







































J. Zhai et al. /Future Technology                                                                                      November 2025| Volume 04 | Issue 04 | Pages 
205-215 

205 

 

 

 

Article 

Research on intelligent regulation mechanisms of 

learner cognitive load in digital learning 

environments 
Jianxu Zhai*, I Gusti Putu Sudiarta, Made Hery Santosa, I Wayan Puja Astawa  

Universitas Pendidikan Ganesha, Jl. Udayana No.11, Banjar Tegal, Singaraja, Kabupaten Buleleng, Bali 81116, Indonesia 

A R T I C L E   I N F O 
 

Article history: 
Received 30 June 2025  
Received in revised form 
08 August 2025 
Accepted 23 August 2025 
 
Keywords:  
Cognitive load regulation,  
Intelligent tutoring systems,  
Adaptive learning technologies,  
Digital learning environments 
 
*Corresponding author:  
Email address: 
undiksha_jianxu@sina.com 
 
 
DOI: 10.55670/fpll.futech.4.4.17 

A B S T R A C T 
 

This research develops an intelligent cognitive load regulation framework for 
digital learning environments in the context of educational policy reforms. After 
China's Double Reduction Policy took effect, tutorial-concentrated schooling 
evolved into technology-facilitated learning, putting unimaginable cognitive 
burdens on students. In response, the research combines cognitive load theory 
with adaptive technologies to resolve these issues through real-time 
recognition of cognitive states and personalized interventions. Based on the 
mixed-methods design with 320 Dongcheng District students, the research uses 
established measures such as NASA-TLX adapted to e-learning environments to 
assess multidimensional patterns of cognitive load. The smart regulation 
system shows significant efficacy with lower socioeconomic students posting 
15.3-point improvements in academic scores, task accomplishment rates 
enhanced by 32%, and the level of cognitive loads decreased by 23.1% on 
average across various types of learners. The system can recognize with 87.3% 
accuracy and respond in 234 milliseconds, thus facilitating timely interventions. 
Self-paced review activities yield 91.2% success rates, while collaborative tasks 
remain problematic at 68.4% success rates. The results extend cognitive load 
theory with dynamic adaptation capacities needed for self-managed digital 
learning. The present study provides evidence-based practice to maximize 
cognitive experiences of e-learning, facilitating education equity objectives 
while developing core self-regulated learning skills in post-reform education 
systems. 

1. Introduction 

China's Double Reduction Policy, implemented in 2021, 
has drastically altered the education sector by capping 
excessive homework and banning profit-making education 
tutoring in major subjects, putting traditional pressures on 
off-stream learning support systems [1]. Changes driven by 
policy have especially heightened the demand for successful 
digital learning solutions as conventional tutoring-intensive 
models make way for technology-enabled pedagogical 
paradigms [2, 3]. The spatial separation inherent in virtual 
learning environments brings into play complex cognitive 
demands linked with multimedia information processing, 
independent wayfinding through digital interfaces, and self-
managed learning administration without explicit 
instructional facilitation [4, 5]. Such cognitive demands are 
radically different from common classroom experiences, 
warranting systematized investigations of how students learn 
to find their way through these technology-enriched learning 
environments. Policy-driven cutbacks in extraneous tutorial 
support have left spectacular scaffolding deficits in learning 

that online systems will need to rectify if they are to remain 
capable of continued provision of educational quality and 
equity [6]. Contemporary online education systems are being 
driven increasingly hard to reconcile mandatory content 
provision with individualized learning needs, especially if 
pedagogic accommodations are constrained by technology 
limitations [7, 8]. The diversity of learner cognitive abilities, 
knowledge levels, and technological proficiency creates such 
immense tensions with the uniformity of delivering digital 
content [9, 10]. Increasingly, schools find themselves unable 
to offer differentiated learning experiences that cater to 
various cognitive requirements within the limitations of 
standardized digital learning environments. Cognitive Load 
Theory offers theoretical explanations of how learners 
mentally process information in computer-aided learning 
systems, yet gaps between theoretical concepts and real-
world implementation in natural learning environments are 
significant [11, 12]. Conventional application of cognitive load 
concepts frequently does not reflect the dynamic, interactive 
processes of today's online learning systems, especially those 

 

 

Future Technology 

Open Access Journal 

https://doi.org/10.55670/fpll.futech.4.4.17 

 

November 2025| Volume 04 | Issue 04 | Pages 205-215 

Journal homepage: https://fupubco.com/futech 

 

ISSN 2832-0379 

mailto:undiksha_jianxu@sina.com
https://doi.org/10.55670/fpll.futech.4.4.17
https://fupubco.com/futech


J. Zhai et al. /Future Technology                                                                                      November 2025| Volume 04 | Issue 04 | Pages 205-215 

206 

 

involving multimedia and complex navigational designs [13]. 
The dynamic cognitive load processes within extended online 
learning sessions are still inadequately researched, thereby 
undermining the development of implementable intervention 
tools for managing cognitive overload situations [14]. 
Existing theoretical models primarily concentrate on static 
instructional design principles but neglect the adaptive needs 
of technology-driven learning environments. Current 
methods for cognitive load management in virtual learning 
environments have noteworthy deficiencies in reflecting real-
time learner cognitive state and differences [15]. Current 
adaptive learning environments mostly attend to 
performance-based adjustments without explicit cognitive 
load measurement, even risking omitting opportune 
moments of timely intervention before the onset of learning 
issues [16]. Most of the available technologies are post-factum 
performance measures instead of anticipatory monitoring of 
brain states, leading to reactive measures that do not pre-
empt cognitive overload conditions [17]. Infrequent 
embedding of smart technologies into contemporary learning 
systems limits their capacity to offer the next generation of 
adaptive support necessary in modern learning 
environments [18]. Understanding how cognitive load 
regulation can be effectively implemented in real-world 
educational contexts where policy-driven changes have 
altered traditional learning support systems remains 
inadequately addressed in current research. Existing studies 
predominantly examine cognitive load in controlled 
laboratory settings rather than investigating authentic 
scenarios where multiple social, technological, and 
pedagogical factors interact simultaneously. The lack of 
comprehensive frameworks for integrating cognitive load 
theory with intelligent technologies specifically designed for 
post-reform educational environments represents a 
significant limitation in current knowledge. Additionally, 
most current research fails to account for the dynamic 
adaptation requirements that emerge when learners 
transition from highly structured external support systems to 
more autonomous digital learning environments.  

To address these critical gaps, the current investigation 
explores how cognitive load theory must adapt to 
contemporary educational realities. Three questions guide 
this work. Integrating real-time detection into digital systems 
remains challenging. Patterns differ sharply between 
students from different socioeconomic backgrounds when 
tutoring ends. Finding the right balance proves crucial, 
especially ensuring technology supports rather than replaces 
teachers. The research focuses on three objectives: 
developing an adaptive framework that integrates machine 
learning with cognitive load theory, testing it in policy-
disrupted schools, and creating human-centered guidelines. 
Continuous monitoring replaces periodic checks while the 
system learns from individual student paths instead of forcing 
predetermined routes. Real classrooms affected by policy 
changes offer authentic testing grounds that laboratory 
studies miss. This work fundamentally shifts cognitive load 
theory from describing what happens to actively intervening 
when students need help. The framework bridges cognitive 
science and educational technology right where learning 
occurs, in classrooms facing real disruption rather than 
controlled settings. This research addresses these limitations 
by developing an intelligent cognitive load regulation 
framework specifically designed for online learning 
environments in post-policy educational contexts. This study 
defines intelligent regulation mechanisms as systems that 
detect learner cognitive states and dynamically adjust 

instructional elements to maintain optimal load. The study 
integrates cognitive load theory with adaptive technologies to 
create responsive learning systems capable of real-time 
cognitive state detection and regulation, addressing the 
specific challenges that emerge when traditional educational 
support structures are transformed by policy reforms. The 
research contributes to advancing personalized online 
education by establishing evidence-based methodologies for 
optimizing cognitive experiences in digital learning 
environments, ultimately supporting broader educational 
goals of equity and effectiveness in technology-enhanced 
learning contexts where learners must develop greater 
autonomy and self-regulation capabilities. 

2. Data and methods 

2.1 Theoretical framework and research design 
This study establishes its theoretical foundation on 

cognitive load theory within digital learning environments, 
integrating empirical insights from the educational reform 
context in Dongcheng District. As illustrated in Figure 1, 
cognitive load theory encompasses three distinct components 
operating within working memory's limited capacity. 
Intrinsic cognitive load arises from the inherent complexity of 
learning materials, element interactivity, and prior 
knowledge requirements, which directly impact learners' 
information processing capabilities in digital environments. 
Extraneous cognitive load emerges from suboptimal interface 
design, navigation complexity, and multimedia elements that 
may impede rather than facilitate learning processes in online 
platforms. Germane cognitive load represents the cognitive 
resources dedicated to schema construction, knowledge 
integration, and skill transfer, ultimately contributing to 
meaningful learning outcomes. 

The theoretical framework illustrated above requires 
careful consideration of how learning phases manifest 
differently in digital contexts. Beyond acquiring and 
automating knowledge, students face particular complexity at 
the transfer level, where they tackle new problems without 
the tutoring support that once guided such applications.  

Contemporary research reveals how digital 
environments reshape cognitive load dynamics [19]. In digital 
learning, students juggle multiple tasks at once: navigating 
interfaces, understanding content, and managing their own 
learning process. This creates overlapping cognitive demands 
that traditional classrooms rarely impose [20]. This approach 
shifts from taking snapshots of cognitive load to following its 
ups and downs throughout learning sessions, making it 
possible to intervene early when students start struggling. 
These considerations highlight why cognitive load theory 
requires adaptation for digital contexts, particularly to 
address temporal dynamics and concurrent cognitive 
demands. The study takes a mixed-methods design, 
integrating quantitative measures of cognitive load using 
validated tools and qualitative content analysis of learning 
experience derived from interviews and observational data. 
Methodological triangulation in this way allows for in-depth 
scrutiny of how intelligent regulation mechanisms can 
streamline cognitive load balance across the three 
dimensions [21]. A quasi-experimental design assesses 
learning outcomes pre- and post the introduction of adaptive 
regulation systems with special emphasis on individual 
differences in cognitive capacity and learning style in the 
target study population. The synthesis of real-time behavioral 
analytics with performance data offers strong evidence for 
the assessment of intervention effect while ensuring 
ecological validity in natural educational settings. 



J. Zhai et al. /Future Technology                                                                                      November 2025| Volume 04 | Issue 04 | Pages 205-215 

207 

 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

2.2 Intelligent Regulation System Architecture 
The intelligent regulation system architecture employs a 

three-tier framework for cognitive load optimization in 
digital learning environments. Real-time monitoring 
mechanisms capture multidimensional behavioral indicators 
through embedded analytics that track task engagement 
patterns, response latencies, and navigation sequences within 
the learning platform. These indicators enable algorithmic 
detection of cognitive states, including attention fluctuation, 
fatigue onset, and comprehension difficulties, providing 
continuous assessment beyond periodic performance 
evaluations [22]. The monitoring infrastructure processes 
streaming data through edge computing nodes to minimize 
latency, ensuring timely intervention when cognitive 
overload indicators emerge. 

The adaptive regulation algorithms utilize learner 
profiles constructed from behavioral patterns and 
socioeconomic stratifications identified in the Dongcheng 
District study, where students' adaptation to reduced 
tutoring support revealed distinct cognitive load patterns 
across different demographic groups. The system implements 
hierarchical difficulty adjustment through content 
decomposition strategies that segment complex materials 
into cognitively manageable units, with granularity 
determined by real-time performance feedback [23]. 
Reinforcement learning works well for personalized 
pathways since Q-learning can balance familiar and 
challenging content as students progress. This adaptive 
approach avoids the need for pre-labeled data that limits 
supervised methods.  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 
 
The scaffolding engine generates contextual support 

through natural language processing, delivering explanations, 
hints, and worked examples calibrated to momentary 
comprehension gaps identified through error pattern 
analysis. Multimodal data fusion integrates disparate 
information streams through ensemble learning methods 
that synthesize behavioral, performance, and self-reported 
indicators into unified cognitive load estimates. The fusion 
architecture employs temporal convolutional networks to 
capture time-dependent patterns in clickstream data, while 
attention mechanisms weight the relative importance of 
different modalities based on task characteristics [24]. For 
processing lengthy clickstream data, TCNs work better than 
recurrent networks. They avoid the memory fade that occurs 
when LSTMs attempt to recall patterns from hours earlier in 

a learning session. Performance metrics incorporate the 
assessment framework established in the parent study, 
enabling direct comparison with traditional learning 
outcomes. Self-report instruments embedded within the 
platform collect subjective cognitive load ratings through 
validated scales, providing calibration points for algorithmic 
predictions. This comprehensive approach enables nuanced 
detection of cognitive states that inform intervention timing 
and intensity, particularly crucial for learners adapting to 
reduced external support structures in the post-reform 
educational landscape. This parallels how cognitive load 
theory describes human learning—multiple information 
streams processed separately, then integrated. The 
algorithms essentially mimic this natural process, handling 

Intrinsic Cognitive Load
·Content Complexity

·Element interactivity

·Prior Knowledge

·Task Difficulty

Extraneous Cognitive Load
·Interface Design

·Navigation Complexity

·Remote Interaction

·Multimedia Elements

Germane Cognitive Load
·Schema Construction

·Knowledge integration

·Skill Transfer

·Automation Processes

Working Memory
Limited Capacity

Cognitive Load Theory in Digital Learning Environments

Digital Learning Environment Context

 
 
Figure 1. Theoretical framework of cognitive load components in digital learning environments 

 



J. Zhai et al. /Future Technology                                                                                      November 2025| Volume 04 | Issue 04 | Pages 205-215 

208 

 

different data types through specialized methods before 
combining results. Table 1 shows how different algorithms 
performed during testing, which led to choosing the ensemble 
method. The ensemble combining RL's adaptive decisions, 
TCN's temporal patterns, and gradient boosting's error 
correction achieved the best balance across all metrics, 
justifying the additional complexity. 

Table 1. Algorithm performance comparison 

Method Accuracy Latency Adaptability 

Q-learning (RL) 86.7% 241ms 0.92 

Random Forest 82.1% 178ms 0.71 

LSTM 87.9% 423ms 0.78 

Ensemble 
(adopted) 

88.3% 267ms 0.89 

 
Model validation employed stratified 5-fold cross-

validation, where each fold served as a validation set once, 
while the remaining 80% was used for training. After 
selecting the best model through cross-validation, the final 
performance was evaluated on a held-out test set (15% of the 
total data). 

2.3 Data collection and analytical procedures 
The data collection protocol builds upon the 

foundational dataset from the Double Reduction Policy 
impact study, which documented tutoring participation 
declining from 75% to 40% and family education spending 
averaging 30% of household income. This dramatic shift in 
educational support patterns provides the context for 
examining how intelligent cognitive load regulation can 
address emerging learning challenges.  As shown in Table 2, 
the research maintains the original sample of 320 students 
across grades 4-9. The multidimensional assessment 
framework incorporates insights from the initial Double 
Reduction Policy evaluation (50 parents and 20 educators 
from the original Dongcheng District study), which was later 
expanded to include 280 parents and 45 teachers during the 
six-month intelligent system implementation phase to track 
adaptation patterns in the post-tutoring era. The 
multidimensional assessment framework incorporates the 
NASA Task Load Index adapted for e-learning environments, 
which demonstrates robust psychometric properties for 
measuring cognitive load across six dimensions, including 
mental demand, physical demand, temporal demand, 
performance, effort, and frustration levels. While the NASA-
TLX was adapted for digital display, its core items remained 
unchanged. Previous implementations in similar e-learning 
contexts reported strong reliability (α > 0.85) [25], 

supporting its use without redundant revalidation. Data 
collection occurred in three phases aligned with the academic 
calendar, capturing baseline measurements, mid-term 
adjustments, and end-of-year outcomes. Students completed 
cognitive load assessments immediately following digital 
learning sessions, ensuring ecological validity of self-reported 
measures. The protocol tracked engagement patterns 
averaging 4.37 hours per week of platform usage, revealing 
substantial variation across socioeconomic strata identified 
in the parent study. Semi-structured interviews with parents 
explored perceptions of their children's adaptation to 
reduced tutoring support, while educator observations 
documented classroom manifestations of cognitive load 
during technology-mediated instruction [26].  

 

Table 2. Participant demographics and data analysis methods 

Characteristic 
Students 
(n=320) 

Parents 
(n=50) 

Educators 
(n=20) 

Analysis 
Method 

Grade Level 
4-9 (M=6.7, 

SD=1.5) 
- - 

Descriptive 
statistics 

Socioeconomic 
Status 

Low: 31.6%, 
Middle: 

43.1%, High: 
25.3% 

Low: 
32.0%, 
Middle: 
42.0%, 
High: 

26.0% 

- 
Stratified 
analysis 

Prior Tutoring 
Participation 

73.4% 
(n=235) 

92.0% 
involved 

95.0% 
observed 

Chi-square 
test 

Cognitive Load 
Measurement 

NASA-TLX 
for e-

learning 
(n=312 

complete) 

Semi-
structured 
interviews 

Classroom 
observation 

protocol 

Mixed-
methods 
analysis 

Digital 
Learning 

Engagement 

4.37 
hrs/week 
(SD=1.42, 
range: 1.5-

8.2) 

- - 
Time-
series 

analysis 

Academic 
Performance 

Standardized 
test scores 
(pre/post) 

- - 
Repeated 
measures 

ANOVA 

Note: Data collected between September 2021 and June 2022, 
building upon the original Double Reduction Policy impact study. 
NASA-TLX = National Aeronautics and Space Administration Task 
Load Index, adapted for educational contexts to measure cognitive 
load in digital learning environments. 

Analytical procedures employ hierarchical linear 

modeling nested students within classrooms within schools, 

controlling for individual (prior performance, device access), 

classroom (technology infrastructure), and school-level 

(socioeconomic composition) variables. Propensity score 

matching balanced comparison groups, while sensitivity 

analyses confirmed robustness. The mixed-methods 

approach integrates quantitative metrics from standardized 

assessments with thematic analysis of qualitative data, 

enabling triangulation of cognitive load indicators. Machine 

learning algorithms process behavioral trace data to identify 

patterns predictive of cognitive overload, though human 

judgment remains central to intervention decisions. 

Statistical analyses control for prior tutoring participation 

rates and socioeconomic factors, ensuring that observed 

effects reflect genuine cognitive load variations rather than 

confounding variables inherent in the post-policy educational 

landscape. 

Given the sensitive nature of collecting cognitive and 
behavioral data from minors, ethical considerations were 
paramount throughout the study. Working with underage 
students meant taking extra precautions. The consent process 
involved parents first, then students separately. Data security 
went beyond basics with facial data processed and deleted 
within hours. Parents could review their child's participation 
anytime, while built-in alerts caught signs of academic or 
emotional strain. 

 
 

 



J. Zhai et al. /Future Technology                                                                                      November 2025| Volume 04 | Issue 04 | Pages 205-215 

209 

 

3. Results 

3.1 Identification of cognitive load characteristics in 
digital learning 
The identification of cognitive load characteristics 

among Dongcheng District students reveals distinct patterns 
that reflect the profound impact of transitioning from 
traditional tutoring-intensive education to technology-
mediated learning environments following the Double 
Reduction Policy implementation. As illustrated in Figure 2a, 
cognitive load distribution demonstrates a clear 
socioeconomic gradient, with students from lower 
socioeconomic backgrounds experiencing mean cognitive 
load scores of 68.4 (SE=2.39), significantly higher than their 
middle-class peers at 58.2 (SE=1.74) and high-income 
counterparts at 52.7 (SE=2.14). This disparity becomes 
particularly pronounced when students engage with digital 
learning platforms requiring simultaneous management of 
multiple information sources, interface navigation, and self-
regulated learning strategies previously scaffolded by 
external tutoring support. 

 

 

Figure 2. Cognitive Load Distribution Patterns in Dongcheng District 
Students. (a) Cognitive Load Distribution by Socioeconomic 
Background; (b) Individual Differences in Cognitive Load Perception 

 

Figure 2b shows that cognitive load perception varies 
significantly within groups, not just between socioeconomic 
categories. Individual differences in digital literacy, online 
experience, and metacognitive awareness create diverse 
cognitive load profiles within demographic groups. Students 
from lower socioeconomic backgrounds display the widest 
distribution range, suggesting that limited access to 
technological resources at home amplifies individual 
differences in adapting to digital learning environments. 
Median cognitive load scores in horizontal bars show that 
although statistically significant differences appear at the 
group level, there is considerable overlap between 
distributions, indicating the multifaceted nature of factors 
influencing cognitive load over and above economic status 
alone. Temporal comparison of variation of cognitive load 
over learning stages, shown in Figure 3, portrays a typical U-
shape pattern consistent with the skill acquisition theory in 
virtual learning systems. The early phase of learning shows 
the highest levels of cognitive load, reaching near 75, which 
indicates students' difficulties with the new content, 
concurrently adjusting to new computer interfaces as well as 
self-paced learning demands. This maximum level of 
cognitive load is consistent with the period shortly after the 
policy start, when students were deprived of formal tutoring 
support and had the twofold challenge of learning subject 
matter and learning to employ the technology. The gradual 
trend downwards during the acquisition of skills stage 
reflects productive construction of cognitive schemas and 
automation of procedural knowledge with steady levels of 
load averaging 62 weeks, 5-10. 

 

Figure 3. Temporal variations in cognitive load across learning 
phases during digital learning adaptation 

The mastery level illustrates the lowest values of 
cognitive load, around 50, that show a good balance of digital 
learning strategies and content knowledge structures. 
However, the transfer level (defined as the cognitive demands 
when applying learned concepts to novel contexts) reveals a 
different pattern. When students tried applying what they 
learned to new situations during the transfer phase, a notable 
shift in cognitive load occurred. This revealed just how 
dependent they had been on tutors walking them through 
difficult problems. The pattern specifically influences 
learners who have been used to repetitive drilling methods 
typical for classical tutoring, since they now have to build 
flexible problem-solving approaches without external 
guidance. The return of cognitive load during transfer tasks 
highlights the necessity for attaining adaptive expertise over 
routine proficiency in computer-based instruction. These 
patterns confirm what the theoretical framework predicted 
about transfer-level challenges in post-tutoring digital 



J. Zhai et al. /Future Technology                                                                                      November 2025| Volume 04 | Issue 04 | Pages 205-215 

210 

 

environments. Analysis of environment-specific factors for 
the digital environment identifies particular sources of 
cognitive challenge not found in the conventional classroom. 
Split-attention effects result as learners toggle between 
instructional video, digital textbook, and practice interface 
screens, generating extraneous cognitive load with a 
disproportionate impact on students with lower working 
memory capacity. The lack of timely instructor feedback and 
peer collaboration in asynchronous online courses adds more 
contextually relevant cognitive load to students because they 
need to actively build meaning via self-explaining and 
monitoring by metacognition. Machine learning algorithms 
monitoring students' patterns of interaction detect vital 
points of cognitive overload via rapid switching between 
resources, long pause times, and patterns of errors, providing 
intervention points for focused interventions. The results 
illustrate that cognitive load in post-policy e-learning 
environments is multidimensional in nature as it is affected 
by socioeconomic characteristics, time learning phases, and 
technology-specific requirements. Elucidation of such trends 
allows for the creation of intelligent support systems that 
modify instructional design dynamically according to real-
time cognitive state detection, with the ultimate aim of 
enabling more equitable learning outcomes within 
heterogeneous populations of students adjusting to policy 
change in education. 

3.2 Validation of the intelligent regulation mechanism 
effectiveness 
The verification of smart cognitive load control 

mechanisms has shown significant improvements in learning 
outcomes among Dongcheng District students transitioning 
to independent digital learning following the implementation 
of the Double Reduction Policy. As shown in Figure 4a, 
scholastic performance is significantly improved across all 
the socioeconomic cohorts after the introduction of 
individualized cognitive load control. The largest 
improvements were registered by lower socioeconomic 
status students, who had previously depended to a large 
extent on tutoring assistance, at 15.3 points, as opposed to 
10.4 and 8.7 points for middle-class and more affluent 
students, respectively. Informal feedback from parents 
consistently highlighted the financial relief of having 
affordable learning support, while teachers noted improved 
persistence among previously struggling students. As shown 
in Table 3, quantitative improvements align with qualitative 
feedback across key metrics. This differential improvement 
pattern suggests that intelligent regulation mechanisms 
particularly benefit learners who lost the most support under 
the new policy framework, thereby contributing to 
educational equity goals. 

Learning efficiency metrics, presented in Figure 4b, 
demonstrate the system's effectiveness in optimizing task 
completion across diverse learning activities. Problem-
solving tasks show the most dramatic improvement, with 
completion rates increasing from 52% to 78% while reducing 
time investment by 38%. Classroom observations revealed 
students spending significantly less time in unproductive 
struggle, with teachers commenting that the system's 
scaffolding appeared well-timed to maintain productive 
challenge without causing frustration. This enhancement 
proves particularly valuable for students transitioning from 
rote memorization approaches typical of traditional tutoring 
to more analytical thinking required in self-directed learning 
environments. Video learning and interactive tasks exhibit 
completion rate improvements exceeding 85%, indicating 

that the system successfully maintains student engagement 
across multiple content modalities without the external 
motivation previously provided by tutors. 

Table 3. Integration of quantitative and qualitative evidence 

 
 

The adaptive nature of the regulation system 
accommodates different learning styles effectively, as shown 
in Figure 4c. While all learner types experience cognitive load 
reductions exceeding 22%, satisfaction ratings reveal 
nuanced responses to system interventions. Most satisfied 
are auditory learners (4.5), possibly a spin-off from audio 
feedback facilities offered by the system, replacing the verbal 
instructions of teachers. Physically-based learners, in spite of 
all their big cognitive load savings from 75 to 58, are less 
satisfied (4.1), which indicates computer settings continue to 
pose a problem for physically-based learning modes. The 
results set out how multimodal support is imperative to 
overcoming varied learning requirements in online teaching 
contexts. 

System performance measures, as articulated in Table 4, 
confirm the technical reliability required to support mass-
scale education change. 87.3% accuracy in identifying 
cognitive load confirms safe detection of struggling student 
moments, and a 234-millisecond response time provides 
timely intervention before frustration or disengagement. The 
8.7% false positive rate is far below the industry benchmark, 
reducing interruptions to learning flow. These technological 
advancements become even more relevant when taking into 
account the 320 students impacted, 75% of whom had 
previously relied on outside tutoring to be supported 
academically.  

The application of intelligent regulation mechanisms 
holds transformation potential beyond the provision of 
improved performance delivery. Metacognitive awareness is 
cultivated in learners through system feedback, with the 
learners progressively internalizing self-regulation strategies 
formerly scaffolded by tutors. The 82.4% adaptation accuracy 
attests to the fact that personalized interventions are highly 
coupled to individual learning pathways, precipitating 
autonomous learning competencies essential to long-term 
academic achievement. The 99.2% system availability 
guarantees support continuity with little disruption, allaying 
fears of the effects of the digital divide on learning continuity. 

 
 

Quantitative Finding 
Qualitative 

Evidence 
Convergence 

Low-SES: +15.3 points 

Parents report 
financial relief 

from affordable 
AI support 

Strong 

Task completion: 
52%→78% 

Teachers observe 
reduced time in 

unproductive 
struggle 

Strong 

Response time: 234ms 

Students 
experience 

timely support 
before 

frustration 

Strong 

Load reduction: >22% 

Learners report 
decreased 

confusion and 
anxiety 

Moderate 



J. Zhai et al. /Future Technology                                                                                      November 2025| Volume 04 | Issue 04 | Pages 205-215 

211 

 

 

 

 
 
Figure 4. Effects of intelligent cognitive load regulation on learning 
Outcomes. (a) Academic achievement before and after regulation, (b) 
Learning efficiency and task completion analysis, (c) Cognitive load 
optimization effects on different learner types 

These validation findings demonstrate that intelligent 
cognitive load management indeed closes the support gap 
generated by policy-driven educational reform. The synthesis 
of enhanced academic performance, efficiency of learning, 
and strong technical performance forms the basis of 
deployability in larger student cohorts experiencing 
equivalent transitions from conventional tutoring-dependent 
paradigms to technology-facilitated self-study learning 
environments. This triangulation demonstrates how 
quantitative improvements translate to lived experiences—
efficiency gains reflect reduced frustration, accuracy metrics 

capture responsive support, and performance improvements 
embody renewed learning confidence. 

Table 4. System accuracy and responsiveness evaluation results 

Note: System evaluated using data from 320 students in Dongcheng 
District following Double Reduction Policy implementation, with 
particular focus on supporting students who previously relied on 
tutoring services (pre-policy participation rate: 75%). 

3.3 Evaluation of educational practice application 
effectiveness 
The analysis of pedagogical practice applications in 

Dongcheng District offers comprehensive insights into 
stakeholder adaptation amid the adoption of smart cognitive 
load management systems in the post-Double Reduction 
Policy era. The critique integrates feedback from 320 
students, 45 educators, and 280 parents who together 
experience the paradigm shift from tutoring-dependent 
learning to independent learning with the support of 
technology. 

As can be seen in Figure 5a, multi-stakeholder 
acceptance levels record steady enhancement during the six-
month implementation duration. Student acceptance went up 
from 65.3% to 82.4%, indicating a gradual adjustment to 
independent learning spaces once contained by extensive 
tutoring sessions. Teacher acceptance showed the greatest 
improvement, from 58.2% to 78.6%, despite initial resistance 
based on fears of technological incorporation undermining 
entrenched pedagogical traditions. Parents maintained the 
highest acceptance levels throughout, increasing from 72.4% 
to 85.2%, driven by relief at finding cost-effective alternatives 
to the expensive private tutoring services described in the 
foundational study. Longitudinal patterns of satisfaction, 
graphed in Figure 5b, reveal rich adaptation dynamics within 
stakeholder groups. The curve for teacher satisfaction shows 
maximum volatility, dipping to a low point of 2.8 during the 
third month before increasing to 4.1, with a showing of an 
episode of maximum adaptation where teachers grappled to 
balance algorithmic suggestion and professional intuition. 
This temporary slump is seconded by qualitative evidence 
from teacher interviews identifying issues in having 
confidence in automated detection systems for specific 
student needs previously covered by face-to-face tutoring 

Performance 
Metric 

Value 
Standard 
Deviation 

Benchmark 
Comparison 

Cognitive Load 
Detection Accuracy 

87.3% ±3.2% 
+12.5% vs. 

baseline 

Response Time 
(milliseconds) 

234 ±45 
-68% vs. 
manual 

False Positive Rate 8.7% ±2.1% 
Industry 

standard: 15% 

Adaptation 
Precision 

82.4% ±4.5% 
+18.3% vs. 

static 

User State 
Prediction F1-

Score 
0.856 ±0.034 

Above 0.8 
threshold 

System Uptime 99.2% ±0.3% 
Meets SLA 

requirements 



J. Zhai et al. /Future Technology                                                                                      November 2025| Volume 04 | Issue 04 | Pages 205-215 

212 

 

sessions. Student satisfaction demonstrates consistent 
improvement from 3.2 to 4.2, modest oscillation indicating 
the phase of transitioning from active recipient of tutoring to 
proactive self-regulated learner. 

 

 

 
Figure 5. Stakeholder acceptance and satisfaction analysis for an 
intelligent cognitive load regulation system (a) Multi-stakeholder 
system acceptance comparison, (b) Long-term usage satisfaction and 
feedback trends 
Note: Data collected from 320 students, 45 teachers, and 280 parents 
in Dongcheng District schools during the six-month implementation 
period following the Double Reduction Policy, which resulted in 
tutoring participation declining from 75% to 40%. Initial data (Month 
1) represents baseline measurements when families were adapting 
to reduced tutoring support, while current data (Month 6) reflects 
post-implementation outcomes with intelligent system support. 
Satisfaction ratings based on a 5-point Likert scale. Shaded areas in 
panel (b) represent 95% confidence intervals. 

System applicability analysis by varying teaching 
contexts, as depicted in Table 5, presents valuable insights 
into the differential effectiveness of the intelligent regulation 
mechanisms. Self-study review exercises have the highest 
ranking in effectiveness rating (9.1/10) with a rate of 
achievement of 91.2%, precisely meeting independent study 
skill development requirements of students lost through the 
provision of structured tutoring. Math problem-solving 
proficiency (8.7/10, 86.3% pass rate) is especially notable 
given the subject's prominence in the middle of Chinese 
academic examinations and previous intense emphasis on 

after-school supplementary instruction averaging 4-6 hours a 
week. 

Table 5. System applicability analysis in different teaching scenarios 

Note: Effectiveness scores based on performance metrics of 320 
Dongcheng District students transitioning from tutoring-dependent 
to self-directed learning. 

The significant performance difference between group 
and individual learning performance implies both promise 
and limitations of existing technology solutions. Exam 
preparation situation scenarios demonstrate superior 
performance (8.5/10, 84.7% pass rate), whereas 
collaborative learning demonstrates the worst performance 
(6.8/10, 68.4% pass rate), which mirrors the difficulty of 
simulating peer learning interactions that occurred 
organically within tutoring center settings. Science 
experiments share the same limitations (7.2/10, 72.8% 
success rate), which implies experiential learning 
components need creative solutions around existing system 
limitations to offset decreased hands-on direction. The 
careful assessment confirms that effective cognitive load 
management systems perfectly fill support gaps caused by 
policy-driven education transformation. Language learning 
exercises (7.9/10, 78.5% correct) are moderately successful 
with natural language processing support but cannot entirely 
substitute spontaneous corrective feedback that was always 
offered by human tutors. The evidence indicates that although 
technology-based solutions cannot substitute for 
individualized attention in conventional tutoring, they 
provide scalable, fair alternatives that foster key independent 
learning competencies. The upward trend of stakeholder 
satisfaction, with differentiated effectiveness for different 
teaching contexts, supports smart regulation systems as the 
path forward for sustainable education practices in the post-
policy era. 

4. Discussion  

This research contributes to cognitive load theory by 
showing its dynamic application in policy-reformed 
education technology-mediated learning environments. 
Adaptive regulation mechanisms expand Sweller's model 

Teaching 
Scenario 

Effectiveness 
Score 

Implementation 
Challenges 

Adaptation 
Requirements 

Success 
Rate 

Mathematics 
Problem 
Solving 

8.7/10 High 
computational 

load for complex 
problems 

Enhanced 
algorithm 

optimization 

86.3% 

Language 
Learning 

7.9/10 Nuanced feedback 
for writing tasks 

NLP model 
integration 

78.5% 

Science 
Experiments 

7.2/10 Limited hands-on 
simulation 

VR/AR 
component 

development 

72.8% 

Collaborative 
Projects 

6.8/10 Group dynamics 
complexity 

Multi-user state 
tracking 

68.4% 

Self-paced 
Review 

9.1/10 Minimal - well-
suited 

Minor UI 
adjustments 

91.2% 

Exam 
Preparation 

8.5/10 Stress factor 
consideration 

Anxiety 
detection 
module 

84.7% 



J. Zhai et al. /Future Technology                                                                                      November 2025| Volume 04 | Issue 04 | Pages 205-215 

213 

 

[15] from static instructional design to include real-time 
adaptive capabilities that adjust to learners' dynamically 
changing states of cognitive load. Traditional research has 
focused on pre-set sequencing of content, and this research 
divulges the processes of how machine learning algorithms 
most effectively redistribute load in real time, with detection 
accuracy 12.5% higher than traditional approaches. The 
results clarify how tailored regulation supports extensive 
learning through the sustenance of cognitive load in optimal 
ranges, avoiding frustration caused by overload and 
disengagement due to underload. Pedagogic innovations 
close structural cracks opened up by the Double Reduction 
Policy disruption of conventional tutoring circuits. Unlike 
studies of the incorporation of technology into ancillary tools 
[27], this study investigates environments where digital 
measures become vital supports to learning. The recorded 
teacher transformation from peddlers of knowledge to 
facilitators of learning reflects more profound pedagogic 
transformations than are normally reported on in studies of 
online education. A shift in acceptance from 58.2% to 78.6% 
by teachers within six months implies long-term professional 
growth can overcome adoption issues, as documented by 
Chen et al. [4], if evidence translates into concrete student 
gains for the post-tutoring learning environments. 

Differential learning performance in teaching 
environments tests the pedagogical suitability of technology. 
Self-directed activities with 91.2% success are set against 
68.4% effectiveness for collaborative projects, while 
confirming Kirschner and De Bruyckere's incredulity [28] 
regarding technology's incompetence to fully replicate social 
processes of learning. This differential is especially critical in 
post-tutoring environments where peer-to-peer interactions 
previously prevalent in tutoring facilities need to be recreated 
virtually. While people's task performance adheres to 
cognitive load theory principles for managing complexity, the 
conclusions highlight necessities for innovative solutions that 
facilitate collective knowledge construction under 
technological constraints. Ethical implications of ongoing 
cognitive monitoring range from privacy to learner agency 
and algorithmic control. Although Williamson [29] addresses 
surveillance capitalism in general, substituting AI for human 
tutors brings in distinctive ethical facets. The trade-off of 
assisting underprivileged students and eschewing 
technological dependence needs to be approached with 
sensitivity. Personalization based on data needs to 
acknowledge that learning entails affective, social, and 
creative aspects that are impenetrable to algorithmic 
simplification, especially when technology replaces human 
pedagogical relationships. 

Many limitations restrict generalizability. The one-
district urban sample can hardly represent China's diverse 
educational landscapes, particularly rural regions with poor 
infrastructure [30]. Expanding the system faces several 
challenges. The NLP components are trained on Chinese text, 
requiring a complete redesign for other languages. Rural 
schools with limited bandwidth (<10 Mbps) cannot support 
real-time features. Future work should develop offline-
capable versions while maintaining core functionality. The 
six-month timeframe cannot identify plateau effects or 
sustainability concerns raised by longitudinal research. 
Assessment of cognitive load may not capture sophisticated 
digital learning processes, especially metacognitive growth 
and transfer capabilities. These limitations necessitate 
caution in extrapolating findings to outside settings without 
considering local technological readiness and cultural 
learning cultures. 

Future work needs to conduct multi-site studies in 
varying contexts with external validity. Longitudinal studies 
following entire education cycles would determine whether 
technology-mediated regulation engenders true autonomy or 
new dependency. New technology integration could resolve 
existing collaborative learning bottlenecks. Studies need to 
investigate the wider implications of algorithmic educational 
support to make sure efficiency gains do not compromise 
humanistic values built into holistic education. Cross-cultural 
implementations need to be pursued as cognitive patterns 
will differ across education systems. The key question is 
whether smart systems can aid education equity in building 
twenty-first-century skills without undermining human 
factors that characterize unique learning experiences during 
the era of the digital economy. 

5. Conclusion  

This study demonstrates that intelligent cognitive load 
regulation mechanisms effectively optimize learning 
outcomes in digital environments following educational 
policy reforms. The research reveals significant 
improvements across multiple dimensions: students from 
lower socioeconomic backgrounds achieved 15.3-point gains 
in academic performance, task completion rates increased by 
32%, and cognitive load levels decreased by an average of 
23.1% across different learner types. The 87.3% detection 
accuracy and 234-millisecond response time validate the 
technical feasibility of real-time cognitive state monitoring in 
educational contexts. These findings extend cognitive load 
theory by incorporating dynamic adaptation capabilities 
essential for self-directed digital learning environments. The 
investigation contributes both theoretical insights and 
practical frameworks for educational technology 
implementation in post-tutoring contexts. The documented 
transformation of 320 Dongcheng District students from 
tutoring-dependent to self-regulated learners provides 
empirical evidence for technology-mediated educational 
equity. While limitations exist regarding single-district 
sampling and a six-month duration, the positive trajectory of 
stakeholder satisfaction (rising from 3.2 to 4.2 for students) 
suggests sustainable adoption potential. Future 
developments should focus on enhancing collaborative 
learning support, addressing the current 68.4% effectiveness 
rate, and expanding cross-regional validation. The 
convergence of cognitive science and educational technology 
demonstrated here offers promising pathways for scaling 
personalized learning support while maintaining pedagogical 
quality in the digital transformation of education.  

Ethical issue 
The authors are aware of and comply with best practices in 
publication ethics, specifically with regard to authorship 
(avoidance of guest authorship), dual submission, 
manipulation of figures, competing interests, and compliance 
with policies on research ethics. The authors adhere to 
publication requirements that the submitted work is original 
and has not been published elsewhere. 

Data availability statement 
The manuscript contains all the data. However, more data will 

be available upon request from the authors. 

Conflict of interest 

The authors declare no potential conflict of interest. 

 

 



J. Zhai et al. /Future Technology                                                                                      November 2025| Volume 04 | Issue 04 | Pages 205-215 

214 

 

References 

[1] A. Skulmowski and K. M. Xu, "Understanding 
cognitive load in digital and online learning: A new 
perspective on extraneous cognitive load," 
Educational psychology review, vol. 34, no. 1, pp. 171-
196, 2022. DOI: https://doi.org/10.1007/s10648-
021-09624-7 

[2]  A.-L. Le Cunff et al., "Cognitive load and 
neurodiversity in online education: a preliminary 
framework for educational research and policy," in 
Frontiers in Education, 2025, vol. 9: Frontiers Media 
SA, p. 1437673. DOI: 
https://doi.org/10.3389/feduc.2024.1437673 

[3] M. BĂNUȚ and D. ANDRONACHE, "Students' cognitive 
load in online education, under the lens of learning 
theories," Studia Universitatis Babes-Bolyai, 
Psychologia-Paedagogia, vol. 68, no. 2, 2023. 

[4] O. Chen and S. Kalyuga, "Exploring factors influencing 
the effectiveness of explicit instruction first and 
problem-solving first approaches," European Journal 
of Psychology of Education, vol. 35, no. 3, pp. 607-
624, 2020. DOI: https://doi.org/10.1007/s10212-
019-00445-5 

[5] R. E. Mayer, "Evidence-based principles for how to 
design effective instructional videos," Journal of 
Applied Research in Memory and Cognition, vol. 10, 
no. 2, pp. 229-240, 2021. DOI: 
https://doi.org/10.1016/j.jarmac.2021.03.007 

[6] F. Paas, A. Renkl, and J. Sweller, "Cognitive load 
theory and instructional design: Recent 
developments," Educational psychologist, vol. 38, no. 
1, pp. 1-4, 2003. DOI: 
https://doi.org/10.1207/S15326985EP3801_1 

[7] L. Chen, P. Chen, and Z. Lin, "Artificial intelligence in 
education: A review," IEEE access, vol. 8, pp. 75264-
75278, 2020. DOI: 
https://doi.org/10.1109/ACCESS.2020.2988510 

[8] R. F. Kizilcec and H. Lee, "Algorithmic fairness in 
education," in The ethics of artificial intelligence in 
education: Routledge, 2022, pp. 174-202. 

[9] R. Sajja, Y. Sermet, M. Cikmaz, D. Cwiertny, and I. 
Demir, "Artificial intelligence-enabled intelligent 
assistant for personalized and adaptive learning in 
higher education," Information, vol. 15, no. 10, p. 596, 
2024. DOI: https://doi.org/10.3390/info15100596 

[10] W. Strielkowski, V. Grebennikova, A. Lisovskiy, G. 
Rakhimova, and T. Vasileva, "AI‐driven adaptive 

learning for sustainable educational transformation," 
Sustainable Development, vol. 33, no. 2, pp. 1921-
1947, 2025. DOI: https://doi.org/10.1002/sd.3221 

[11] E. Gkintoni, H. Antonopoulou, A. Sortwell, and C. 
Halkiopoulos, "Challenging cognitive load theory: The 
role of educational neuroscience and artificial 
intelligence in redefining learning efficacy," Brain 
Sciences, vol. 15, no. 2, p. 203, 2025. DOI: 
https://doi.org/10.3390/brainsci15020203 

[12] C. Halkiopoulos and E. Gkintoni, "Leveraging AI in e-
learning: Personalized learning and adaptive 
assessment through cognitive neuropsychology—A 
systematic analysis," Electronics, vol. 13, no. 18, p. 
3762, 2024. DOI: 
https://doi.org/10.3390/electronics13183762 

[13] W. Holmes et al., "Ethics of AI in education: Towards a 
community-wide framework," International Journal 
of Artificial Intelligence in Education, vol. 32, no. 3, 

pp. 504-526, 2022. DOI: 
https://doi.org/10.1007/s40593-021-00239-1 

[14] K.-J. Laak and J. Aru, "AI and personalized learning: 
bridging the gap with modern educational goals," 
arXiv preprint arXiv:2404.02798, 2024. DOI:  
https://doi.org/10.48550/arXiv.2404.02798 

[15] J. Sweller, "Cognitive load theory and educational 
technology," Educational technology research and 
development, vol. 68, no. 1, pp. 1-16, 2020. DOI:  
https://doi.org/10.1007/s11423-019-09701-3 

[16] O. Zawacki-Richter, V. I. Marín, M. Bond, and F. 
Gouverneur, "Systematic review of research on 
artificial intelligence applications in higher 
education–where are the educators?," International 
journal of educational technology in higher education, 
vol. 16, no. 1, pp. 1-27, 2019. DOI: 
https://doi.org/10.1186/s41239-019-0171-0 

[17] B. Rienties and L. Toetenel, "The impact of learning 
design on student behaviour, satisfaction and 
performance: A cross-institutional comparison across 
151 modules," Computers in human behavior, vol. 60, 
pp. 333-341, 2016. DOI: 
https://doi.org/10.1016/j.chb.2016.02.074 

[18] R. Alfredo et al., "Human-centred learning analytics 
and AI in education: A systematic literature review," 
Computers and Education: Artificial Intelligence, vol. 
6, p. 100215, 2024. DOI: 
https://doi.org/10.1016/j.caeai.2024.100215 

[19] J. Sweller, J. J. Van Merriënboer, and F. Paas, 
"Cognitive architecture and instructional design: 20 
years later," Educational psychology review, vol. 31, 
no. 2, pp. 261-292, 2019. DOI: 
https://doi.org/10.1007/s10648-019-09465-5 

[20] J.-C. Hong, M.-Y. Hwang, K.-H. Tai, and C.-R. Tsai, "An 
exploration of students’ science learning interest 
related to their cognitive anxiety, cognitive load, self-
confidence and learning progress using inquiry-based 
learning with an iPad," Research in Science 
Education, vol. 47, no. 6, pp. 1193-1212, 2017. DOI: 
https://doi.org/10.1007/s11165-016-9541-y 

[21] R. E. Mayer, "The past, present, and future of the 
cognitive theory of multimedia learning," Educational 
Psychology Review, vol. 36, no. 1, p. 8, 2024. DOI: 
https://doi.org/10.1007/s10648-023-09842-1 

[22] C.-C. Lin, A. Y. Huang, and O. H. Lu, "Artificial 
intelligence in intelligent tutoring systems toward 
sustainable education: a systematic review," Smart 
learning environments, vol. 10, no. 1, p. 41, 2023. DOI: 
https://doi.org/10.1186/s40561-023-00260-y 

[23] T. Huang, J. Geng, H. Yang, S. Hu, Y. Chen, and J. Zhang, 
"Long short-term attentional neuro-cognitive 
diagnostic model for skill growth assessment in 
intelligent tutoring systems," Expert Systems with 
Applications, vol. 238, p. 122048, 2024. DOI:  
https://doi.org/10.1016/j.eswa.2023.122048 

[24] A. Létourneau, M. Deslandes Martineau, P. Charland, J. 
A. Karran, J. Boasen, and P. M. Léger, "A systematic 
review of AI-driven intelligent tutoring systems (ITS) 
in K-12 education," npj Science of Learning, vol. 10, 
no. 1, p. 29, 2025. DOI: 
https://doi.org/10.1038/s41539-025-00320-7 

[25] Y. Choi and H. Lee, "Psychometric properties for 
multidimensional cognitive load scale in an E-
learning environment," International Journal of 
Environmental Research and Public Health, vol. 19, 



J. Zhai et al. /Future Technology                                                                                      November 2025| Volume 04 | Issue 04 | Pages 205-215 

215 

 

no. 10, p. 5822, 2022. DOI: 
https://doi.org/10.3390/ijerph19105822 

[26] G. Gutierrez, I. O. Lunsky, S. Van Heer, A. Szulewski, 
and T. Chaplin, "Cognitive load theory in action: e-
learning modules improve performance in 
simulation-based education. A pilot study," Canadian 
journal of emergency medicine, vol. 25, no. 11, pp. 
893-901, 2023. DOI: 
https://doi.org/10.1007/s43678-023-00586-z 

[27] T. De Jong and A. Lazonder, "The guided discovery 
principle in multimedia learning," The Cambridge 
handbook of multimedia learning, vol. 2, pp. 371-390, 
2005. 

[28] P. A. Kirschner and J. J. Van Merriënboer, "Do learners 
really know best? Urban legends in education," 
Educational psychologist, vol. 48, no. 3, pp. 169-183, 
2013. DOI: 
https://doi.org/10.1080/00461520.2013.804395 

[29] B. Williamson, "Governing software: Networks, 
databases and algorithmic power in the digital 
governance of public education," Learning, Media and 
Technology, vol. 40, no. 1, pp. 83-105, 2015. DOI: 
https://doi.org/10.1080/17439884.2014.924527 

[30] H. Xie, H.-C. Chu, G.-J. Hwang, and C.-C. Wang, "Trends 
and development in technology-enhanced 
adaptive/personalized learning: A systematic review 
of journal publications from 2007 to 2017," 
Computers & Education, vol. 140, p. 103599, 2019. 
DOI: 
https://doi.org/10.1016/j.compedu.2019.103599 

 

  

This article is an open-access article distributed under the 

terms and conditions of the Creative Commons Attribution 

(CC BY) license 

(https://creativecommons.org/licenses/by/4.0/). 

https://creativecommons.org/licenses/by/4.0/

