Design and Optimization Path of Interactive Intelligent Learning Platform in Physical Education Teaching Informatization System in Colleges and Universities

With the rapid development of information technology, sports in colleges and universities also ushered in the wave of informationization. Informatization system of college physical education is not only the demand of the times, but also an important way to improve the quality of college physical education teaching and to promote the physical and mental health development of students [1]. At the same time, in order to adapt to the needs of society, as a college physical education, should be coordinated with moral education so that students learn the basic techniques of sports, improve the basic skills of sports at the same time, learn to work together with their peers around them, and take this education as a main line throughout [2]. Therefore, to cultivate students’ spirit of solidarity and cooperation and mutual help has become a focus of daily teaching, interactive learning is increasingly important, fully embodies the students as the main body of the teaching, the teacher as the leading, emphasizing teacher-student interaction, student interaction, is to cultivate students’ spirit of cooperation in a good mode [3-4].

Interactive learning in physical education is mainly reflected in the students’ right to choose their own learning programs, teachers and students to discuss and solve the difficulties encountered in learning. In the teacher-student interaction, teachers teach students according to their abilities, so that students can adjust their learning progress according to their own learning process and learning methods, and form a good mutual help relationship with classmates, indirectly promoting collaborative learning among classmates [5]. For this reason, interactive learning platform is becoming a hot topic in the field of education. It integrates the Internet, artificial intelligence and education technology, which can help students learn better and also assist teachers to educate better [6-8]. Interactive learning platform is a brand new educational technology, which is different from traditional education. The platform can not only better meet the learning needs of students, but also improve the quality of teaching, teaching efficiency, collaboration ability, and interaction effect [9-10]. It can provide corresponding learning materials, such as videos and exercises, according to students’ different needs, levels, and interests, and recommend them to students through the learning management system to realize effective online and offline interactive collaboration [11-12]. At the same time, teachers can formulate appropriate assessments and recommendations based on students’ performance, thus realizing personalized education [13].

However, the existing e-learning platforms generally have problems such as heavy content construction and light application depth, poor integration of information technology and teaching, poor interactivity, and poor student learning autonomy [14-16]. The reason for this is that the platform is often based on artificial design rules for students’ learning feedback, single mode, inaccurate content recommendation, which can not effectively enhance students’ learning initiative, resulting in a waste of resources. Aiming at the above problems, the paper designs an interactive intelligent learning platform based on the university sports teaching information system and its optimization path.

The study designs an interactive online teaching resource management platform based on deep learning, and then proposes a machine reading comprehension model, which is mainly divided into five layers, including input embedding layer, embedding coding layer, article-question focus layer, model coding layer, and output layer. The model is used to build an intelligent Q&A interactive system, and an intelligent Q&A system based on deep learning interactive has been established. The learners visit the system, ask questions, and search by using the system, and the system provides feedback answers for the learners by using the knowledge base, teaching materials, and forum data to help the learners to answer the questions, which improves the motivation of learning and improves the learning effect.

2

Interactive Intelligent Learning Platform Design for College Sports

2.1

Interactive Intelligent Learning Platform

2.1.1

Design of the overall architecture of the platform

Based on deep learning and NET as the development platform, the interactive online teaching resource management platform has been designed, and the structure is shown in Fig. 1.

2.1.2

Database design

1)

Content Collection

The prerequisite for building an interactive online teaching resource management platform is to gather the content needed by the platform.

Collect the basic data of students and interactive teaching resources, construct a database, clarify the group teaching task, collect that task at the same time, and precisely analyze the platform content through the literature analysis method, the process is as follows: (1) $G_{x z d} = \frac{b}{E_{r} \times g}$

Among them, the content of the teaching platform is G_xzd teaching content is g ; the basic information of students is E_r.

Utilize formula (1) to collect the required content of the platform, take it as the premise, formulate relevant teaching objectives and design learning objectives according to the actual situation of students.

2)

Construct database

Taking the content collection and teaching objectives as premise, the platform database is designed to store internally the collected content, with both search and use functions. The interactive object of the module is students, teachers and administrators, when entering the module, users need to apply for access first, the module will audit the relevant ID cards, in the case of failure to match the user’s resources, the module automatically exits; after the audit, the user enters the corresponding interface according to their needs, the module according to the user’s needs, to provide them with the required interface.

2.2

Automated Grouping Module for Interactive Teaching Resources

The grouping process of the Interactive Teaching Resources Automation Grouping Module is shown in Figure 2.

Grouping is achieved by automated grouping algorithm with the following steps:

Step 1: Clean the data with the following formula: (2) $D = \sum_{e}^{o} i^{*} \frac{m}{f + n}$

Where the student data is D; each student indicator is $\frac{m}{f + n}$ ; and the cleaning factor is $\sum_{e}^{o} i$ .

Step 2: According to the gap of $\frac{m}{f + n}$ , analyze the basic differences, formulate the mixed-mode group of men and women, so that the performance of each group has a balanced nature, and measure the similarity of the comprehensive quality of each student through the cosine similarity algorithm to improve the accuracy of the grouping, so that M₁,M₂,⋯,M_r and N₁,N₂,⋯,N_r are the indicators of 2 students, then the cosine similarity is as follows: (3) $\cos θ = {\begin{array}{l} \frac{n * r}{\sqrt{N_{1}, N_{2}, \dots, N_{r}}} \\ \frac{m * r}{M_{1}, M_{2}, \dots, M_{r}} \end{array}$

Where the indicator number is r; the number of indicators is m,n. The value interval of cosθ is [0, 1], if cosθ ≈ 1, it means that M and N are more similar; if cosθ ≈ 0, it means that the gap between M and N is larger, so it can be divided into the same group to make up for their respective shortcomings, the similarity matrix is as follows: (4) $[\begin{matrix} \cos θ_{1} & \cos θ_{2} & \dots & \cos θ_{n} \\ \cos θ_{11} & \cos θ_{12} & \dots & \cos θ_{1 n} \\ ⋮ & ⋮ & ⋮ \\ \cos θ_{m 1} & \cos θ_{m 2} & \dots & \cos θ_{m n} \end{matrix}]$

Step 3: Sort the rows (columns) within the matrix, obtain the value with the least similarity within each row (column), expand the sorting according to the comprehensive ability of the students, so that the student with a slightly stronger ability is the leader of the group, and the value of the comprehensive ability of the students is calculated by the following formula: (5) $Q = \sqrt{X_{t} + X_{y} + \dots + X_{n}}$

Sort the Q-values in descending order to realize the election of group leaders.

Step 4: Introduce the automated grouping algorithm to accomplish the automated grouping with the following formula: (6) $M = \log (\frac{v}{v c * b})$

Where the grouping factor is M ; the similarity ordering is $\frac{v}{v c * b}$ .

The students with the smallest similarity are assigned to the corresponding groups to realize the automated grouping function.

2.3

Technology for building interactive intelligent question and answer modules for sports

Deep learning [17] is utilized to simulate the learning ability of the human brain to design an interactive intelligent question and answer module. The module focuses on emphasizing knowledge feedback learning, controlling the variables of the learning algorithm by setting parameter weights, and obtaining training data through the user’s basic information in the database.

The active classification of questions and answers is more important within the interactive intelligent Q&A module, which not only clarifies the user’s request, but also understands the topic characteristics of the database, provides the user with question-answer retrieval functions, reduces the difficulty of retrieval, and shortens the time of information screening.

3

Design and implementation of sports quiz based on machine reading comprehension

3.1

Data Segmentation and Preprocessing

3.1.1

Data Segmentation Processing

Model for data processing is first of all the participle operation, this paper calls jieba participle and pkuseg participle to the same paragraph for participle operation, compare the results of jieba participle and pkuseg participle, the role of the participleizer is used to analyze each sentence, and the selected data is the book “Physical Education Curriculum and Teaching Theory”.

3.1.2

Data loading

In the actual operation of natural language processing, the amount of data read is often very large, taking into account the limited computer memory and IO speed is very slow, so when loading the data can not be loaded into memory all at once, you need to consider multi-threaded or iterative loading, which needs to be based on the actual situation to define a loader.

Loader needs to be an iterable object, internal configuration parameters iter, by calling iter can return an iterator.

3.1.3

Article Preprocessing

1) Firstly, the data are cleaned and filtering operations such as de-stopping words are performed.

2) According to the TF-IDF formula, get the TF-IDF vectors of questions and articles.

3) Multiply the TF-IDF vectors of questions and articles, and select the five vectors with the largest values from them, and the corresponding articles are the most relevant text contents with questions in the final output.

3.1.4

Labeling answers

In this paper, it is assumed that the answer corresponds to a certain span in the text, and the task of machine reading comprehension is to predict this span. Although each Q&A pair in the training set has a paragraph, a question, and a corresponding answer, the location of the answer in the paragraph is unlabeled, and the task of MRC is supervised, so the answer needs to be labeled in the paragraph during preprocessing.

3.2

RNNTRANS modeling

In this paper, a machine reading comprehension model based on RNN transformer network is proposed. The overall framework of the machine reading comprehension algorithm based on neural network architecture is mainly composed of the following modules respectively: input coding module, article-question interaction module, and answer prediction module. The model contains five layers: the input embedding layer, the embedding coding layer, the article-question focus layer, and the model coding layer.

3.2.1

Input Embedding Layer

For the embedding of ELMo, it is first necessary to pre-train a bi-directional language model biLM, which, given an n-word sentence, needs to first compute a context-independent representation of the word $x_{k}^{L M}$ , which is the input to a multi-layer bi-directional long short-term memory. At each position k the forward long-short-term memory layer generates a context-relative representation ${\vec{h}}_{k, i}^{L M}$ , while a backward long-short-term memory layer outputs a representation ${\overset{\leftarrow}{h}}_{k, i}^{L M}$ in a similar manner. Thus, each word wk passing through a layer of bi-directional long-short-term memory network possesses 2L + 1 representations, as shown in equation (7).

(7)

R_{k} = {x_{k}^{L M}, {\vec{h}}_{k, i}^{L M}, {\overset{\leftarrow}{h}}_{k, i}^{L M} | i = 1, ..., L} = {h_{k}^{L M} | i = 1, ..., L}

Finally ELMo projects the outputs of all the layers in BiLM into a single vector by a linear combination as shown in equation (8).

(8)

E L M o_{k} = γ \sum_{i = 1}^{L} s_{i} (w) h_{k, i}^{L M}

Both w and Y are parameters obtained from training, s_i(w) is the activation function and h is the hidden layer.

The purpose of adding the fast network is to make the neural network better trained, for a single network layer, the output is determined by the word representation and the network layer weights. The fast network calculation for two layers is shown in equation (9):

(9)

y = H (x, W_{H}) \times T (x, W_{H}) + x \times (1 - T (x, W_{H}))

where H is the nonlinear function, x is the word representation input, and W_H is the weight of the network layer.

3.2.2

Embedding the coding layer

In this paper, a two-way gate control recursive one-member network is added to the embedded coding layer, which is a variant of rnn, embodying temporal interactions between coding words and connecting the outputs of two GRUs, i.e., forward output $\vec{h}$ and backward output $\vec{h}$ , at each time step, with the formula for the forward as shown in Eq. (10) and for the backward as shown in Eq. (11), and the final outputs of the two-way gate control unit are computed as shown in Eq. (12).

(10)

\vec{h} = G R U [χ_{w}; χ_{c}; χ_{e}]

(11)

\overset{\leftarrow}{h} = G R U [χ_{w}; χ_{c}; χ_{e}]

(12)

H_{t} = [\vec{h}; \overset{\leftarrow}{h}]

where x_w is the output of the word embedding algorithm Glove, x_c is the output of the character embedding CharCNN, and x_e is the output of the deep contextualization ELMo.

The self-attention mechanism [18] maps Q = K = V so that long-range dependencies can be efficiently captured. For the multi-attention mechanism, Q,K,V will be mapped into several different Q,K,V through the parameter matrix and calculate Attention separately, and finally the results will be spliced together, and the expression of the multi-attention mechanism is shown in Equation (13).

(13)

h e a d_{i} = A t t e n t i o n (Q W_{i}^{Q}, K W_{i}^{K}, V W_{i}^{V})

where Q is the problem, K is the key value, V is the weight value, and $W_{i}^{Q} \in R^{d_{m o k l} \times d_{k}}, W_{i}^{K} \in R^{d_{m o k l} \times d_{k}}, W_{i}^{V} \in R^{d_{m o d e l} \times d v}$ . The coding blocks embedded in the coding layer are composed from bottom to top containing: positional coding, convolutional layer, self-attention layer, and feed-forward network fnn layer, where each convolutional, self-attention, and fnn operation is performed in the residual block.

3.2.3

Article-Issue Focus Layer

This layer is mainly used to compute the context-to-question attention and question-to-context attention based on the coded representations of articles and questions obtained from the embedded coding layer to obtain the interaction information between the context and questions. The encoded articles and questions are represented as c and q, respectively, and a trilinear function is used to compute the similarity between contexts and questions. The similarity matrix S is calculated. Where the similarity matrix S is calculated as shown in equation (14).

(14)

f (q, c) = W_{o} [q, c, q \otimes c]

W_o is a trainable parameter, and q,c is the intermediate representation of the question and the article respectively. The rows and columns of S are normalized with softmax, and the vector of questions that are attended to in the whole article can be computed by the function formula A, and the specific computation of A is shown in Eq. (15). Due to the high similarity between query words and question words, in order to express which context words are crucial for answering the question, the query-context attention can also be calculated by the function formula, denoted as B. The specific calculation of B is shown in Equation (16).

(15)

A = s o f t \max (α (C_{i}, Q_{i})) Q^{T}

(16)

B = s o f t \max ({(α (C_{i}, Q_{j}))}^{T}) C^{T}

where α is a trilinear function, C_i represents the article, and Q_j represents the problem.

3.2.4

Model coding layer

The model coding layer consists of 2 model coding blocks, each of which consists of three coding blocks stacked on top of each other, with the 2 model coding blocks sharing parameters, and the inputs to this layer are of the form [c,a,c ⊗ a,c ⊗ b], a and b being the rows of matrices A and B, respectively.

3.2.5

Output layer

The goal of this layer is to predict the probability of each position in the article as an answer start or end. In order to achieve this goal, the outputs of the two model coding blocks in the previous layer need to be subjected to softmax operations, and the results obtained are denoted as M₀ and M₁. The probability of the answer start position is calculated as shown in Equation (17), and the probability of the answer end position is calculated as shown in Equation (18).

(17)

p^{b} = s o f t \max (W_{1} [M_{0}; M_{1}])

(18)

p^{e} = s o f t \max (W_{2} [M_{0}; M_{1}])

p^b and p^e are the probabilities of the word starting and ending as an answer, respectively. W₁ and W₂ are the trainable parameters, M₀ and M₁ represent the Encoder Blocks output from low to high respectively, and finally the parameters are corrected step by step by minimizing the loss function to get the optimal result.

4

Feedback on the Application of Interactive Intelligent Learning Platform for Physical Education and Sport Teaching

4.1

Topic-related analysis

Topic association analysis: the topics of statements in interactive Q&A reflect the subject areas that users are concerned about, and topic association analysis of interactive Q&A not only facilitates the analysis of the topic structure of interactive Q&A, but also helps to explore the user’s intention in the interactive process.

For the topic association analysis of interactive Q&A, this paper first uses the interactive scene data in the corpus knowledge base to construct the topic model HMM_dt = (A_dt,B_dt,π_dt) based on HMM; then, based on the topic transfer probability $a_{d t}^{i j}$ in the topic state transfer matrix A_dt, we calculate the degree of association between different topics $v_{d t}^{i j}$ , and construct the topic association matrix for interactive Q&A $V_{d t} = {[v_{d t}^{i j}]}_{N \times N}$ . In this paper, the degree of association is defined as the maximum probability of a topic change occurring between different topics (dt_i,dt_j). Maximum probability, and the calculation formula is as follows: (19) $v_{d t}^{i j} = \max (a_{d t}^{i j}, a_{d t}^{j i}), i = 1, 2, ..., N; j = 1, 2, ..., N$

Where N is the number of topic states; finally, the topic association structure in interactive Q&A is analyzed based on the topic association matrix V_dt, and Fig. 3 shows the topic association analysis results for online customer service Q&A in this paper.

Dialogue behavior analysis: Dialogue behavior in interactive Q&A reflects the potential semantic relationship between interactive Q&A statements, and the analysis of dialogue behavior in interactive Q&A is beneficial to deciphering the user’s intention and realizing the structural division of interactive Q&A in the interactive process.

Similar to topic association analysis, this paper firstly uses the interactive scene data in the corpus knowledge base to construct the HMM-based dialog behavior model HMM_da = (A_da,B_da,π_da; and then realizes the parsing of the dialog behavior structure based on the state transfer probability matrix of dialog behaviors $A_{d} a = {[a_{d a}^{i j}]}_{N \times N}$ , where N is the number of dialog behavior states. However, there is a temporal dependency of the dialog behaviors in interactive Q&A, so in this paper, the relationship weights of the dialog behaviors are given different relationship weights based on the transfer direction of the dialog behaviors in the parsing of the dialog behavior structure, i.e., $(v_{d a}^{i j} \neq v_{d a}^{j i})$ . In this paper, we take the state transfer probability of the dialog behaviors as the relationship weight of the dialog behaviors, i.e., $v_{d a}^{i j} = a_{d a}^{i j}$ , and analyze the structure of the dialog behaviors of interactive Q&A by the state transfer probability matrix of dialog behaviors, A_da, Fig. 1; Fig. 2; Fig. 3 is the number of dialog behaviors. Q&A’s conversation behavior structure, Figure 4 shows the results of this paper for the online Q&A’s conversation behavior analysis.

By analyzing the dialog behavior structure of the Q&A shown in the figure, it can be found that the dialog behavior changes in interactive Q&A have the following characteristics:

4.2

Effectiveness and Evaluation of the Operation of Interactive Q&A Platform for Sports Learning

Ten Physical Education students were selected to test the application of the interactive learning platform proposed in this paper by applying the primary Q&A as Q&A through the Knowledge Base and the secondary Q&A through Q&A with the help of Physical Education Curriculum and Teaching Theory.

4.2.1

Evaluation of the text-based question-and-answer approach

1)

Q&A Accuracy Test

The text Q&A results of 10 students are shown in Table 1. It can be seen through the table, based on deep learning interactive intelligent Q&A system can solve most of the problems of sports majors, although the average accuracy rate of the primary Q&A is only 75%, but in combination with the secondary Q&A, the average accuracy rate can reach 92%, and because the process of the secondary Q&A is constantly improving the knowledge base of the primary Q&A, which ultimately can make the learners solve the problems only through the primary Q&A.

Table 1.

Test for text accuracy

Student number	Number of subjects	Correct number	One time accuracy	The correct number of questions	Total accuracy	Failed to correct the number of questions
1	14	10	71.44%	2	85.71%	2
2	19	15	78.95%	3	94.74%	1
3	29	21	72.41%	5	89.66%	3
4	18	13	72.22%	3	88.89%	2
5	26	20	76.92%	5	96.15%	1
6	24	19	79.17%	3	91.67%	2
7	27	17	62.96%	7	88.89%	3
8	13	10	76.93%	2	92.31%	1
9	29	25	86.21%	3	96.55%	1
10	22	16	72.74%	5	95.45%	1

2)

Q&A time efficiency test

System response time is the learner from the input question to the system to return the answer to the time that is the time consumed by the question-answering process, the system statistics of ten students in a question-answering and secondary question-answering process time, the learner’s question-answering process, the program response time that is and statistics to get the results as shown in Table 2.

Table 2.

Test of text response efficiency

Student number	An example of an answer problem	Once time room(ms)	Examples of secondary answers	Secondary ring Time room (ms)
1	What is the connotation of physical education course	155	What is the sports theory	147
2	What is the lesson of PE class	151	Who put the sports theory	151
3	What is the main content of modern sports	157	What is the meaning of informationization education	153
4	What is the journey of sports	164	What is the role of intelligent teaching	153
5	What are the elements of sports teaching design	162	What is the role of physical education	161
6	What is the process of teaching optimization	163	How to guide the theory of teaching	147
7	What are the characteristics of sports	140	How interactive smart platforms are used in class	145
8	What is the content of sports teaching	143	What is the interactive smart platform	149
9	What is the form of physical resources	165	How to ensure the fairness of the sports test	139
10	What are the characteristics of sports teaching	160	Can the role of interactive smart platform be able to achieve the main effect	148

As can be seen from the table, students in the text mode of question and answer through the system, for the average waiting time for a question and answer 156ms, the average waiting time for the second question and answer 149.3ms, it can be seen that the system’s response time is very fast, do not have to learners to go to a long time to wait for the learner to ensure that learners and the system a good interaction between the two.

4.2.2

Evaluation of the video question-and-answer approach

1)

Q&A Accuracy Test

The video Q&A accuracy rate of 10 students using the deep learning interactive based intelligent Q&A system is shown in Table 3.

Table 3.

Video accuracy test

Student number	Problem number	Correct number	Accuracy	Failed to correct the number of questions
1	14	10	71.43%	4
2	20	16	80%	4
3	30	21	70%	9
4	19	13	68.42%	6
5	29	20	68.97%	9
6	24	14	58.33%	10
7	26	18	69.23%	8
8	16	13	81.25%	3
9	30	22	73.33%	8
10	25	19	76%	6
Average accuracy			71.7%

It can be seen from the table that under the condition that the students’ questions do not change, the average answer accuracy of the video Q&A mode is only 71.7%, which is mainly due to the small number of fragmented videos in the video Q&A database, but the video Q&A mode is an auxiliary role to the text Q&A mode, and the Q&A accuracy of 71.7% can basically meet the needs of students.

2)

Q&A efficiency test:

The results of the video Q&A efficiency test are shown in Table 4. As can be seen from the table, the average waiting time of video Q&A is 101.9ms, which belongs to the faster response to students’ questions, and can meet the needs of learners in terms of time efficiency.

Table 4.

Video response efficiency test

Examples	Answer time (ms)
What is the connotation of physical education course	93
What is the journey of sports	112
What is the theoretical basis of physical education	102
What is teaching informationization	87
What is the characteristic of teaching informationization	97
What is the definition of the interaction of sports teaching	110
What is modern sports teaching	105
What is the cause of physical education	126
What is the role of intelligent teaching	91
What are the characteristics of sports teaching	96

Through the above four experimental test results, it can be seen that the use of deep learning based interactive intelligent Q&A system, whether in the form of text or video Q&A in the accuracy of Q&A and Q&A efficiency can meet the needs of students, can quickly and accurately solve the scholars’ problems, and provide learners with a good learning interactive environment.

4.3

Change and Analysis of Physical Education Learning Skills

A total of 100 male and female students with similar physical forms and personal qualities were selected from physical education majors. Among them, 50 students in the experimental class were taught using the method proposed in this paper, and 50 students in the control class were taught using the traditional method.

4.3.1

Comparative Analysis of Students’ Interest in Physical Education Learning

The experimental group and the control group belong to independent observations before and after the experiment, and both come from the total that conforms to the normal distribution, so it meets the prerequisites of the paired t-test.

In order to explore whether the two groups of students before and after the experiment is able to produce functional changes, as well as whether the differences in the changes are significant, from the four dimensions in the interest in physical education learning to carry out investigations as shown in the table, statistical experimental data and paired samples t-test, the results are shown in Table 5 and Table 6.

Table 5.

Comparison of sports learning interest in pre-experimental laboratory class

Content	Premeasurement	Posttest	P value
Positive interest	31.14±0.57	33.27±0.88	0.034*
Negative interest	31.00±0.75	28.00±0.84	0.038*
Sports participation	12.13±0.58	15.29±0.62	0.005**
Independent exploration and learning	14.00±0.88	18.00±0.55	0.035*

*

Note: indicates P≤0.05, a significant level;

**

indicates P≤0.01, a highly significant level.

Table 6.

Comparison of physical learning interest in pre-experiment comparison class

Content	Premeasurement	Posttest	P value
Positive interest	31.00±0.32	32.00±0.37	0.822
Negative interest	31.00±0.65	29.00±0.61	0.006**
Sports participation	11.78±0.20	15.36±0.53	0.105
Independent exploration and learning	14.57±0.54	18.00±0.21	0.135

*

Note: indicates P≤0.05, a significant level;

**

indicates P≤0.01, a highly significant level.

Through comparative analysis, it can be seen that positive interest in the experimental pre-test scores of 31.14±0.57, the experimental post-test scores of 33.27±0.88, P=0.034<0.05, with significant differences; in terms of negative interest, the experimental pre-test scores of 31.00±0.75, the experimental post-test scores of 28.00±0.84, experimental scores declined, the negative interest is weakened, the P=0.038<0.05, a significant difference. In terms of independent inquiry and learning, the performance of the pre-experiment test was 14.00±0.88, and the performance of the post-experiment test was 18.00±0.55, P=0.035<0.05, indicating that there is a significant difference between independent inquiry and learning before and after the experiment. It has been shown that the intelligent question-answering system that uses deep learning interactive can enhance students’ interest, prevent negative interest to a certain extent, and improve their involvement in sports learning.

It can be seen from the table that the pre-test scores of the students in the control class in terms of positive interest were 31.00±0.32, and the post-test scores were 32.00±0.37, P=0.822>0.05, although the results also improved, but none of them reached statistical significance. In terms of negative interest, the pre-test score was 31.00±0.65, and the post-test score was 29.00±0.61, P=0.006<0.01, which was very significant. Overall, the students in the control class also improved in all dimensions before and after the experiment, but did not reach statistical significance. The results indicate that traditional teaching has no significant impact on the improvement of interest in learning physical education.

After the independent samples t-test of the post-experimental test data as shown in Table 7, the results of the students in the experimental class were contrasted with the results of the students in the control class and in terms of positive interest, the test scores of the experimental class were 36.21±0.57 and the test scores of the control class were 30.64±0.45 with a significant difference of P=0.038<0.05. In terms of negative interest, the test scores of the experimental class were 27.22±0.51 and the test scores of the control class were 31.27±0.88, p=0.275>0.05, and there was an increase in all dimensions, but none of them reached a statistical difference. In terms of the degree of sports participation, the test scores of the experimental class were 25.09±0.50 and the test scores of the control class were 17.55±1.44, P=0.041<0.05, with a significant difference; in terms of independent inquiry and learning, the test scores of the experimental class were 25.00±0.45 and the test scores of the control class were 17.78±1.57, P=0.025<0.05, with a significant difference. The results indicate that the intelligent question-answering system based on deep learning interactive is more effective in improving students’ interest in physical education learning compared to the traditional teaching method.

Table 7.

Comparison of students’ interest in students after the experiment

Content	Laboratory class	Cross-reference class	P value
Positive interest	36.21±0.57	30.64±0.45	0.038*
Negative interest	27.22±0.51	31.27±0.88	0.275
Sports participation	25.09±0.50	17.55±1.44	0.041*
Independent exploration and learning	25.00±0.45	17.78±1.57	0.025*

4.3.2

Comparative analysis of students’ independent learning ability

In order to experiment better, questionnaires were distributed to the two groups of students before the experiment, and the results of the survey were subjected to the independent samples t-test as shown in Table 8, p > 0.05, which showed no significant difference between the groups, proving that comparative analysis can be carried out and that the results of the experiment are valid.

Table 8.

Cooperation ability and independent learning ability test data analysis

Group	Cooperative ability	Autonomous learning ability
Experimental group	124.3±6.24	70.12±3.58
Control group	121.1±5.24	68.3±2.85
T value	1.109	1.09
P value	0.184	0.206

*

Note: indicates P≤0.05, a significant level;

**

indicates P≤0.01, a highly significant level.

The results of the pre- and post-experimental questionnaires on the autonomous learning ability of the two groups of students were subjected to paired-sample t-tests, and the results of the tests are shown in Figure 5. Comparative judgments on the scaffolding teaching method for independent learning ability are analyzed. As can be seen from the table, the t-value of the test scores before and after the experiment of the experimental group is -6.042, and the test scores show a very significant difference by the t-test p < 0.01. The t-value of the test scores before and after the experiment of the control group is -1.78, p < 0.05 by t-test, and the experimental test scores are significantly different.

Through the above analysis, there is a large difference between the two groups of tests after the experiment, and the independent learning ability of the students after using the deep learning interactive based intelligent question and answer system for teaching is obviously better than the traditional teaching method. A questionnaire survey was conducted before the experiment, and the results showed that there was no difference between the experimental group and the control group. After the teaching experiment, the experimental group had a big change, the control group also had a change, but the change was less, while the independent learning ability of the experimental group was obviously improved, while the control group’s change was not as obvious as that of the experimental group. Therefore, the students’ independent learning initiative and motivation in general under the traditional teaching method resulted in the failure to achieve the best independent learning ability.

5

Conclusion

In this paper, we first design the interactive sports teaching platform based on deep learning, propose the machine reading comprehension model based on RNN transformer network, integrate the model into the interactive sports teaching platform, and finally get the intelligent Q&A system based on deep learning interactive. The system provides two Q&A modes: text Q&A and video Q&A, so the accuracy and efficiency of these two types of Q&A are tested separately. For text Q&A, the average accuracy rate can reach 92%, which is suitable for physical education learning content Q&A. In the comparison experiment of physical education teaching, the influence of the intelligent Q&A system based on deep learning interaction on students’ independent learning ability is positive, but the traditional teaching method is not prominent enough to improve the effect.

Język:: Angielski

Częstotliwość wydawania:: 1 razy w roku
Dziedziny czasopisma:: Nauki biologiczne, Nauki biologiczne, inne, Matematyka, Matematyka stosowana, Matematyka ogólna, Fizyka, Fizyka, inne

Kanał RSS czasopisma

Design and Optimization Path of Interactive Intelligent Learning Platform in Physical Education Teaching Informatization System in Colleges and Universities

Xu Yuan

Jingwen Hu

Data publikacji: 21 mar 2025

Otrzymano: 30 paź 2024

Przyjęty: 14 lut 2025

DOI: https://doi.org/10.2478/amns-2025-0628

Słowa kluczoweDeep learning, Machine reading comprehension, Interactive question and answer, Physical education informatization

© 2025 Xu Yuan, published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Słowa kluczowe
Deep learning, Machine reading comprehension, Interactive question and answer, Physical education informatization