BASIC DIRECTION AND REALIZATION PATH
OF PE TEACHING INNOVATION IN PSS
BASED ON DEEP LEARNING MODEL
Huiming Ke
College of Fine Arts, Guangdong Polytechnic Normal University, Guangzhou
Guangdong, 510665, China
yangwang_320@sina.com
Yang Wang
School of Materials Design and Engineering, Beijing institute of fashion
technology, Beijing, 100029, China
Reception: 02/11/2022 Acceptance: 29/12/2022 Publication: 23/01/2023
Suggested citation:
K., Huiming and W., Yang. (2023). Basic direction and realization path of PE
teaching innovation in PSS based on deep learning model. 3C Tecnología.
Glosas de innovación aplicada a la pyme, 12(1), 70-85. https://doi.org/
10.17993/3ctecno.2023.v12n1e43.70-85
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
70
BASIC DIRECTION AND REALIZATION PATH
OF PE TEACHING INNOVATION IN PSS
BASED ON DEEP LEARNING MODEL
Huiming Ke
College of Fine Arts, Guangdong Polytechnic Normal University, Guangzhou
Guangdong, 510665, China
yangwang_320@sina.com
Yang Wang
School of Materials Design and Engineering, Beijing institute of fashion
technology, Beijing, 100029, China
Reception: 02/11/2022 Acceptance: 29/12/2022 Publication: 23/01/2023
Suggested citation:
K., Huiming and W., Yang. (2023). Basic direction and realization path of PE
teaching innovation in PSS based on deep learning model. 3C Tecnología.
Glosas de innovación aplicada a la pyme, 12(1), 70-85. https://doi.org/
10.17993/3ctecno.2023.v12n1e43.70-85
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
ABSTRACT
At present, the traditional model of PE in PSS (PSS)has seriously affected the quality
of PE teaching in PSS and the perception of PE among primary and secondary school
students. Because of the urgent need for innovation in PE in PSS, this study proposes
the LSTM model to achieve an accurate prediction of the innovation direction of PE in
PSS. Based on the LSTM model, the user behavior is classified by extracting the
important features of the innovation direction. Expression to achieve accurate
prediction of the future development direction of PE. Using the data confusion matrix
to estimate the prediction accuracy of the LSTM model, the four evaluation indicators
of Accuracy, Precision, F1, and AUC are 0.0532~0.2323 higher than the baseline
model. The prediction results of PE teaching innovation in PSS from three aspects of
teaching thought, teaching content, teaching objectives and essence are output,
which has obvious guiding significance for the overall optimization of PE classrooms
in PSS. This result shows that the LSTM prediction model has important practical
value.
KEYWORDS
PE; Artificial intelligence; LSTM network; Model analysis; Innovate
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
71
PAPER INDEX
ABSTRACT
KEYWORDS
1. INTRODUCTION
2. RELATED WORKS
2.1. Recommendation algorithm based on deep learning and its application
2.2. Recommendation algorithm and application based on LSTM
3. RELATED CONCEPTS
3.1. RNN network structure
3.2. LSTM network structure
3.3. Prediction result output layer
4. INNOVATION DIRECTION MODEL BASED ON LSTM MODEL
5. MODEL EVALUATION
6. RESULT ANALYSIS
6.1. Baseline model parameter setting
6.2. Innovative development direction of PE
6.2.1. Innovation of teaching ideas
6.2.2. Essence of PE teaching process and innovation of main objectives
6.2.3. Innovation of teaching content system
7. CONCLUSION
8. CONFLICT OF INTEREST
REFERENCES
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
72
PAPER INDEX
ABSTRACT
KEYWORDS
1. INTRODUCTION
2. RELATED WORKS
2.1. Recommendation algorithm based on deep learning and its application
2.2. Recommendation algorithm and application based on LSTM
3. RELATED CONCEPTS
3.1. RNN network structure
3.2. LSTM network structure
3.3. Prediction result output layer
4. INNOVATION DIRECTION MODEL BASED ON LSTM MODEL
5. MODEL EVALUATION
6. RESULT ANALYSIS
6.1. Baseline model parameter setting
6.2. Innovative development direction of PE
6.2.1. Innovation of teaching ideas
6.2.2. Essence of PE teaching process and innovation of main objectives
6.2.3. Innovation of teaching content system
7. CONCLUSION
8. CONFLICT OF INTEREST
REFERENCES
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
1. INTRODUCTION
Ball games, track and field, and some sports have converted the governments of
PE in PSS. Methods such as explanation and demonstration have greatly affected
teachers' teaching quality and students' cognition of PE[1]. Such as demonstration
and explanation, teaching students to start, long jump, and pull up. Make students
directly participate in imitation, resulting in a poor experience and poor effect[2-3].
Therefore, PE teachers need to enrich their teaching methods, improve their teaching
ability, make teaching innovations, and flexibly use their innovative methods to teach
students, so that students can enjoy high-quality PE courses[4]. Therefore, the reform
of PE in schools is domineering.
At present, a considerable part of the exploration of sports innovative education
focuses on the innovation of teaching methods[5]. However, if we only innovate in
teaching methods all the time, it is difficult for PE to have a leap-forward development.
Therefore, the overall reform of PE teaching is highly praised by researchers[6]. To
cultivate students' comprehensive sports ability and innovative sports consciousness,
researchers propose that PE should become a kind of lifelong education, and the
teaching model at this stage must be reformed[7]. According to the characteristics and
differences between modern sports and traditional sports, some researchers have
made a comparison from the aspects of innovative teaching environment, inducing
students' original interest, teachers' innovative teaching technology, teaching
evaluation, and extracurricular activities[8-9]. This paper expounds on the design of
an innovative education model in PE from four aspects. In addition, some scholars
pointed out that in the long-term PE teaching, it is necessary to reform the "systematic
learning" mode (traditional teaching mode). Only by combining organically in PE
teaching can we innovate. Some researchers also believe that cultivating students'
innovative spirit, improving the interest of monks, and resonating through students'
innovation, to build a set of general and innovative PE innovation modes in the new
era with innovative function and positive thinking[10].
Based on the development of the deep learning model[11], relevant grounds have
developed explosively[12-13], and these developments have promoted the innovative
development of other industries and other fields. Among them, LSTM is mainly used
for the processing of time series. It can accurately predict the most suitable behavior
mode for users according to the characteristic data with obvious time input by model
users. If the user selects other options, it can make a selection based on the current
prediction[14-15]. However, when choosing an innovation direction, it is often related
to the existing direction, that is, there is a certain opportunity. According to this
characteristic, this study applies the LSTM network to predict the innovation direction
of PE teaching in schools of primary and secondary and determines the
implementation content and path according to the prediction direction.
2. RELATED WORKS
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
73
2.1. RECOMMENDATION ALGORITHM BASED ON DEEP
LEARNING AND ITS APPLICATION
Deep learning technology models can accurately capture attributes or features and
promote them to a higher level of representation[16]. Early such technologies were
limited by the Boltzmann machine (RBM)[17-18]. Hinton et al. Used the Boltzmann
machine for modeling according to the data and optimized the fitting efficiency of the
Boltzmann machine by using the contrast divergence algorithm[19]. The results
showed that the optimized method can be well applied to Netflix. Song et al. proposed
to use the NNM of DNN to extract Netflix user information, which is based on a
recommendation model[20-21].
RNN is commonly used to process sequence data. Hidari[22] uses the neural
network system to take the sequence data of the user's click items in the session
record as the input data of RNN. If the quantity of data is large and concentrated, the
prediction effect of RNN is very accurate. According to the above research, some
researchers[23-24] took the historical behavior of news users as input and used RNN
for a recommendation. The research found that it has good results. Liu et al.[25-26]
used nearly 15 different RNN algorithms to process user information. On this basis,
they found a new deep learning algorithm, which has a two-way RNN structure.
2.2. RECOMMENDATION ALGORITHM AND APPLICATION
BASED ON LSTM
LSTM network is improved based on RNN hidden layer unit and has long-term
memory function. Generally speaking, the problem that RNN can solve is the problem
that LSTM can handle and perform well. At present, the LSTM network is mainly used
for natural language processing, speech recognition, and image understanding.
Graves[27] et al. Took the lead in applying the LSTM network to word prediction. After
training in English and French databases, the accuracy of word prediction is 8%
higher than that of standard RNN. Li et al. [28] proposed a Twitter tag
recommendation system based on the LSTM network. The system first uses the skip-
gram model to generate vocabulary, then uses CNN to generate each sentence in the
article into a sentence vector, and finally uses this sentence vector to train the LSTM
network. The experimental consequences show that the recommendation based on
LSTM achieves better results than the recommendation model of standard RNN and
Gru. A large number of research results show that LSTM network is suitable for time
series information flow modeling.
3. RELATED CONCEPTS
3.1. RNN NETWORK STRUCTURE
RNN is a kind of time recurrent network, which is considered to be the result of
repeated and alternating on the same timeline in a neural network architecture. The
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
74
2.1. RECOMMENDATION ALGORITHM BASED ON DEEP
LEARNING AND ITS APPLICATION
Deep learning technology models can accurately capture attributes or features and
promote them to a higher level of representation[16]. Early such technologies were
limited by the Boltzmann machine (RBM)[17-18]. Hinton et al. Used the Boltzmann
machine for modeling according to the data and optimized the fitting efficiency of the
Boltzmann machine by using the contrast divergence algorithm[19]. The results
showed that the optimized method can be well applied to Netflix. Song et al. proposed
to use the NNM of DNN to extract Netflix user information, which is based on a
recommendation model[20-21].
RNN is commonly used to process sequence data. Hidari[22] uses the neural
network system to take the sequence data of the user's click items in the session
record as the input data of RNN. If the quantity of data is large and concentrated, the
prediction effect of RNN is very accurate. According to the above research, some
researchers[23-24] took the historical behavior of news users as input and used RNN
for a recommendation. The research found that it has good results. Liu et al.[25-26]
used nearly 15 different RNN algorithms to process user information. On this basis,
they found a new deep learning algorithm, which has a two-way RNN structure.
2.2. RECOMMENDATION ALGORITHM AND APPLICATION
BASED ON LSTM
LSTM network is improved based on RNN hidden layer unit and has long-term
memory function. Generally speaking, the problem that RNN can solve is the problem
that LSTM can handle and perform well. At present, the LSTM network is mainly used
for natural language processing, speech recognition, and image understanding.
Graves[27] et al. Took the lead in applying the LSTM network to word prediction. After
training in English and French databases, the accuracy of word prediction is 8%
higher than that of standard RNN. Li et al. [28] proposed a Twitter tag
recommendation system based on the LSTM network. The system first uses the skip-
gram model to generate vocabulary, then uses CNN to generate each sentence in the
article into a sentence vector, and finally uses this sentence vector to train the LSTM
network. The experimental consequences show that the recommendation based on
LSTM achieves better results than the recommendation model of standard RNN and
Gru. A large number of research results show that LSTM network is suitable for time
series information flow modeling.
3. RELATED CONCEPTS
3.1. RNN NETWORK STRUCTURE
RNN is a kind of time recurrent network, which is considered to be the result of
repeated and alternating on the same timeline in a neural network architecture. The
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
structural characteristics of RNN determine that it is more conducive to processing
time-series. RNN structure is shown in Fig.1, where a is the processing unit of the
RNN hidden layer, XT is the input value of the current time, and HT
is the output value
of the current time hidden layer. As can be seen from Fig.1, HT
is determined by the
current input value XT and the output value HT-1 of the previous time. HT
will affect the
output the next time, that is, each output difference is not only related to the current
input value. It is also related to the output difference of the previous time.
Theoretically, RNN can process any length of time series data. Pascanu[29] and
others used detailed digital reasoning to explain the causes of these phenomena, that
is, the traditional RNN mode usually changes according to the correct direction of the
weight at the end of the time series in the training environment. However, the longer
the input time interval, the smaller the impact on the correct change of connection
weight. Therefore, the network system is more inclined to input new data and does not
have the function of long-term memory.
Figure 1. RNN network structure
3.2. LSTM NETWORK STRUCTURE
LSTM completes the problem of gradient disappearance and gradient explosion of
the RNN model and retains information for a long time. LSTM and RNN have similar
network structures, but the structure of the hidden layer is more complex, as exposed
in Fig.2
y
(t-1)
y
(t)
y
(t+1)
L
(t-1)
L
(t)
L
(t+1)
o
(t-1)
o
(t)
o
(t+1)
h
(-)
h
(t-1)
h
(t)
h
(t+1)
h
(-)
x
(t-1)
x
(t)
x
(t+1)
W W W W
U U U
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
75
Figure 2. LSTM Structure neurons
There are three control doors inside the LTSM, which are input gate it, output gate
ot and forgetting gate ft. The input xt at each moment and the output Ht-1
at the
previous moment jointly determine that the state value of each gate unit at the current
time has been the intermediate unit Ct. At time t, the update formula of each door is as
follows [30].
3.3. PREDICTION RESULT OUTPUT LAYER
The output layer of the prediction results is a two-layer full connection layer: each
node of the first layer is connected to all the data features output by the LSTM unit to
realize the integration of local features; Each node of the second layer is connected to
the second layer, and each node is fully connected. The integration feature is
calculated and the predicted value is output. The calculation is shown in formula (7).
4. INNOVATION DIRECTION MODEL BASED ON LSTM
MODEL
The recommendation mode can be divided into input part, processing part, and
output part according to function. The input part converts the user's original education
σ
σ
tanh σ
tanh
C
t- 1
h
t-1
w
f
b
f
w
i
b
i
w
c
b
c
w
a
b
a
i
t
f
t
x
t
O
t
h
t
h
t
C
t
(1)
(2)
(3)
(4)
(5)
(6)
Ct=ft*Ct−1+it*˜
Ct
[ ]
( )
1
,
o o
t t t
o w h x bσ
−
=⋅+
[ ]
( )
1,
f f
t t t
f w h x bσ−
=⋅+
[ ]
( )
1,
i i
t t t
i w h x bσ−
=⋅+
( )
tanh
t t t
h o C=∗
˜
Ci=tanh(wc⋅[ht−1,xt]+
（7）
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
76
Figure 2. LSTM Structure neurons
There are three control doors inside the LTSM, which are input gate it, output gate
ot and forgetting gate ft. The input xt at each moment and the output Ht-1 at the
previous moment jointly determine that the state value of each gate unit at the current
time has been the intermediate unit Ct. At time t, the update formula of each door is as
follows [30].
3.3. PREDICTION RESULT OUTPUT LAYER
The output layer of the prediction results is a two-layer full connection layer: each
node of the first layer is connected to all the data features output by the LSTM unit to
realize the integration of local features; Each node of the second layer is connected to
the second layer, and each node is fully connected. The integration feature is
calculated and the predicted value is output. The calculation is shown in formula (7).
4. INNOVATION DIRECTION MODEL BASED ON LSTM
MODEL
The recommendation mode can be divided into input part, processing part, and
output part according to function. The input part converts the user's original education
σ
σ
tanh σ
tanh
C
t- 1
h
t-1
w
f
b
f
w
i
b
i
w
c
b
c
w
a
b
a
i
t
f
t
x
t
O
t
h
t
h
t
C
t
(1)
(2)
(3)
(4)
(5)
(6)
Ct=ft*Ct−1+it*˜
Ct
[ ]
( )
1
,
o o
t t t
o w h x bσ
−
=⋅+
[ ]
( )
1,
f f
t t t
f w h x bσ−
=⋅+
[ ]
( )
1,
i i
t t t
i w h x bσ−
=⋅+
( )
tanh
t t t
h o C=∗
˜
Ci=tanh(wc⋅[ht−1,xt]+
（7）
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
method into the numerical form required for calculation through the LSTM network,
and the education vector representation used by each user is shown in Fig.3. The
processing part processes the input data through the LSTM network to obtain the
output result. The structure of the LSTM network needs to be determined, including
the number of network layers, time step, and connection settings between layers. We
used the educational method as the number of eigenvalues, defines the dimensions of
input and output data, and determines the structure of the whole LSTM network
model. The softmax layer maps the value of the output vector of the LSTM processing
layer to the (0,1) region. The output part takes the last dimension of the processing
results of the softmax layer to get the final development direction.
Figure 3. Output model
This study constructs an LSTM classification model. It can be used to identify the
index categories of three main sports innovation methods and provide information for
the construction of sports innovation development direction. In this paper, the LSTM
model adopts a three-tier structure. The number of the intermediate network are 70,
50, and 25 respectively. The "CNN" algorithm is used for gradient training and
optimization of network functions. Maxepochs is 70, minibacksize is 30, and the
learning rate is 0.001. For each type of sports innovation direction, 70% of data are
randomly selected as the data of the training set and 30% as data for testing. An
LSTM model is constructed on the MATLAB software platform, and the model is
established and trained by using the deep learning function package. The NVI 7732
processor and NVI 7790 processor equipped with Intel are one set of experiment
environments.
5. MODEL EVALUATION
Compare the predicted classification results of the LSTM model with the actual
labeling results[31] to estimate the presence of the LSTM model. For the model, the
binary classification confusion matrix is calculated[32]. The category recognition
model uses a one-to-many method to define the confusion matrix[33]. Table 1 lists the
"one to many" method of binary classification based on the classical matrix.
Table 1. Confusion matrix of training data set
1
n
r i
t
f w x b
=
⎛ ⎞
−
⎜ ⎟
⎝ ⎠
∑
∑
=
−
n
1t
ir
bxw
22
xw
f
Class IOn class I
Correct prediction as a positive example (TPI) Incorrectly predicted as a counterexample (FNI)
Incorrectly predicted as positive (FPI) Correctly predicted as a counterexample (TNI)
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
77
TPI is the class I positive sample correctly classified by the model; FNI
is the first type
of positive sample of model misclassification; FPI is another class i
samples of model
misclassification; TNI is the class i
other samples correctly classified by the model.
The average accuracy, average accuracy, average recall, average Kappa coefficient,
F1,
and area AUC are calculated to appraise the classification performance of the two
BILSTM models[34], as follows:
Where p0i is the accuracy and N is the total number of records.
The individual leveling rate index is not very accurate. F1
score represents the
effectiveness of classifier recognition positive classification[35]. Kappa coefficient is an
index for conformance testing[36-37].
(8)
(9)
(10)
(11)
(12)
(13)
k
1
0
Kappa / k
1
i i
ii
p pe
pe
=
⎛ ⎞
−
=⎜ ⎟
−
⎝ ⎠
∑
k
1
/ k
i i
ii i i i
TP TN
AA
TP TN FP FN
=
⎛ ⎞
+
=⎜ ⎟
+ + +
⎝ ⎠
∑
k
1
/ k
i
ii i
TP
AP
TP FP
=
⎛ ⎞
=⎜ ⎟
+
⎝ ⎠
∑
k
1
/ k
i
ii i
TP
AR
TP FN
=
⎛ ⎞
=⎜ ⎟
+
⎝ ⎠
∑
1
2
score AP AR
F
AP AR
⋅
=+
k k
1 1
1/ k / k
2
i i
i i
i i i i
TP TN
AUC
TP FN TN FP
= =
⎛ ⎞
= +
⎜ ⎟
+ +
⎝ ⎠
∑ ∑
(14)
(15)
(16)
(17)
1
2
macro P marco R
macro F
macro P marco R
× − × −
−=−+−
1
1
n
i
i
macro P P
n
=
−=∑
accuracy
T
N
E
N
=
1
1
n
i
i
macro R R
n
=
−=
∑
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
78
TPI is the class I positive sample correctly classified by the model; FNI is the first type
of positive sample of model misclassification; FPI is another class i samples of model
misclassification; TNI is the class i other samples correctly classified by the model.
The average accuracy, average accuracy, average recall, average Kappa coefficient,
F1, and area AUC are calculated to appraise the classification performance of the two
BILSTM models[34], as follows:
Where p0i is the accuracy and N is the total number of records.
The individual leveling rate index is not very accurate. F1 score represents the
effectiveness of classifier recognition positive classification[35]. Kappa coefficient is an
index for conformance testing[36-37].
(8)
(9)
(10)
(11)
(12)
(13)
k
1
0
Kappa / k
1
i i
ii
p pe
pe
=
⎛ ⎞
−
=⎜ ⎟
−
⎝ ⎠
∑
k
1
/ k
i i
ii i i i
TP TN
AA
TP TN FP FN
=
⎛ ⎞
+
=⎜ ⎟
+ + +
⎝ ⎠
∑
k
1
/ k
i
ii i
TP
AP
TP FP
=
⎛ ⎞
=⎜ ⎟
+
⎝ ⎠
∑
k
1
/ k
i
ii i
TP
AR
TP FN
=
⎛ ⎞
=⎜ ⎟
+
⎝ ⎠
∑
1
2
score AP AR
F
AP AR
⋅
=+
k k
1 1
1/ k / k
2
i i
i i
i i i i
TP TN
AUC
TP FN TN FP
= =
⎛ ⎞
= +
⎜ ⎟
+ +
⎝ ⎠
∑ ∑
(14)
(15)
(16)
(17)
1
2
macro P marco R
macro F
macro P marco R
× − × −
−=−+−
1
1
n
i
i
macro P P
n
=
−=∑
accuracy
T
N
E
N
=
1
1
n
i
i
macro R R
n
=
−=
∑
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
Where NT is the number of environments, N is the total index, gi is the classification
result, and pi is the classification result predicted by the model[38-39].
6. RESULT ANALYSIS
6.1. BASELINE MODEL PARAMETER SETTING
To prove the efficiency of the projected model, we choose some machine learning
methods based on artificial feature engineering to extract features. These methods
are often used in decision prediction, Including DT, NB, LDA, LR, SVM, GBDT, and
RF. In addition, we also selected two prediction models based on deep learning CNN
and CNN RNN as comparison methods. The settings of some baseline model
parameters are shown in Table 2, and the other model parameters without initial
values are the default values.
Table 2. Setting of baseline model parameters
We could see from Table 3 that the model proposed in this study is 0.0532 ~
0.2323 higher than the baseline model in terms of accuracy, precision, F1, and AUC,
and 0.0422 higher than the most competitive CNN-RNN on average. Among the
traditional machine learning algorithms, GBDT and RF have the best average
performance on the five evaluation indexes, because GBDT and RF are classifiers
based on the idea of decision tree integration, and the final result is determined by
multiple trees. Better prediction. In the application model of deep learning, CNN-RNN
has better performance than CNN, because the CNN-RNN model can not only obtain
the locally relevant information between learning behaviors but also capture the time
relationship between learning behaviors. To a great extent, it captures potentially
important information and improves prediction accuracy. Compared with the deep
learning models CNN-RNN and CNN, our CLNN model performs better in the
prediction of sports innovation direction. This is mainly because the LSTM model can
effectively solve long-standing problems. The simple recurrent neural network (RNN)
can make the error transfer through the time and layer gate mechanism when the
level is more constant, allowing the periodic network to learn multiple time steps,
establish the long-term cause-effect relationship, and expand the prediction
presentation of the model.
(18)
( )
2
1
RMSE
N
i i
i
g p
EN
=
−
=
∑
Baseline method Baseline parameters Value
SVM C, y C=1,y=1/210
DT criterion gini'
GBDT n_estimator 500
RF n_estimator 500
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
79
Table 3. Performance of different models on different evaluation indicators
6.2. INNOVATIVE DEVELOPMENT DIRECTION OF PE
Through the performance of different models on different evaluation indicators, it is
found that CNN has better accuracy than other models, but its decision-making
accuracy is low. Therefore, CNN-LSTM used in this study is the CLNN model. CNN-
RNN is optimized to improve the accuracy of prediction and decision-making. This
study selects five systems, including the teaching thought system, teaching process
nature, and main goal system, teaching content system, teaching evaluation system,
and sports text introduction system. According to the parameter comparison of model
evaluation indicators in Section 6.1, CNN-LSTM is selected for prediction, and finally,
three innovative development directions are determined: innovative teaching concept,
innovative nature and main objectives of PE teaching process, and innovative
teaching content system.
6.2.1. INNOVATION OF TEACHING IDEAS
Establish the educational thought of "seeking knowledge and innovation" and
"health first" facing the future. The thought of "health first" emphasizes the content and
methodology of the combination of PE and health education, closely combines the
thought of "health first" with the construction of PE discipline and expands the benefits
of maintaining and promoting health. In the field of lifelong PE, we should clarify the
special role of PE in quality service education and give new directions to the teaching
content. PE in PSS should comprehensively promote quality education, and establish
the guiding ideology of PE teaching of "seeking knowledge and innovation" and
"health first" so that students can master basic physical skills and form a good habit of
adhering to physical exercise. physical exercise. PE reform has also changed our
thinking set of taking PE as the educational carrier and education as the goal, re-
understand the goal, function, content, means, and methods of PE, and building a
new PE teaching system for PSS in the 21st century.
Method Accuracy Precision Recall F1 AUC
DT 0.8394 0.8506 0.9666 0.9049 0.6636
NB 0.8388 0.8800 0.9217 0.9004 0.7241
GBDT 0.8792 0.8887 0.9615 0.9237 0.7694
LR 0.8542 0.8583 0.9768 0.9137 0.6848
RF 0.8622 0.8727 0.9668 0.9172 0.7174
SVM 0.8620 0.8977 0.9422 0.9194 0.7689
CNN 0.8724 0.8717 0.9678 0.9224 0.7156
CNN-RNN 0.8562 0.8932 0.9528 0.9241 0.7601
CLNN 0.9363 0.8862 0.9624 0.9602 0.8870
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
80
Table 3. Performance of different models on different evaluation indicators
6.2. INNOVATIVE DEVELOPMENT DIRECTION OF PE
Through the performance of different models on different evaluation indicators, it is
found that CNN has better accuracy than other models, but its decision-making
accuracy is low. Therefore, CNN-LSTM used in this study is the CLNN model. CNN-
RNN is optimized to improve the accuracy of prediction and decision-making. This
study selects five systems, including the teaching thought system, teaching process
nature, and main goal system, teaching content system, teaching evaluation system,
and sports text introduction system. According to the parameter comparison of model
evaluation indicators in Section 6.1, CNN-LSTM is selected for prediction, and finally,
three innovative development directions are determined: innovative teaching concept,
innovative nature and main objectives of PE teaching process, and innovative
teaching content system.
6.2.1. INNOVATION OF TEACHING IDEAS
Establish the educational thought of "seeking knowledge and innovation" and
"health first" facing the future. The thought of "health first" emphasizes the content and
methodology of the combination of PE and health education, closely combines the
thought of "health first" with the construction of PE discipline and expands the benefits
of maintaining and promoting health. In the field of lifelong PE, we should clarify the
special role of PE in quality service education and give new directions to the teaching
content. PE in PSS should comprehensively promote quality education, and establish
the guiding ideology of PE teaching of "seeking knowledge and innovation" and
"health first" so that students can master basic physical skills and form a good habit of
adhering to physical exercise. physical exercise. PE reform has also changed our
thinking set of taking PE as the educational carrier and education as the goal, re-
understand the goal, function, content, means, and methods of PE, and building a
new PE teaching system for PSS in the 21st century.
Method
Accuracy
Precision
Recall
F1
AUC
DT
0.8394
0.8506
0.9666
0.9049
0.6636
NB
0.8388
0.8800
0.9217
0.9004
0.7241
GBDT
0.8792
0.8887
0.9615
0.9237
0.7694
LR
0.8542
0.8583
0.9768
0.9137
0.6848
RF
0.8622
0.8727
0.9668
0.9172
0.7174
SVM
0.8620
0.8977
0.9422
0.9194
0.7689
CNN
0.8724
0.8717
0.9678
0.9224
0.7156
CNN-RNN
0.8562
0.8932
0.9528
0.9241
0.7601
CLNN
0.9363
0.8862
0.9624
0.9602
0.8870
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
6.2.2. ESSENCE OF PE TEACHING PROCESS AND INNOVATION
OF MAIN OBJECTIVES
PE is not equal to physical exercise. PE is not fitness. PE alone cannot solve the
problem of strengthening the physique. At present, we should combine imparting
knowledge and skills with cultivating consciousness, skills, and habits, and pay
attention to cultivating students' self-learning and self-habit consciousness, to make
students make achievements in PE and lay the foundation of lifelong PE. At the same
time, PE should recognize the transformation from "the main purpose of PE is to
improve physique" to "health first", and establish the main objectives of PE in PSS: (1)
make students have a basic understanding and positive attitude towards PE,
understand the original intention of physical exercise, and establish a healthy concept
of physical exercise; (2) Master knowledge and correct methods of fitness, and be
able to exercise regularly by using a variety of basic sports skills and fitness methods;
(3) Exercise independence and the habit of peaceful coexistence.
6.2.3. INNOVATION OF TEACHING CONTENT SYSTEM
Establish the educational content and curriculum system of PSS in China in the
21st century, and strive to form a diversified and comprehensive PE content system. It
includes: (1) the combination of PE thought education and physical exercise
education; (2) The innovation of primary and secondary school teachers in the
teaching methods and contents of PE and the innovation of the PE system; (3) The
innovative evaluation system of primary and secondary school sports is constantly
changing with the development of the times.
7. CONCLUSION
This deep learning technology can extract the attributes or features of the data and
abstract them into higher-level representations, and use them to predict the direction
of PE innovation, which has long-term guiding significance. This research predicts the
direction of PE teaching in PSS based on the LSTM model. First, it analyzes and
identifies the innovation direction data set, and then uses the LSTM model to extract
text context features from both forward and backward directions to predict the
innovation direction and implementation path of PE. Finally, the basic direction of the
innovation and development of PE is determined, the research is carried out and the
following conclusions are drawn: (1) The LSTM model proposed in this study is
0.0532 higher than the baseline model in the four evaluation indicators of Accuracy,
Precision, F1, and AUC. ~0.2323, indicating that the LSTM model has excellent
prediction accuracy and effect; (2) The PE system is first of all completely dynamic,
and the development direction of PE innovation is essentially the direction jointly
selected by teachers, students, and schools. PE teaching innovation in PSS should be
carried out from three aspects: teaching thought, teaching content, teaching goal, and
essence.
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
81
8. CONFLICT OF INTEREST
The authors declared that there is no conflict of interest.
REFERENCES
(1) Al-Hazzaa, H. M., Abahussain, N. A., Al-Sobayel, H. I., Qahwaji, D. M., &
Musaiger, A. O. (2011). Physical activity, sedentary behaviors and dietary
habits among Saudi adolescents relative to age, gender and region. The
International Journal of Behavioral Nutrition and Physical Activity, 8. https://
doi.org/10.1186/1479-5868-8-140.
(2) A, C. W., A, X. W., Jz, A., Liang, Z. A., Xiao, B. A., Xin, N. B.,... Ehd, A. (2021).
Uncertainty Estimation for Stereo Matching Based on Evidential Deep
Learning. https://doi.org/10.1016/j.patcog.2021.108498
(3) Aubert, S. (2020). Active Healthy Kids Global Alliance "Global Matrix"
Initiative: Process, Results, Impact and Evaluation. https://doi.org/10.20381/
ruor-24426.
(4) Anthamatten, P., Brink, L., Lampe, S., Greenwood, E., Kingston, B., & Nigg, C.
(2011). An assessment of schoolyard renovation strategies to encourage
children's physical activity. International Journal of Behavioral Nutrition &
Physical Activity, 8(1), 27. https://doi.org/10.1186/1479-5868-8-27.
(5) Aguilar-Farias, N., Cortinez-O'Ryan, A., Sadarangani, K. P., Oetinger, A. V., &
Cristi-Montero, C. (2016). Results From Chile's 2016 Report Card on
Physical Activity for Children and Youth. Journal of Physical Activity and
Health, 13(2), S117-S123. https://doi.org/10.1123/jpah.2016-0314.
(6) Russell, R. D. (1975). Health Education. Project of Joint Committee on Health
Problems in Education of the National Education Association and the American
Medical Association. communication quarterly.
(7) Miao, J., Wang, Z., Ning, X., Xiao, N., Cai, W., & Liu, R. (2022). Practical and
secure multifactor authentication protocol for autonomous vehicles in 5G.
Software: Practice and Experience. https://doi.org//10.1002/SPE.3087
(8) Ayers, W. (2018). The Shifting Ground of Curriculum Thought and Everyday
Practice: Thinking About Schools.
(9) Ning, X., Gong, K., Li, W., & Zhang, L. (2021). JWSAA: joint weak saliency
and attention aware for person re-identification. Neurocomputing, 453,
801-811. https://doi.org/10.1016/j.neucom.2020.05.106
(10) Heyneman, S. P., & Lee, B. (2016). International organizations and the future
of education assistance. International Journal of Educational Development,
48(3), 9-22. https://doi.org/10.1016/j.ijedudev.2015.11.009.
(11) Hao, X., Zhang, G., & Ma, S. (2016). Deep Learning. International Journal of
Semantic Computing, 10(03), 417-439. https://doi.org/10.1142/
S1793351X16500045
(12) Qiu, L., Liu, Y., Hu, Q., & Liu, Y. (2018). Student dropout prediction in
massive open online courses by convolutional neural networks. Soft
Computing, 23. https://doi.org/10.1007/s00500-018-3581-3
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
82
8. CONFLICT OF INTEREST
The authors declared that there is no conflict of interest.
REFERENCES
(1) Al-Hazzaa, H. M., Abahussain, N. A., Al-Sobayel, H. I., Qahwaji, D. M., &
Musaiger, A. O. (2011). Physical activity, sedentary behaviors and dietary
habits among Saudi adolescents relative to age, gender and region. The
International Journal of Behavioral Nutrition and Physical Activity, 8. https://
doi.org/10.1186/1479-5868-8-140.
(2) A, C. W., A, X. W., Jz, A., Liang, Z. A., Xiao, B. A., Xin, N. B.,... Ehd, A. (2021).
Uncertainty Estimation for Stereo Matching Based on Evidential Deep
Learning. https://doi.org/10.1016/j.patcog.2021.108498
(3) Aubert, S. (2020). Active Healthy Kids Global Alliance "Global Matrix"
Initiative: Process, Results, Impact and Evaluation. https://doi.org/10.20381/
ruor-24426.
(4) Anthamatten, P., Brink, L., Lampe, S., Greenwood, E., Kingston, B., & Nigg, C.
(2011). An assessment of schoolyard renovation strategies to encourage
children's physical activity. International Journal of Behavioral Nutrition &
Physical Activity, 8(1), 27. https://doi.org/10.1186/1479-5868-8-27.
(5) Aguilar-Farias, N., Cortinez-O'Ryan, A., Sadarangani, K. P., Oetinger, A. V., &
Cristi-Montero, C. (2016). Results From Chile's 2016 Report Card on
Physical Activity for Children and Youth. Journal of Physical Activity and
Health, 13(2), S117-S123. https://doi.org/10.1123/jpah.2016-0314.
(6) Russell, R. D. (1975). Health Education. Project of Joint Committee on Health
Problems in Education of the National Education Association and the American
Medical Association. communication quarterly.
(7) Miao, J., Wang, Z., Ning, X., Xiao, N., Cai, W., & Liu, R. (2022). Practical and
secure multifactor authentication protocol for autonomous vehicles in 5G.
Software: Practice and Experience. https://doi.org//10.1002/SPE.3087
(8) Ayers, W. (2018). The Shifting Ground of Curriculum Thought and Everyday
Practice: Thinking About Schools.
(9) Ning, X., Gong, K., Li, W., & Zhang, L. (2021). JWSAA: joint weak saliency
and attention aware for person re-identification. Neurocomputing, 453,
801-811. https://doi.org/10.1016/j.neucom.2020.05.106
(10) Heyneman, S. P., & Lee, B. (2016). International organizations and the future
of education assistance. International Journal of Educational Development,
48(3), 9-22. https://doi.org/10.1016/j.ijedudev.2015.11.009.
(11) Hao, X., Zhang, G., & Ma, S. (2016). Deep Learning. International Journal of
Semantic Computing, 10(03), 417-439. https://doi.org/10.1142/
S1793351X16500045
(12) Qiu, L., Liu, Y., Hu, Q., & Liu, Y. (2018). Student dropout prediction in
massive open online courses by convolutional neural networks. Soft
Computing, 23. https://doi.org/10.1007/s00500-018-3581-3
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
(13) Yang, B., Lei, Y., Liu, J., & Li, W. (2016). Social Collaborative Filtering by
Trust. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1-1.
https://doi.org/10.1109/tpami.2016.2605085
(14) Naeen, H. M., & Jalali, M. (2019). A decentralized trust-aware collaborative
filtering recommender system based on weighted items for social tagging
systems. https://doi.org/10.48550/arXiv.1906.05143
(15) Yan, C., Pang, G., Bai, X., Liu, C., Xin, N., Gu, L., & Zhou, J. (2021). Beyond
triplet loss: person re-identification with fine-grained difference-aware
pairwise loss. IEEE Transactions on Multimedia. https://doi.org/10.1109/
TMM.2021.3069562
(16) Khatun, A., Denman, S., Sridharan, S., & Fookes, C. (2020). Joint
identification-verification for person re-identification: A four stream deep
learning approach with improved quartet loss function. Computer Vision
and Image Understanding, 197-198, 102989. https://doi.org/10.1016/
j.cviu.2020.102989
(17) Cai, W., Zhai, B., Liu, Y., Liu, R., & Ning, X. (2021). Quadratic polynomial
guided fuzzy C-means and dual attention mechanism for medical image
segmentation. Displays, 70, 102106. https://doi.org/10.1016/
j.displa.2021.102106
(18) Roch, S. (2011). Phase Transition in Distance-Based Phylogeny
Reconstruction. Computer Science. https://doi.org/10.48550/arXiv.1108.5781
(19) RUMELHART, D. E., Hinton, G. E., & Williams, R. J. (1988). Learning Internal
Representations by Error Propagation: Readings in Cognitive ence. https://
doi.org/10.1007/978-0-387-39940-9_3246
(20) Theodoros K. Dikaliotis, Tracey Ho, Sidharth Jaggi, Svitlana Vyetrenko, Hongyi
Yao, Michelle Effros, Jörg Kliewer, Elona Erez. (2010). Multiple-access
Network Information-flow and Correction Codes. IEEE Transactions on
Information Theory, 57(2), 1067-1079, https://doi.org/10.1109/TIT.2010.2095130.
(21) Ning, X., Duan, P., Li, W., & Zhang, S. (2020). Real-time 3D face alignment
using an encoder-decoder network with an efficient deconvolution layer.
IEEE Signal Processing Letters, 27, 1944-1948. https://doi.org/10.1109/
LSP.2020.3032277
(22) Hidasi, B., Karatzoglou, A., Baltrunas, L., & Tikk, D. (2015). Session-based
Recommendations with Recurrent Neural Networks. Computer ence. https://
doi.org/10.48550/arXiv.1511.06939
(23) Haruki Okamura, Hiroko Tsutsui, Toshinori Komatsu, Masuo Yutsudo, Akira
Hakura, Tadao Tanimoto, Kakuji Torigoe, Takanori Okura, Yoshiyuki Nukada,
Kazuko Hattori, Kenji Akita, Motoshi Namba, Fujimi Tanabe, Kaori Konishi,
Shigeharu Fukuda & Masashi Kurimoto. (1995). Cloning of a new cytokine
that induces IFN-gamma production by T cells. Nature, 378(6552), 88.
https://doi.org/10.1038/378088a0
(24) Shan, W. (2022). Digital streaming media distribution and transmission
process optimisation based on adaptive recurrent neural network.
Connection Science, 34(1), 1169-1180. https://doi.org/
10.1080/09540091.2022.2052264
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
83
(25) Szegedy, C., Liu, W., Jia, Y., Sermanet, P., & Rabinovich, A. (2014). Going
Deeper with Convolutions. IEEE Computer Society. https://doi.org/10.1109/
CVPR.2015.7298594
(26) Ying, L., Nan, Z. Q., Ping, W. F., Kiang, C. T., Pang, L. K., Chang, Z. H.,... Nam,
L. Adaptive weights learning in CNN feature fusion for crime scene
investigation image classification. Connection Science. https://doi.org/
10.1080/09540091.2021.1875987
(27) Graves, A. (2012). Long Short-Term Memory. Springer Berlin Heidelberg.
https://doi.org/10.1007/978-3-642-24797-2_4
(28) Li, W., & Abiad, V. (1990). Institutions, Institutional Change and Economic
Performance. Social Science Electronic Publishing, 18(1), 142-144. https://
doi.org/10.1017/CBO9780511808678
(29) Pascanu, R., Mikolov, T., & Bengio, Y. (2012). On the difficulty of training
Recurrent Neural Networks. JMLR.org. https://doi.org/10.1007/
s12088-011-0245-8
(30) Le, Q. V., & Mikolov, T. (2014). Distributed Representations of Sentences and
Documents. arXiv e-prints.
(31) Zhou, Y., Daamen, W., Vellinga, T., & Hoogendoorn, S. P. (2019). Ship
classification based on ship behavior clustering from AIS data. Ocean
Engineering, 175(MAR.1), 176-187. https://doi.org/10.1016/
j.oceaneng.2019.02.005.
(32) Meng, W., Wei, Y., Liu, P., Zhu, Z., & Yin, H. (2019). Aspect Based Sentiment
Analysis With Feature Enhanced Attention CNN-BiLSTM. IEEE Access, 7,
167240-167249. https://doi.org/10.1109/ACCESS.2019.2952888.
(33) Yu, S., Liu, D., Zhu, W., Zhang, Y., & Zhao, S. (2020). Attention-based LSTM,
GRU and CNN for short text classification. Journal of Intelligent and Fuzzy
Systems, 39(1), 1-8. https://doi.org/10.3233/JIFS-191171.
(34) Ma, J., Jia, C., Yang, X., Cheng, X., & Zhang, C. (2020). A Data-Driven
Approach for Collision Risk Early Warning in Vessel Encounter Situations
Using Attention-BiLSTM. IEEE Access, 8, 188771-188783. https://doi.org/
10.1109/ACCESS.2020.3031722.
(35) Ma, X., & Hovy, E. (2016). End-to-end Sequence Labeling via Bi-directional
LSTM-CNNs-CRF. Paper presented at the Proceedings of the 54th Annual
Meeting of the Association for Computational Linguistics, 1. https://doi.org/
10.13140/RG.2.1.2182.5685.
(36) Chen, G., Lin, H., Hu, H., Yan, Y., Wan, Y., Xiao, T., & Peng, Y. (2022). Research
on the Measurement of Ship's Tank Capacity Based on the Monte Carlo
Method. Chemistry and Technology of Fuels and Oils, 58(1), 232-236. https://
doi.org/10.1007/s10553-022-01371-x.
(37) Sheng, K., Liu, Z., Zhou, D., He, A., & Feng, C. (2017). Research on Ship
Classification Based on Trajectory Features. Journal of Navigation, 1-17.
https://doi.org/10.1017/S0373463317000546.
(38) Dewani, A., Memon, M. A., y Bhatti, S. (2021). Development of computational
linguistic resources for automated detection of textual cyberbullying
threats in Roman Urdu language. 3C TIC. Cuadernos de desarrollo aplicados
a las TIC, 10(2), 101-121. https://doi.org/10.17993/3ctic.2021.102.101-121
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
84
(25) Szegedy, C., Liu, W., Jia, Y., Sermanet, P., & Rabinovich, A. (2014). Going
Deeper with Convolutions. IEEE Computer Society. https://doi.org/10.1109/
CVPR.2015.7298594
(26) Ying, L., Nan, Z. Q., Ping, W. F., Kiang, C. T., Pang, L. K., Chang, Z. H.,... Nam,
L. Adaptive weights learning in CNN feature fusion for crime scene
investigation image classification. Connection Science. https://doi.org/
10.1080/09540091.2021.1875987
(27) Graves, A. (2012). Long Short-Term Memory. Springer Berlin Heidelberg.
https://doi.org/10.1007/978-3-642-24797-2_4
(28) Li, W., & Abiad, V. (1990). Institutions, Institutional Change and Economic
Performance. Social Science Electronic Publishing, 18(1), 142-144. https://
doi.org/10.1017/CBO9780511808678
(29) Pascanu, R., Mikolov, T., & Bengio, Y. (2012). On the difficulty of training
Recurrent Neural Networks. JMLR.org. https://doi.org/10.1007/
s12088-011-0245-8
(30) Le, Q. V., & Mikolov, T. (2014). Distributed Representations of Sentences and
Documents. arXiv e-prints.
(31) Zhou, Y., Daamen, W., Vellinga, T., & Hoogendoorn, S. P. (2019). Ship
classification based on ship behavior clustering from AIS data. Ocean
Engineering, 175(MAR.1), 176-187. https://doi.org/10.1016/
j.oceaneng.2019.02.005.
(32) Meng, W., Wei, Y., Liu, P., Zhu, Z., & Yin, H. (2019). Aspect Based Sentiment
Analysis With Feature Enhanced Attention CNN-BiLSTM. IEEE Access, 7,
167240-167249. https://doi.org/10.1109/ACCESS.2019.2952888.
(33) Yu, S., Liu, D., Zhu, W., Zhang, Y., & Zhao, S. (2020). Attention-based LSTM,
GRU and CNN for short text classification. Journal of Intelligent and Fuzzy
Systems, 39(1), 1-8. https://doi.org/10.3233/JIFS-191171.
(34) Ma, J., Jia, C., Yang, X., Cheng, X., & Zhang, C. (2020). A Data-Driven
Approach for Collision Risk Early Warning in Vessel Encounter Situations
Using Attention-BiLSTM. IEEE Access, 8, 188771-188783. https://doi.org/
10.1109/ACCESS.2020.3031722.
(35) Ma, X., & Hovy, E. (2016). End-to-end Sequence Labeling via Bi-directional
LSTM-CNNs-CRF. Paper presented at the Proceedings of the 54th Annual
Meeting of the Association for Computational Linguistics, 1. https://doi.org/
10.13140/RG.2.1.2182.5685.
(36) Chen, G., Lin, H., Hu, H., Yan, Y., Wan, Y., Xiao, T., & Peng, Y. (2022). Research
on the Measurement of Ship's Tank Capacity Based on the Monte Carlo
Method. Chemistry and Technology of Fuels and Oils, 58(1), 232-236. https://
doi.org/10.1007/s10553-022-01371-x.
(37) Sheng, K., Liu, Z., Zhou, D., He, A., & Feng, C. (2017). Research on Ship
Classification Based on Trajectory Features. Journal of Navigation, 1-17.
https://doi.org/10.1017/S0373463317000546.
(38) Dewani, A., Memon, M. A., y Bhatti, S. (2021). Development of computational
linguistic resources for automated detection of textual cyberbullying
threats in Roman Urdu language. 3C TIC. Cuadernos de desarrollo aplicados
a las TIC, 10(2), 101-121. https://doi.org/10.17993/3ctic.2021.102.101-121
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
(39) Hong Chen,Xuefei Peng & Caiyun Chen. (2022). Corporate social
responsibility fulfilment, product-market competition and debt risk:
Evidence from China. Applied Mathematics and Nonlinear Sciences, 7(2),
757-772. https://doi.org/10.2478/AMNS.2021.2.00163.
https://doi.org/10.17993/3ctecno.2023.v12n1e43.70-85
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254-4143
Ed.43 | Iss.12 | N.1 January - March 2023
85
