The Construction of Blended Teaching Model of College English in Applied Colleges and Universities Based on Neural Networks

The social requirements for English teaching in contemporary colleges and universities are to cultivate talents with English practice ability and industry use ability, and the conventional college English classroom cannot fully meet such requirements [1-2]. With the development of cloud computing, artificial intelligence, mobile Internet, wireless networks and other information technologies, the digital teaching of information technology provides powerful technical support and promotes the rapid development of the digitalization of college English teaching in applied colleges and universities.

The blended teaching mode is a brand-new mode that integrates traditional classroom teaching and emerging teaching methods based on modern information technology [3]. It can not only further play the leading role of education among teachers but also better reflect the social subject position of students themselves [4-5]. By integrating pre-course preparation, classroom activities and post-course guidance through online and offline resources, the blended teaching mode broadens the time and space of traditional classroom teaching, constituting a double-line effectively driven cycle system [6-9]. The blended teaching mode is also applicable to the teaching of English courses, which provides a new way of learning and rich learning resources for English courses in colleges and universities and is able to provide targeted guidance based on the personalized needs of students in the course, which is conducive to the promotion of student independent learning, creative learning and collaborative learning [10-13]. However, the current blended teaching does not fully realize students’ independent learning; the classroom is still the teacher teaching as the main activity, and the learning of college English courses is also completed according to the planning of the English teacher and does not reflect the students’ autonomy, initiative and innovation [14-16]. Although the blended teaching mode has changed the teaching form to a certain extent, it actually continues the traditional English classroom teaching mode [17]. In addition, as a new teaching mode combining tradition and innovation, blended teaching has high technical requirements and needs the full support of information technology and hardware equipment so as to design a more recognized teaching mode [18-19].

In order to better complete the construction of the English blended teaching mode for college students, this paper applies the parallel K-means (ACS-Kmeans) clustering algorithm with adaptive cuckoo search to cluster the collected data on the English teaching effect. A blended English teaching evaluation index system was designed from three aspects: pre-course learning evaluation, in-course learning evaluation, and post-course learning evaluation. Then, the GA-BP neural network was used for blended teaching evaluation, and a better evaluation model was obtained through the optimised GA-BP network with a network structure, initial connection weights, and thresholds. The blended teaching model with this evaluation model as the main method is constructed from four aspects: teaching goal setting, teaching pre-course preparation, teaching process implementation and teaching evaluation, thus completing the construction of the university English blended teaching model based on GA-BP neural network.

2

ACS-KMeans-based clustering of English language teaching data

In order to better complete the evaluation of English online and offline hybrid teaching effectiveness, a series of indicators for hybrid English teaching effectiveness evaluation are designed and used in the process of English teaching effectiveness evaluation. In order to make the collected data better serve the evaluation of hybrid English teaching effectiveness, the Parallel K-means (ACS-Kmeans) clustering algorithm with Adaptive Cuckoo Search is applied to the collected English teaching effectiveness data for data clustering.

2.1

Programming algorithm and fitness function

If q is used to represent the dimensionality of the sample vectors of data related to the effectiveness of English language teaching and the requirement of data clustering is to cluster the data by clustering algorithm into having k categories. The essence of the coding algorithm in the operation of the parallel K-means clustering algorithm for adaptive cuckoo search is to vectorially solve for the coordinates of the k sets of vectors. By using the k data clustering centres of the EFL data to represent the locations of the bird’s nests in the algorithm, the locations of the bird’s nests in the EFL dataset are in fact matrices about k × q. The specific coding method can be expressed as Eq: (1) $A = [\begin{matrix} n e s t_{11} & n e s t_{12} & \dots & n e s t_{1 q} \\ n e s t_{21} & n e s t_{22} & \dots & n e s t_{2 q} \\ \dots & \dots & \dots & \dots \\ n e s t_{k 1} & n e s t_{k 2} & \dots & n e s t_{k q} \end{matrix}]$

In this formula, the q-dimensional vectors of the 1st clustering centre, the 2nd clustering centre and the k th data clustering centre are represented by nest₁₁, nest₁₂, ⋯, nest_1q, nest₂₁, nest₂₂, ⋯, nest_2q and nest_k1, nest_k2, ⋯, nest_kq respectively.

Using the Euclidean distance to confirm the category of the sample, with x and y representing the two vectors, the Euclidean distance between x and y can be expressed as Eq: (2) $D_{x y} = \sqrt{\sum_{i = 1}^{n} {(n e s t_{x i} - n e s t_{y i})}^{2}}$

Where the difference obtained by subtracting vectors x and y in dimension i is represented by nest_xi − nest_yi and satisfies 1 ≤ x ≤ k and 1 ≤ y ≤ k.

In the parallel K-means clustering algorithm for adaptive cuckoo search [20], the level of the fitness value evaluates the degree of goodness of an individual, and the magnitude of its value is positively correlated with the degree of goodness of an individual. Usually, the fitness function is expressed by the distance within the class and the number of clustering points [21], but this leads to the number of iterations and the degree of goodness or badness of the individual in the process of solving the fitness is not good enough, in order to change such a situation, in the use of adaptive cuckoo search of parallel K-means clustering algorithm for clustering the data of the effectiveness of English language teaching the fitness function is expressed by the formula as: (3) $F i t (i) = \frac{C N_{i}}{D S_{i}} i = 1, 2, \dots, k$

In the formula, the number of data points of English teaching effect samples in the i th cluster and the sum of the distances between each data point in the i nd class and the clustering centre y are represented by CN_i and DS_i, respectively, and DS_i satisfies Eq. $D S_{i} = \sum_{x \in y} D_{x y} (x = 1, 2, \dots, C N_{i}; x \neq y)$ ; CN_i In the case of obtaining the class by sample aggregation, the average distance of all points to the class centre is represented by $F i t (i)$ , $F i t (i)$ is used as a fitness function of the parallel Kmeans clustering algorithm for adaptive cuckoo search and the magnitude of its value is positively correlated with the effect of clustering the English teaching effectiveness data. The fitness values of the k classes are summed to get the value of total fitness G, and again, the magnitude of its value is positively correlated with the clustering effect. The total fitness G can be expressed as using the formula: (4) $G = \sum_{i = 1}^{k} F i t (i)$

2.2

Clustering Algorithm Flow for English Teaching Data

Combined with the discussion on programming algorithms and fitness functions in the previous subsection, the algorithmic flow of the parallel K-means (ACS-Kmeans) clustering algorithm for adaptive cuckoo search can be summarised as follows: 1)

Use Iterator_max and P_a to represent the maximum number of iterations and the probability of discovery; step_max and step_min to represent the maximum and minimum step sizes; and perform input operations on the English teaching effectiveness sample data set to be clustered, the number of clusters, Iterator_max, P_a, step_max, and step_min, in that order.

2)

Perform the initialisation operation on the locations of the bird’s nests with the number of k.

3)

Use K-means clustering algorithm to perform cluster division operations to solve for the optimal bird’s nest location for each English teaching effectiveness sample data.

4)

Solve the value of each bird’s nest adaptation according to formula (3) to determine the most suitable location of the bird’s nest.

5)

Find the total fitness G through equation (4).

6)

The bird’s nest locations from the previous iteration process are kept and other bird’s nest locations are updated.

7)

In the update operation of other bird’s nest locations, r and P_a are obtained. where r represents a random number that satisfies $r \in [0, 1]$ . The value of r is compared with the value of P_a. When P_a > r returns (6) and vice versa the nest is discarded and a new nest construction operation is performed.

8)

After constructing a new nest, perform operations (3) to (5), compare its G value with the value of G of the previous generation, if the G value of the new nest is greater, determine the nest combination as the updated nest combination, and vice versa return to step (7).

9)

The algorithm ends when the number of iterations is maximum and vice versa return to (6).

2.3

Blended teaching quality evaluation index system

In order to achieve an effective evaluation of the effectiveness of English online and offline blended teaching, based on the teaching data obtained by applying the ACS-KMeans data clustering algorithm to cluster the data related to blended English teaching, the blended English teaching evaluation index system, including the evaluation of pre-course learning, the evaluation of learning in the course of the class and the evaluation of post-course learning is designed. In this paper, starting from the actual teaching situation, the evaluation indexes are designed as shown in Table 1, under which there are three first-level indexes, which cover 20 second-level indexes, and the content of these indexes is used to reflect the specific implementation process of blended teaching. Based on the establishment of the evaluation index table for blended teaching, the blended evaluation index system established in this paper is shown in Table 2.

Table 1.

Evaluation index table of mixed teaching

Dimension	Primary indicator	Secondary indicator
Mixed teaching evaluation	Pre-class study evaluation	Number of students check-in
		Watch the number of video
		Frequency of interaction
		Online test number
		Clear task target
	Teaching evaluation in class	Classroom discipline (test attendance, etc.)
		Group cooperation
		Classroom performance (discussion interaction)
		The teaching is serious and sensitive
		Attitude is rigorous and refined
		counselling
		Focus on guidance and inspiration
		Heavy difficulty
		Attention concentration
		Theoretical combination
	Teaching evaluation in class	Online work
		Online test
		Online discussion interaction
		Basic knowledge
		Online test scores

Table 2.

Evaluation index system of mixed teaching

Dimension	Primary indicator	Secondary indicator
Pre-class study evaluation	X₁	Number of students check-in
	X₂	Watch the number of video
	X₃	Frequency of interaction
	X₄	Online test number
	X₅	Clear task target
Teaching evaluation in class	X₆	Classroom discipline (test attendance, etc.)
	X₇	Group cooperation
	X₈	Classroom performance (discussion interaction)
	X₉	The teaching is serious and sensitive
	X₁₀	Attitude is rigorous and refined
	X₁₁	counselling
	X₁₂	Focus on guidance and inspiration
	X₁₃	Heavy difficulty
	X₁₄	Attention concentration
	X₁₅	Theoretical combination
Teaching evaluation in class	X₁₆	Online work
	X₁₇	Online test
	X₁₈	Online discussion interaction
	X₁₉	Basic knowledge
	X₂₀	Online test scores

3

Establishment of a hybrid teaching quality evaluation model based on GA-BP

3.1

Collection and pre-processing of raw data

A questionnaire survey is a common way to obtain information. This paper combined the theory and practice of blended teaching, extracted the factors affecting the evaluation results, in which the offline classroom indicators of data are the design of the questionnaire, and then issued the questionnaire and the results of the questionnaire to analyse the results of the screening obtained. The questionnaire is designed according to the content of the evaluation indicators, and the open-ended questions are not counted in the total score of the questionnaire from the aspect of data analysis. The online teaching platform collects index data for the online part, which can be derived from the platform according to the evaluation system. After the results of the questionnaire and the data exported from the platform are screened and integrated to obtain valid data, in order to reduce the difficulty of correcting the weights due to the large magnitude of change in the input data, it is necessary to normalise the scoring data to the interval [0,1] after collecting the original data. The normalisation function used in this paper is the maximum-minimum method, which can retain the original meaning to a greater extent in the process of a linear transformation of the data, and the information is not easily lost. The normalization formula for the max-min method is as follows: (5) $X = \frac{I - I_{\min}}{I_{\max} - I_{\min}}$

Where X is the normalised BP neural network input value, I refers to the original data collected, I_min represents the minimum value of the original score and I_max represents the minimum value of the original score.

3.2

BP neural network model structure design

The structure and function of biological neural networks can be compared to artificial neural networks, and the method is commonly used to evaluate mathematical or computational models. Neural networks form a complex network structure by connecting a large number of neurons, which can be adaptive and can change internal operations based on external information. Thus, neural networks are also a practical application of mathematical and statistical methods, which allow us to obtain local structures that can be expressed in large spaces. On the other hand, neural networks can use mathematical statistics to solve decision-making problems in applications (i.e., statistical methods allow artificial neural networks to make simple decisions).

Error back propagation neural network is referred to as a BP network [22]. The network structure is a three-layer feed-forward layering consisting of an input layer, a hidden layer, and an output layer. Based on the propagation of errors, they can be divided into forward and backward propagation networks. The working mode of the BP network is that information is sent from the input layer to the unit in the hidden layer. The data generated is calculated and sent by the hidden unit to the output layer. This mode is called feed forward mode. Then the print is compared with the expectation, and if it does not meet the expectation, it is converted to error back propagation, which passes back the error along the original path and reduces the error signal by modifying the connection weights of the neurons in each layer. Through continuous forward and backward alternating “memory training” until the expected error reaches the desired requirements, so as to obtain a stable network model. Specifically, the following key steps are included. 1)

Determination of the number of neurons. The number of input neurons to the model is usually determined by taking the number of indicators in the indicator system as the number of neurons. For example, if the system has 25 sub-indicators, then the number of neurons in the input layer is set at 25.

2)

Determination of the number of neurons in the output layer. The output layer of the model represents the final assessment result of the hybrid model, and the number of neurons in the output layer can be set based on the assessment result.

3)

Determination of the number of implicit layers of the network. The number of hidden layers can be none or one or more layers. The meanings of the representations are: no hidden layer represents a linearly separable function or decision. 1 hidden layer represents a continuous mapping from one finite space to another finite space. 2 hidden layers can represent any decision boundary to arbitrary precision using rational activation functions, and can approximate any smooth mapping to any precision. 2 or more hidden layers can learn complex features.

Usually, the number of hidden layers depends on the number of input layers, output layers and input samples. In this paper, the number of hidden layers is determined based on the following formula: (6) $h = \frac{s}{c (m + n)}, c \in [2, 10]$

Where, n is the input layer size, m is the output layer, s is the sample size, and c is a constant. 4)

Determination of the number of neurons in the hidden layer.

The convergence of the network determines the number of neurons in the hidden layer. The number of neurons in the hidden layer will bring about performance problems. Too few will make the training is not enough, resulting in the network is not “robust” and the error tolerance rate being reduced; on the contrary, if too many neurons’ training time is long, the performance will be reduced, and the error can not be improved much. Therefore, most of the neuron parameters are determined using genetic algorithms or particle algorithms to find the optimal parameters, and finally get an empirical formula to determine. The following equations are several widely used empirical formulas: (7) $s = {(0.43 m n + 0.12 m^{2} + 2.54 n + 0.77 m + 0.35)}^{1 / 2} + 0.51$ (8) $s = {(m + n)}^{1 / 2} + a, a \in [1, 10]$ (9) $h = \frac{s}{c (m + n)}, c \in [2, 10]$

Where, n is the input layer size, m is the output layer, s is the sample size, c and a are constants. 5)

Neuron conversion function. There are many kinds of neuron conversion functions for the BP neural network, such as the Sigmoid function, Hard Sigmoid function, Swish function, ReLU function and so on. Among them, the Sigmoid function expression is shown in equation (10). (10) $f (x) = \frac{1}{1 + e - λ x}$

Where the coefficient λ indicates the degree of compression of the function. 6)

Model structure and training process.

BP neural network has a strong nonlinear mapping ability and can approximate the nonlinear function with arbitrary accuracy, but because of its method of gradient descent algorithm, the selection of many parameters in the training process does not have a theoretical basis, so it has certain limitations: (1)

Slow convergence of the error in the learning process and easy to fall into the local minimum, BP neural network is a nonlinear optimization method based on gradient descent method, so for some complex problems, the training process may last a long time due to the slow convergence, from the training process, it is along the slope of the surface of the error down the approximation, and the error surface in the actual problem is generally complex and irregular, with many local distributions of the surface. irregular, with many local minima, which can cause the network to fall into local minima.

(2)

The selection of the parameters of the BP neural network (such as the number of layers of the hidden layer, the number of neurons in the hidden layer, and the learning rate, etc.) does not have a clearer theoretical basis so far, and it is generally determined by empirical formulas or constant training experiments, which may lead to a long learning time and low efficiency.

(3)

The training, learning and memory functions of the network are unstable. When the sample changes, the network model that has been trained will have to retrain the network, affecting the samples that have been previously learned.

3.3

GA-BP neural network evaluation model

In the training process of the BP neural network, the network structure and the choice of initial connection weights and thresholds have a great influence on the network training, but it is impossible to obtain accurately, so this paper aims at these shortcomings of BP neural network, GA-BP neural network is used for the evaluation of blended teaching, weights and thresholds optimised BP network helps us to get the effect of better evaluation model establishment.

A genetic algorithm is an optimization tool that simulates biological evolution. It simulates the collective evolutionary behavior of a population. Each individual represents an approximate solution to the search space of the problem. The genetic algorithm starts from any initial population, and through individual inheritance and mutation, it effectively implements a stable and optimal breeding and selection process so as to enable the population to evolve to a better range of the search space. Genetic Algorithm Optimisation of the BP Neural Networks is mainly divided into the following three parts: 1)

Determine the structure of the BP neural network, determine the number of nodes in the input layer according to the hybrid teaching index system constructed in Chapter 2, determine the number of output neurons according to the evaluation results, and then determine the number of nodes in the implied layer according to the empirical formula l=2m+1 (m is the number of neurons in the input layer).

2)

Genetic algorithm optimises the BP neural network weights and thresholds to randomly generate a population whose individuals represent the network weights and thresholds, and then the fitness function is used to calculate the fitness value, and finally, the optimal individual is found through selection, crossover and mutation operations.

3)

Prediction using GA-BP neural network [23], after initialisation with the optimal individuals, the weights and thresholds of the BP neural network can be locally optimised again during the training process, and the optimised BP neural network has a better prediction accuracy and prediction efficiency for English teaching data.

According to the above steps, the hybrid teaching quality evaluation model based on the GA-BP neural network is established, as shown in Figure 1: 1)

After determining the structure of the neural network, the preprocessed data are used as the input values of the neural network. The BP network structure designed in the GA-BP neural network model of this paper is 20-41-1.

2)

Determine the population size S = 50 and the maximum number of iterations as 2000, select the crossover probability and variance probability, and set the upper and lower bounds of the weights and thresholds to generate the initial population.

3)

Determine the fitness according to the simplified fitness function, as shown in Equation (11), where x_i is the target output value and $x_{i}^{'}$ is the actual output value. (11) $F = \frac{1}{\sum_{i = 1}^{N} {| x_{i}^{'} - x |}_{i}}$

4)

Generate a new generation of population through selection, crossover, mutation and other processes, and obtain the maximum number of iterations or the minimum error through iterative calculation.

5)

Substitute the weights and thresholds optimised by the genetic algorithm into the BP neural network, train the GA-optimised neural network with sample data until it meets the error requirements, and reverse normalise the output of the network to get the evaluation results of the test samples, and complete the construction of the evaluation model of the quality of blended teaching of university English in colleges and universities.

The GA-BP neural network algorithm flow is shown in Figure 2:

4

Evaluation model-based blended learning model

In order to improve the education mode of college students, enrich the blended education resources, and create a personalised cultivation mode for college students, based on the integration of blended teaching and data-driven related theories, and taking the implementation process of teaching activities as the guideline, the blended teaching model is proposed to be based on the principle of “Teaching Goal Setting → Teaching Preparedness → Teaching Process Implementation → Teaching Evaluation and”. Based on the process-induced “prepare first, learn first, teach later and evaluate later”, a hybrid teaching model based on the hybrid teaching quality evaluation model of GA-BP was constructed.

4.1

Teaching and Learning Objective Setting

To establish the precise mapping relationship between students’ abilities and teaching objectives, the operation steps are as follows: first, quantify the difficulty coefficient of teaching objectives. Secondly, quantify the multidimensional abilities, such as knowledge base, learning ability, and learning attitude, that students already possess. Third, establish the mapping relationship between teaching objectives and multidimensional student characteristics. And fourth, adjust in both directions to achieve a high degree of matching between the teaching objectives and the characteristics of the students.

4.2

Preparation for teaching sessions

According to the teaching content and student characteristics, personalised teaching materials are pushed to students before class to achieve the preparation for the teaching class. The operation steps are as follows: firstly, the prepared teaching resources, including videos, courseware, exercises and other materials, are stratified and refined; secondly, a comprehensive judgement is made according to the students’ characteristics, so as to establish a corresponding relationship with the teaching content; thirdly, the corresponding teaching content is pushed to the corresponding students according to the judgement results; and fourthly, the teacher adjusts the teaching content appropriately according to the pushed content.

4.3

Implementation of the teaching and learning process

The online part of blended teaching is divided into “independent learning” and “topical discussion”, while the offline part is divided into “classroom lecture” and “group discussion”, the online part. In the online part, we obtain data on students’ learning attitudes, learning effects, and interactive behaviours in “independent learning” and “thematic discussion” and accurately record students’ learning behaviours and performances; in the offline part, we accurately design the lecture plan according to the determined teaching objectives and teaching contents and refine the common difficulties in teaching. The offline part, according to the determined teaching objectives and teaching content, accurately designs the teaching plan, refines the common difficulties in teaching, and classify the individual problems, uses group discussion to solve the different individual problems of the students, and finally, according to the overall ability of the students, carry out the second push to improve and consolidate the content learnt, the “four links”+“two pushes” are connected, and the learning objectives and tasks of the course are completed in a step-by-step manner.

4.4

Evaluation of teaching and learning

Offline teaching evaluates students’ performance and involvement in class and group discussions. The degree of students’ mastery of fixed knowledge points is obtained through offline exams. Online teaching evaluates students’ learning performance according to the data information obtained from the learning platform, including data on homework submission, interactive evaluation, and video playback, integrating online and offline data information. After the hybrid teaching quality evaluation model based on GA-BP, students’ learning process is accurately evaluated, corresponding interventions are taken in a targeted manner, and then feedback is given back to the formulation of teaching objectives and content, forming a virtuous cycle of “formulation-implementation-feedback-amendment” to improve the whole process of teaching.

5

Effectiveness of GA-BP Hybrid Modelling in English Language Classes in Colleges and Universities

5.1

The Effectiveness of the Implementation of Hybrid Modelling in College English Teaching

The teaching experiment was conducted by choosing two classes of the same year of a university’s college English programme, both with 50 students. Class 1, as the baseline group, was taught using a traditional teaching model, while Class 2, as the experimental group, was taught using a blended teaching model with a course duration of about five months. The data on students’ English learning levels and engagement were analysed using SPSS 22.0. The changes in running time of the k-means algorithm and ACS-Kmeans algorithm were compared. The results of the running time comparison are shown in Fig. 3. The biggest improvement in this case is the improvement of the initial clustering. The same set of data was calculated by using the k-means algorithm and the ACS-Kmeans algorithm, and from the time statistics, it can be seen that overall, the ACS-Kmeans algorithm has a shorter running time than the original k-means algorithm. K-means algorithm. A comparison of the accuracy of the clustering operation is shown in Figure 4. The k-means algorithm has high and low accuracy in the case of an unstable number of data iterations, which seriously affects the clustering results, while the ACS-Kmeans algorithm still maintains its accuracy in a stable state and reaches a high value of 93% in the case of an unstable number of iterations.

The students’ English learning level after implementation includes sub-stage scores, pre-implementation test scores and post-implementation measured scores, and the statistical results of the two groups’ English levels before and after the implementation of the teaching model are shown in Table 3. As can be seen from Table 3, after a complete course of study in College English, the English scores of the students in both groups have improved. By comparing the mean values of the grades before and after the implementation of the model, it was found that the overall mean value of the students in the benchmark group increased by 12.5% throughout the semester, and the overall mean value of the experimental group increased by 19.2% throughout the semester, and in terms of the growth rate of the grades, the overall growth rate of the grades of the students in the experimental group was higher than that of the benchmark group. The experimental results indicate that the students who used the blended teaching model gained a greater improvement in their English proficiency during the semester of study.

Table 3.

The two groups carry out the English level of the teaching model

Group	Measurement type	Mean of achievement	Standard deviation	The difference is 95% confidence interval
Group	Measurement type	Mean of achievement	Standard deviation	Lower limit	Upper limit
Experimental group	Pre-model	80.26	2.178	10.451	13.478
Experimental group	After the model is implemented	95.67	3.115	10.451	13.478
Reference group	Pre-model	81.09	2.124	7.665	10.745
Reference group	After the model is implemented	91.23	3.415	7.665	10.745

5.2

Analysis of teaching quality evaluation model based on GA-BP neural network

The experiment takes the teaching effect survey data of English majors in a university in a city as the experimental object and 800 training samples of teaching effect survey data are obtained after the English teaching effect survey data are sorted out. 400 of these samples are used as training samples for the GA-BP neural network English teaching effect evaluation model, and 400 are used as test samples for the model. Matlab software is used to simulate the evaluation process of the GA-BP neural network in English online-offline hybrid teaching effect. Verify the performance when evaluating the effectiveness of English online-offline hybrid teaching.

When the range of values of the actual assessment value is set at 0.4-0.6, the evaluation grade is satisfactory. When the range of values is 0.6-0.8, the evaluation grade is medium. When the value range is 0.8 to 0.9, the evaluation grade is good. Values that are in the range of 1 are considered excellent. A value range below 0.4 is poor. The evaluation results of teaching effectiveness obtained from the hybrid teaching quality evaluation of English teaching effectiveness data in this university are presented in Table 4. From Table 4, it can be seen that the method based on the GA-BP neural network can well complete the evaluation of the English blended teaching effect, and the obtained network training value is very close to the actual training value. The difference between the training value and the actual value of the student’s performance of No. 15 is only 0.01. It verifies the feasibility and reliability of the method in the evaluation of English online and offline blended teaching effects.

Table 4.

Mixed teaching evaluation results

Number	Network training value	Actual evaluation	Grade
1	0.911	0.9	good
2	0.689	0.7	medium
3	0.878	0.9	good
4	0.716	0.7	medium
5	0.987	1	superior
6	0.716	0.7	medium
7	0.525	0.5	passing
8	0.414	0.4	passing
9	0.656	0.6	medium
10	0.978	1	superior
11	0.667	0.7	superior
12	0.545	0.5	passing
13	0.567	0.5	passing
14	0.787	0.7	medium
15	0.301	0.3	bad

Using MATLAB software to simulate and analyse the teaching quality evaluation model of the GA-BP neural network, 30 students were randomly selected from two groups, respectively, to predict their grades. The number of network training is set to 2000 times, and the training target error is set to 0.0001. After network training and substitution into the test set, the output simulation evaluation results and actual evaluation results are obtained, as shown in Figure 5. It can be seen from the figure that GA-BP and BP are able to predict students’ performance more accurately than the former. The prediction accuracy of GA-BP and BP is 95.23% and 90.41%, respectively. That is to say, the hybrid university English teaching quality evaluation model based on the GA-BP neural network is able to predict students’ English performance through teaching, which in turn facilitates the judgement of whether the hybrid teaching mode is suitable for university English translation teaching, and then makes timely adjustments.

The performance of the GA-BP evaluation model is further analysed. Figures 6 and 7 show the response speed and CPU occupancy of GA-BP and BP for predicting student performance, respectively. As can be seen from Figures 6 and 7, the GA-BP evaluation model has a high prediction accuracy along with a fast reaction speed, and when the number of students reaches 30, the reaction speed of BP in predicting students’ grades is slower than that of the GA-BP evaluation model by 9.5s, and its overall CPU occupancy is also higher than that of the GA-BP evaluation model. That is, the hybrid university English teaching quality evaluation model based on the GA-BP neural network achieves progress in teaching evaluation technology in the hybrid teaching mode and facilitates university English teaching.

6

Conclusion

Applying the appropriate neural network optimization algorithm to a reasonable teaching quality evaluation system can result in a more scientific teaching quality evaluation model. It is conducive to the construction of a more scientific hybrid teaching model, so this paper proposes that the GA-BP neural network be used to establish a hybrid teaching model. 1)

In terms of running time, the ACS-Kmeans algorithm is shorter than the k-means algorithm. The k-means algorithm is unstable in terms of accuracy when the number of data iterations is not stable, and the ACS-Kmeans algorithm is able to achieve an accuracy of 93% and is very stable when the number of iterations is not stable. The average English achievement of students in the benchmark group applying the traditional teaching model increased by 12.5% throughout the semester, and the students in the experimental group applying the blended teaching model constructed in this paper improved their English achievement by 6.7% more than the students in the benchmark group, indicating that the use of the blended teaching model enables students to obtain greater improvement in English learning, and verifying that the blended teaching model based on the GA-BP evaluation model proposed in this paper has a positive effect on students’ English achievement improvement.

2)

The network training values obtained by the method based on the GA-BP neural network are very close to the actual training values. It shows that the GA-BP evaluation model not only has high prediction accuracy but also has fast response speed. When the number of students is 30, the response speed of the GA-BP model is 9.5s faster than the BP model, and its overall CPU occupancy rate is also optimized. The GA-BP neural network-based hybrid teaching quality evaluation model for college English demonstrates the advancement of teaching evaluation technology, which can be gradually implemented and applied in college English teaching.

Language:: English

Publication timeframe:: 1 times per year
Journal Subjects:: Life Sciences, Life Sciences, other, Mathematics, Applied Mathematics, General Mathematics, Physics, Physics, other

Journal RSS Feed

The Construction of Blended Teaching Model of College English in Applied Colleges and Universities Based on Neural Networks

Manli Jia

Published Online: Feb 05, 2025

Received: Sep 17, 2024

Accepted: Dec 21, 2024

DOI: https://doi.org/10.2478/amns-2025-0062

KeywordsGA-BP neural network, Blended teaching, K-means clustering, Teaching effectiveness evaluation

© 2025 Manli Jia, published by Sciendo

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

Keywords
GA-BP neural network, Blended teaching, K-means clustering, Teaching effectiveness evaluation