so, it is undeniable that the learning concept of deep learning has been highly recognized and respected in the field of education. However, it is interesting to summarize the deep learning literature by asking the following questions. What is the status of deep learning research? How to effectively promote sustained development education? This study will carry out research on the status quo of deep learning through bibliometrics, providing references for scholars’ follow-up research.

2. Background

The concept of deep learning was proposed by Ference Marton and Roger Saijo, who demonstrated the strong correlation between the learning process and the hierarchy of learning outcomes through a series of experimental studies, and found differences in the characteristics of learning methods through learners’ descriptions of learning outcomes [14,15]; Biggs et al. divided learning styles into shallow learning and deep learning, arguing that learners would choose different learning methods for different learning tasks, resulting in different learning outcomes [16]; after intensive research, Biggs et al. published a series of results, the two most famous of which are the SPQ Learning Process Scale, which measures deep/shallow learning strategies and motivation. In addition, scholars such as Nelson Laird, Kolb, and Weige have defined the connotation of deep learning from different perspectives. Meanwhile, some scholars have revealed the basic characteristics of deep learning based on shallow and deep learning, for example, Eric Jensen and LeAnn Nickelsen described the basic characteristics of deep learning as higher-order thinking, deep processing, deep understanding, active construction, and problem-solving.

With the wide attention of the international education community, research on deep learning has entered a rapid development stage. In 1995, Hoon et al. of Nanyang Technological University in Singapore applied deep learning strategies to a high school chemistry class to facilitate students to visualize abstract concepts and explore the connections among numerous chemical facts. In 2004, the American Association for Educational Communications and Technology (AECT) redefined educational technology and concluded that surface learning cannot meet the requirements of social development and identified deep learning as the basic concept and direction of effort for future educational technology development. Cultivating innovative talents is one of the core values pursued in the 21st century, and international organizations and countries around the world have devoted themselves to exploring what kind of talents should be cultivated in the new century. In 2010, the Hewlett Foundation launched a 15-year strategic plan for deep learning (The William and Flora Hewlett Foundation, 2012), and the National Research Council (NRC) proposed several approaches in its 2012 report. In 2012, the University of Victoria in Canada launched the deep learning Global Initiative, a collaboration with more than 1000 schools in ten countries seeking solutions for deep learning change [17]. In 2015, Noel Entwistle of Lancaster University in the UK conducted a study on deep learning in students from an internal perspective of student learning. This study combined factors from the environmental correlation methodology held by Biggs, Marton, and Salchow with student personality factors [18], and the United States released a document making deep learning a national policy for education in the 21st century [19]. In 2017, the 2017 Horizon Report (Higher Education Edition) published by the New Media Consortium stated that deep learning will be a key factor in decision-making at higher education institutions in the next five years and beyond [20].

Research on deep learning ranges from conceptual features, intrinsic mechanisms, model construction, strategy exploration, assessment development, and then disciplinary applications. With the rapid development of digital intelligence, promoting high-quality and comprehensive development of talents is necessary for global economic development, deep learning opens up new ways of education, and the future application and development of deep learning have become the focus of attention in the field of education. At present, the research direction of deep learning is diversified, and the research results are remarkable, but there is a lack of systematic and objective organizational research. In this context, it is necessary to conduct an effective in-depth analysis of the field. Literature reviews are considered to be an effective way to gain insight into a particular research area [21]. In order to understand the current state of research in this field, it is necessary to conduct a systematic analysis of the field using scientometric software (CiteSpace 6.1.R2) [22].

There are also many publications on deep learning worldwide, which indicates that research results on deep learning are available worldwide. Therefore, this study systematically combs the existing research through CiteSpace software to show the overall situation, research hotspots and evolution rules of deep learning. This will give researchers and policymakers new ideas and perspectives to guide sustainability education in future research and policy. By collecting relevant materials and literature on deep learning research, this study takes the development of deep learning as a whole, and reveals the overall situation, research hotspots and evolution trends of deep learning research from five dimensions: development trends, countries, research institutions, researchers and keywords. This study can partially cover what is not covered by other studies. The main contributions of this study include the following four points.

(1)

Identifying publication trends of deep learning research using big data.

(2)

Finding out which countries, regions, research institutions, and scholars have the most impact on big data-based deep learning research and how they work together.

(3)

Analyzing the research hotspots of deep learning.

(4)

Analyzing the research frontiers and research trends in deep learning research.

The rest of the paper is organized as follows: Section 2 focuses on the data collection and research methodology. Section 3 analyzes the authors, institutions, and national visualization results of deep learning. Section 4 describes the theoretical foundations and new developments in deep learning. Section 5 discusses the research results of deep learning. Finally, Section 6 summarizes the research and describes the next steps.

3. Research Materials and Methods

3.1. Data Collection

First of all, compared with databases such as SCl and Scopus, SSCl publications mainly cover literature on humanities and social sciences, and deep learning belongs to this category. Secondly, studies in the SSCI database are more influential and representative. SSCI covers a wide range of journals, it was founded in 1956, and in 1999 SSCI included 1809 of the world’s most important social science journals in full text, while Scopus is a database launched by Elsevier in November 2004, Scopus covers more journals but has less impact and is limited to recent articles [23]. Therefore, in order to enhance the directivity of the study and ensure the comprehensiveness of the data sample, this study used the SSCI database from the Web of Science (WOS) core collection as the data source, with the subject search term “Deep Learning”, and to improve the accuracy of the data, the subject area was limited to “Education Educational Research”, the language was limited to “English”. The first article on deep learning in the SSCI database was recorded in 1992. In order to minimize the omission of important basic research from earlier years, the data was extracted from the period 1 January 1992 to 1 September 2022. The literature was output in plain text format and a total of 2951 research articles were retrieved. The retrieved data were filtered and excluded to obtain 2827 papers related to deep learning.

3.2. Analytical Methods

3.2.1. Analysis Path

In order to comprehensively analyze the current situation and development trend of deep learning research, this study analyzes the path based on the characteristics of the CiteSpace analysis tool from the following three aspects:

(1)

Analysis of basic information on deep learning. This allows us to have a general overview of deep learning, including the number of publications, authors, countries, institutions, etc.

(2)

Analysis of deep learning research hotspots. Through the analysis of the keyword co-occurrence graph and time zone graph, we can understand the main areas, basic contents and research hotspots of deep learning research.

(3)

Research frontier and trend analysis of deep learning. By analyzing high co-citation terms and burst analysis, the current research frontier of deep learning can be understood based on these high co-citation terms. Burst analysis is used to identify citation bursts, which refers to the intensity of the sudden appearance or disappearance of citations in a certain research topic in a certain research field in a certain period of time, and to a certain extent represents the development direction of a certain research trend [24]. Based on the observed sudden changes in citations over time, trends in research topics over time can be inferred.

3.2.2. CiteSpace and Settings

In this study, a bibliometric approach was used to analyze the data sources using CiteSpace (6.1.R2), an econometric analysis tool developed by Professor Chaomei Chen’s research team, for multivariate, time-slice, and dynamic visualization [25]. The Java-based CiteSpace software can use algorithms and mathematical models to draw a knowledge map of a certain research field and visualize the literature, so as to reveal research hotspots and evolutionary trends [26]. Therefore, CiteSpace software can reveal the overall situation of a certain research field. Through the comprehensive application of quantitative and qualitative analysis, researchers can collect more information, so that the research results are in line with both subjective experience and objective data, and is more scientific and accurate, and helps researchers make accurate judgments. In recent years, CiteSpace has been used by scholars to conduct quantitative and qualitative analysis of literature in different research fields, systematically revealing the research hotspots and evolutionary rules in each research field [27,28,29,30,31,32]. Of course, these findings are useful in helping researchers understand the development of a particular study. To import the data source into CiteSpace, the time was set to “1 January 1992–1 September 2022” and the time slice was set to 1 year. The Title, Abstract, Author Keywords, and Keywords plus attributes under the Terminology Resources menu bar were selected, and the Selection Criteria under the Top N% column was set to 25%. The co-occurring knowledge network was pruned using Pathfinder, Minimum Spanning Tree, and Merged Network pruning methods. It was mainly presented in the form of nodes and lines, N = number of nodes, and E = number of lines. The size of nodes reflects the frequency of relevant data references or occurrences, the lines indicate the relationship between nodes, and the thickness of the lines between nodes reflects the strength of the connection between data. The aggregation effect is measured by modularity and silhouette, the Q value represents the degree of modularity, Q greater than or equal to 0.3 indicates a high degree of modularity of the network, and as the Q value increases, the clustering effect of the network will also improve. The silhouette (S) measures the homogeneity of the network; if S is greater than or equal to 0.5, it means the clustering result is reasonable, and the more S converges to 1, the higher the homogeneity of the network. Based on co-citation analysis theory, pathfinding network algorithm, and minimum spanning tree algorithm, the literature data are analyzed econometrically and mapped to discover the current situation, research hotspots, and evolutionary trends of the field. The analysis includes author collaboration analysis, institutional analysis, country co-occurrence network analysis, and keyword co-occurrence mapping analysis.

4. Results and Analysis

4.1. Overview of Deep Learning Research

4.1.1. Statistical Analysis of the Number of Published Papers

The statistical analysis of the number of published papers in the field of deep learning research (Figure 1), which shows the data of the past 30 years, can clearly reflect the popularity of deep learning research over the years. In general, the number of research papers related to “deep learning” has been increasing, showing a gradual upward trend, with small fluctuations in most years and only some years with larger fluctuations. Through graphical data and literature analysis, the research history of deep learning is divided into three stages.

(1)

Budding growth stage (1992–2006). The amount of research is relatively small, and the year-on-year growth rate is not obvious, with 288 papers published, and the average annual publication volume less than 19 papers. The research in this stage mainly focuses on the theoretical basis, basic characteristics, occurrence conditions, and implementation mode of deep learning. The main research authors are Ference Marton, Roger Saijo, Biggs, Nelson Laird, and Kolb. Laird, Kolb, and Weige, among others.

(2)

Rapid expansion phase (2007–2019). Deep learning research has entered a rapid growth phase since 2007, with the number of papers published peaking in 2019 (288). This phase focuses on exploring the influencing factors that hinder deep learning and uncovering strategies to promote deep learning by combining examples. For example, problem-driven, reflective learning, hands-on participation, interactive learning environments, cooperative/collaborative learning, self-directed learning, and formative assessment are effective methods to promote deep learning, and deep learning strategy research supports the applied research on deep learning.

(3)

Gradual reduction stage (1 January 2020 to 1 September 2022). From 2020 to 2021, the amount of literature related to deep learning research began to decrease gradually. The reason could be the impact of the COVID-19 outbreak in 2019. After 2020, the average number of publications per year is about 227. The epidemic has made people start to change their learning methods, such as micro-classes, MOOCs and blended teaching, etc. The concept of intelligent technology-enabled education has become an emerging research trend, and the use of various information technology tools to support and optimize the learning environment of students’ deep learning has been the main direction of deep learning research in recent years. This indicates that deep learning research topics will be more diversified in the future, and deep learning research will develop towards a sustainable and stable direction of high enthusiasm.

4.1.2. Authors of Core Articles

By showing important authors in the field of deep learning research, it is easier for scholars to find representative authors and publications with great research impact in sustainable development education research. In this study, CiteSpace is used to count the authors of deep learning in the research field and the important authors are displayed in the table. Generally speaking, the identification of core authors can help discover the “knowledge map” of the field and also promote academic communication and cooperation in the field. Currently, the internationally accepted method for determining core authors is based on Price’s Theory. Price’s Law formula: M = 0.749(N_max)1/2 (where M is the minimum number of core authors and N_max is the number of articles by the most published authors). The co-occurrence mapping analysis of authors was conducted (Figure 2), and as shown in

Table 1. Deep learning research core author statistics.

Author	Number of Papers Published	Serial Number
TSAI C	23	1
ELLIS R	14	2
LIANG J	11	3
CHEN Y; HWANG G	9	4
CHAN C	8	5
KEMBER D; HAN F	7	6
LIN C; PHAN H; BARAOFFIO A; WANG Y; LI Y; WANG H	6	7
CASE J; BROWN G; ABBIATI M	5	8
PROSSER M; RICHARDSON J; LINDBLOM-YLANNE S; KRAJCIK J; DOLMANS D; CANO F; SILEN C; ANDERSON E; BALASOORIYA C; DOCHY F; RUMMEL N; CHEN S; PARPALA A; LIU H. XIAO Y; BAO L	4	9

, a core group of authors represented by TSAI C, ELLIS R, LIANG J, CHEN Y, and HWANG G has been formed in the field of deep learning research, but the density of cooperation between author nodes (Density = 0.0012), indicating that the cooperation is relatively fragmented. In addition, the statistics of the number of author papers published in deep learning research (Table 1) found that the formula of core authors shows that M ≈ 4, that is, the minimum number of core authors is four. TSAI C has published the most papers in the field of deep learning research, with 23 papers.

4.1.3. Distribution of Study Countries

The number of articles published by a country or region reflects the importance, influence and contribution of the country or region in the field. Deep learning research has gained the attention of scholars around the world. As shown in Figure 3, the analysis of the cooperation network between countries shows N = 99, E = 275, Density = 0.056, which means that a total of 275 inter-country cooperation lines are formed among 99 countries, and most of the countries are closely connected with each other, among which the cooperation and communication between the United States, Australia, England, China, and other countries are stronger. Statistical deep learning research country publication volume (

Table 2. Statistics on the number of national publications on deep learning research.

Number of Articles Issued	Centrality	Starting Year	Country
830	0.35	2002	USA
357	0.26	2002	AUSTRALIA
271	0.32	2002	ENGLAND
203	0.12	2003	PEOPLES R CHINA
140	0.13	2002	CANADA
123	0.01	2005	CHINA
110	0.06	2002	NETHERLANDS
105	0.09	2006	GERMANY
86	0.11	2003	SPAIN
71	0	2003	ISRAEL
64	0.03	2003	FINLAND
61	0	2002	SWEDEN
60	0.05	2007	NEW ZEALAND
54	0	2002	SOUTH AFRICA
50	0	2008	TURKEY

) collates the development degree and contribution of different countries in deep learning research in education, and the top five countries include: the USA (830 articles), Australia (357 articles), England (271 articles), China (203 articles), and Canada (140 articles), and a total of 15 countries have published more than 50 articles. Research on deep learning is diverse and the content and results vary considerably across countries and regions; for example, Asian students show significantly higher use of deep motivation, surface strategies, and achievement strategies, while Australian international students show higher use of deep strategies and surface motivation [33]. The majority of Singaporean students prefer teaching and learning methods that encourage deep learning [34]. Many students in the UK are used to surface or surfing approaches to learning [35]. Research in the United States focused on educational model building, which is mainly related to economic development and the importance of educational models [36,37,38]. Research on deep learning in China focuses on the integration of educational theory and practice [39,40,41,42]. Through the cooperation between countries, we can find countries with strong influence in the field of deep learning research and their degree of cooperation, which provides more resource channels for sustainable development education research.

4.1.4. Research Institutions

Spatial distribution and analysis of collaborating institutions using CiteSpace can provide effective information for countries and institutions to find partners. CiteSpace shows that N = 470, E = 362, Density = 0.0033, i.e., 470 research organizations and 362 network partnerships, with a density of 0.0033, indicating that there is little collaboration between institutions. According to the statistics, 16 organizations published more than 20 articles in this field. Taiwan University of Science and Technology published the largest number of papers, with 43 articles in the WoS database. The top five organizations include: the Chinese Taiwan University of Science and Technology (43 articles), the University of Sydney (42 articles), the University of Hong Kong (40 articles), Griffith University (34 articles), and Monash-Union University (30 articles), reflecting their importance and influence in the field (Figure 4). Comparative analysis showed a direct correlation between the number of papers published by researchers in specific organizations and the number of papers in the countries where these organizations are located. For example, researchers at Michigan State University, University of Illinois, Arizona State University, Columbia University, and Indiana State University in the United States, and the University of Sydney, Griffith University, and Monash University in Australia published a large number of papers on deep learning, which explains the high number of publications in the United States and Australia.

4.1.5. Topics and Areas of Deep Learning Research

The keywords were effectively summarized and refined, pointing to the core of the article, and the frequent keywords were used to identify the main topics in the research field, so as to provide research ideas for sustainable development education. Scholars could start from the main research topics to analyze. According to the results of keyword co-occurrence analysis, after excluding the search terms, the five keywords that were used for a long time and with high frequency (frequency over 200) were “student” (396), “education” (343), “knowledge” (241), “science” (228), and “higher education” (217). The keywords “knowledge” (241) and “science” (228) and “higher education” (217), can be considered as the main areas and basic contents of deep learning (

Table 3. Deep learning research themes.

Word Frequency	Centrality	Starting Year	Keywords
396	0.13	1992	student
343	0.1	1995	education
241	0.06	1992	knowledge
228	0.04	1992	science
217	0.05	1994	higher education
194	0.06	1993	performance
157	0.07	1993	strategy
147	0.02	1994	motivation
145	0.07	1998	model
123	0.06	1994	perception
116	0.08	1996	classroom
110	0.02	2001	experience
101	0.06	1998	achievement
100	0.02	1994	environment

), among which, higher education is the main research object of international deep learning. In addition, the co-occurrence frequency of keywords such as performance, strategy, motivation, model, perception, experience, and environment all exceeded 100. Among them, student and education have relatively high centrality values and are the deep learning research topics in each area; we used the keywords of “deep learning” according to the citation. In each topic area, we selected representative articles that are more closely related to the topic according to the citation frequency from high to low, and analyzed their research contents.

Deep learning research is rich (Figure 5) in topics and involves four major modules: (1) developing a conceptual framework for deep learning, (2) exploring the factors influencing deep learning, (3) uncovering effective strategies for deep learning, and (4) evaluating the effectiveness of deep learning. The four core research themes are further elaborated below:

(1)

Establishing a deep learning conceptual framework

The conceptual framework of deep learning laid the foundation for research on deep learning practices. In 1976, Marton et al. introduced the concept of deep learning, arguing that students who adopt a deep approach are motivated to learn, focus on understanding, and emphasize meaning and knowledge integration [14]. After the introduction of the deep learning concept, researchers began to conduct a lot of research on the concept of deep learning. Biggs and Tang et al., identified two types of learning styles for students, namely shallow learning and deep learning [43]. Then, Entwistle et al. proposed three types of learning methods: surface learning methods, deep learning methods, and achievement learning methods [44]. Nelson Laird, Kolb, and Weige defined the connotations of deep learning from different perspectives. Subsequently, Eric Jensen and LeAnn Nickelsen revealed the basic characteristics of deep learning. Dart et al. demonstrated that students with qualitative and empirical concepts preferred deep learning methods, while students with quantitative concepts preferred surface learning methods [45]. Thus, it is clear that in the context of the time, research on deep learning tended to take the educational process as the research anchor point, actively exploring educational issues in social change from the micro level of classroom teaching. The study of deep learning concepts played an important role in the subsequent research progress and was the foundation of deep learning research.

(2)

Exploring deep learning influencing factors

The research on deep learning mainly revolves around the process and results of deep learning, among which, the study of deep learning influencing factors is also one of the hot spots in deep learning research. The results of a large number of empirical studies show that the deep learning approach has different connotations under different conceptual frameworks, and its influencing factors are far more complex than we think [46], and understanding the factors affecting deep learning can lead to the design of appropriate courses, which may successfully cultivate high-quality learners [47]. The factors that influence students’ deep learning can be divided into four dimensions: (1) learning environment. Different learning environments influence the approach students take in a course [48], and “deep” activities motivate internal learning rather than just the completion of task requirements [49]. A suitable environment for deep learning should provide students with continuous feedback and encourage the application of learned theories, concepts, and knowledge to solve new problems. (2) Students themselves. Individual factors such as learning time and effort commitment, interest in learning, expected course goals, and self-regulation ability can provide a better explanation for changes in deep learning styles than learning environment factors [50]. For example, students’ own cognitive profile [51], students’ perceptions of learning [47], students’ satisfaction with the quality of the course, and students’ own personalities may influence students’ learning styles [52]. (3) Teachers. Learning approaches are not immune to the influence of teachers [53,54]. (4) Assessment methods [55]. Inappropriate assessment may hinder deep learning [56], students’ perceptions of assessment can change students’ approaches to learning [57], and process, performance, and formative assessments contribute to deep learning [58,59]. Assessment has an important role in guiding and regulating students’ learning; if teachers adopt assessment strategies that encourage students to become critical and innovative thinkers, they will trigger deep learning, and conversely, if they adopt assessment strategies that emphasize rote learning and memorization, students will choose surface approaches. Thus, teaching and assessment methods should foster active and long-term student engagement in learning tasks.

(3)

Digging deep learning effective strategies

With the keywords “performance”, “strategy”, “motivation”, “perception”, “experience”, and “environment”, the theme of deep learning revolves around methods to promote deep learning. Lynch introduced self-evaluation, peer evaluation, and feedback systems in an educational course, and the results proved that the method encouraged students to reflect on, evaluate, and critique the learning process of self and peers, effectively improving learners’ academic performance and critical thinking skills [60]; Dummer proposed an innovative and flexible deep learning for teaching field trips in geography classes’ assessment method, the reflective fieldwork journal, and the results showed that it enabled teachers to understand students’ learning processes and enhance students’ critical self-reflection skills and written communication skills [61]; Pegrum et al. used a structured task format to introduce creative podcasts in the course, and comparative experiment results showed effective promotion of deep knowledge understanding and effective retention [62]; Rozendaal et al. found that teachers who implemented collaborative and interactive teaching strategies promoted deep cognitive processing in students [63]; Cope and Staehr used student perceptions of learning environments suitable for deep learning and found a statistically significant increase in the proportion of students using deep learning methods [64]. Kek, Megan Yih Chyn A (2011) found that problem-based learning helps to enhance students’ critical thinking skills [65]; peer-to-peer tutoring (RPT) and case study learning (CBL), as proposed by Rainer Lueg and Kulak, can increase student engagement and excitement and facilitate deep learning [66]. In addition, it has been shown that the combination of self-evaluation, peer assessment and feedback facilitates higher quality learning outcomes and the development of critical thinking skills [67]; formative assessment may contribute to students’ deeper learning approach, while summarization may contribute to their surface approach [68]. Recently, new pedagogical approaches such as problem-based learning and project-based learning have been shown to facilitate students’ deep learning. For example, Loyens, Sofie M. (2013), conducted an empirical study based on students’ learning approaches and academic outcomes in a PBL setting and showed that the problem-based learning (PBL) model can effectively promote students’ deep learning more than other instructional approaches [69,70,71]. Furthermore, increasing students’ awareness of deep learning is a key part of effective teaching and learning [72].

(4)

Evaluating the effectiveness of deep learning

The effectiveness of deep learning is well supported by empirical studies, highlighting higher levels of academic achievement, better knowledge understanding and mastery, more enjoyable emotional experiences, and better development of higher-order thinking and competencies. Several studies have shown a positive correlation between deep learning and academic achievement [73,74,75,76,77], with deep learning leading to better learning outcomes [78]. In addition, a study by Sheard showed a positive correlation between learning engagement and academic achievement. Students who adopted deep learning were strongly engaged in their learning and remained engaged in their learning [79]. In fact, deep learning is usually associated with academic achievement and learning satisfaction [80,81]. As described by Ward, a deep and strategic approach is associated with academic success, while a superficial approach leads to poorer understanding [82]. Furthermore, Heijne-Penninga found that deep learning open-book exam times were negatively correlated, students scored lowest on closed-book questions about open-book topics [83], closed-book exams motivated deep learning approaches more than open-book exams [84], and students who were motivated to learn in-depth consistently performed better on exams [85]. Thus, deep learning methods provide educational value in terms of short-term academic performance and sustained motivation [86]. Deep learning has a positive impact on students’ self-efficacy levels and moral reasoning development [87], promoting better higher-order thinking and competence development. In addition, some findings suggest that deep learning methods play a crucial role in improving cognitive and non-cognitive gains of college students [88]. Deep learning effectiveness studies focus on the cognitive dimension, and other dimensions beyond cognition need to be observed for deep learning in a contextualized view.

In general, deep learning influencing factors, deep learning strategies, and deep learning effectiveness have received high attention. The deep learning conceptual framework is the substrate of deep learning. Deep learning influencing factors have also been the focus of scholars’ attention, deep learning strategies are the main difficulty of this research due to various influencing factors. Deep learning strategies are the way to apply many deep learning results. The effectiveness of deep learning guarantees the continuous development of deep learning research and is the key to the practical application of deep learning. Researchers keep exploring the occurrence mechanism of deep learning from multiple perspectives, exploring specific strategies to promote deep learning, and trying to make scientific measurements of deep learning effects. However, researchers are somewhat deficient in deep learning resource construction and motivation stimulation, and there is less relevant research literature, which may restrict the effective application of deep learning in education and teaching.

5. Theoretical Basis and New Dynamics of Deep Learning Research

5.1. Theoretical Basis of the Analysis Study

High-frequency co-cited words play a key role in the knowledge flow network, which is the basis of subject knowledge research. The research on deep learning mainly provides a direction for how to improve sustainable development education through the measurement of students ‘learning strategies and performance, and the analysis of related influencing factors. As shown in

Table 4. Deep learning high-frequency co-cited words.

Cluster ID	Size	Silhouette	Year	Terms
0	107	0.702	2005	Approaches to learning; higher education; academic achievement; learning approaches; learning strategies;
1	72	0.678	2011	Gross anatomy education; medical education; first-year undergraduate/general; undergraduate education; system;
2	68	0.586	2010	Deep learning; active learning; machine learning; educational data mining; sentiment analysis;
3	67	0.69	2011	Professional development; professional learning; early adolescence;
4	64	0.781	2005	conceptual change; education; knowledge; teachers; simulation
5	63	0.768	2008	Learning disability; curriculum design; pre-service teachers; online learning; context;
6	49	0.718	2011	Media in education; teaching/learning strategies; interactive learning environments; cooperative/collaborative learning; confusion;
7	47	0.766	2005	Science education; instruction; physics; biology; example;
8	41	0.779	2001	Strategy; student; performance; time; blended learning
9	34	0.908	2003	Memory; acquisition; 2nd language; vocabulary; retention
10	31	0.772	2013	Formative assessment; withholding answers; clinical placements; visualization; physical manipulation

, from these high-frequency co-cited words “approaches to learning”, “learning strategies”, “curriculum design”, “sentiment analysis”, “learning disability” and “performance”, deep learning is divided into two main research directions: theory and practice. Theoretical research mainly focuses on the basic connotation, inner mechanism, and influencing factors of deep learning. For example, Roberts et al. explored the impact of theories of deep learning on students’ engagement in reading [89]. Kek et al. analyzed the role of problem-based learning in digital learning environments to enhance student’s critical thinking skills [65]. Mayhew discussed the impact of deep learning on students’ development of moral reasoning [86]. Empirical research includes deep learning strategies and deep learning effectiveness. For example, creating high-quality learning environments to facilitate deep learning by designing effective interactive learning applications [90]. Loyens, Sofie M. (2013) conducted an empirical study based on students’ learning approaches and academic outcomes in a PBL setting [68]. Rozina and Shaar (2012) used a questionnaire to conduct an investigation focused on analyzing the study [91]. Heijne-Penninga (2010) and others explored the relationship between deep learning, cognitive demand, and test scores by using moderate information processing to test learners’ learning levels and combining the results of the cognitive demand scale [92], is a source of information for students’ deep learning [93].

5.2. Frontier and Trending Topics in Deep Learning Research

5.2.1. Frontiers of Deep Learning Research

In order to better show the evolution trend of mathematical literacy research, we identify the frontier fields of deep learning by interpreting the basic topics of research of deep learning in the past 30 years and the derivation of various research branches based on the keyword co-occurrence map (Figure 6). We identified the following three main research paths and sorted out the core content to guide further research of sustainable development education.

Firstly, theoretical research; deep learning theory research takes students’ learning process and learning methods as the research path, mainly analyzes the theoretical basis, occurrence conditions, and promotion strategies of deep learning, and explores several implementation models of deep learning.

Second, practical research; first, applies theoretical research results of deep learning to the teaching practice of specific subject courses; then, explores the inner mechanism and effective strategies of deep learning with subject teaching examples; and finally, uses measurement tools to test the effect of deep learning according to the practice results. Among them, deep learning strategies have been a great concern, and deep learning strategy research provides a more scientific and comprehensive analytical perspective for sustained development education based on student participation [94], the role of teachers [95], and the need to perceive the characteristics of deep learning from the perspective of teaching and learning.

Third, technical support research, where researchers use various information technology tools to support and optimize learning environments that facilitate deep learning for students, and research tools, methods, and techniques used to identify the characteristics, patterns, structures, and evolutionary mechanisms of deep learning are constantly updated and integrated. Starting from 2016, with the emergence of big data analytics, the field of deep learning research has seen the emergence of MOOC learning groups [96], flipped classrooms [97], peer-to-peer tutoring (RPT) [33], online learning technologies (SAOLT) [98], computer-supported collaborative learning (CSCL) scripts [99,100,101], and peer feedback in online courses. These are all new teaching and learning models based on intelligent technologies for deep learning. How to develop sustainable education in a technology-supported teaching environment is a difficult challenge.

5.2.2. Trending Topics in Deep Learning Research

The temporal graphs of high-frequency words and their outbreaks in deep learning are visualized in the analysis of noun term outbreaks, and the Figure shows the 25 highest frequency groups of terms (Figure 7). There is a very close relationship between these high-frequency words and other high-frequency words in recent years, indicating that these keywords lay the foundation for subsequent research.

As shown in Figure 6, the top three most explosive keywords are qualitative difference (2006–2011, strength = 7.87), strategy (1999–2002, strength = 6.78), and interactive learning environment (2010–2015, strength = 6.35), indicating that these three keywords are representative in the field of deep learning research. The strategy research topic of deep learning has been popular in deep learning during 1992–2022, which represents a continuous frequency. In terms of time and content, before 2010, deep learning research mainly focused on the intrinsic mechanisms of deep learning. Since 2010, with the emergence of the field of learning science research, the number of deep learning studies has increased substantially, and the factors influencing deep learning have become the main area of research, especially the influence of student perceptions and learning environment on deep learning, as well as the analysis and evaluation of these issues. With the diversification of research, deep learning strategies and other issues related to deep learning became the new research focus, and deep learning strategy research enhanced the importance of deep learning in educational research. Continuing into 2015, strategy research promoting deep learning became increasingly focused due to the advancement and application of deep learning and began to shift to research on deep learning effectiveness assessment and collaborative research across disciplinary domains. Four basic trends in deep learning research, in general, are predicted to be likely.

(1)

An empirical study focusing on the development of deep learning in authentic classrooms.

The research results of deep learning are applied to the teaching of sub-disciplinary courses and interdisciplinary integrated learning, and the mechanisms of deep learning are further explored with the help of specific examples of course teaching. The study also provides an empirical study of the effects of these learning models in real classrooms.

(2)

Designing an environmental support system suitable for the contextual nature of deep learning.

With the deepening influence of information technology on deep learning, it begins to shift from exploring individual-specific operation strategies to constructing generalized operation models, and the study of factors influencing deep learning in technological environments is of great significance to teaching practice. Big data, distance education technology, and artificial intelligence technology can provide scientific support for teaching and learning, integrating simulated learning or virtual evaluation systems of online information technology, and setting up simple and efficient student activities. Integrating multiple competency requirements is an effective means of addressing deep learning research.

(3)

Innovate the evaluation method of deep learning and improve the efficiency of deep learning evaluation.

Although deep learning research has received much attention in the field of education, there is no standard evaluation paradigm to measure the effectiveness of deep learning due to its abstractness and complexity. At present, the evaluation methods to test the effectiveness of students’ deep learning are mainly at the cognitive level, i.e., the internalization and transfer of knowledge, and mainly use research methods such as questionnaires, classroom observations, and mini-tests, which cannot reveal the real effect of deep learning comprehensively, scientifically and effectively, and are also prone to bias in evaluation orientation. A large number of studies have found that appropriate evaluation methods have an obvious role in promoting deep learning. Therefore, it is necessary to strengthen the understanding of the essence of deep learning and pay attention to the ability of reflection and criticism as well as the ability of social interaction, and carefully identify the difference between deep learning and other learning concepts in order to build a comprehensive, scientific and unified evaluation standard.

(4)

Focus on deep learning interdisciplinary and multidisciplinary analysis.

Deep learning is comprehensive, diverse, and closely linked to students’ self-awareness, teachers’ pedagogy, social structures, and cultural awareness. Students’ subjective consciousness influences their learning patterns, teachers act as guides and activators of learning, and positive teacher–student relationships are part of a changing and significantly innovative educational landscape [102]. Therefore, deep learning should be discussed in the context of the various factors that drive changes in teaching and learning and lead to diverse forms of research that should focus on multidisciplinary collaboration and analysis in the future. Deep learning research objects are more internationally focused on students in colleges and universities as well as adults, and relatively little research has been conducted at the basic education level, where deep learning is still maladaptive and needs to be combined with different cultural backgrounds and characteristics of students at different levels, to enable a research breakthrough for continued research on deep learning. Therefore, the next research focus of deep learning will revolve around the direction of basic education, assessment, etc.

6. Conclusions

In this paper, we take 2827 articles of English literature in the field of deep learning included in the Web of Science core set (with SSCI as the search focus) as an example and use the CiteSpace (6.1.R2) tool to perform econometric and scientific statistical analysis of research papers related to deep learning and present them in a visual way, which is useful for us to analyze the current status of the field of deep learning research helpfully. The article focuses on the visual analysis of authors, countries, journal institutions, keywords, and co-citations. The data results show that deep learning research has formed a research network system with core authors as the leading authors, multiple countries and institutions interconnected, and a wide range of research topics.

At the author, country, and institution levels, Asia was found to have the highest number of publications and more frequent contacts between research countries, but institutions and authors were relatively scattered. Therefore, communication and cooperation between institutions and authors should be strengthened in the follow-up research. Based on the visual analysis of high-frequency cited words and keywords, it was found that the hot topics of research on deep learning mainly focused on the conceptual framework, influencing factors, implementation strategies, and effectiveness of deep learning. From the conceptual framework, the empirical research on deep learning attaches great importance to the connotation definition of core concepts, among which the conceptual framework proposed by Biggs, Marton and Sergio, Entwistle and Ramsden has been more recognized. In terms of influencing factors, students’ self-efficacy, learning motivation, learning environment, instructional assessment models, and other factors may have direct or indirect effects on deep learning approaches. In terms of implementation strategies, problem-based learning, interactive teaching and learning, case study learning, and appropriate assessment are all effective strategies to promote deep learning. From the research results, deep learning can improve academic performance, promote meaningful knowledge construction, gain enjoyable emotional experiences, and develop higher-order thinking and reflective skills. There was also a significant correlation between deep learning and learning effect. Deep learning was helpful to improve learning literacy, promote the integration of deep learning into teaching, and facilitate the realization of the concept of sustainable development education. With the development of smart education, recently, attention has been paid to the issue of analyzing the learning nature of deep learning, and optimizing measurement methods and tools in order to accurately assess the effectiveness of deep learning as a new dynamic for future research [103].

In general, deep learning research takes the education process as the research point, makes positive and specific responses to the problems of sustainable development education from multiple dimensions of teaching, and constantly returns to the essence of learning while responding to the needs of sustainable development education. Facing the accelerated speed of knowledge turnover and the development of economic globalization and informatization, deep learning with technical support is the trend of the times. Whether technology and resources can enhance the effect of deep learning, promote learners’ knowledge construction and flexible application, and ultimately solve practical problems is a problem that needs to be solved in the international education community. The purpose of sustained development education is to develop students’ insight, analytical skills, problem-solving, and higher-order thinking skills, and there is no doubt that these skills can help students adapt to the rapid changes in society. As deep learning with technological support receives further widespread attention, researchers have shifted their horizons from the internal cognition of learning to the continuation of external contexts [104], and there is a tendency to integrate multiple research perspectives in complex learning environments.

Statistical analysis shows that there is a lack of comprehensive bibliometric research on deep learning in the field of education. This study systematically analyzes the theoretical basis, research status and development trend of deep learning in the field of education, and provides relevant information about the core authors, research institutions and research countries of deep learning research. Finally, this paper provides an objective forecast of the research trend in deep learning to provide a reference for subsequent research.

7. Limitations and Future Work

Firstly, this study reveals the development of deep learning research and that deep learning has experienced three stages: germination, rapid expansion and the gradual decline of deep learning. Secondly, it defines the topic of deep learning research: deep learning research mainly focuses on the conceptual framework, influencing factors, implementation strategies and effectiveness of deep learning. Finally, it analyzes the latest progress in deep learning: deep learning research has begun to focus on how to promote deep learning with the help of intelligent technology and explore the internal structure and implementation mechanism of deep learning. The research finds that deep learning is no longer a specific way of learning, but a teaching idea, which is the unified expression of deep learning environment, deep teaching and deep learning. Teachers guide students to participate deeply in learning, appropriately adopt advanced learning strategies, promote the development of students ‘higher-order thinking, realize the application of this knowledge and ability in real situations, and realize the transformation from traditional education to sustainable development education.

Although this study systematically analyzes the recent progress of deep learning research, there are still some limitations; we only analyzed the English literature in the SSCI database, ignoring the articles included in other languages and other databases, and the depth and comprehensiveness of the analysis is not enough. Second, we only analyzed the literature and lacked empirical evidence of some literature findings. In addition, the depth and comprehensiveness of the analysis are lacking due to the complexity of the relevant disciplines and the limitations of the authors’ knowledge. For more and more specific research paths, a close reading of the literature is needed, and a more in-depth study and analysis based on it. These shortcomings will be further improved and analyzed in a subsequent study.