| Originally published in the 11th issue of “Educational Measurement and Evaluation” in 2019
Author Introduction
Zhang Shaojun/ Teacher at the College of Education Science, Hunan Normal University, PhD in Education, main research directions include curriculum and pedagogy, and basic education curriculum reform.
Chen Mingying/ Master’s student at the College of Education Science, Hunan Normal University, main research direction is curriculum and pedagogy.
【Abstract】With the rise of information technology and learning sciences, “deep learning” has become a hot research topic in the field of education. Using “deep learning” as a keyword, we retrieved 1827 journal articles and 174 master’s and doctoral theses from the CNKI database over the past fifteen years. Based on the dimensional statistics of literature quantity and research content, the main research hotspots in China’s deep learning studies over the past fifteen years are concentrated on the connotation, characteristics, strategies, and evaluation of deep learning. This article categorizes and organizes the aforementioned aspects of deep learning using content analysis, further clarifying its research ideas and dynamics. To better serve China’s practical teaching and promote student growth in future deep learning research, new research directions such as the deep theoretical value of deep learning, deep teaching strategies, and multi-dimensional dynamic evaluation urgently need to be explored.
【Keywords】Deep Learning; Connotation; Characteristics; Strategies; Evaluation
“Deep learning” originated from the machine learning method of multi-layer neural networks in artificial intelligence. Research in the education field can be traced back to the mid-1950s, when Ference Marton and Roger Säljö at the University of Gothenburg in Sweden proposed the concepts of deep learning (deep learning) and surface learning (surface learning) in their experimental research on student learning processes, as presented in the paper “The Essence of Learning: Outcomes and Processes” (1976). China started this research relatively late, with He Ling and Li Jiahou being the first to propose the concept of deep learning in “Promoting Students’ Deep Learning” (2005). After the release of the “National Medium- and Long-Term Education Reform and Development Plan Outline (2010-2020)” (2010) and the “State Council’s Implementation Opinions on Deepening the Examination and Enrollment System Reform” (2014), the research process related to deep learning accelerated, with a surge in literature quantity, diverse research perspectives, and varied research methods across a wide range of disciplines. Based on this, to better grasp the dynamics of deep learning research, this article adopts content analysis, following the principles of objectivity and comprehensiveness, to conduct quantitative analysis and systematic sorting of relevant research literature published in China over the past fifteen years, aiming to benefit future deep learning research in further serving China’s educational practice.
1. Research Sampling and Methods
1 Research Subjects
By entering the keyword “deep learning” in the China Journal Full-text Database and the China Excellent Master’s and Doctoral Thesis Full-text Database, selecting literature directory classification from the Social Sciences II series, and setting the time span from 2005 to June 2019, a total of 2127 Chinese literature pieces were retrieved, mainly involving three major disciplines: computer science (29 pieces), control engineering (32 pieces), and education (2001 pieces). This study focuses on the education field, selecting 2001 pieces of literature (1827 academic papers and 174 master’s and doctoral theses) as research samples.
2 Research Methods
This article mainly adopts content analysis as the research method, which is one of the primary research methods in education, analyzing the amount of information contained in the literature and its changes, conducting reproducible and effective inferences from the content of the literature, and achieving the effect of seeing the essence through in-depth analysis. The specific implementation steps of content analysis are three:First, select research samples based on research questions and design analysis dimensions;Second, choose analysis units, design category tables, and classify statistics based on the designed categories;Third, use statistical analysis tools to provide objective, systematic, and quantitative descriptions of the statistics and draw conclusions. Based on this, this article focuses on the connotation, characteristics, strategies, and evaluation of deep learning from the dimensions of literature quantity and research content, combining quantitative and qualitative analysis.
2. Variables of Literature and the Breadth of Impact
1 Yearly Variables of Research Literature
Statistics of relevant papers on “deep learning” yield Figure 1. In Figure 1, the horizontal axis represents the year, and the vertical axis represents the number of papers published each year, with two curves representing academic papers and master’s and doctoral theses, respectively. Figure 1 shows that although the total literature from 2005 to 2010 remained low, both academic papers and theses showed a growth trend, indicating an overall upward trend in deep learning research. In 2010, the “National Medium- and Long-Term Education Reform and Development Plan Outline (2010-2020)” stated, “Attention should be paid to cultivating students’ autonomous learning ability, emphasizing the cultivation of students’ initiative, independence, creativity, and problem-solving abilities.” Therefore, in 2011, the number of academic papers on deep learning doubled compared to the previous year. In 2014, the “State Council’s Implementation Opinions on Deepening the Examination and Enrollment System Reform” was introduced, emphasizing that promoting students’ healthy growth and success should be the starting and ending point of the reform, “focusing on assessing students’ independent thinking and ability to analyze and solve problems using the knowledge they have learned.” In the same year, the Ministry of Education’s Basic Education Curriculum Textbook Development Center launched a research project on “deep learning” teaching improvement in multiple experimental zones, inspiring more scholars to explore deep learning in conjunction with educational practice, leading to an explosive growth in the number of papers after 2015. Among them, the number of papers in 2016 reached 219, more than double the 74 papers in 2015, with the total literature quantity approaching that of previous years combined; the number of papers in 2018 was close to half of the total literature from all previous years; and by the first half of 2019, the number of papers had reached 525.
Figure 1 Analysis of Literature Quantity Related to Deep Learning from 2005 to 2019
2 Breadth of Impact of the Literature
（1）Wide Range of Academic JournalsBy organizing and analyzing the journals where the papers are published, it was found that there are a total of 379 journals involving papers on “deep learning”, with 12 journals having published 25 or more papers (see Table 1). As research delves into different disciplines, the number of papers published in various journals has also increased.Table 1 Distribution of Published Papers by Journal
（2）Dispersal of Theses Across Many UniversitiesQuantitative analysis of the universities where the theses are affiliated shows that among the 174 theses, they involve 72 universities. Among them, 8 universities have published 5 or more theses (see Table 2), with the highest number reaching 11.Table 2 Distribution of Theses by University
Clearly, the relevant research on “deep learning” theses mainly originates from normal universities, indicating that normal universities are more timely and sensitive in tracking educational research hotspots and recognizing the trends of teaching reform. Table 2 also shows that research on the theme of “deep learning” not only expands the breadth of research subjects in universities but also deepens related research.
3. Content Categories of Research
Based on the interpretation and analysis of the literature, the main content of deep learning research focuses on four aspects: connotation, characteristics, strategies, and evaluation (as shown in Table 3).Table 3 Research Content Categories of Literature Related to Deep Learning from 2005 to 2019
1 Research on the Connotation of Deep Learning
Research on the connotation of deep learning mainly has two perspectives. First, from the perspective of learning levels. Professor Li Jiahou and others were the first to introduce the concept of deep learning in China, and based on Bloom’s taxonomy of educational objectives, they divided teaching objectives into six levels: recall, understanding, application, analysis, synthesis, and evaluation; shallow cognitive levels remain at the first two levels, while the cognitive level of deep learning corresponds to the latter four levels, meaning that based on understanding learning, learners can critically learn new ideas and facts, integrating them into their existing cognitive structures, making connections between various ideas, transferring existing knowledge to new situations, making decisions, and solving problems in learning.[6]Some scholars also base their views on Dewey’s educational philosophy of cultivating students’ good thinking habits, combined with the requirements of the times for learning in the context of information technology, stating that deep learning is a concept corresponding to surface learning, where the process of deep learning triggers students’ higher-order thinking and cognition, ultimately cultivating students’ critical thinking and innovative consciousness.[7]
Second, from the perspective of rationality and non-rationality. The “depth” of deep teaching refers to the level of knowledge interpretation and the richness of student development. Deep teaching transcends instrumental teaching, not controlling the teaching process through technology or procedures, nor regarding the acquisition of textbook knowledge as the only task of teaching, but returning to the essence of teaching, focusing on context, process, values, and meanings, emphasizing guiding students to transcend superficial symbolic knowledge learning, entering the thoughts, methods, logic, values, and meanings behind knowledge symbols, elevating symbolic learning to deep-level meaning acquisition, making students’ learning full of value care and meaning care.[8]Therefore, deep learning encompasses multiple aspects of learning, involving students’ perception, thinking, emotions, will, and values in a comprehensive and engaged manner, aiming at the comprehensive development of specific, social individuals, and forming the basic path to cultivate students’ core competencies.[9]From the perspective of rationality and non-rationality, researchers generally agree on the connotation of deep learning, expressing a basic understanding of the essence of learning: deep learning is not only an explicit behavior of learning, nor is it limited to the mastery of knowledge and acquisition of intelligence, but advocates a meaningful learning that is proactive and critical, emphasizing attention to students’ inner world, such as emotional qualities and personality psychology.
2 Research on the Characteristics of Deep Learning
Transitioning from shallow to deep learning is efficient, manifested in five characteristics: the primary characteristic is cognitive understanding, the inherent characteristic is higher-order thinking, the essential characteristic is holistic connectivity, the necessary characteristic is creative criticism, and the trending characteristic is expert construction.[10]These characteristics manifest in three aspects of deep learning: first, closely related to students’ explicit and implicit series of quality goals in achieving cultivation objectives and outcomes; second, requiring time, effort, and determination for cognitive processing in acquiring complex thinking skills; third, necessitating students’ full participation in behavioral, emotional, and cognitive dimensions.[11]When judging students’ performance in deep learning and its occurrence patterns, Guo Hua believes that it can be examined from five pairs of characteristics: association and structure (the mutual transformation of knowledge and experience), activity and experience (students’ learning mechanisms), essence and variation (deep processing of learning objects), transfer and application (simulating social practices in teaching activities), value and evaluation (the implicit factors of human growth), with the core characteristic being “activity and experience.”[9]In contrast to Guo Hua’s perspective on deep learning characteristics from the viewpoint of student “subjectivity,” Zhang Hao and others propose characteristics of deep learning based on cognitive theoretical foundations, emphasizing critical understanding, information integration, knowledge construction, attention to transfer and application, problem-solving orientation, and advocacy for proactive lifelong learning.[12]Regarding learning itself, the relevant characteristics of deep learning mainly manifest in rich and meaningful learning activities, high-quality problem-driven learning, attentive listening learning, deep expressive learning, and deep understanding of texts.[13]Additionally, there are studies exploring the characteristics of deep learning based on the differences in learning behaviors of deep and shallow learners in blended learning environments.[14]The aforementioned research on the characteristics of deep learning exhibits multiple layers and dimensions, with both cognitive psychological theoretical foundations and affective dynamic investigations, also focusing on the learning process itself and the specific behaviors of teachers in teaching.
3 Research on Deep Learning Strategies
First, research on theoretical guidance and practical strategies. In response to the superficial teaching issues in curriculum concepts, teaching objectives, learning methods, curriculum structure, teaching methods, and evaluation of effects, some scholars propose using strategies that promote students’ continuous, understanding, critical, exploratory, experiential, and reflective learning to facilitate the occurrence of deep learning.[15]Some scholars advocate for a guide for teachers to adjust their beliefs and a series of teaching behaviors:establish teaching objectives that develop higher-order thinking, guide students to deeply understand; integrate meaningful learning content, guide students to critically construct; create authentic situations that promote deep learning, guide students to actively experience; choose evaluation methods that maintain focus, guide students to reflect deeply.According to Professor Cui Yunhu, an important strategy to promote students’ deep learning is for teachers to do a “curriculum case,” as teacher guidance can lead to faster, more meaningful learning than complete autonomy, providing professionally designed opportunities for deep learning.[17]Regarding overall practical strategies, deep teaching can be promoted through shaping the learning character of cultural practices, carrying out innovative learning activities, and implementing quality-oriented learning evaluations.[18]Second, research on teaching strategies for different disciplines or specific paradigms. Each discipline has its corresponding “depth” teaching strategies; for example, in English, the focus is on constructing a deep teaching model based on enhancing core competencies in English, i.e., “focusing on thematic significance, setting overall teaching objectives for units, deeply reading texts, and integrating core teaching content; targeting core competencies, implementing deep teaching activities.”[19]In mathematics, the emphasis is on forming students’ awareness of problem-solving strategies, i.e., “stimulating thinking by creating good contexts, designing effective activities to promote thinking, thus entering the state of deep learning.”[20]From the perspective of specific teaching paradigms, some scholars point out that learning activity design in the flipped classroom teaching paradigm should “promote meaningful knowledge organization based on problem systems and strategies based on explicit thinking and interpersonal interaction,” while also coordinating teaching and learning activities, reasonably selecting media and teaching strategies to enhance students’ thinking quality.[21]Some scholars explore constructing an effective scenario to promote students’ deep learning in the context of situational teaching paradigms, avoiding the current superficial teaching tendency of “emphasizing knowledge over ability, emphasizing noise over thinking, and emphasizing tasks over generation.”[22]Third, research on deep teaching strategies for different educational stages. At the preschool stage, deep learning can be effectively promoted by carrying out scientific activities, creating rich scientific discovery rooms and activity areas, and integrating scientific activity ideas into children’s daily life and play activities.[23]In primary and secondary schools, deep learning can be facilitated by inducing cognitive, emotional, and value conflicts among students;[24]or by reconstructing deep learning models and proposing feasible plans for classroom reform to address the “learning difficulties” of students.[25]At the university stage, corresponding teaching content, methods, and processes can be explored based on the requirements of deep learning;[26]in the graduate stage, classroom design can be conducted from four aspects: “self-understanding before class, sharing understanding during class, reflecting on understanding after class, and applying understanding at the end of class”[27]to enhance deep learning motivation and engagement.
Additionally, research on deep learning strategies also involves different theoretical foundations and significances, such as the educational significance of “research-based learning (or scientific inquiry), multi-dimensional representation learning, learning through thinking, and active learning”[28]and research on adult online deep learning based on immersion theory.[29]
4 Research on Deep Learning Evaluation
With the deepening of research on deep learning, evaluation research on deep learning has increasingly attracted the attention of educational theorists and practitioners, with two research orientations: student learning outcome orientation and classroom teaching process orientation.Research oriented towards student learning outcomes mainly manifests in the construction of evaluation systems and experimental studies. Studies on system construction include the deep learning evaluation system proposed in 2014, based on Biggs’ SOLO taxonomy, Bloom’s cognitive objective taxonomy, Simpson’s psychomotor skills objective taxonomy, and Krathwohl’s affective objective taxonomy, constructing a four-dimensional evaluation system integrating cognition, thinking structure, psychomotor skills, and affect.[12]Experimental evaluations mainly focus on reflective deep learning experimental studies, dividing learning content into well-structured and non-well-structured problems, measuring and analyzing learning outcomes in five dimensions: total scores, shallow knowledge, deep knowledge, thinking structure, and works evaluation.[30]There are also studies based on transfer theory and the SOLO level classification method, constructing the evaluation model of deep learning into two aspects: the basis of deep learning and the degree of deep learning, as well as three dimensions: understanding of new knowledge, internal association transfer, and external expansion transfer, i.e., the “3+2” evaluation model, changing traditional evaluation methods for maintaining and transferring deep learning.
Research oriented towards classroom teaching process mainly refers to evaluation studies based on student development. To this end, teachers should reflect the “process” standards of deep teaching in the teaching process: deeply interpreting knowledge, connecting knowledge with life experiences, immersing students in ideological and cultural experiences, cultivating students’ core competencies, and so on.[8]At the same time, from the perspective of deep learning, it is suggested to initially construct a smart classroom evaluation index system that organically combines pre-class, in-class, and post-class aspects[32]and to follow the diversification of evaluation subjects and the diversification of evaluation forms and methods when organizing evaluations.[33].
4. Research Outlook
Deep learning emphasizes critical thinking, knowledge construction, and problem-solving orientation, which has attracted the deep attention of China’s educational authorities and educational theorists and practitioners. In recent years, research and practical attempts on “deep learning” have brought many benefits to educational reform and teaching practice in China. However, aspects such as the deep theoretical value of deep learning, teachers’ “deep” teaching strategies, and evaluation mechanisms for deep learning urgently need to be explored further.
1 The Deep Theoretical Value of Deep Learning Needs to be Further Explored
On one hand, how to transcend the general theoretical explanatory framework of deep learning’s concepts, principles, and functions, and further explore the principles of deep learning’s occurrence mechanisms, deep processing mechanisms, implementation conditions, as well as students’ methodologies, motivations, and expected learning effectiveness in deep learning, are issues that future deep learning research cannot overlook. In other words, how to promote students to pay attention to the internal connections of knowledge and the integration of information at a higher level, emphasize metacognition and critical reflection, and other higher-order thinking abilities, as well as the importance of high emotional and behavioral engagement, requires in-depth research.[34]After all, deep learning is not only about superficial information processing learning but also concerns higher-level issues such as students’ life growth and moral emotions, where “some deep learning methods are more conducive to moral growth than others.”[35].On the other hand, regarding the practical orientation of deep learning theory, how the “deep” theoretical value of deep learning is recognized by both teachers and students, how it effectively guides students’ learning behaviors and is experienced by students, and subsequently effectively enhances students’ higher-order thinking abilities such as analysis, synthesis, and evaluation, is also worthy of further attention.
2 Teachers’ Deep Teaching Strategies Still Require Comprehensive and Multi-Perspective Exploration
Under the concept of deep learning, classroom teaching practice is not a simple unidimensional interaction between teachers and students, but is based on shared teaching experiences and life development between teachers and students, encompassing: teacher professional development and growth, students’ mental health development and personality development, utilization and creation of teaching resources, teaching strategies and methodologies, theories of learning, etc. From the perspective of the teacher subject, it also involves teachers’ self-learning strategies, teaching design strategies, organizational teaching strategies, communication teaching strategies, and teaching evaluation strategies. The purpose of employing these strategies by teachers is to “foster active learning in and out of the classroom”[36]and “promote critical thinking and deep learning”.[37].
In addition to general strategy research, deep teaching strategies based on cognition in different subjects or fields also need to be closely examined. For example, in primary and secondary school mathematics, research on enhancing students’ “self-regulated learning strategies and explicit instruction of metacognition”[38], research on “metacognitive and metacomprehension knowledge for deep understanding of texts, especially for understanding inferential information” in reading instruction[39], and research on “inquiry as the central strategy for all science courses” in science instruction[40], etc.
3 Multi-Dynamic Evaluation Mechanism Construction Based on Student Development is Needed
Conventional curriculum and teaching evaluations find it difficult to effectively assess students’ learning conditions in the context of deep learning, primarily due to being constrained by the evaluation ideology centered on “classroom teaching,” which regards classroom teaching activities and their outcomes as the evaluation objects. As previously mentioned, deep learning evaluation based on classroom teaching activities mainly manifests in the orientation of student learning outcomes and the orientation of classroom teaching processes. Under the context of the “Internet+” era, the time and space, fields, processes, methods, paths, and goal positioning, content categories, and behavioral representations of autonomous learning are becoming increasingly personalized and even “fragmented”.①Therefore, it is necessary to attempt to construct a multi-dynamic evaluation mechanism based on student development under conditions of random changes in learning time and space:for instance, a stage evaluation mechanism based on “individual learning” focusing on cognitive processes, learning behaviors, and emotional engagement; an immediate evaluation mechanism based on “network learning” focusing on learning environments, personalized curricula, and learner experiences; and a multi-modal semantic evaluation mechanism based on the “five modalities”② of information media, information input (multiple symbols), and participant experiences; etc. Notes:① The issue of fragmentation mainly manifests in six aspects: “time fragmentation, reading fragmentation, learning fragmentation, knowledge fragmentation, information fragmentation, and intelligence fragmentation.” Refer to Hu Zhuanglin’s article “The Era of Multi-modal Fragmentation”.② The “five modalities” refer to visual modality, auditory modality, tactile modality, olfactory modality, and gustatory modality. Refer to Zhu Yongsheng’s article “The Theoretical Basis and Research Methods of Multi-modal Discourse Analysis”.
References:
You may also be interested in
▶ Construction, Application, and Insights of the Evaluation System for Middle School Students’ Deep Reading Literacy
▶ Reflecting on Practical Text Reading Questions in the Gaokao from International Large-Scale Reading Tests
▶Reading Ability Value-Added Evaluation Based on IRT Equivalence
▶Application of Mind Maps in Whole Book Reading and Its Formative Evaluation
▶ Mind-Oriented Whole Book Reading and Its Formative Evaluation
▶Regional Monitoring of Practical Text Reading Ability of Eighth Grade Students
Special Statement:
♦ This article is sourced from the journal “Educational Measurement and Evaluation”. Please indicate the source if reprinted. (WeChat ID: jyclypj)
♦ The journal “Educational Measurement and Evaluation” has never commissioned any intermediary organizations to compile articles, and has never charged any fees for accepted manuscripts.
