Available via license: CC BY 4.0
Content may be subject to copyright.
Applied Mathematics and Nonlinear Sciences, 9(1) (2024) 1-20
Applied Mathematics and Nonlinear Sciences
https://www.sciendo.com
†Corresponding author.
Email address: luochong2024@163.com
ISSN 2444-8656
https://doi.org/10.2478/amns-2024-1569
© 2023 Chong Luo, published by Sciendo.
This work is licensed under the Creative Commons Attribution alone 4.0 License.
Analysis of Literature Education Strategies and Learning Behaviors in the Age of Big
Data
Chong Luo1,†
1. Henan Geology Mineral College, Zhengzhou, Henan, 450000, China.
Submission Info
Communicated by Z. Sabir
Received February 19, 2024
Accepted May 14, 2024
Available online July 2, 2024
Abstract
In the contemporary era of big data, the educational landscape is undergoing significant transformations. Literature
education, as a vital component, faces both emerging opportunities and challenges. This study develops a framework for
analyzing learning behaviors specific to literature education. It employs both an enhanced K-means clustering algorithm
and a refined Apriori algorithm for mining data on student learning behaviors. Through cluster analysis and the
investigation of association rules, this research explores the interconnections between students’ learning behaviors and
their literary education. The findings categorize students into four distinct groups based on their learning behaviors.
Students in Category 1 are identified as the most proficient learners, consistently achieving test scores above 85.
Conversely, Category 2 students display the least motivation and effectiveness, with their examination scores not
exceeding 70. Students in Categories 3 and 4 exhibit comparable levels of performance. Crucially, the analysis reveals
that the most significant predictors of students’ literary achievement are their regular and examination scores, with
correlation coefficients of 0.627 and 0.653, respectively. This segmentation and analysis of student behaviors facilitate
the early detection of atypical learning patterns by educational practitioners, enabling timely intervention strategies to
enhance academic outcomes.
Keywords: K-means; Apriori; Learning behavior; Literature education.
AMS 2010 codes: 97P20
Chong Luo. Applied Mathematics and Nonlinear Sciences, 9(1) (2024) 1-20
2
1 Introduction
The Chinese language and literature education orientation program occupies an important position in
the Chinese education system, which is not only a key link in cultivating students’ language ability
but also carries the mission of inheriting and promoting Chinese culture [1]. With the continuous
development and change of society, the Chinese language and literature education orientation
program also needs to be constantly adjusted and reformed to meet the needs of the new era [2].
With the rapid development of information technology and the wide application of the Internet, we
have stepped into a brand new era - the “Internet + era” [3]. This era has not only changed people’s
way of life but also profoundly affected the field of education, especially in the development and
utilization of educational resources of ancient Chinese literature. Ancient Chinese literature, as an
important part of the long-lasting culture of the Chinese nation, carries rich historical information and
cultural value and is an indispensable part of the inheritance and development of Chinese culture [4-
5].
In the context of the Internet + era, how to effectively develop and integrate these valuable literary
resources is not only the key to improving the quality of education, but also an important way of
cultural inheritance [6]. The Internet+ era provides new opportunities for the development and
application of educational resources of ancient Chinese literature. The application of Internet
technology makes educational resources more abundant and diverse, learning methods more flexible
and convenient, and the time and space boundaries of education can be broken [7]. However, this
change also brings a lot of challenges. How to effectively integrate and utilize these resources while
maintaining the essence of literary resources has become a problem that educators and researchers
must face [8].
Although the Internet+ era has brought new opportunities for ancient Chinese literature education, it
is also accompanied by many challenges. Developing and integrating high-quality educational
resources, as well as innovating educational methods and means, have become the main tasks faced
by educators at present. Truman, S. E. combines the influence theory, critical race academic research
methodology, and research results on whiteness, pointing out that although there are diversities in
literary education in English literature, mainstream literary education still focuses on white literature
[9]. Asonova E et al. explored and analyzed the concept and framework of literary education in depth,
as well as explored relevant theories and information about teaching and learning about literature,
concluding that higher education literature education and library services have a very limited role to
play in the improvement of literary cognition, and arguing that other avenues for acquiring and
learning about literature can be explored [10]. Montiel, I et al. built a categorization model of
sustainability teaching and learning centered on literary genres based on the rootedness theory in
order to help students improve their general literacy and affective cognition in corporate sustainability
management [11]. MAHON, ÁINE, et al. summarized the ideas and results of Rorty’s and Cavell’s
research in the field of literature, emphasizing the positive role of literature in moral education [12].
Bell, M describes the definition of thought at the literary and philosophical levels and the
development of the teaching of thought. It points out that research on fine arts literature continues to
evolve and move forward, but a universal analytical framework for analyzing the nature of literature
continues to be missing [13]. Zhao, J et al. explored the role played by classical Chinese poetry in
literary history and related studies and envisioned a time-series entropy weighting method to explore
the correlation between the cultural influence of the authors of the poems and the time as well as the
changes, which could help to understand the evolution of the poems at the historical level [14]. Kearns,
E. et al. investigated the role played by big data technology in literary analysis and, based on empirical
analysis, learned that machine learning tools effectively recovered textual relational networks, which
facilitated the continued role of literary and cultural influences [15]. Yuan, Y attempted to build a
Analysis of Literature Education Strategies and Learning Behaviors in the Age of Big Data
3
psychoanalytic cross-framework using artificial intelligence technology and big data technology in
order to explore and analyze the characterization of the characters in the classic Chinese literature
Dream of Red Mansions, which is of positive significance for a deeper understanding of the literary
context and humanistic spirit [16].
This paper constructs a framework for analyzing students’ learning behaviors based on literature
education, improves the K-means algorithm through high-density clustering and average weighting,
improves the Apriori algorithm by using the weights and matrices, data division, and influence, and
uses the enhanced HDWA-Kmeans algorithm and the WMDE-Apriori algorithm for the data mining
of the student’s learning behavior system. Then, 50 students from the Chinese Department of
University M were selected as the research subjects for the experiment, and the students were divided
into four categories by recording and analyzing their learning behaviors in the literature education
course. The learning motivation, time commitment, and effectiveness of the four categories of
students were compared separately to analyze the differences in the learning behavior characteristics
of each category. Finally, a Pearson correlation analysis was conducted between students’ learning
behaviors and their literary achievements to explore the correlation between students’ learning
behaviors and literary education.
2 A framework for analyzing learning behaviors based on literature education
2.1 Data mining process
The process of most data mining involves determining business objectives, collecting data, preparing
data, creating a data model, evaluating the model, and publishing the results. The premise of data
mining is that a huge amount of data already exists in the database. Still, the massive data information
of the student behavior system contains a large number of detailed lists of data. Also, it contains some
null values, and all the irregularities in these massive data will affect the efficiency of the decision
support system and affect the normal operation of the system, so we have to select, transform, and
integrate a large amount of accumulated research data in order to form the Valuable data, these
valuable data are summarized to create a data warehouse, and this topic will be the data warehouse
system as the main data source of data mining.
Data mining is considered to be an advanced computer technology that discovers valuable laws from
a large amount of data, and it is a high-tech process of selecting the correct, representative, beneficial,
and well-expressed patterns from a large amount of data. The data mining process is shown in Figure
1. From raw data to the knowledge mining process for the following six steps, this process needs to
be repeatedly verified and assessed to finally achieve decision-making support.
Data
Target data
Preprocessed
data
Transformation
data
Patterns
Knowledge
PreprocessingSelection Data Mining
Interpretation
Evaluation
Transformations
Figure 1. Data mining process diagram
From the data mining process diagram, we can see that data mining involves traversing through a
series of links in a cycle, analyzing the operation of human-computer interaction. The following are
the specific explanations:
Chong Luo. Applied Mathematics and Nonlinear Sciences, 9(1) (2024) 1-20
4
1) Used in a particular research direction, including predictive knowledge and predictive results
used in the process.
2) To create a dataset that you want to get, integrate one or more datasets for processing to get a
final dataset.
3) To the existence of unreasonable, noisy, useless, null-valued data in the huge amount of data,
and the need to sort out the temporal transformations of the data and the sequence of the time
occurrence, etc.
4) Data transformation is to discover what characteristics the data have, classify and redefine
these data, and reduce the number of useful variables. For example, classify a customer as
“high-value”, “medium-value”, or “low-value”. The treatment plan given by the doctor to the
patient is divided into “treatment plan 1”, “treatment plan 2”, and “treatment plan 3”.
Customers applying for credit cards are divided into “safe customers” and “dangerous
customers”.
5) Determine the ultimate goal of data mining and find appropriate algorithms and techniques to
achieve it.
6) Pattern formation, which is the use of data mining techniques to generate a pattern or a
collection of data of particular interest.
7) Interpretation, which is the process of explaining the discovered patterns in detail, eliminating
the useless and unrealistic patterns, and transforming them into a practical and correct pattern
that meets reality.
8) Validation of knowledge, the results of the data mining to carry out practical testing,
verification of the mining of these laws in the reality of the contribution made, or to verify the
discovery of the hidden information is correct, the cycle of dealing with the conflict found in
the process.
The data mining process is cyclic, a method to continuously improve the results of mining is a process
of digging to discover the characteristics of the potential data, then creating a data model, and finally
repeatedly verifying and making reasonable adjustments and then applying the process.
2.2 Improved Cluster Mining
In the implementation of clustering algorithms, one of the problems that always has to face is how to
determine the K value and how to determine the initial clustering center. Random selection will
produce unpredictable results, so in order to be able to obtain a better initial clustering center, the K-
means algorithm is improved based on the high density of the algorithm, the algorithm with the idea
that high-density objects are more likely to be the center of the clusters and more effective in obtaining
the initial Clustering center. In order to be able to differentiate each data object and cluster them faster
into the cluster to which they belong, that is, the cluster with the closest distance, and at the same
time to ensure that the algorithm finally converges, the clustering algorithm was improved by
recalculating the center of mass. The HDWA-Kmeans algorithm was finally created.
Analysis of Literature Education Strategies and Learning Behaviors in the Age of Big Data
5
2.2.1 K-means algorithm based on high-density clustering
Due to the K-means algorithm effect being obvious, the idea is simple, and other advantages are
widely used in a variety of life fields. Still, the selection of the initial clustering center problem is a
very obvious defect of the K-means algorithm. If the selection is not appropriate, then the final
clustering effect often appears to be the result of the local optimum, which can not achieve our
requirements, that is, the global optimum. In order to optimize this situation, this paper uses an
improved algorithm with the idea that high-density objects are more likely to become the clustering
center when selecting the initial clustering center.
Let initial data data set:
12
, , , n
D x x x=
,
k
cluster classes:
12
, , , k
C C C C=
,
m
sets:
12
, , , m
M M M M=
. have the following definitions:
Definition 1 The formula for Euclidean distance is expressed as:
( ) ( )
2
1
,n
i j il jl
l
d x x x x
=
=−
(1)
In Equation (1),
i
x
,
i
x
are data objects,
il
x
is the 1st feature attribute of
i
x
, and
jl
x
is the 1st
feature attribute of
j
x
.
Definition 2 The clustering center (Centerk) of a clustered cluster is denoted as:
1
ik
ki
xC
k
Center x
C
=
(2)
In Equation (2),
k
Center
represents the center of the
k
nd cluster,
k
C
represents the
k
th cluster,
and
ik
xC
represents all data objects belonging to cluster
k
C
.
Definition 3 point sample distance (distance between the data object and sample data set), i.e., the
distance between the data object and the mean value of the data object within the data set is denoted
as:
( )
tan ,
i
im
xD
dis ce d x centerM
=
(3)
In Equation (3),
M
denotes the set consisting of the two points with the shortest distance,
D
denotes the original dataset that remains after deleting the data objects in set
M
, and
m
centerM
denotes the average of the
m
th set
m
M
.
Definition 4 The formula for calculating the sum of the distances between data object
i
X
and other
clusters that do not contain the object is denoted as:
( )
( )
,
jk
i j i
xC
sum d x x x D
=
(4)
Chong Luo. Applied Mathematics and Nonlinear Sciences, 9(1) (2024) 1-20
6
In the use of the K-means algorithm for cluster analysis, often a bunch of unknown data objects for
cluster analysis, at the same time these objects are likely to be unable to intuitively find the
distribution of its laws, but the higher the density of the data objects are more likely to be clustered
together. Therefore, using the density idea to find
k
initial clustering centers can improve the effect
and performance of clustering.
Here consider two special cases: the first is that when selecting from the nearest point, there may be
more than one pair of the same nearest distance to the data object point collection. Combined with
the idea that high-density data is easier to aggregate, it is proposed that in this case, according to
formula (3) can be derived from the point sample distance of each pair of nearest-distance points and
the nearest-distance point corresponding to the smallest value of the point sample distance can be
calculated and compared as the initial point of the next set
M
. The second case is similar. That is
when there is an existing sample set of initial points, the nearest data objects that need to be added to
the dataset are added together when there is more than one. The construction process of the K-means
algorithm based on high-density clustering to obtain the initial clustering center is shown in Fig. 2.
Start
Calculate the Euclidean distance between two of the remaining
data objects in the dataset according to equation (3-1)
Find the two data objects
with the shortest distance that are
unique or not
Calculate the distances of all points with equal shortest distances to
the rest of the data in the dataset according to equation (3 -3)
The two data objects with the shor test distances form a sam ple set Mm
(0<mk), and remove the two data with the shortest distances from
the data set D
Calculate the mean value of all data objects within the sample set Mm
Calculate the distance between each data object in the leftover data of the data set
D and the sam ple set Mm according to form ula (3-3), find the point with the closest
distance to be added to the set Mm , and if there are more than one point with the
closest distance to be added together, and remove it (them) from the data set D
The number of data objects in
the sample set Mm is α(n/k), 0<α1.
Number of sample sets mk, i.e.,
the number of sample sets is equal to the number of
cluster classes
Calculate the arithmetic mean of all sample sets M1, M2,
and Mm as the initial cluster center, respectively
End
N
Y
N
Y
N
Figure 2. Initial clustering center process based on high-density clustering of K-means algorithm
Analysis of Literature Education Strategies and Learning Behaviors in the Age of Big Data
7
2.2.2 K-means algorithm based on weighted average
For a given dataset
12
, , , n
D x x x=
,
( )
1,2, ,Xi i n=
represents the
i
th data object in dataset
D
. For dataset X, each data object in it is different. If the value of the difference between each data
object and the others can be precisely quantified, then each data object can be accurately differentiated.
Thus the objects in the dataset can be clustered more quickly. After obtaining a better initial clustering
center, a threshold coefficient is added to the formula for recalculating the center of mass of each
cluster, i.e., The K-means algorithm based on the weighted average method is used to quantify
differences between data objects.
The purpose of the weighted average-based K-means algorithm in the clustering process is to assign
the data objects to the cluster with the highest similarity, where the similarity is measured by the
Euclidean distance, as previously stated. The weighted average of all clusters is compared with the
Euclidean distance of each data object to achieve this.
A better initial clustering center is obtained based on the idea of high-density clustering. The way to
get it is to calculate the arithmetic mean of each sample collection separately, and each arithmetic
mean is the initial clustering center point. That is
12
,, k
D D D
. The
k
clusters of clustering are
denoted as
12
, , , k
C C C
. The sum of data objects of all clustered clusters is the original data object
that is:
Definition 1: The sum of the number of all data objects in the data set is equal to the sum of the
number of individual data clusters:
1
k
i
i
NN
=
=
(5)
Definition 2
*
j
Expressions:
( )
*
1
tan ,
i
N
jj
j
j
dis ce x y
N
=
=
(6)
Equation (6),
1, 2, ,j Ni=
,
N
denotes the total number of all data objects in the data set
D
, and
( )
tan ,
jj
dis ce x y
denotes the Euclidean distance between data objects
i
x
and
i
y
.
Definition 3 The expression for the weight of each data object in data set
D
is denoted as:
*
*
1
j
jN
j
j
=
=
(7)
In Equation (7), it can be concluded that the similarity of data objects in dataset
D
is inversely
correlated with the weights, i.e., the smaller the weights, the smaller the Euclidean distance, the higher
the similarity, and vice versa.
Chong Luo. Applied Mathematics and Nonlinear Sciences, 9(1) (2024) 1-20
8
Definition 4 The weighted average of all data objects in the
i
nd clustering cluster
Ci
is
*
i
C
, and
the formula is calculated as follows:
*
1
1
ii
N
i j j
j
ixC
Cx
N
=
=
(8)
Equation (8),
j
x
represents the
j
rd data object in cluster
Ci
, and
j
represents the weight of the
data object
j
x
in cluster
Ci
.
In order to make the K-means algorithm based on the weighted average method can ensure the
clustering results do not converge or the results are not optimal, compared with the traditional
objective function, the improved algorithm adds a threshold coefficient
i
to represent the weights,
which is expressed in the formula:
i
i
N
N
=
(9)
In Equation (9), it can be seen that the threshold coefficient here is the ratio of the number of clusters
per cluster to the number of original data objects. Thus, the final expression for the objective criterion
function can be obtained, calculated as:
( )
2
**
1
tan ,
ji
kN
i j i
i x C
J dis ce x C
=
=
(10)
Equation (10),
( )
2
*
tan ,
ji
dis ce x C
denotes
j
x
and
*
i
C
.
2.2.3 Flow and implementation of the HDWA-Kmeans algorithm
The K-means algorithm based on high-density clustering and the K-means algorithm based on the
weighted average method are named as HDWA-Kmeans algorithm. The flow of the algorithm is
shown in Fig. 3.
Analysis of Literature Education Strategies and Learning Behaviors in the Age of Big Data
9
Initialization
Calculate the center of mass of each
cluster class to repartition the data set
Calculate J* value to update the center of
mass
Converge or not
converge sufficient convergence
condition
End
Y
Obtaining Initial Cluster Centers Based
on High Density Clustering Algorithm
Data pre-processing
N
Figure 3. HDWA-Kmeans algorithm process
2.3 Improved association rule mining
In order to solve the performance problem of the Apriori algorithm when generating a large number
of candidate item sets and repeatedly scanning the database, this paper proposes an improvement
method that utilizes weights and matrices. To improve the efficiency of the algorithm and alleviate
the bottleneck caused by running pressure, another improvement method based on data division is
proposed. In order to achieve strong and effective association rules, it is suggested to improve the
influence degree. Finally, the flow, implementation, and experiments of the improved algorithm are
analyzed to verify its efficiency.
2.3.1 Apriori algorithm based on weights and matrices
The performance bottleneck of the Apriori algorithm is the need to generate a large number of
candidate itemsets and the need to repeatedly scan the database. In order to solve the performance
bottleneck of the Apriori algorithm, this paper proposes an improved method based on weights and
matrices.
Let the transaction database
D
, there are
m
transactions and
n
transaction items, from which a
mn
-order Boolean transaction matrix can be constructed:
11 12 1
21 22 2
12
n
n
m m mn
M
=
(11)
1, 1, 2, , ; 1,2, ,
0,
ij i
ij
ij i
Ti m j n
T
= = =
(12)
Chong Luo. Applied Mathematics and Nonlinear Sciences, 9(1) (2024) 1-20
10
In Equation (11), the
i
nd record in the transaction dataset
T
is denoted by
i
T
. The
m
transaction
and
n
transaction items correspond to the rows and columns in the Boolean transaction matrix
M
,
respectively.
Let the set of items contained in transaction database
D
be set
12
, , , h
I I I I=
and the
probability that item set
( )
1,2, ,
j
I j h=
occurs in
D
be
( )
j
PI
, which is given by Eq:
( )
j
l
PI m
=
(13)
In Equation (13), the number of occurrences of
j
I
in the transaction set
T
is denoted by
l
, i.e.,
the number of 1’s in column
j
of matrix
M
, and the number of transactions is denoted by
m
. The
weight is denoted as
( )
j
WI
, which is calculated by the formula:
( ) ( )
1
j
j
m
WI l
PI
==
(14)
In Equation (14), the number of occurrences of
j
I
in the transaction set
T
is denoted by
l
, i.e.,
the number of 1’s in column
j
of matrix
M
, and the number of transactions is denoted by
m
.
The
i
th record in the transaction dataset is
i
T
, then the weight of
i
T
is the average weight of the
items contained in
i
T
, i.e., averaged over all
( )
j
WI
of
( )
1 1,2, ,
ij jn
==
:
( )
( )
1
&ji
n I T j
iji
WI
WT T
=
=
(15)
In Equation (15), the number of items contained in transaction
i
T
is denoted by
i
T
.
The support of items based on weights is denoted by
sup port
W
. Then,
sup port
W
denotes the proportion
of the weights of the transaction containing item
S
to the weights of all the transactions, and from
matrix
M
, all the transactions containing item
S
are included. Therefore, the formula for
sup port
W
is:
( ) ( )
( )
&
1
sup
1
i
m S T
i
i
port m
i
i
WT
WS
WT
=
=
=
(16)
In Equation (16),
S
represents any item in the transaction database.
Analysis of Literature Education Strategies and Learning Behaviors in the Age of Big Data
11
2.3.2 Apriori algorithm based on data partitioning
The Apriori algorithm generates a large number of candidate item sets, which not only increases the
space overhead but also increases the time overhead. Therefore, in order to reduce the overhead in
time and space and facilitate parallelized processing, this paper adopts the idea of division, which
divides the transaction database into independent and non-interfering partitions based on certain data
division criteria. Partitioning firstly facilitates parallel mining and improves computation and mining
efficiency. In addition, it reduces the spatial overhead when mining in only one partition.
The transaction database
D
is divided into
n
partitions based on certain data division criteria (e.g.,
this paper’s analysis of students’ one-card spending behavior can be divided according to criteria such
as college, gender, etc.):
12
12
12
, , , n
n
n
D D D D
D D D D
D D D
=
=
=
(17)
On each partition
( )
1,2, ,
i
D i n=
, the mining analysis is performed separately using the Apriori
algorithm and the result of mining the frequent itemset is
( )
1,2, ,
i
P i n=
. All the frequent itemsets
are merged into the final frequent itemset:
12 n
P P P P=
(18)
2.3.3 Influence-based Apriori algorithm
The Apriori algorithm has two main steps: the first step is to generate frequent item sets. The second
step is to generate association rules. Algorithm improvements based on data division and weights and
matrices are used to create frequent item sets. In contrast, algorithm improvements based on influence
degree are used to generate association rules from frequent item sets.
The effectiveness of association rules generated by the Apriori algorithm is not sufficient to rely on
the three evaluation criteria of support, confidence, and lift. The support degree guarantees that the
rules are valuable, the confidence degree guarantees the strength of the regulations, and the
enhancement degree guarantees the validity of the rules to some extent, but all three have certain
limitations. For this reason, this paper proposes the use of influence degree to analyze the strength
and effectiveness of rules more precisely.
For a rule
XY
,
( )
P Y X
denotes the confidence level of the rule and
( )
PY
denotes the
percentage of occurrences of itemset
Y
in dataset
D
, define the influence level of
XY
as:
If
( )
( )
P Y X P Y
, then:
( )
( )
( )
( )
1
P Y X P Y
effect X Y PY
−
= −
(19)
If
( )
( )
P Y X P Y
, then:
Chong Luo. Applied Mathematics and Nonlinear Sciences, 9(1) (2024) 1-20
12
( )
( )
( )
( )
( ) ( ) ( )
( ) ( ) ( )
1 1 1
0 1 1
P Y X P Y
effect X Y PY
effect X Y when P Y and P X
effect X Y when P Y and P X
−
=
= = =
= =
(20)
According to the definition, when the influence degree is positive, it takes the value range of
0,1
.
When
( )
( )
P Y X P Y=
, the influence degree is 0, when
( )
1P Y X =
, the influence degree is 1.
When the influence degree is negative, it takes the value range of
1,0−
, when
( )
( )
P Y X P Y=
,
the influence degree is 0, and when
( )
0P Y X =
, the influence degree takes -1.
When the rule supports, the confidence level is greater than the threshold, and the influence degree is
greater than 0, then the rule is a strong, valid rule.
2.3.4 WMDE-Apriori Algorithm Flow
The Apriori algorithm, based on weights and matrices, data partition, and influence improvement, is
named the WMDE-Apriori algorithm. The algorithm flow is shown in Fig. 4.
InitializationData pre-processing
Define minimum support and partition the transaction
database based on the partitioning criteria D
Scan each partition of D to generate a Boolean
transaction matrix
Calculate W(Ij ) and W(Ti)
Generate candidate item sets
Compute support(s)
Generate frequent itemsets
Whether the frequent itemset is empty
Merge all frequent itemsets
End
Frequent items et
self-join, generate
candidate itemset
Y
N
Figure 4. WMDE-Apriori algorithm flowchart
Analysis of Literature Education Strategies and Learning Behaviors in the Age of Big Data
13
3 Results and Discussion
In this study, 50 students from the Chinese Department of School M were selected as research subjects,
to whom questionnaires were distributed, and their daily study behaviors were observed and recorded
in order to analyze the students’ study behaviors and their correlation with their literary achievements.
3.1 Cluster Analysis of Student Learning Behavior
Through cluster analysis, student learning behavior as an indicator can be categorized into four
categories of learners, and the results show that there are obvious differences between each category
of learning group. In order to analyze the specific performance of behavioral differences, first of all,
by comparing the quantitative mean value of learning behavior in each category, the mean value of
student learning behavior is shown in Table 1. Based on the mean values of students’ literary
education course learning performance, learning hours, and days of learning delay, the following
definitions were made for each category of students:
1) Students in Category 1 performed well overall and achieved the highest regular grades (98.65)
and test scores (94.23), but students in this category had higher study procrastination and were
comparable to Category 3 in terms of motivational performance.
2) The overall performance of students in category 2 is poorer. Observation of Table 1 reveals
that the test efficacy of this learning group is higher. Still, the specific analysis is due to the
fact that this category of students has the lowest average test length among the four categories,
with a test length of only 1,345.65 hours. This category had the highest procrastination among
all students, with 14.056 days of study procrastination and 7.012 days of task procrastination.
3) The overall performance of students in Category 3 is comparable to that of Category 4, and
they are able to achieve good grades and are more motivated to complete their tasks.
4) Students in category 4 achieved excellent overall results, had the highest test efficacy of the
four categories, and had the lowest academic procrastination of all categories, with 8.954 days
of academic procrastination and 2.134 days of task procrastination.
Table 1. The students’ mean of study behavior
Category
Study effect
Time input
Study initiative
Exercise
score
Normal
grade
Exam
grade
Test
efficiency
Test
duration
Total duration
of tasks
Study
delay days
Task delay
days
1
60.432
98.65
94.23
0.012%
5523.34
13123.54
11.363
5.834
2
53.546
93.00
77.67
0.062%
1345.65
9142.66
14.056
7.012
3
58.531
95.15
90.42
0.034%
2654.86
10753.73
10.518
3.824
4
71.842
91.45
92.54
0.194%
3154.62
14023.82
8.954
2.134
3.1.1 Comparison of students’ motivation to learn
Research has proved that there are significant differences among the four categories of students in
each learning behavior indicator. In order to analyze the differences between each category of students
in each category of learning behavior, this study uses ANOVA to explore the differences between
each category of learning behavior using the LSD method. In order to investigate the behavioral
characteristics of each category of students in-depth, the researcher analyzes and compares data on
learning behaviors with more significant differences among different student categories.
Chong Luo. Applied Mathematics and Nonlinear Sciences, 9(1) (2024) 1-20
14
The results of ANOVA for different students’ learning motivation behaviors are shown in Table 2.
The analysis shows that in the learning motivation dimension, except for Category 1 and Category 3
students who do not have significant differences in the behavioral indicators of days of study delay,
the remaining categories of students have substantial differences in the two behavioral indicators of
learning motivation. It is worth mentioning that although there is a difference in the number of days
of task procrastination between Category 3 and Categories 1 and 2, it is not very significant because
of the Sig. The value is close to 0.05.
The statistical results of the number of days of study delay for different categories of students are
shown in Figure 5 in terms of days. From the figure, it can be seen that compared with the other three
learning groups, the overall learning interval of category 4 is smaller, most of which is in the range
of 0-5 days, indicating that the students in this category are actively involved in learning once the
course is released. Compared to the other three categories of students, they had learning delays of
more than 5 days. In addition, since the course assignments were released two weeks after the course
started, the analysis found that Category 2 and some Category 1 students began learning only near
the release of the assignments, which shows that the assignments or learning tasks motivate students
to learn and become task-driven learners, and how to enhance the learning initiative of this group of
students has become one of the topics to be explored in the future.
Table 2. Different categories of student study behavior LSD test in study initiative
Sig.
Study behavior
Category
2
3
4
Study delay days
1
0.012
0.184
0.001
2
0.005
0.015
3
0.006
Task delay days
1
0.001
0.046
0.000
2
0.032
0.002
3
0.001
Figure 5. Line diagram of study delay days and student number
Analysis of Literature Education Strategies and Learning Behaviors in the Age of Big Data
15
3.1.2 Comparison of Student Time Commitment
The study analyzed the differences in learning behaviors of different categories of students in the
category of time commitment and the results of ANOVA are shown in Table 3, which revealed that
there are significant differences between all categories of students. Category 4 and Category 3 have
a smaller difference in the test duration index compared to other groups.
The statistical results of the total time of completing the task for different categories of students are
shown in Figure 6. It was analyzed that compared with the other three categories of students, Category
4 students had the longest study hours, with more than 30 of them studying for more than 14,000
hours, indicating that they had the highest commitment to study. On the contrary, category 2
consumed the shortest number of hours, with only one person studying for more than 10,000 hours.
Combined with the analysis in Table 3, although the test efficacy of students in this category is
relatively high, the low number of hours consumed in the test results in low motivation in this category
of students.
Table 3. Different categories of student study behavior LSD test in time input
Sig.
Study behavior
Category
2
3
4
Total duration of
tasks
1
0.001
0.001
0.001
2
0.002
0.000
3
0.000
Test duration
1
0.001
0.001
0.001
2
0.000
0.000
3
0.015
Figure 6. Line diagram of the total duration of tasks and student number
3.1.3 Comparison of student learning outcomes
This section examines the differences between different categories of students on the dimension of
learning effectiveness, and the results of the ANOVA are shown in Table 4. The analysis found that
among the indicators of learning effectiveness, the differences between categories were relatively
small.
Chong Luo. Applied Mathematics and Nonlinear Sciences, 9(1) (2024) 1-20
16
Specifically, the analysis found that in practice problem scores, there were differences only between
category 4 and categories 1 and 2. The usual scores indicate significant differences between category
4 and the other three learning groups but no significant differences between the other three learning
groups. Category 2 had a significant difference in test performance compared to the other three
learning groups, while no significant differences were found between the other three learning groups.
Significant differences were observed in test efficacy between all three categories of learning groups,
with category 2 being the most significant. The use of learning effectiveness as the sole criterion for
assessing students is not reasonable, and learning achievement only categorizes students without
indicating the differences among them.
The statistical results of the test scores of students in each category are shown in Figure 7. The peak
of the test scores of Category 1 students is concentrated on the right side of 85. In contrast, the test
scores of Category 3 and Category 4 students are not very different and are more evenly distributed.
In addition, the test scores of Category 2 are significantly different from the other three categories,
with a peak of 70 points, which is located on the leftmost side of the four categories, indicating that
students in this category do not achieve the desired learning outcomes. Combined with the above
analysis of each category of students’ learning motivation and time commitment, it can be seen that
by clustering students and analyzing the differences between different categories of students, it is easy
for teachers to find abnormal behavior in time and take targeted measures, for example, for category
2 students, the teacher should first prompt students to take the initiative to engage in learning, increase
their interest in the content of the course and learning commitment, and promote the emergence of
their learning behaviors.
Table 4. Different categories of student study behavior LSD test in study effect
Sig.
Study behavior
Category
2
3
4
Exercise score
1
0.365
0.763
0.015
2
0.612
0.025
3
0.094
Normal grade
1
0.089
0.311
0.001
2
0.548
0.000
3
0.002
Exam grade
1
0.001
0.314
0.075
2
0.001
0.000
3
0.603
Test efficiency
1
0.003
0.039
0.034
2
0.002
0.001
3
0.041
Analysis of Literature Education Strategies and Learning Behaviors in the Age of Big Data
17
Figure 7. Line diagram of exam grade and student number
3.2 Analysis of association rules between learning behavior and literary education
In order to investigate the correlation between students’ behaviors and the effectiveness of literary
education, this study conducted a Pearson correlation analysis between students’ learning behaviors
and students’ literary education achievement, and the results of the analysis are shown in Table 5.
From the data in the table, it can be seen that among the attributes of students’ learning behaviors, the
ones that have a moderate correlation with their literary achievement are usual marks and test scores,
with correlation coefficients of r=0.627 (p<0.05) and r=0.653 (p<0.05), respectively. Among the
dimensions of students’ learning gains, test efficacy had a moderate positive correlation with it, with
a correlation coefficient of r=0.515 (p<0.05), and the length of the test and the number of days of
study delay both had a low negative correlation with it (p<0.05).
Chong Luo. Applied Mathematics and Nonlinear Sciences, 9(1) (2024) 1-20
18
Table 5. The correlation results of students’ study behavior and literary achievement
Literary achievement
Study experience
Study harvest
Exercise score
Pearson correlation
0.231
0.011
-0.063
Sig.(double tail)
0.040
0.870
0.612
N
50
50
50
Normal grade
Pearson correlation
0.627
0.055
0.072
Sig.(double tail)
0.022
0.056
0.031
N
50
50
50
Exam grade
Pearson correlation
0.653
0.076
0.056
Sig.(double tail)
0.045
0.598
0.099
N
50
50
50
Test efficiency
Pearson correlation
0.316
0.041
0.515
Sig.(double tail)
0.089
0.612
0.827
N
50
50
50
Test duration
Pearson correlation
0.273
0.053
0.354
Sig.(double tail)
0.414
0.965
0.028
N
50
50
50
Total duration of
tasks
Pearson correlation
0.056
0.216
0.033
Sig.(double tail)
0.238
0.023
0.485
N
50
50
50
Study delay days
Pearson correlation
0.012
-0.087
-0.331
Sig.(double tail)
0.209
0.734
0.423
N
50
50
50
Task delay days
Pearson correlation
0.009
-0.297
-0.082
Sig.(double tail)
0.011
0.014
0.36
N
50
50
50
4 Conclusion
This study constructs a student behavior analysis framework based on literary education, uses the
improved K-means algorithm and the improved Apriori algorithm for data mining of student behavior,
and explores the association between student learning behavior and literary education.
1) Based on the cluster analysis of students’ learning behaviors, students are divided into 4
categories. Category 1 students had good overall performance but high levels of study
procrastination. Overall, Category 2 students performed poorly and had the most
procrastination among the four categories. Category 3 and Category 4 students performed
similarly, and Category 4 students were the most motivated among the four categories.
2) Category 4 students performed best in terms of motivation. Their study intervals were
generally short, usually less than five days. The other three categories of students had study
intervals that were greater than five days. Category 4 students had the most time commitment
to study, with over 30 studying over 14,000 hours. The time commitment of Category 2
students was the lowest, with only one student studying for more than 10,000 hours. Category
Analysis of Literature Education Strategies and Learning Behaviors in the Age of Big Data
19
1 students were the most effective in terms of learning effectiveness, with most of them
scoring over 85 on their exams. Only one of the Category 2 students scored more than 75 on
the exam.
3) Pearson correlation analysis between students’ learning behaviors and their literature
performance showed that the higher correlation with literature performance was between
students’ regular performance (r=0.627) and exam performance (r=0.653). The total number
of hours spent completing the task was the only factor that showed a positive correlation in
terms of learning experience. In terms of learning gains, only learning efficacy showed a
moderate positive correlation with it (r=0.515).
References
[1] Kan, W., & Dechao, L. (2022). Are translated chinese wuxia fiction and western heroic literature similar?
a stylometric analysis based on stylistic panoramas. Digital Scholarship in the Humanities(4), 4.
[2] Wee, L. H. (2017). Review: the sound and sense of chinese poetry. Journal of Chinese Linguistics, 45(1),
231-237.
[3] John, R., & David. (2019). Two cultures versus general education. European Review.
[4] Manuel, J., & Carter, D. (2019). Resonant continuities: the influence of the newbolt report on the
formation of english curriculum in new south wales, australia. English in Education, 53.
[5] Horner, B. (2018). Futures for english studies: teaching language, literature and creative writing in higher
education. edited by ann hewings, lynda prescott, philip seargeant. 2016. basingstoke: palgrave macmillan,
xi + 268.: reviews. world englishes, 37(2), 386-388.
[6] Nishihara, T. (2019). Literary expressions in authorized high school textbooks for english communication
i, ii and iii. The Journal of Educational Research, 24.
[7] Vomero, S. K. (2024). Teaching social justice through shakespeare: why renaissance literature matters
now. edited by hillary eklund and wendy beth hyman. Shakespeare Quarterly(1), 1.
[8] Cascardi, A. J. (2018). Literary education in a free society. Journal of Philosophy of Education, 52(4),
723-741.
[9] Truman, S. E. (2020). White déjà vu: troubling the certainty of the english canon in literary education.
English in Australia.
[10] Asonova, E., Borusyak, L., & Romanicheva, E. (2020). Literary education: what key stakeholders think.
Educational Studies.
[11] Montiel, I., Antolin-Lopez, R., & Gallo, P. J. (2018). Emotions and sustainability: a literary genre-based
framework for environmental sustainability management education. Academy of Management Learning
& Education, 17(2), 155-183.
[12] MAHON, ÁINE, & O’Brien, E. (2018). Resonance, response, renewal: literary education in rorty and
cavell. Journal of Philosophy of Education, 52(4), 695-708.
[13] Bell, M. (2018). Aesthetics and value in literary education: reflections on f. r. leavis. Journal of Philosophy
of Education, 52(4), 709-722.
[14] Zhao, J., Wei, Y., & Wu, B. (2022). Analysis of the social network and the evolution of the influence of
ancient chinese poets. Social science computer review.
[15] Kearns, E., & Morin, C. (2023). Irish minerva writers and the affordances of big data: some preliminary
findings. Irish University Review, 53(1), 27-47.
[16] Yuan, Y. (2021). Analysis of fictional characters in the context of artificial intelligence and big data:
taking a dream of red mansions for example. Digital Scholarship in the Humanities(1).
Chong Luo. Applied Mathematics and Nonlinear Sciences, 9(1) (2024) 1-20
20
About the Author
Chong Luo, Master of Arts, Lecturer. Graduated from the Yanbian University in 2013. Worked in
Henan Geology Mineral College. His research interests include Chinese Language and Literature and
Chinese traditional culture.
