Intelligent teaching ability of contemporary college talents based on BP neural network and fuzzy mathematical model

Abstract

In the evaluation of traditional college talents’ teaching ability, the importance of evaluation indicators lacks evaluation, and the evaluation results are relatively random. In order to improve the evaluation efficiency of university scientific research talents, this study combines BP neural network and fuzzy mathematical theory to build an evaluation model. Combining the talent training process and ability requirements of colleges and universities, a secondary index system is proposed, and the weight of the evaluation index is determined by combining data collection. This paper first normalizes the samples, determines the training and test samples, and then uses trial and error to determine the number of hidden layer neurons. Then use fuzzy mathematics theory to construct fuzzy similarity matrix to describe the fuzzy relationship between factor domain and judgement domain. Calculate membership to get comprehensive evaluation results. Finally, this paper uses statistical methods to draw the results into statistical charts and combines the simulation results to obtain performance comparison results. The feasibility of the model is verified by experimental research, and the model can be applied to practice, and can provide theoretical reference for subsequent related research.

Keywords

BP neural network fuzzy mathematical evaluation model college talent scientific research

1 Introduction

The talents cultivated in colleges and universities must eventually move toward society. Whether graduates from colleges and universities meet social needs, whether they can achieve employment, whether they can find jobs suitable for their profession and specialty, whether they have comprehensive qualities and work ability to adapt to the needs of jobs as soon as possible, is an important topic that Chinese colleges and universities generally face [1]. With the acceleration of China’s economic development, the demand for high-level and high-quality talents has been expanding. The cultivation and reserve of excellent talents is largely related to the comprehensive competitiveness and status of a country and a nation in the world [2]. As a high-level talent in China, scientific research talents in colleges and universities shoulder the historical responsibility of contributing to China’s scientific and technological progress, cultivating innovative top talents for China’s prosperity, creating high-level scientific research achievements, and providing high-level social services [3]. If the Chinese nation wants to stand in the undefeated forest of the world, it must rely on a large number of high-level and high-quality talents with both ability and political integrity. According to statistics from relevant experts, since the reform and opening up to the present, after more than 30 years of development, China’s scientific research personnel education has quickly entered the ranks of the world’s major educational countries by the small educational countries [4].

The scientific development talent education industry has such a good development momentum, it is very worthy of recognition and gratification. However, we must also realize that with the expansion of scientific research talents in universities, the number of scientific research talents will increase proportionally every year, and colleges and universities will face more and more pressure on the management and education of scientific research talents. For example, how to ensure the large-scale training of colleges and universities while taking into account the quality of scientific research personnel training, how to improve the comprehensive quality of scientific research talents, how to achieve the individualized training of scientific research talents, etc., these have become very important issues in the current management training of scientific research personnel in China [5]. In order to improve the teaching quality of talents in contemporary colleges and universities, all departments of the school should take effective measures from various aspects of morality and intelligence, and formulate scientific, reasonable and effective training management programs [6]. Establish a relatively complete scientific and reasonable evaluation index system and evaluation method that can truly reflect the comprehensive quality of scientific researchers, to guide and encourage scientific researchers to develop in a comprehensive and healthy direction [7]. The establishment of a comprehensive evaluation index system for scientific research talents in colleges and universities not only affects the personal interests of scientific research personnel, but also helps school leaders and management personnel to comprehensively and accurately grasp scientific research results. On the other hand, it also provides more comprehensive and accurate information for most employers when selecting scientific research talents. Therefore, a scientific and reasonable scientific talent evaluation system and its effective implementation is an effective supervision and guarantee of the teaching quality of college talents.

Therefore, this paper proposes to evaluate the teaching ability of contemporary college talents based on BP neural network and fuzzy mathematical model. This article first standardized the samples. After determining the training and test samples, use trial and error to determine the number of hidden layer neurons. Introduce fuzzy mathematical theory and construct fuzzy similarity matrix to describe the fuzzy relationship between factor domain and judgment domain. Calculate membership to get comprehensive evaluation results. Finally, this paper uses statistical methods to plot the results into statistical graphs and combines simulation results to obtain performance comparison results.

2 Related work

Throughout the history of foreign education, education has a long history and relevant work experience is also quite rich. Based on the importance of higher education, people’s evaluation of the quality of higher education also has a high degree of attention. At present, many countries in the world have set up evaluation institutions for higher education, such as the international INOAAHE – International Higher Education Quality Assurance Organization Alliance, APQN established in the Asia-Pacific region – Education Quality Assurance Organization Alliance. They are all alliance organizations established to guarantee the quality of higher education, and these organizations meet regularly every year to discuss the problems encountered in the education evaluation process [8]. In addition, many developed countries have also conducted extensive and in-depth research and practice on the quality of education. For example, some countries represented by the United Kingdom and the United States have established a relatively comprehensive evaluation system for colleges and universities and applied the system to all disciplines in the country, in which large-scale educational quality assessment was carried out [9]. Foreign graduate students pay more attention to the examination of the thesis in the comprehensive quality evaluation, and the independent research ability and innovation ability are also reflected in the graduate students’ writing papers [10].

The emergence of neural networks has provided a direction for related work in higher education. Zhang Lin applied neural networks to higher education and found that students were short of learning. After analyzing the students’ hobbies and knowledge systems, she predicted their behavior. Evaluate teachers ‘teaching effects and students’ learning effects, find out information that promotes or hinders teaching, and make targeted adjustments to improve teaching results [11]. At present, the application of neural networks in the field of education mainly focuses on the research and analysis of data in university information management systems, including student basic information systems, student selection and grade systems, and book lending systems. However, most of the current major researches remain at the theoretical stage, mainly discussing the feasibility of the implementation of neural networks in the information management system and the possible effects. There are not many mining, and analysis combined with actual data [12]. In “Application of Neural Network Technology in Teaching Management System”, Cheng Junjing et al. used decision tree algorithm to study the data of students’ organic processing practice. The study found that the relationship between students’ internship scores and gender, theoretical achievement and actual operation time can better guide teachers in the process of teaching according to the characteristics of various students to achieve the effect of improving students’ performance. However, their research data sample size is only 500, and the data dimension is only 4 dimensions. The data volume is relatively small, which is not enough to reflect the advantage of neural network in discovering knowledge from a large amount of data [13]. Others have discussed the many applications of neural networks in the field of education, including Chen Haiyu et al., “Application Research of Neural Network in Employment Guidance in Higher Vocational Colleges” [14], Xia Tao et al., “Application of Neural Network and Expert System in English Teaching” [15], Zhang Yue, “Application of Neural Network Technology in Data Analysis of College Entrance Examination” [16], Peng Yue III1 “Exploration Research on Neural Network Technology in Postgraduate Cultivation” and Huang Wenzhong’s “Research on Library Circulation Neural Network Model”, etc. [17].

In recent years, many scholars in China have conducted a lot of research on the comprehensive quality evaluation of graduate students and have accumulated a lot of successful experiences. Among them, the representative one is the fuzzy evaluation method of college students’ talents proposed by Professor Li Lijuan of Tongji University. The theory points out that the talent structure can be divided into three elements: knowledge, ability and personality, and expands the basic elements and individualized elements into 32 evaluation indicators, according to which the volume theory is proposed. Professor Li Lijuan constructed a fuzzy evaluation system for talents. The system can carry out various evaluations such as personality evaluation, module evaluation, combination evaluation and overall evaluation, and can also carry out collective evaluation, and can also carry out staged self-evaluation at any time according to the actual situation of students [18]. In addition, on the basis of fuzzy analytic hierarchy process and fuzzy comprehensive evaluation, Cheng Zhan, etc., took Wuhan University of Technology as an example to study the comprehensive quality evaluation of college graduates [19]. Jiang Dejun et al. used fuzzy mathematics theory to construct a quadratic quantitative model based on fuzzy comprehensive evaluation and carried out a more targeted comprehensive quality evaluation research [20]. Jiang Fei et al. combined the Delphi method with the analytic hierarchy process to construct a theoretical model of the graduate evaluation system [21] These studies have achieved some results in some aspects. However, the comprehensive quality evaluation of graduate students involves many influencing factors, both qualitative and quantitative, and the factors are related to each other. Therefore, any evaluation method cannot guarantee the comprehensiveness and scientific of the evaluation. In addition, compared with some developed countries abroad, there are still some gaps between theoretical research and practical application of the comprehensive quality evaluation methods for graduate students in China. At the same time, some scientific research theories have not yet reached the height of practical application, and various evaluation methods are also independently solved the practical problems encountered and did not rise to systematic integrated research.

3 Theoretical analysis

3.1 Capacity evaluation model

The model is an abstract description of the characteristics of things. It is a tool that combines real problems with scientific problems such as mathematics and computers and uses the methods and theories of related disciplines to realize the analysis, research and development of problems.

Model-based evaluation is generally divided into three aspects, including the object to be evaluated, evaluation indicators and evaluation methods. The subject of the evaluation is evaluated by what kind of criteria; the evaluation index is evaluated by what kind of criteria: the evaluation method is used to achieve the evaluation, as shown in Fig. 1.

Fig. 1

Structure of the capability evaluation model.

3.2 Evaluation index system

The influencing factors of college students’ scientific research ability include external factors such as training environment, and internal factors such as their own ability. At present, China has basic external factors for the cultivation of college students. Internal factors play a major role in the impact of scientific research capabilities, and external factors are related to certain internal factors, which indirectly affects students’ research ability. Therefore, this paper is limited to evaluating the intrinsic ability of college students. According to the basic process of the cultivation of scientific research ability of college students and the process of mastering knowledge, as shown in Figs. 2 and 3, this paper divides the evaluation of scientific research ability of college students into three aspects: learning ability, practical ability and innovation ability. The learning ability mainly runs through the process of learning the knowledge of the predecessors. It is mainly divided into self-learning ability and passive learning ability, and the passive learning ability is mainly reflected in the professional basic knowledge that it has mastered; As far as scientific research is concerned, the factors influencing the scientific research ability of college students include external factors such as the cultivation environment, and internal factors such as their own capabilities. At present, China has basically various external factors for the cultivation of college students. Internal factors play a major role in the impact of scientific research capabilities, while external factors are related to certain internal factors, which indirectly affect students’ scientific research capabilities. Therefore, this paper is limited to the evaluation of college students’ intrinsic ability.

Fig. 2

Analysis image of the scientific research ability.

Fig. 3

The process of mastering knowledge.

Practice is the process of solving scientific research problems, and it is a cyclical process of discovering problems, analyzing problems, trying to solve problems, and solving problems. The overall inductive practice ability can be divided into problem discovery ability, problem analysis ability and problem-solving ability. Finding problems is the re-recognition or discovery of students’ objective knowledge by using existing knowledge, analyzing problems is the process by which students comprehensively summarize or classify problems by comparing various problems, and solving problem is the process by which the student changes the current state of the problem to the target state by means of a method.

The ability to innovate is reflected in the ability of new ideas, new theories, new methods, and new inventions. Whether it has the ability to innovate should be tested with scientific research results, and the basis for evaluating scientific research results is the “Overview of Chinese Core Journals”. The book is the responsibility of Peking University. The selection method of quantitative ordering and qualitative combination is the result of evaluation by nearly 2,000 scientific research experts, which is the evaluation basis of Chinese universities and research institutes. This paper selects the indicators that are mainly related to college students as the index items of this paper. In addition, the dissertation is regarded as a systematic research and training process, and the process from the topic selection to the thesis defense is an important means of cultivating scientific research consciousness and ability, as shownin Fig. 4.

Fig. 4

Evaluation index system of college research ability.

4 Construction of evaluation model based on BP neural network and fuzzy mathematics

At present, there are two main directions for the design of BP neural network model: one is the optimization of network structure; the other is network weight and threshold optimization algorithm. The optimization of weights and devaluations usually uses the gradient descent method. The advantage of the gradient descent method is that in the local optimization and the direction along the error is small, the speed is faster, but it is easy to fall into the local minimum value, which will affect the final result, and the error probability is high. This paper designs an evaluation model based on BP neural network from two aspects of network structure and weight threshold optimization.

4.1 Establishment of BP neural network model structure

The main content of this part is to reasonably select the number of hidden layers and the number of neurons in the input layer, hidden layer and output layer. The single hidden layer BP neural network has strong nonlinear mapping ability, and the evaluation of students’ scientific research ability belongs to a relatively medium network scale. Therefore, this paper chooses to use single hidden layer BP neural network. The input layer neurons used the 2nd level indicator as the final input quantity (n = 8), and the output layer neurons used the final rating result (p = 1). If the number of hidden layer neurons is small, the network accuracy is not high, the fitting is poor, and the training time is long. If the number of hidden layer neurons is larger, although the network accuracy is higher, the network generalization ability is poor. According to experience, the number of hidden layer neurons is generally $m = \frac{n + p}{2}$ or $m = \sqrt{n + p} + a (a \in [0, 10])$

The trial-and-error methods are used to determine the number of neurons in the hidden layer.

According to the collected data, the output value range is (0, 1), and the input value range is [0, 1], so the transfer function of the output layer uses logsig, and the transfer function of the hidden layer uses tansig. The mathematical relationship between the layers is as follows:

Hidden layer: $yi = tan sig ({net}_{j})$ (1) ${net}_{j} = \sum_{i = 1}^{n} w_{ij} x_{i}$ (2)

Output layer: $o_{k} = log sig ({net}_{k})$ (3) ${net}_{k} = \sum_{j = 1}^{m} v_{jk} y_{i}$ (4)

Transfer function: $log sig (x) = \frac{1}{1 + e^{- x}}$ (5) $tan sig (x) = \frac{2}{1 + e^{- 2 x}} - 1$ (6)

Among them, y_i is the output of the j-th neuron of the hidden layer, net_j is the input of the j-th neuron in the hidden layer, x_i is the input of the i-th neuron in the input layer, w_ij is the connection weight between the i-th neuron of the input layer and the j-th neuron of the hidden layer, o_k is the output value of the kth neuron in the output layer, net_k is the input of the kth neuron in the output layer, and v_jk is the connection weight between the j-th neuron of the hidden layer and the kth neuron of the output layer.

i ∈ [1, n] , j ∈ [1, m] , n = 8, m is the number of neurons in the hidden layer. Through the trial and error method, m = 12 is determined. Because the output layer neurons have only one, p = 1, so k = p. Equations (1) to (6) constitute the BP neural network model of the single hidden layer of this paper.

Figure 5 is a BP neural network model structure for this paper. Since the connection lines between the layers are too dense, the connection weight between the layers is not indicated in Fig. 5. The connection weight between the input layer and the hidden layer is W, which is a matrix of 8 rows and 12 columns, and the connection weight between the hidden layer and the output layer is V, which is a matrix of 8 rows and 1 column.

Fig. 5

BP neural network model structure for college students’ scientific evaluation.

4.2 Fuzzy mathematics

Considering that the evaluation of talent teaching ability has great ambiguity, in order to better deal with such fuzzy information, fuzzy mathematics is introduced here. The mathematical method is used to abstractly describe the fuzzy phenomenon and reveal the nature and laws of the fuzzy phenomenon. A fuzzy set is a collection used to express the concept of ambiguity. If the domain is U, the mapping is u_A : U → [0, 1] , x| → u_A (x) ∈ [0, 1], then the fuzzy subset A on U is determined, the mapping u_A is called the membership function of A, and u_A (x) is called the membership degree of x to A. Among them, if the u_A (x) is closer to 0, it means that the degree to which x belongs to A is smaller, otherwise the degree of membership is larger.

(1) Construct fuzzy similarity matrix

To evaluate an object, you can start with the factor set U and the judgement set V. The fuzzy relationship between the factor domain and the judgement domain can be expressed by the evaluation matrix R: $R = (\begin{matrix} r_{11} \dots r_{1 n} \\ r_{21} \dots r_{2 n} \\ ⋮ \dots ⋮ \\ r_{m 1} \dots r_{mn} \end{matrix}) .$

In which, r_ij is the membership degree of the factor u_i to the level v_j, and thus the i-th row R_i = (r_i1, r_i2, ⋯ , r_in) of the matrix R is the single factor evaluation of the i-th factor u_i. It is a fuzzy subset on V.

(2) Calculate the results of fuzzy comprehensive evaluation

After the fuzzy similarity matrix R is determined, the comprehensive evaluation result B is obtained by B = A • R, and the fuzzy comprehensive evaluation model is $b_{j} = \sum_{k = 1}^{m} (a_{k} \cdot r_{kj}), j = 1, 2, \dots, n$ . a_i should satisfy $\sum_{i = 1}^{m} a_{i} = 1$ so that the coefficient of importance of each factor u_i can be characterized.

4.3 Standardization method for indicator data

In order to make the data of different metrics comparable, it is necessary to standardize the data so that the processed data is the same as the metric, so that it is used as a sample for the object evaluation index calculation. Standardized functions usually require strict monotony, and the range of values is clear. Based on the above questions, the standardization of indicators in the evaluation of students’ scientific research ability adopts the min - max method, which can keep the input data at [0, 1].The method is as follows: $\begin{matrix} yij = \frac{x_{ij} - min x_{ij}}{max x_{ij} - min x_{ij}}, \\ i = 1, 2, \dots, n, j = 1, 2, \dots m \end{matrix}$ (7)

Among them, x_ij is the data of the j-th attribute of the i-th evaluation object. minx_ij and maxx_ij represent the minimum and maximum values of the attributes in all data, respectively.

4.4 Design of evaluation index and corresponding grade

The ability evaluation index of this paper is the weighted sum of the indicator data and its weight, and the evaluation index is represented by D. According to the evaluation index range of collected data, the evaluation grades are divided into excellent, good, medium and poor. When O ∈ (0, 0.25], the evaluation level is poor; when O ∈ (0.25, 0.5], the evaluation level is medium; when O ∈ (0.5, 0.75], the evaluation level is good; when O ∈ (0.75, 1), the evaluation level is excellent.

4.5 Sample design

The principles to be followed in the sample design: The number of samples in each class should be the same or similar, and the network uneven training should be avoided, and the number of test and training samples should be the same or similar. The sample design is shown in Tables 1 and 2. The evaluation index in Table 1 is obtained by weighted averaging the weights obtained by combining the analytic hierarchy process and the coefficient of variation method. The acquisition of the weights of each index combines the expert’s point of view and the original data of the evaluated object. Compared with the AHP method, the method combines objective reality and the result is closer to reality.

Table 1
Training samples (partial)

Serial number a b c d

Professional foundation 1 0.87 0.33 0.03

Self-learning 0.97 0.59 0.43 0.21

Problem found 0.9 0.58 0.5 0.1

Analyse problem 0.77 1 0.45 0.09

Solve the problem 0.84 0.81 0.47 0.11

Thesis 0.81 0.63 0.44 0.48

Research project 0.17 0.85 0.02 0

Publication 0.93 1 0.5 0.11

Evaluation index 0.87 0,75 0.5 0.24

Evaluation level Excellent Good Medium Poor

Serial number	a	b	c	d
Professional foundation	1	0.87	0.33	0.03
Self-learning	0.97	0.59	0.43	0.21
Problem found	0.9	0.58	0.5	0.1
Analyse problem	0.77	1	0.45	0.09
Solve the problem	0.84	0.81	0.47	0.11
Thesis	0.81	0.63	0.44	0.48
Research project	0.17	0.85	0.02	0
Publication	0.93	1	0.5	0.11
Evaluation index	0.87	0,75	0.5	0.24
Evaluation level	Excellent	Good	Medium	Poor

Table 2

Test sample (partial)

Serial number	1	2	3	4
Professional foundation	1	0.81	0.28	0.06
Self-learning	0.94	0.6	0.56	0.11
Problem found	0.94	0.59	0.58	0.1
Analyse problem	0.78	1	0.56	0.05
Solve the problem	0.98	0.96	0.69	0.08
Thesis	1	0.63	0.15	0.48
Research project	1	0.85	0	0
Publication	0.89	1	0	0.11

Table 3

Relationship between the number of hidden layer neurons and the maximum value of network mean square error

Quantity	Gradient descent	Quasi-Newton method	LM algorithm
1	0.0713	0.2760	0.4500
2	0.4128	0.3996	0.2772
3	0.3132	0.4380	0.3072
4	0.0232	0.2364	0.1022
5	0.0520	0.2412	0.2388
6	0.1728	0.3132	0.0214
7	0.2172	0.2136	0.3588
8	0.3660	0.0973	0.2436
9	0.0439	0.0324	0.3816
10	0.0197	0.1096	0.4548
11	0.3648	0.1320	0.2400
12	0.2568	0.0894	0.0193

According to the principle of sample design, this paper selects 10 sets of data as training samples from four levels. The choice of test samples is: 5 sets of new samples and 5 sets of trained samples are selected in turn. The number of training samples is 40, and the number of test samples is 40. In this way, the feasibility and adaptability of the BP neural network model to the problem to be solved in this paper is verified.

5 Model verification

Based on the BP neural network and fuzzy mathematics of LM algorithm, gradient descent method and quasi-Newton method, this paper uses the trial and error method to determine the number of hidden layer neurons. According to the various functions, graphical interfaces and simulation tools provided by the neural network toolbox of matlabR2012a, this paper evaluates the BP neural network for evaluating the scientific research ability of college students, which can reduce the workload of the program and improve the work efficiency. The above three algorithms are used for learning, and the maximum number of iterations is set 1000 times, and the initial weight and threshold are randomly given. E(w) is the network mean square error after all samples are iterated once, as in Equation (8). In this paper, the maximum value of the hidden layer neurons is determined by using the maximum value of E (w) in the iterative process. $E (w) = \frac{1}{2} \sum_{p = 1}^{P} {(d_{p} - o_{p})}^{2}$ (8)

Among them, p is the sample number, P is the sample size, d_p is the expected output of sample P, o_p is the actual output of sample P, and w is the vector consisting of the network weights and thresholds.

Figure 6 shows the change effect of the maximum error of the network as the number change of neurons in the hidden layer. When the number of hidden layer neurons is 6 and 12, the maximum error of the network reaches the local minimum. This paper takes the average sum of the three algorithm values as a reference. This paper takes the average sum of the three algorithm values as a reference. When m = 6, the average value of E (w) is 0.1691; when m = 12, the average value of E (w) is 0.1218. Therefore, the number of hidden layer neurons in the BP neural network model is 12, and the BP neural network structure is 8-12-1.

Fig. 6

Relationship image between maximum error of the network and the number of hidden layer neurons.

This paper compares the LM algorithm with the gradient descent method and the quasi-Newton method from four aspects: iteration number, network error, generalization ability and prediction accuracy. Therefore, the selection of some parameters should be consistent, which are shown in Figs. 7–9. The single hidden layer BP network of 8-12-1 is used. The maximum number of iterations is 1000, the target error is 0 by default, and the objective function adopts the network total mean square error. The training data of Table 1 is used, and the constructed BP neural network is learned, and the training is ended when the network reaches the maximum number of trainings or target errors. The weight and threshold matrix of the network are preserved, and it is used as the optimal network model for evaluating the scientific research ability of college students, and this model is used to evaluate the scientific research ability of the subsequent college students.

Fig. 7

Performance curve of BP network training process based on gradient descent method.

Fig. 8

Performance curve based on the training process of quasi-Newton BP network.

Fig. 9

Performance curve of BP network training process based on LM algorithm.

The feasibility of the standard BP neural network based on the gradient descent method for the problem is verified. When the experimental analysis uses this method to achieve convergence, the network mean square error is 0.0106, and the number of iterations is 1000.

The feasibility of the BP network based on the quasi-Newton method for the problem in this paper is verified. When the experimental analysis uses this method to achieve convergence, the network mean square error is 1.371e-7, and the number of iterations is 350.

The feasibility of the BP network based on LM algorithm for the problem of this paper is verified. When the experimental analysis uses this method to converge, the network mean square error is 6.35e-7, reaching 43 convergences.

Figure 10 is the network prediction accuracy based on the three algorithms. It can be seen that the network generalization ability based on LM method is 99.50%, and the overall network recognition rate is 92.50% : the network generalization ability based on the quasi-Newton method is 85.00%, and the overall recognition rate of the network is 87.50%; the network generalization ability based on the gradient descent method is 30.50%, and the overall recognition rate of the network is 32.50%.

Fig. 10

Network prediction accuracy based on three algorithms.

According to the analysis, the network generalization ability and the overall recognition rate based on the gradient descent method are relatively poor; it is analyzed that the BP neural network based on the gradient descent method can basically recognize the trained samples, and the untrained samples are basically not recognized; The network based on the quasi-Newton method has a poorer network generalization ability and overall network recognition rate than the LM algorithm. According to the calculation, the sample-based mean square error based on the LM algorithm and the quasi-Newton algorithm is obtained. Although the mean square error of the network based on LM algorithm is smaller than the mean square error of the network based on quasi-Newton method, the network mean square error of both can satisfy the prediction of the scientific research ability evaluation index. Therefore, from the perspective of iteration number and network mean square error, network overall recognition rate and network generalization ability, this paper uses LM-based BP neural network evaluation model to evaluate the scientific research ability of college students.

On this basis, its network parameters are set: The number of iterations is 1000, based on the minimum error generated during the model verification process, the network mean square error is set to 0.0001, the number of network layers is 1 layer, the number of input neurons is 8, the number of neurons in the output layer is 1, a the number of neurons in the hidden layer is 12. At the same time, the initial weights and thresholds are randomly generated, and the transfer function uses Equations (5) and (6).

Through the above analysis, the evaluation model based on BP neural network and fuzzy mathematics satisfies the evaluation of scientific research ability of college students in terms of training efficiency, prediction accuracy and sample verification. In particular, the BP neural network evaluation index prediction rate based on LM optimization algorithm reaches 99.5%, which indicates that the BP neural network-based evaluation model of college students’ scientific research ability is feasible to solve the practical problem. So far, an evaluation model for the research ability of college students is constructed.

6 Results discussion and analysis

Based on the above analysis, this paper constructs a BP neural network and fuzzy mathematics-based evaluation model for college students’ scientific research ability.

First of all, the evaluation index system of college students’ scientific research ability is constructed. The indicator system is divided into two levels, and the first level mainly includes learning ability, practical ability and innovation ability. The second level includes professional basic ability, self-learning ability, problem finding ability, problem analysis ability, problem solving ability, project participation, dissertation, and published papers. Then, the rules for scoring the scientific research ability of college students are constructed. In order to enhance the credibility of the scoring rules, the reliability and validity of the content of the scale can be tested. The test indicates that the content of the scale in the scoring rules is more credible. Finally, the combination weighting method of AHP and coefficient of variation method is used to weight the evaluation index. It combines objective weighting and subjective weighting method to make the evaluation result closer to reality and provide a basis for training samples.

In order to make each quantitative index comparable, the min-max method is used to normalize the indicator data so that the data is at [0, 1]. Data were collected for seniors and graduate students, and the data was standardized. The weighted sum of the sample data and the index weight is specified as the evaluation index of the evaluation object, thereby making a basis for the BP neural network-based training sample. According to the collected samples, four evaluation grades of excellent, good, medium and poor were designed and according to the design rules of the sample, 40 training samples and 40 test samples were selected as experimental data for model construction.

Through the trial and error method, the number of neurons in the hidden layer is determined. According to the network mean square error and the number of iterations, the weight threshold of the BP neural network optimized by the LM algorithm is determined. At the same time, according to the network generalization ability and prediction accuracy compared with the results of the weighted evaluation method, the 8-12-1 single hidden layer BP neural network evaluation model based on LM method is determined.

7 Conclusion

Many colleges and universities have gradually added courses to train students’ scientific research ability, but the management of students’ scientific research activities only focuses on training and training. As for the evaluation of students’ scientific research ability, there is no specific standard. Based on BP neural network and fuzzy mathematics, this study builds an evaluation index system for college students’ scientific research ability according to the process of scientific research ability cultivation and knowledge acquisition process. At the same time, this paper studies the evaluation of students’ scientific research ability from three aspects: evaluation index system, evaluation model and rating prototype system, and forms the basic research ideas and achieves the basic effects. In addition, this paper studies the evaluation model of college students’ scientific research ability. In this paper, a single hidden layer BP neural network is used as the evaluation model, which specifies that the number of neurons in the input layer is a level 2 index, and the number of neurons in the output layer is an evaluation index. At the same time, based on the construction model, the evaluation model of college students’ scientific research ability is verified. The constructed model has certain benefits. Using the designed training samples and test samples, BP neural network was trained based on gradient descent method, quasi-Newton method and LM algorithm, and the benefit of the model was verified by model comparison.

Footnotes

Acknowledgments

This research was supported by China Humanity and Social Science Program Foundation of the Ministry of Education of People’s Republic of China under Grant No. (20YJC740109), with the name of “English Teachers’ Interactional Competence Development”.

References

Zhang

, Research and Application of Capability Evaluation Model Based on BP Neural Network Educational Technology—Taking the Problem-Based Learning Teaching Process Learning as an Example[J], Journal of Computational and Theoretical Nanoscience 13(9) (2016), 6210–6217.

Peng , Lifang , Lingling , A service innovation evaluation framework for tourism e-commerce in China based on BP neural network[J], Electronic Markets 24(1) (2014), 37–46.

Liu

, Zhang

L.U.

, et al., Bitterness intensity prediction of berberine hydrochloride using an electronic tongue and a GA-BP neural network[J], Experimental & Therapeutic Medicine 7(6) (2014), 1696.

Cao

W.D.

, Yan

C.P.

, Ding

, et al., A continuous optimization decision making of process parameters in high-speed gear hobbing using IBPNN/DE algorithm[J], International Journal of Advanced Manufacturing Technology 85(9–12) (2016), 2657–2667.

Zhang

and Wang

, A Study on Readers’ Satisfaction of University Library Based on BP Neural Network[J], Journal of Computational and Theoretical Nanoscience 13(9) (2016), 6005–6010.

Sahoo

, Jha

and Prashar

, A novel approach for Spam Email Filtering Using Machine Learning, Journal of Cybersecurity and Information Management 2(2) (2020), 44–57.

Bai

, Meng

and Yao

, A forecasting method of forest pests based on the rough set and PSO-BP neural network[J], Neural Computing & Applications 25(7–8) (2014), 1699–1707.

Alhabib

and Salama

A.A.

, Using Moving Averages To Pave The Neutrosophic Time Series, International Journal of Neutrosophic Science 3(1) (2020), 14–20.

You

and Cao

, Study of Liquid Lithium Coolant Interaction Based on BP Neural Network Optimized by Genetic Algorithm[J], Journal of Fusion Energy 34(4) (2015), 918–924.

10.

Abdel-Kader

, Abd-El Salam

and Mohamed

, Hybrid Machine Learning Model for Rainfall Forecasting, Journal of Intelligent Systems and Internet of Things 1(1) (2020), 5–12.

11.

Pan

, Yang

, Shi

, et al., BP Neural Network Application Model of Predicting the Apple Hardness[J], Journal of Computational & Theoretical Nanoscience 12(9) (2015), 2802–2807.

12.

Tang

S.L.

, Zhang

X.X.

, Tang

S.L.

, et al., Research on early-warning and evaluation of traditional Chinese medicine industry security based on BP neural network[J], Chinese Traditional & Herbal Drugs 48(2) (2017), 406–418.

13.

Guan

, Shi

and Qi

, Objective evaluation of fabric pilling based on bottom-up visual attention model[J], Journal of the Textile Institute Proceedings & Abstracts 108(4) (2016), 597–604.

14.

, Rong

, Chang

, et al., Decision-making of feature operation chain considering processing requirements and manufacturing stability[J], International Journal of Advanced Manufacturing Technology 87(5–8) (2016), 1–13.

15.

and Cheng

, Heat treatment process evaluation method based on BP neural network optimized by genetic algorithm[J], Heat Treatment of Metals 39(6) (2014), 151–154.

16.

Shi

, Li

, Yang

, et al., A double CPM for project duration estimates by BP neural networks[J], Journal of Xian University of Architecture & Technology 46(2) (2014), 276–265.

17.

Improving event-based rainfall-runoff simulation using an ensemble artificial neural network based hybrid data-driven model[J], Stochastic Environmental Research and Risk Assessment 29(5) (2015), 1345–1370.

18.

Qiu

and Song

, Predicting the Direction of Stock Market Index Movement Using an Optimized Artificial Neural Network Model[J], Plos One 11(5) (2016), e0155133.

19.

Zhao

, Jin

, Guo

, et al., Dynamic risk assessment model for flood disaster on a projection pursuit cluster and its application[J], Stochastic Environmental Research & Risk Assessment 28(8) (2014), 2175–2183.

20.

Tong

, Zhao

, Ye

, et al., Research on energy efficiency evaluation for overhead crane[J], Kybernetes 45(5) (2016), 788–797.

21.

Zhang

, Jiang

, Ma

, et al., Evaluating soil reinforcement by plant roots using artificial neural networks[J], Soil Use and Management 31(3) (2015), 408–416.