Available via license: CC BY 4.0
Content may be subject to copyright.
Applied Mathematics and Nonlinear Sciences, 10(1) (2025) 1-17
Applied Mathematics and Nonlinear Sciences
https://www.sciendo.com
†Corresponding author.
Email address: CC06078845@163.com
ISSN 2444-8656
https://doi.org/10.2478/amns-2025-0411
© 2025 Chen Chen, published by Sciendo.
This work is licensed under the Creative Commons Attribution alone 4.0 License.
Research on the Construction of Smart Education Platform Driven by Artificial
Intelligence Technology in the New Era Higher Education Self-study Examination
Chen Chen1,†
1. The School of Visual Arts, Hunan Mass Media Vocational and Technical College, Changsha,
Hunan, 410100, China.
Submission Info
Communicated by Z. Sabir
Received October 11, 2024
Accepted February 7, 2025
Available online March 19, 2025
Abstract
In the context of the new era, the rapid development of Internet technology, artificial intelligence and other technologies,
the field of education will enter the “era of intelligent education”. The research centers on the self-study examination of
higher education, and builds an artificial intelligence-driven intelligent teaching platform, which integrates the functions
of intelligent management, intelligent examination, intelligent education, intelligent learning, intelligent evaluation,
intelligent service and so on. A feature extraction module is proposed based on convolutional neural network and
knowledge graph, which utilizes the powerful feature extraction capability of CNN to extract the implicit features of
courses and users, constructs a course resource recommendation model, and analyzes the application of artificial
intelligence in the smart teaching platform. The accuracy (0.857 and 0.791) and area under the ROC curve (0.876 and
0.804) results of the model on different datasets are better than those of the comparison model, reflecting better resource
recommendation performance. After using the smart education platform, more than 57% of learners have positive
attitudes towards the platform, their learning experience, and their learning outcomes. The smart education platform
constructed in the article can enhance learners’ interest, motivation, performance, and efficiency in learning, and help
them achieve intelligence and efficiency in self-study exam preparation.
Keywords: Artificial intelligence; Knowledge graph; Convolutional neural network; Resource recommendation;
Intelligent education platform.
AMS 2010 codes: 97B20
Chen Chen. Applied Mathematics and Nonlinear Sciences, 10(1) (2025) 1-17
2
1 Introduction
Self-study examination for higher education refers to a part-time higher education examination
system under the supervision of the Central Television University. The self-study examination is set
up for people who work, expand themselves and improve their qualifications. Its distinctive feature
lies in its high degree of autonomy and flexibility. Candidates can independently choose the majors
and courses according to their own time, ability and interest. This autonomy allows candidates to
better combine their own actual situation, develop a personalized learning plan, and give full play to
their personal strengths and potential. In addition, the college, undergraduate, master’s degree,
doctoral courses offered by the self-study examination are diverse, and the registration conditions
are relaxed, and the study time of the self-study examination is independently arranged, and there is
enough time to balance work and study, so the construction of the intelligent education platform of
artificial intelligence has an important role in the self-study examination [1-4].
Intelligent education platform is an important development trend in the field of education. It
provides more flexible and diversified learning methods for self-study candidates by utilizing the
Internet and intelligent technology. The smart education platform not only changes the traditional
learning mode, but also promotes the development of education equity and personalized education.
The intelligent education platform also provides rich learning resources and teaching tools, and
candidates can access paper and electronic books, course videos, online teaching materials, etc.
through the Internet. Compared with the traditional learning mode, the smart education platform
pays more attention to the personalized needs and interests of candidates, which is conducive to
improving the knowledge base of self-study candidates [5-8].
The higher education self-study examination is an important form of education in China’s higher
education system. As a flexible form of education, the Self-study Examination for Higher Education
(SSE) provides an independent and autonomous learning pathway for students in the workplace and
society who wish to further their studies. Literature [9] emphasizes the importance of self-study
examinations. The demand for self-study examination information is analyzed by applying the idea
of cloud computing, and a cloud-oriented information construction scheme for higher education
self-study examination, i.e., self-study examination as a service, is proposed. And benefit cloud
computing ideas on the architecture and application mode of cloud computing are discussed.
Literature [10] examined the self-study examination of art and design majors. It is emphasized that
in order to realize the cultivation goal of professional applied talents, it is very necessary to explore
the education mode combining self-study examination and skill training. Teaching reform was
elaborated based on changing teaching concepts and other aspects. This is not only in line with the
direction of cultivating skill-oriented talents in higher education, but also has special practical
significance. Literature [11] emphasized the promotion role of China’s higher education self-study
examination on education reform and economic discovery, based on the analysis of China’s higher
education talent cultivation mode, and put forward the coping strategies for its problems. Literature
[12] pointed out that in recent years, the self-study examination system has made proud
achievements, but with the introduction of various forms of higher education, the self-study
examination has suffered a heavy blow, and its problems have been highlighted, and only by solving
these problems can we realize the healthy and sustainable development of the self-study
examination. Literature [13] explored the operation of educational services of self-study higher
education examination. Based on questionnaires, interviews and other methods, it was concluded
that the educational objectives of the self-study higher education examination are more implicit, but
due to the poor applicability, low efficiency and other problems it is subjected to measures and
weakened. Literature [14] states that self-study higher education examination has been well
developed since its inception, but it faces many challenges in its further development. By analyzing
the challenges faced by the self-study examination, strategies are proposed to cope with them.
Research on the Construction of Smart Education Platform Driven by Artificial Intelligence Technology in the
New Era Higher Education Self-study Examination
3
Smart Education Platform is an education platform based on artificial intelligence technology. It
mainly analyzes students’ learning, behavior and performance to recommend teaching resources
suitable for each student, so as to improve students’ learning effect and teaching quality. Literature
[15] creates a new intelligent and efficient intelligent device classroom teaching model, which
combines the current situation and development of the intelligent device classroom and examines
the shortcomings of traditional college language teaching. Its findings provide reference for the
implementation of intelligent classroom teaching. Literature [16] introduces the AI-assisted
interactive smart education framework, which aims to improve the interaction between students in
higher education smart education. The framework provides students with reliable learning materials
and feedback systems to analyze the learning performance of smart education. The experimental
results point out that the proposed framework improves student-teacher interactions and shows high
accuracy in analyzing students’ academic performance skills. Literature [17] examined the design
and application of a digital intelligent teaching cloud platform based on AI algorithms. And through
case studies and data research, it was found that the intelligent teaching platform helps to improve
teaching efficiency and students’ learning experience, but there are also challenges such as data
privacy. Ideas are provided for the development of intelligent teaching cloud platform. Literature
[18] used interviews and case studies to examine the new ideas of AI in the elements of teaching
and learning activities, which lie in the promotion of the intelligent development of education. By
applying AI in physical education, personalized teaching was achieved by constructing Agent layer
and data service layer. Comparative experiments yielded that the system improved students’
physical education performance and sports interest. Literature [19] aims to develop a framework for
educational AI learning platforms and evaluate its applicability. One of the AI educational learning
platform frameworks was found to have particularly good applicability evaluation results. Literature
[20] built an AI-based personalized intelligent learning service platform that improves the efficiency
of users and provides open, on-demand education where knowledge can be accessed from resources
at any time, which is realized to promote the development of scholars.
Based on the self-study examination of higher education, the application architecture of the
intelligent education platform is proposed from the realization of functions such as management
services, examination and teaching affairs, and learning evaluation to realize the information
sharing of learning resources and improve the convenience of management and services. At the
same time, considering the integration of artificial intelligence technology into the platform
application, a course resource recommendation algorithm based on knowledge graph and
convolutional neural network is proposed, using the convolution-based feature extraction model
module to deeply mine the historical information of the interaction with the user, extracting feature
vectors representing the attributes of the course as well as feature vectors representing the attributes
of the user to complement the vectors of the original user as well as the course, and conducting
more angles of the Higher-order aggregation between the vectors, so as to obtain more accurate
embedded vector representations of users and courses. Different datasets and resource
recommendation models are selected for comparison experiments to explore the effect of course
resource recommendation on the proposed KGCNN model. Finally, a questionnaire survey on
learners’ feedback on the use of the smart education platform is conducted from the three
dimensions of platform satisfaction, learning attitude and learning effect, so as to test the actual
application effect of the smart education platform.
2 Construction of a smart education platform based on self-study exams
Self-study examination is a form of higher education that combines individual self-study with
national examination, mainly recruiting socially-typed candidates. Combined with the current status
quo of the informationization construction of higher education self-study examination, the
Chen Chen. Applied Mathematics and Nonlinear Sciences, 10(1) (2025) 1-17
4
intelligent education platform construction architecture scheme is proposed from the realization of
management services, examination and teaching affairs, learning and evaluation functions, etc. The
application architecture of the intelligent education platform is shown in Fig. 1, and the specific
functional modules of the platform are as follows.
Intelligent
education
examination
management
platform
(PC+APP+
wechat)
Intelligent
management Intelligent
evaluation
Intelligent
service Intelligent
learning
Intelligent
educational
administration
Intelligent
examination
Professional training
management
Curriculum management
Graduation practice
management
Graduation thesis defense
School records
management
Professional training
program inquiry
Professional course plan
inquiry
Register online
Admission ticket inquiry
Remote online
examination
Score inquiry
Online application for
exemption
Online transfer application
On-line examination
Test bank information
Intelligent test paper
management
Online intelligent patrol
exam
Identification and video
analysis
Application for exemption
Performance management
Office automation
Candidate information
management
Teacher information
management
Professional information
management
Curriculum information
management
Cloud computing
Big data
Internet of Things
AI technology
Artificial intelligence
Mobile Internet
technology
Application consultation
Policy publicity
Self-service payment
Certificate inquiry
File query
Data download
Questionnaire survey
Study group
Learning forum
Teaching management
Network education resource library
Network teaching platform
Cloud course
MOOC courses
Online course selection learning
platform
Teaching material information platform
Offline face-to-face course inquiry
Online autonomous learning
Offline face-to-face teaching
Learning evaluation
analysis system
Graduation review system
Figure 1. The application architecture of the intelligent education platform
2.1 Intelligent management
According to the work demand and workflow of the Education Examination Center, the
construction system includes office automation system module, candidate information management
system module, teacher information query management system module, professional information
query management system module, course information query management system module, etc., to
assist in the management of various types of information such as candidates, teachers, professions,
courses, etc., as well as the business coordination with the colleges and universities, learning centers,
study aids, etc., to achieve all kinds of office approval processes online to achieve “paperless”
office. It also realizes online processing of all kinds of office approval processes, so as to achieve
“paperless” office and improve the level and efficiency of the management of higher education
self-study examinations.
Research on the Construction of Smart Education Platform Driven by Artificial Intelligence Technology in the
New Era Higher Education Self-study Examination
5
2.2 Intelligent Examination
The platform is equipped with functions such as online application, pass inquiry, remote online
examination, result inquiry, online exemption application, online transfer application to meet the
needs of candidates, and at the same time, it builds an integrated intelligent examination
management system to adapt to the actual needs of the examination management and make the
management of higher education self-study examination more intelligent and efficient.
2.3 Intelligent Academic Affairs
Self-study exams are mainly self-study, the part is mainly to integrate resources for candidates to
provide professional, curriculum and other types of teaching information query services, so this part
of the function to achieve the focus is to facilitate the education and examination administrators to
provide candidates with a more professional through the construction of professional training
management, curriculum management, graduation practice management, graduation dissertation
defense management, management of student records as one of the teaching service system, It
facilitates educational examination administrators to provide candidates with more professional,
accurate and efficient academic information services.
2.4 Smart Learning
To provide self-study students with intelligent learning services mainly in the form of online
learning services, supplemented by offline face-to-face teaching, through the construction of
network teaching management, network education resource library, network selection of online
learning platforms, network teaching service platforms, cloud courses, MOOC courses, teaching
materials and teaching information service platform, offline face-to-face course information service
platforms and other multi-functional platforms for the integration of the intelligent learning and
education service system. The system integrates and optimizes the educational and teaching
resources of higher education institutions, student aid organizations and social education institutions,
expanding the learning channels of self-study students in terms of time and space, and making
learning more independent and efficient.
2.5 Intelligent evaluation
Provide managers and candidates with an evaluation system of candidates’ learning situation,
through linkage with several other platforms, real-time collection and acquisition of data on
candidates’ learning situation, practical internships, graduation theses, teachers’ evaluations, etc.,
combined with the professional training program, intelligently analyze the comprehensive situation
of candidates’ academics, timely analyze and discover potential problems of candidates’ academics,
accurately provide personalized learning programs and learning suggestions, and timely push the
proposed graduation candidates’ reports to the managers. It also sends the report of candidates to be
graduated to the administrators in time and handles them in a timely manner to ensure that the
candidates graduate on time.
2.6 Intelligent Services
As the basic module of the intelligent system, it mainly provides candidates with convenient
services such as application consultation, policy propaganda, self-service payment, pass query, file
query, data download, questionnaire survey, study group, study forum, and so on, and conveniently
Chen Chen. Applied Mathematics and Nonlinear Sciences, 10(1) (2025) 1-17
6
and effectively solves the practical needs encountered by self-testing students in the process of
examination and study by concentrating all kinds of services in the platform of online office hall.
3 Recommendations for artificial intelligence-driven platform course resources
The Higher Education Self-study Examination’s intelligent education platform uses artificial
intelligence technology to provide interactive and efficient education services. Through Internet
data statistics and intelligent analysis, the platform can grasp learners’ cognitive development trends,
knowledge application, and comprehension ability, and provide diagnostic evaluation. At the same
time, the platform can analyze the characteristics of learners and push relevant learning content and
programs to them, truly achieving tailor-made education from a technical perspective. This chapter
proposes a platform course recommendation algorithm (KGCNN) based on knowledge graph and
convolutional neural network, which better realizes course resource recommendations by mining
the deep information hidden in user interaction records.
3.1 KGCNN algorithm framework
The general framework of the KGCNN algorithm is shown in Figure 2. The algorithm uses the
feature extraction module to update the embedding vectors of the user and the course, then performs
message aggregation and link propagation along the links in the knowledge graph, and then uses the
embedding vectors of multiple layers to fuse to obtain higher-order information, and finally splices
the embedding vectors of the user and the course obtained from each layer to obtain the final
embedding vectors of the user and the course, and finally does the inner product of the embedding
vectors of the user and the course to get the recommendation result for the user to treat the
recommended course.
Floor 1
Floor 2
Floor 3
Information
aggregation
Feature extraction module
Floor 1
Floor 2
Floor 3
Information
aggregation
Curriculum
characteristics User
characteristics
++
Merge
Merge
Course
vector
User
vector
Fusion
course
vector
Fusion
user
vector
Splicing aggregate information
Splicing aggregate information
Forecast result
Figure 2. The overall framework of the KGCNN algorithm
Research on the Construction of Smart Education Platform Driven by Artificial Intelligence Technology in the
New Era Higher Education Self-study Examination
7
3.2 Convolutional Neural Networks
Convolutional Neural Network (CNN) is one of the most representative neural networks in the field
of deep learning. Convolutional Neural Network (CNN) uses a layered approach to extract features,
where each layer represents the dimensions of the tensor in a three-dimensional matrix (H, W, C),
which is called the feature map. Where H denotes the size or dimension of the tensor along the
vertical direction, W denotes the size or dimension of the tensor along the horizontal direction and
C denotes the number of feature channels contained in the tensor. The feature map is a
superposition of C tensors of size H × W, each representing the spatial distribution of one image
feature. The input is the feature map and the output vector to achieve classification. Each
component of the output vector corresponds to a feature type, indicating the probability that the
recognized object belongs to that feature.
Convolutional neural networks contain three main layer structures: convolution, pooling, and fully
connected. These layers are connected to each other through activation functions to construct
complex convolutional neural networks. The neuron layer in a convolutional neural network
consists of neurons in three dimensions, i.e., the spatial dimensions of the input: height, width, and
depth.
1) Convolutional Layer
The convolutional layer is one of the core elements in a convolutional neural network and is
used to extract features from the input data. Strictly speaking, the convolutional layer is a
misnomer, as the operations it expresses are actually inter-correlation operations, relying on
dot-multiplication of matrices for summation rather than convolutional operations.
The parameters of the convolutional layer include, given an input image
I
and a
convolutional kernel
K
, a boundary padding size, a step size, and assuming that the input
image has a size of
in in in
W H D
and the size of the convolutional kernel is
w h in out
k k D D
(where
K
is the size of the convolutional kernel,
in
D
is the depth of
the input image, and
out
D
is the depth of the output feature map), the output feature map
o
has a size of
out out out
W H D
:
1 1 1
, , , , , , ,
0 0 0
w h in
k k D
i j k i m j n l m n l k
m n l
O I K
− − −
++
= = =
=
(1)
Where
,,i j k
O
is the value of the
,ij
rd position of the
k
nd channel of the output feature map,
,,i m j n l
I++
is the value of the
,i m j n++
th position of the
l
th channel of the input image and
, , ,
m
K n t k
is the weight of the
,mn
th position of the
k
th channel of the convolution kernel.
Filling is a way to deal with the problem of losing edge pixels when applying multilayer
convolution, mainly by performing a complementary zero operation on the image edges. When this
paper is supplemented with
h
p
rows and
w
p
columns, the output shape will be:
( 1) ( 1)
in w w in h h
W k p H k p− + − − + −
(2)
Further, in calculating the inter-correlation, the convolution window starts from the upper left
corner and slides down and to the right, by default sliding one element at a time. By setting the step
Chen Chen. Applied Mathematics and Nonlinear Sciences, 10(1) (2025) 1-17
8
size to slide multiple elements at a time, efficient computation and reduced sampling can be
achieved. For vertical step
h
s
and horizontal step
w
s
, the output shape is:
[( )/ ] [( ) / ]
in w w h h in w w w w
W k p s s W k p s s− + + − + +
(3)
2) Pooling layer
The pooling layer is a kind of downsampling layer in the convolutional neural network used
for downsampling and feature compression of the input data. The pooling layer extracts the
main features in the feature map by spatially downsampling the feature map output from the
convolutional layer, reduces the number of parameters in the model to reduce the
computational cost and memory consumption, and helps the subsequent layers of the
network to learn and represent the data features more effectively.
Common types of pooling include maximum pooling and average pooling. Maximum
pooling operation is to select the maximum value as the output within a given pooling
window. While average pooling operation is to calculate the average value within the
pooling window and use it as the output.
3) Fully connected layer
Each neuron in the fully connected layer is connected to all the neurons in the previous layer,
and each connection has a weight parameter, so the number of parameters in the fully
connected layer is large. The fully connected layer is located in the last layers of the model
for performing classification or regression tasks, and its role is to flatten the features
extracted from the previous convolutional layer or other feature extraction layers and weight
and sum all the feature connections, and then pass them to the Softmax classifier.
4) Softmax layer
In deep learning, the Softmax layer is characterized by mapping each element in the input
vector to a probability value such that the sum of all probability values is 1. It is usually used
as the output layer of a neural network model for multiple classification tasks. It is able to
output the probability distribution for each category thus helping to determine the final
classification result.
The Softmax layer formula is as follows:
1
j
j
n
kn
j
e
softmax
e
=
=
(4)
where
j
n
denotes the
j
nd element of the original output vector and
K
is the total number of
categories.
3.3 Convolution-based feature extraction module
In the course recommendation domain courses can be composed of segmented attributes such as
videos owned by the course, knowledge points related to the course, the domain to which the course
belongs, the publisher of the course, and the main instructor of the course. In order to better mine
Research on the Construction of Smart Education Platform Driven by Artificial Intelligence Technology in the
New Era Higher Education Self-study Examination
9
the deeper user and course feature information hidden under the record of user-course interaction,
this paper proposes a feature extraction module based on convolutional neural network, which is
designed to extract course and user features.
In this paper, the dimension of the input vector is 64, and the first layer of the module is a
convolutional activation layer with 1 input channel and 3 output channels, and each convolutional
layer is followed by an activation layer, and in this paper, the ReLU function is used as the
activation function. A maximum pooling layer is used after the first convolutional activation layer
and the sliding window size is set to 2. Immediately after this, the module uses a second
convolutional layer with a number of input channels of 3 and a number of output channels of 3. A
pyramid pooling layer is added before the fully-connected layer as the total number of videos,
knowledge points, and domains owned by different courses is not exactly the same. This extracts
the features of the embedding vectors from different perspectives while unifying the output sizes,
and finally the features output from the pyramid pooling layer are mapped using the fully connected
layer to obtain the final embedding vector representations of the courses and users.
3.4 KGCNN aggregation layer
Using the interaction graph between users and courses, we can obtain higher-order information
between users and other courses, and use this higher-order information to model the embedding
vectors of users and courses more effectively. Considering the following cases, user
1
u
has studied
course
0
i
and user
2
u
has studied
0
i
and
1
i
, although
1
u
has not studied course
1
i
but
2
u
who has similar interest with user
1
u
is interested in course
1
i
, some features of course
1
i
can be
considered as contributing to the modeling of user
1
u
preference. Therefore, inspired by the KGAT
algorithm, the information aggregation layer used in this paper uses a fusion of multiple aggregation
approaches. Define the set of embedding vectors for users as
u
e
, the set of embedding vectors for
courses as
i
e
, and
e
p
as the splicing of user vectors and course vectors with the following
formula:
e u i
p e e=
(5)
where II is the splice-by-row operation. Definition
n
e
p
is the layer
n
aggregation vector of the
information aggregation layer, where only neighboring nodes are aggregated at each aggregation,
where the information aggregation formula is as follows:
1 1 1 1
( ) ( )()
n n n n n
e relu u e s u e s
p f f p p f p p
− − − −
= + +
(6)
where is the inner product operation,
()
relu
f
is a nonlinear function, and
1n
e
p−
and
1n
s
p−
are
related layer
1n−
aggregation vectors, where
()
u
f
is formulated as follows:
()
u
f x wx b=+
(7)
where
w
is the weight parameter and
b
is the corresponding bias. The corresponding
s
n
p
equation is as follows:
*
nn
se
p A p=
(8)
Chen Chen. Applied Mathematics and Nonlinear Sciences, 10(1) (2025) 1-17
10
Where
A
is the record of the user’s interaction with the course, the fusion of information between
different entities is achieved by left-multiplying the interaction matrix
A
by the entity vector
n
e
p
,
while
0
e
p
is the corresponding initial
e
p
vector, which is formulated as follows:
0(( ) ( ))
cnn cnn
e u u i i
p e e e e
= + +
(9)
Where
cmn
u
e
and
cmn
i
e
are the feature vector representations extracted by the CNN-based feature
extraction module for users and courses, respectively, and
and
are the corresponding
weight parameters, both of which are 0.4 in this paper’s algorithm
and
.
3.5 KGCNN Loss Function
BPR loss is a loss function used in recommender systems, which is based on the Bayesian
Personalized Ranking (BPR) model. The BPR loss is used in KGCNN algorithm as the optimization
objective of the algorithm, which is formulated as follows:
2
( , , )
ˆ ˆ
()
BPR ui uj
u i j O
Loss ln y y
= − − +
(10)
where
( , , )u i j
is two sets of training data,
i
is the course that user
u
has actually interacted
with,
j
is the course that user has not interacted with,
ˆui
y
is the corresponding user
u
‘s
prediction score for course
i
, and
2
|| ||
is the regularization term, which generally prevents the
model from overfitting.
3.6 Experimentation and Analysis
3.6.1 Data sets
Deep learning algorithms require more data than traditional machine learning algorithms. In order
to meet this demand, this chapter adopts the MOOCCube and Criteo datasets, which are derived
from the open datasets of the Schoolhouse Online website.
MOOCCube is a free data warehouse that provides access to information about 706 courses and
hundreds of thousands of course selection records from 190,000 users in online education.
According to the description on the official website of MOOCCube dataset, the dataset can be used
in several MOOC-related research areas such as course recommendation, student behavior
prediction, and course concept extraction. In this chapter, the CourseRecommend dataset, obtained
after appropriate processing of the MOOCCube dataset, is used as the data support for course
recommendations. The ratio of the divided training set, test set, and validation set is 8:1:1.
Criteo, a digital company specializing in online performance marketing, makes publicly available
the Criteo dataset, which is widely used to evaluate the performance of recommender systems. In
this experiment, 1 million pieces of data were randomly selected from the huge Criteo dataset and
rigorously divided into 10% for the test set, 10% for the validation set, and 80% for the training set.
Such a division method can effectively simulate actual application scenarios, and model training
and performance evaluation can be carried out on this basis.
Research on the Construction of Smart Education Platform Driven by Artificial Intelligence Technology in the
New Era Higher Education Self-study Examination
11
3.6.2 Contrasting models
In order to evaluate the performance of the KGCNN model proposed in this paper, it is validated
using comparative experiments using the following comparative models: the Wide&Deep model,
the DeepFM model, the DCN model, and the XDeepFM model.
3.6.3 Analysis of experimental results
For the course resource recommendation experiments on the smart education platform, this paper
selects the area under the ROC curve and the accuracy rate to evaluate the experimental results. Fig.
3 and Fig. 4 show the accuracy and loss function changes of the KGCNN model after 200 rounds of
training in the CourseRecommend dataset, respectively, and the number of horizontal coordinate
rounds in the figure represents the number of rounds of training. As the number of rounds increases,
the accuracy of the training and validation sets is increasing, the loss function is decreasing, and the
variation decreases after about 50 rounds of training, and the overall phenomenon of convergence is
shown.
Figure 3. The accuracy of the KGCNN model changes
Figure 4. The loss of the KGCNN model changes
0
20
40
60
80
100
120
140
160
180
200
0.0
0.2
0.4
0.6
0.8
1.0
Training set
Test set
Accuracy
Epoches
Training set Test set
200
180
160
140
120
100
80
60
40
20
0
0.0
0.2
0.4
0.6
0.8
1.0
Training set
Test set
Loss
Epoches
Training set
Test set
Chen Chen. Applied Mathematics and Nonlinear Sciences, 10(1) (2025) 1-17
12
Figure 5 shows the accuracy results for each model on the CourseRecommend dataset and Criteo
dataset, and Figure 6 shows the area under the ROC curve results for each model on the
CourseRecommend dataset and Criteo dataset. The Wide&Deep model, which is the base model, is
generally lower in terms of performance than the other models. Whereas, the XDeepFM model and
DCN model, which use higher-order display feature interactions, outperform the DeepFM model
and Wide&Deep model, which do not use higher-order display feature interactions. The accuracy of
the KGCNN model proposed in this chapter on CourseRecommend dataset and Criteo dataset is
0.857 and 0.791, and the area under the ROC curve is 0.876 and 0.804, respectively, which have the
optimal accuracy and area under the ROC curve in the experimental results. The KGCNN model in
this paper has better course resource recommendation performance in the smart education platform
based on self-study exam, and it can recommend learning resources for higher education self-study
exam students in line with their preferences.
Figure 5. Comparison of the accuracy of different models
Figure 6. Comparison of the ROC curve of different models
4 Evaluation and analysis of the Intelligent Education Platform
The main purpose of this study is to build a smart education platform for self-study exams in higher
education. After the design and realization of the functional requirements for self-study exams are
completed, the smart education platform has to test the application effect of the system and modify
the problems of the platform according to the test results to further improve the system. In order to
Wide&Deep
DeepFM
DCN
XDeepFM
KGCNN
0.72
0.76
0.80
0.84
0.88
CourseRecommend
Criteo
Accuracy
Models
CourseRecommend Criteo
Wide&Deep
DeepFM
DCN
XDeepFM
KGCNN
0.70
0.72
0.74
0.76
0.78
0.80
0.82
0.84
0.86
0.88
CourseRecommend
Criteo
The area under the ROC curve
Models
CourseRecommend Criteo
Research on the Construction of Smart Education Platform Driven by Artificial Intelligence Technology in the
New Era Higher Education Self-study Examination
13
verify whether the intelligent education platform improves the learning efficiency of learners and
recommends appropriate course learning resources according to the needs of learners to a certain
extent, the application effect of the platform is verified through the questionnaire survey method.
4.1 Questionnaires
The target of this questionnaire was selected from freshman to sophomore students of University of
M. Feedback was gathered from learners after using the platform. The questionnaire on the use
effect of the smart education platform was designed to evaluate the system evaluation indexes from
the three dimensions of satisfaction with the platform, learning attitude, and learning results. The
total number of questionnaires distributed was 194 and only 182 were effectively collected.
4.2 Analysis of application effects
4.2.1 Analysis of platform satisfaction
In the questionnaire on the use effect of the smart education platform, questions 1, 2 and 3 of the
survey questions analyze the satisfaction survey of the self-study exam learners with the platform,
mainly from the three aspects of whether the learners support the behavioral data to be accessed and
used, whether the platform’s function is designed to be comprehensive, and whether the
recommended resources are in line with the personalized needs. The results of the satisfaction
survey of the platform are shown in Figure 7, with A~E corresponding to very compliant, relatively
compliant, generally compliant, not very compliant, and very non-compliant (same as below).
23.1% of learners strongly support the use of learner behavioral data, 38.9% support the use of
behavioral data, and the majority of learners support the use of behavioral data, to varying degrees,
to analyze learners’ personalized characteristics and recommend learning resources for them.25.1%
of learners believe that the platform functionality is very comprehensive, and 32.8% believe that the
system’s resources are more comprehensive, and, in general. Overall, most learners think that the
platform functions in the system are comprehensively designed, but a few learners think that more
can be added. Meanwhile, 30.9% and 35.4% of the learners think that the course resources
recommendation meets the needs to different degrees, indicating that most learners think that the
course resources recommendation function of the platform can basically meet the personalized
needs of the learners, and has a good recommendation effect. In general, most learners acknowledge
the satisfaction of the smart education platform to varying degrees, and the learner satisfaction
survey analysis indicates that the smart education platform has certain effectiveness.
Chen Chen. Applied Mathematics and Nonlinear Sciences, 10(1) (2025) 1-17
14
Figure 7. Results of platform satisfaction survey
4.2.2 Analysis of learning attitudes
The 4th, 5th, and 6th questions in the questionnaire survey analyze the learners’ learning attitudes
after having used the Smart Education Platform, mainly from the three aspects of whether the
system is a powerful tool to assist self-study exams, whether the system can improve learning
interest, and whether the system improves learning motivation.
The results of the survey on learning attitudes after using the platform are shown in Figure 8. 26.2%
of the learners and 35.4% of the learners thought that the smart education platform was a better tool
to assist self-study exams to different degrees, and overall the platform could help learners prepare
for self-study exams to a certain extent. 32.4% of the learners strongly agreed that the platform
could increase learners’ interest in learning, and 38.2% of the learners think that the platform can
improve learning interest, and there are a few learners who think that the platform can’t help
learners to improve their learning interest. 22.6% and 39.3% of learners agree to varying degrees
that the platform can improve learning motivation, and in general, the intelligent education platform
has a good and positive effect on self-study exams.
Figure 8. The results of the study attitude survey after the platform use
A
B
C
D
E
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
Support for behavioral data
Full functional design
Reasonable recommendation
Evaluation result
Proportion
Support for behavioral data
Full functional design
Reasonable recommendation
A
B
C
D
E
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
Assisted self-study examination
Improve learning nterest
Improve learning initiative
Evaluation result
Proportion
Assisted self-study examination
Improve learning nterest
Improve learning initiative
Research on the Construction of Smart Education Platform Driven by Artificial Intelligence Technology in the
New Era Higher Education Self-study Examination
15
4.2.3 Analysis of learning outcomes
Questions 7 and 8 of the questionnaire are entitled Analyzing Learners’ Learning Effectiveness
Survey of the Smart Education Platform, which is mainly analyzed in terms of whether the platform
can help learners improve their grades and whether the platform can improve their learning
efficiency.
The results of the survey on the learning effect of the platform after its use are shown in Figure 9.
The survey results show that 31.1% of the learners and 38.3% of the learners think that the platform
can help them improve their academic performance to different degrees. The platform’s
effectiveness in improving learning efficiency was recognized by 64.0% of learners. A small group
of learners believe that the platform has no impact on the learning of self-study exams. Overall,
learners recognize that the intelligent education platform has a certain role in helping learners
prepare for self-study exams.
Figure 9. The results of the study effect of the platform
5 Conclusion
With the progress of science and technology, utilizing intelligent teaching platforms for learning is
an inevitable trend in the process of students’ self-study examination. Based on the self-study
examination of higher education, the study constructs a smart education platform that integrates the
functions of management, learning, evaluation, and service. Combined with knowledge graph and
convolutional neural network, the course resource recommendation model available on the smart
education platform is proposed. Finally, the smart education platform is evaluated through a
questionnaire survey.
The study constructs a course resource recommendation model to analyze the use of artificial
intelligence in the smart education platform. Through experiments, it is found that the KGCNN
model proposed in this paper has good recommendation accuracy and convergence effect, and the
accuracy and area under the ROC curve are higher than the comparison model on
CourseRecommend dataset and Criteo dataset, in which the accuracy is 0.857 and 0.791, and the
area under the ROC curve is 0.876 and 0.804, respectively.
A
B
C
D
E
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
Improve grades
Improve learning efficiency
Evaluation result
Proportion
Improve grades
Improve learning efficiency
Chen Chen. Applied Mathematics and Nonlinear Sciences, 10(1) (2025) 1-17
16
The smart education platform passed validation of effectiveness, and over 57% of learners had a
positive attitude towards the platform in the three dimensions of satisfaction, learning attitude, and
learning effect. After applying the wisdom education platform for self-study exam learning, learners’
interest in learning, learning motivation, learning performance and learning efficiency are all
improved, and they think that the platform’s course resource recommendation is more reasonable
and is a better self-study exam support tool.
The establishment of an open and flexible learning platform for self-study exams with strong
applicability is not only to meet the learning demands of the candidates and adapt to the needs of
education informatization, but also the inherent needs of the reform and development of the
self-study exam system itself. The application of the intelligent education platform to the self-study
examination has changed the traditional self-study examination learning mode, and the diversified
learning resources can mobilize students’ enthusiasm and make the self-study examination learning
more efficient.
Funding:
This research was supported by the 2023 Higher Education Scientific Research Planning Project of
the Chinese Association of Higher Education: Research on the Construction of Smart Educational
Resources for Self-study Examinations in Higher Education in the New Era (No.: 23ZXKS0404).
References
[1] Li, R., Zheng, H., Guo, H., & Zhang, L. (2024, February). Research on Academic Evaluation Methods
of Self study Exams from the Perspective of Military Vocational Education in the New Era. In
Proceedings of the 3rd International Conference on Education, Language and Art (ICELA 2023) (Vol.
831, p. 210). Springer Nature.
[2] Chawinga, W. D., & Zozie, P. A. (2016). Increasing access to higher education through open and
distance learning: Empirical findings from Mzuzu University, Malawi. International Review of
Research in Open and Distributed Learning, 17(4), 1-20.
[3] Moskal, P. D., Dziuban, C., & Hartman, J. (2010). Online learning: A transforming educational
environment for adults in higher education. In Web-based education: concepts, methodologies, tools and
applications (pp. 852-866). IGI Global.
[4] Zhao, Y. B. (2012). Analysis on Factors Influencing Sustainable Development of Self-study
Examination of Agricultural Science Disciplines. Asian Agricultural Research, 4(1), 74-77.
[5] Díaz-Parra, O., Fuentes-Penna, A., Barrera-Cámara, R. A., Trejo-Macotela, F. R., Ramos-Fernández, J.
C., Ruiz-Vanoye, J. A., ... & Rodríguez-Flores, J. (2022). Smart Education and future trends.
International journal of combinatorial optimization problems and informatics, 13(1), 65.
[6] Jangjarat, K., Klayklung, P., Chocksathaporn, P., & Maskran, P. (2023). The impact of smart education
on learning outcomes in the digital era: A systematic review. Advance Knowledge for Executives, 2(2),
1-9.
[7] Simić, K., Despotović-Zrakić, M., Bojović, Ž., Jovanić, B., & Knežević, Đ. (2016). A platform for a
smart learning environment. Facta universitatis-series: Electronics and Energetics, 29(3), 407-417.
[8] Palanivel, K. (2020). Emerging technologies to smart education. Int. J. Comput. Trends Technol, 68(2),
5-16.
[9] Lei, D. Y. (2013, June). A Cloud-Vased Self-Study Examination Information Construction Program. In
2013 International Conference on Computational and Information Sciences (pp. 532-534). IEEE.
Research on the Construction of Smart Education Platform Driven by Artificial Intelligence Technology in the
New Era Higher Education Self-study Examination
17
[10] Zhang, J., & Lu, J. (2014, May). Study on the Teaching Mode of Combining Self-study Examination
with Skill Training for Art Design. In International Conference on Education, Language, Art and
Intercultural Communication (ICELAIC-14) (pp. 87-91). Atlantis Press.
[11] Chen, D. (2011). Practices and Explorations on Talent-Training Mode of Self-Taught Higher Education
Examination. International Education Studies, 4(3), 36-40.
[12] Xu, Q. (2020). On the Current Status and Future of Self-Taught Higher Education Examination. Journal
of Contemporary Educational Research, 4(11).
[13] Gao, G. (2003). The Educational Administration of the Self-Taught Higher Education Examination in
China: Findings from Field Research. Higher Education Policy, 16, 463-477.
[14] Ni, C. (2019, August). Research on Scale Retrogression and Development Countermeasures of
Self-study Examination in China—Taking the Jiangsu Province as an Example. In 2019 International
Conference on Education Innovation and Economic Management (ICEIEM 2019) (pp. 112-116).
Atlantis Press.
[15] Qi, Y., & Huang, Y. (2022). A chinese intelligent teaching platform for colleges based on cloud
computing. Mobile information systems, 2022(1), 3487248.
[16] Zhang, Y., Qin, G., Cheng, L., Marimuthu, K., & Kumar, B. S. (2021). Interactive Smart Educational
System Using AI for Students in the Higher Education Platform. Journal of Multiple-Valued Logic &
Soft Computing, 36.
[17] Wang, D. (2024, August). Research on Smart Teaching Cloud Platform Based on Artificial Intelligence
Algorithms. In Proceedings of the 2024 4th International Conference on Internet of Things and Machine
Learning (pp. 396-402).
[18] Li, F. (2021, April). Information teaching platform of college physical education based on artificial
intelligence technology. In Journal of Physics: Conference Series (Vol. 1852, No. 2, p. 022030). IOP
Publishing.
[19] Thongprasit, J., & Wannapiroon, P. (2022). Framework of Artificial Intelligence Learning Platform for
Education. International Education Studies, 15(1), 76-86.
[20] Zhao, B. (2023, September). Construction of Personalized Intelligent Learning Service Platform for
Universities Based on Artificial Intelligence. In 2023 International Conference on Industrial IoT, Big
Data and Supply Chain (IIoTBDSC) (pp. 389-394). IEEE.
