Integration of deep learning and vocational college education evaluation index system model construction

Evaluation of higher education is an important means to improve the quality of teaching in higher education, and is an important tool for internal self-management and external supervision of higher education. Constructing a scientific and reasonable evaluation index system of higher education is an important way to guarantee the improvement of the quality of higher education, and a complete evaluation index system should contain the indicators of teaching quality, student development, social service, school conditions and other aspects [1-4].

First of all, teaching quality is one of the most core tasks of higher education, and its evaluation index should focus on students’ learning effect and ability cultivation, which can be carried out from the aspects of curriculum, teaching methods, teaching resources, teaching results and students’ evaluation, and can also consider students’ evaluation of teaching effect and students’ employment after graduation, etc. [5-8]. Secondly, the core of student development in higher education is to cultivate the ability of students to develop comprehensively, and the development of students becomes an important basis for evaluating the quality of higher education. Student development includes not only the cultivation of academic ability, but also the cultivation of students’ sense of innovation, practical ability, social responsibility and lifelong learning ability [9-12]. Therefore, the evaluation indexes can include students’ participation in scientific research projects, students’ internship experience, the cultivation of social practice ability and students’ participation in community organizations. In addition, excellent faculty is the key to the improvement of education quality in colleges and universities. The evaluation indexes can include teachers’ academic level, teaching ability, practical experience, ability to cultivate high-level talents, and the stability of the faculty team [13-16]. And it can consider the academic structure, academic structure, title structure of the faculty as well as the long-term mechanism for training the faculty. Of course, the evaluation indexes of college education also include school conditions, scientific research achievements and other aspects [17-19].

Zhao,Y. examined the problems and improvement measures in the evaluation process of physical education teaching in colleges and universities. Based on the theory of multiple intelligences, the teaching evaluation index system was constructed from teaching, management and other aspects, and a fuzzy comprehensive evaluation model was proposed, and the effectiveness of the model was proved through case study [20]. Hu,C.et al. emphasized the importance of the teaching evaluation and created a teaching quality assurance index system of colleges and universities based on the CIPP model, and carried out an empirical study with colleges and universities as a case study. The survey and experimental results verified that the teaching quality evaluation index system of colleges and universities based on the CIPP model has good applicability [21]. Yan,J. combined with the educational needs of the new era and the teaching characteristics of research undergraduate universities, constructed a classroom teaching evaluation index system Ming, in order to guide teachers and students to change the concept of teaching, improve the teaching methodology, improve the quality of teaching and promote the common development of teachers and students [22]. Yang,S. explored the construction of distance education teaching quality index system. Through the questionnaire survey and literature analysis, the distance education teaching quality index system is given, including teaching service, teaching effect evaluation and other indicators. And based on the hierarchical analysis method, the weight of each index is revealed, and the results of distance education teaching quality index system are calculated [23]. Liu,X.et al. constructed the evaluation index system of entrepreneurship education in colleges and universities based on the evaluation model of context and process, and investigated the pilot university of entrepreneurship education as the research object, and put forward corresponding countermeasures against the evaluation results, which is of some reference significance to the research of entrepreneurship education and evaluation in colleges and universities [24]. Wang,S. discussed the construction of distance education teaching quality index system through questionnaire survey and literature analysis, including teaching service, teaching effect evaluation and other indexes. Wang, C. et al. took the theory of scientific knowledge mapping as the guiding idea, and for the current situation of physical education teaching evaluation in colleges and universities, applied the method of experimental research to carry out empirical research in several colleges and universities. The results showed that teachers and students were relatively satisfied with the new evaluation method [25]. Huang,M. et al. based on the inspiration of PBL theory, outlined the application of PBL theory in music teaching by analyzing the evaluation index system of music classroom teaching in colleges and universities and constructed an environmental impact information system. The results of the study emphasize that PBL has a positive impact on students’ learning and provides a theoretical basis for the reform of the music classroom environmental information system [26]. Luo,T. explores the evaluation index system of colleges and universities in the midterm of the “14th Five-Year Plan” and its role and significance, proposes a method for the construction and optimization of the index system, and argues that this system can be used in actual teaching and learning based on the case study. Based on the case study, it demonstrates the application of this system in the actual evaluation. Provide reference for the scientific and standardized evaluation of colleges and universities [27].

This paper establishes the education assessment index system of vocational colleges and universities according to the deep learning theory, and constructs the comprehensive assessment model of vocational colleges and universities’ education based on the ANP-CRITIC-Cloud model. The ANP method is used to determine subjective weights, the CRITIC method is used to determine objective weights, and a more scientific combination of weights is determined with the help of deep learning ideas. Using the comprehensive assessment model of education in vocational colleges and universities, the education scale of vocational colleges and universities is divided into four categories, and the cloud similarity measurement is carried out on the basis of this model, so that the obtained education assessment results are more accurate. Finally, school A was selected to verify the feasibility of the education evaluation model.

2

Construction of Evaluation Indicator System for Vocational Education Integrating Deep Learning

2.1

Deep learning features

1)

Student-centered active learning

Students are the main body of learning activities, only when students really take the initiative of learning and actively participate in the learning process, they can have a high sense of responsibility and commitment to their own learning, so student-centered active learning is the basis for realizing deep learning is even more necessary. The primary difference between deep learning and shallow learning is the motivation to learn, and the prerequisite for realizing deep learning is the initiative of the learners and their strong interest in the content to be learned, which will support them to complete the entire learning process, solve the complex problems faced in the study one by one and realize the construction of knowledge, and ultimately realize the enhancement of higher-order thinking skills.

2)

Complex information processing process

Deep learning [28] is a complex information processing process, including deep understanding of knowledge, integration of information resources, knowledge construction, migration and application, reflection and criticism. First of all, deep learning is a kind of learning based on understanding, if the cognitive process of the learner is only simple memorization and repetition, but lack of understanding, the knowledge in the mind can not form a long-lasting memory and is very easy to be confused. Therefore, deep learning emphasizes learning on the basis of understanding, requiring learners to memorize on the basis of clarifying concepts and grasping the essence of the content, and only through deep processing and understanding can knowledge be better mastered.

The next step is to integrate information resources. No knowledge exists in isolation, more or less connected with other subjects and fields, and the knowledge gained from books or teachers’ teaching process alone cannot meet the requirements of deep learning, and the integration of knowledge from various disciplines and fields can help learners better understand the knowledge. In addition, deep learning also emphasizes the integration between old and new knowledge, linking new ideas and knowledge with previous knowledge and integrating them into the original cognitive structure to form a new and more complete knowledge system, so as to achieve a long-lasting memory of knowledge.

The third part is knowledge construction. On the basis of obtaining information resources in multiple ways and from multiple angles, learners should learn to analyze and judge the information obtained, assimilate and conform to it, link new knowledge and new ideas with the original cognitive structure, construct new knowledge on the basis of the original cognitive structure, so that the knowledge system is constantly improved and expanded. Finally, migration and application. Migration and application can realize the transformation from theory to practice, the purpose of learning is for better application, and deep learning emphasizes more on the ability of learners to flexibly apply what they have learned to real situations, especially the ability to solve complex problems.

3)

Development and enhancement of higher-order thinking, problem solving and other higher-order abilities

Compared to shallow learning, deep learning is a type of higher-order learning that emphasizes the development and enhancement of learners’ higher-order thinking, problem-solving, and other higher-order abilities. Higher-order thinking refers to a higher level of cognitive ability, which corresponds to the four levels of “analyzing, synthesizing, evaluating, and creating” in Bloom’s cognitive classification goals. Higher-order thinking is the key to realizing deep learning, and it is also the result of the pursuit of deep learning. Through deep understanding of knowledge, integration of information from various aspects and establishment of interconnections, learners construct a more comprehensive and integrated reticulated knowledge structure system, and flexibly apply what they have learned to real-world problems or new problem situations, ultimately realizing the results of deep learning in terms of the development and enhancement of higher-order thinking and problem-solving abilities.

2.2

Establishment of teaching evaluation indicators for vocational colleges and universities

Through the use of relevant literature, teaching evaluation should encompass the entire teaching process of teachers, including teaching objectives, teaching design, teaching process, teaching effects, and other teaching elements. In order to maintain the consistency of the descriptions among the indicators at all levels, this paper comprehensively considers the relevant literature review, the concept of “learning evaluation and teaching” and the characteristics of teaching mode, and expresses the first-level indicators of teaching evaluation as “learning design”, “learning environment”, “learning process” and “learning effect”, and the results of the evaluation indicators are shown in Table 1.

Table 1.

Education evaluation index system

	Primary indicator	Secondary indicator
First category	Learning design(U1)	Learning goal(V1)
		Learning strategy(V2)
		Learning method(V3)
Second category	Learning environment(U2)	Learning support(V4)
Second category	Learning environment(U2)	Learning resources(V5)
Third category	Learning process(U3)	Learning link(V6)
		Learning content(V7)
		Learning participation(V8)
		Learning assessment(V9)
Fourth category	Learning effect(U4)	Learning will(V10)
		Learning ability(V11)
		Learning quality(V12)

1)

“Learning Design” Level 1 Indicator

Learning design indicator refers to the pre-preparation of students’ learning by teachers in the process of deep learning teaching, which guides the development of the whole teaching activities and is the basis of the whole teaching activities. Therefore, it is reflected in three aspects: learning objectives, learning strategies, and learning methods.

2)

“Learning Environment” Level 1 Indicator

The learning environment indicator refers to the physical environment in the deep learning teaching mode, which provides support for the development of deep learning teaching. The influence of this learning environment on the quality and effectiveness of deep learning teaching cannot be ignored. Therefore, it is reflected in two aspects: learning support and learning resources.

3)

“Learning Process” Level 1 Indicators

The learning process indicator covers the content, behavior, and assessment of both teachers’ teaching and students’ learning in all aspects of deep learning teaching. Combining the characteristics of the deep learning teaching mode, this indicator reflects the four aspects of the learning process: learning content, learning participation, and learning assessment.

4)

“Learning Effectiveness” Level 1 Indicator

Learning effect indicators include student learning and teacher teaching. Student learning is reflected in terms of willingness to learn and ability to learn, while teacher teaching is reflected in terms of quality of learning.

3

Deep Learning Education Evaluation Model Construction

3.1

Methodology for determining the weights of evaluation indicators

3.1.1

Analysis and determination of subjective weights

The ANP method is applied to analyze and determine the subjective weights.

The principle and steps of ANP operation are as follows:

1)

Construct ANP structure

ANP [29] has a hierarchical structure similar to a complex network structure, which is divided into a control layer and a network layer. The control layer consists of the decision goal and criterion layers. The network layer is mainly composed of element groups, elements within them, and connected elements. According to the mutual influence and affiliation between the elements, the elements are aggregated and combined at different levels to form a multilevel network structure model as shown in Figure 1.

On the basis of the constructed evaluation indicator system, the mutual influence relationship between indicators is determined and the ANP structure is formed. Judge the relevance of the secondary indicators in the evaluation indicator system through the results of the correlation analysis of the influencing factors and the use of expert group questionnaires.

2)

Determine the degree of superiority

After constructing the evaluation ANP structure, in order to reflect the relative importance of the elements of the network layer of the structure to each criterion, the concept of dominance is introduced, and the element that contributes more or is more important to criterion i has a greater degree of dominance. The degree of dominance is categorized into direct and indirect dominance.

Direct dominance degree: For the i th criterion, the direct comparison of the degree of influence of two elements in the network layer on it can obtain the direct dominance degree of these two elements on the criterion, and according to the results, determine the relative importance weights or the relative order of superiority of each element in the network layer on each criterion in the digging system layer.

Indirect dominance: For the i rd criterion, compare the degree of influence of two elements in the network layer on one other element (also known as the sub-criterion) governed by that criterion, and obtain the degree of indirect dominance of those two elements over that element.

In the ANP structure, the determination of the relative importance of indicators takes the degree of direct dominance. The elements in the ANP structure are not independent of each other, in in a feedback relationship, so the indirect dominance degree is used to determine the relative weights. Comparison of two two elements uses 1-9 rating scale method.

3)

Construction of judgment matrix and test

Let the criterion layer has P₁,P₂,...,P_m criterion, the network layer has n element groups, denoted as B_ij(j = 1,2,...,n), and the elements in each element group are denoted as B_ij(j = 1,2,...,n). Compare the indirect dominance degree of the elements with feedback relationship, take P_i in the criterion layer as the primary criterion, and take a B_ij in B_i as the secondary criterion, and use the 1-9 scoring scalar method to compare all the elements in B_i in two-by-two comparison, and get the influence of any two elements in B_i under the dominance of P_i on the secondary criterion B_ij The degree of influence. As a result, a normalized eigenvector matrix W_ij is constructed with B_ij as the sub-criterion.

If element B_ia in element group B_i is used as the sub-criterion, the corresponding comparison matrix and normalized eigenvectors are obtained.

For element group B_i with n elements, there are n judgment matrices and normalized eigenvectors, and all the eigenvectors are formed into a new matrix W_ij, i.e., the judgment matrix under the condition of B_i as the main criterion is obtained as follows.

(1)

W_{i j} = [\begin{matrix} W_{i 1}^{(j 1)} & W_{i 1}^{(j 2)} & \dots & W_{i 1}^{(j n)} \\ W_{i 2}^{(j 1)} & W_{i 2}^{(j 2)} & \dots & W_{i 2}^{(j n)} \\ ⋮ & ⋮ & ⋱ & ⋮ \\ W_{i n}^{(j 1)} & W_{i n}^{(j 2)} & \dots & W_{i n}^{(j n)} \end{matrix}]

\[{{W}_{ij}}=[\begin{matrix} W_{i1}^{(j1)} & W_{i1}^{(j2)} & \cdots & W_{i1}^{(jn)} \\ W_{i2}^{(j1)} & W_{i2}^{(j2)} & \cdots & W_{i2}^{(jn)} \\ \vdots & \vdots & \ddots & \vdots \\ W_{in}^{(j1)} & W_{in}^{(j2)} & \cdots & W_{in}^{(jn)} \\ \end{matrix}]\]]

The consistency test is performed on the judgment matrix, where CI = 0 indicates full consistency, and the larger the value of CI, the worse the consistency.

(2)

C I = \frac{λ_{\max} - n}{n - 1}

\[CI=\frac{{{\lambda }_{\max }}-n}{n-1}\]]

Eq:

CI - the index of coherence.

λ_max - Maximum eigenvalue of the judgment matrix.

n - judgment matrix order.

To measure the magnitude of CI , a random consistency indicator is introduced RI . The consistency ratio is calculated CR. A consistency ratio of CR ≤ 0.1 means that the consistency test is passed and the result is acceptable.

(3)

C R = \frac{C I}{R I}

\[CR=\frac{CI}{RI}\]]

Eq:

CR - Ratio of consistency.

RI - Randomized consistency index.

4)

Construct the unweighted supermatrix

The unweighted supermatrix governed by the corresponding criterion is obtained by forming all W_ij(i = 1,2,...,n;j = 1,2,,...,n) into a chunked matrix W of order N×N, W_ij being a submatrix of W.

The supermatrices W are all non-negative matrices, and the normalized eigenvectors obtained only when there are no interrelationships between the elements form a 0-matrix of W_ij.

5)

Calculation of weighted supermatrices

Taking the criterion layer P_i(i = 1,2,...,n) as the primary criterion and the network layer B_i(i = 1,2,...,n) as the secondary criterion in turn, the 1-9 scale method is used to compare the indirect dominance of any other two-element group to obtain the weight matrix MA as follows.

(4)

M = [\begin{matrix} M_{11} & M_{12} & \dots & M_{1 n} \\ M_{21} & M_{22} & \dots & M_{2 n} \\ ⋮ & ⋮ & ⋱ & ⋮ \\ M_{n 1} & M_{n 2} & \dots & M_{n n} \end{matrix}]

\[M=[\begin{matrix} {{M}_{11}} & {{M}_{12}} & \cdots & {{M}_{1n}} \\ {{M}_{21}} & {{M}_{22}} & \cdots & {{M}_{2n}} \\ \vdots & \vdots & \ddots & \vdots \\ {{M}_{n1}} & {{M}_{n2}} & \cdots & {{M}_{nn}} \\ \end{matrix}]\]]

The weighted supermatrix M is used to weight the supermatrix to obtain the weighted supermatrix $\bar{W}$ $\bar{W}$], which is calculated as shown below.

(5)

\begin{array}{l} \bar{W_{i j}} = M_{i j} W_{i j} \\ \bar{W} = [\begin{matrix} M_{11} W_{11} & M_{12} W_{12} & \dots & M_{1 n} W_{1 n} \\ M_{21} W_{21} & M_{22} W_{22} & \dots & M_{2 n} W_{2 n} \\ ⋮ & ⋮ & ⋱ & ⋮ \\ M_{n 1} W_{n 1} & M_{n 2} W_{n 2} & \dots & M_{n n} W_{n n} \end{matrix}] \end{array}

\[\begin{array}{*{35}{l}} \overline{{{W}_{ij}}}={{M}_{ij}}{{W}_{ij}} \\ \bar{W}=[\begin{matrix} {{M}_{11}}{{W}_{11}} & {{M}_{12}}{{W}_{12}} & \cdots & {{M}_{1n}}{{W}_{1n}} \\ {{M}_{21}}{{W}_{21}} & {{M}_{22}}{{W}_{22}} & \cdots & {{M}_{2n}}{{W}_{2n}} \\ \vdots & \vdots & \ddots & \vdots \\ {{M}_{n1}}{{W}_{n1}} & {{M}_{n2}}{{W}_{n2}} & \cdots & {{M}_{nn}}{{W}_{nn}} \\ \end{matrix}] \\ \end{array}\]]

6)

Calculate the limit supermatrix to obtain subjective weights

The step k dominance degree between elements is obtained by k powering $\bar{W}$ $\bar{W}$] and is denoted as ${\bar{W}}^{k}$ ${{\bar{W}}^{k}}$]. To obtain the importance ranking of all elements for the whole network, the final limit ranking is obtained by cumulative influence to obtain the limit supermatrix ${\bar{W}}^{\infty}$ ${{\bar{W}}^{\infty }}$], i.e., the final weight ranking of all the elements in the ANP network, which is computed as shown below.

(6)

{\bar{W}}^{\infty} = \lim_{N \to \infty} \frac{\sum_{k = 1}^{N} {\bar{W}}^{k}}{N}

\[{{\bar{W}}^{\infty }}=\underset{N\to \infty }{\mathop{\lim }}\,\frac{\underset{k=1}{\overset{N}{\mathop \sum }}\,{{{\bar{W}}}^{k}}}{N}\]]

If the obtained limit converges and is unique, the eigenvector of the limit supermatrix is the desired subjective weight u_j.

3.1.2

Analysis and determination of objective weights

In the educational evaluation model, the CRITIC method [30] is used to analyze and determine the objective weights.The principle of operation and steps of the CRITIC assignment method are as follows:

1)

Data standardization

According to the quantification of qualitative and quantitative indicators, the indicator value of each indicator of the evaluation indicator system is obtained.

Assuming that there is a total of m measurement point, n indicators, and x_ij represents the value of the j th indicator for the i th measurement point, the evaluation matrix can be expressed as follows: (7) $X = [\begin{matrix} x_{11} & x_{12} & \dots & x_{1 n} \\ x_{21} & x_{22} & \dots & x_{23} \\ ⋮ & ⋮ & ⋱ & ⋮ \\ x_{m n} & x_{m 2} & \dots & x_{m n} \end{matrix}]$ \[X=[\begin{matrix} {{x}_{11}} & {{x}_{12}} & \cdots & {{x}_{1n}} \\ {{x}_{21}} & {{x}_{22}} & \cdots & {{x}_{23}} \\ \vdots & \vdots & \ddots & \vdots \\ {{x}_{mn}} & {{x}_{m2}} & \cdots & {{x}_{mn}} \\ \end{matrix}]\]]

There are positive and negative indicators in the indicators, which need to be standardized, and the larger the indicator value of the positive indicator, the more favorable it is, and it needs to be done positively, as shown in Equation (8), and conversely, the smaller the indicator value of the negative indicator, the more favorable it is, and it needs to be done negatively, as shown in Equation (9).

(8)

x_{i j}^{'} = \frac{x_{i j} - \min (x_{j})}{\max (x_{j}) - \min (x_{j})}

\[x_{ij}^{'}=\frac{{{x}_{ij}}-\min ({{x}_{j}})}{\max ({{x}_{j}})-\min ({{x}_{j}})}\]]

(9)

x_{i j}^{'} = \frac{\max (x_{j}) - x_{i j}}{\max (x_{j}) - \min (x_{j})}

\[x_{ij}^{'}=\frac{\max ({{x}_{j}})-{{x}_{ij}}}{\max ({{x}_{j}})-\min ({{x}_{j}})}\]]

Eq:

$x_{i j}^{'}$ $x_{ij}^{\prime }$] - Standardized indicator value.

max(x_j) - Maximum limit of the value of the j rd indicator.

min(x_j) - The lowest limit of the value taken by the j th indicator.

2)

Calculation of indicator variability

Indicator variability refers to the size of the difference between the values of the same indicator at different measurement points, which is reflected in the form of standard deviation θ. The larger the standard deviation, the greater the volatility of the indicator, i.e., the larger the difference in values between measurement points, the greater the comparative strength of the indicator, and the higher the weight assigned to it, as shown in the following calculation.

(10)

\bar{x_{i j}} = \frac{1}{m} \sum_{i = 1}^{m} x_{i j}

\[\overline{{{x}_{ij}}}=\frac{1}{m}\underset{i=1}{\overset{m}{\mathop \sum }}\,{{x}_{ij}}\]]

(11)

θ_{j} = \sqrt{\sum_{i = 1}^{m} {(x_{i j} - \bar{x_{i j}})}^{2} / (m - 1)}

\[{{\theta }_{j}}=\sqrt{\underset{i=1}{\overset{m}{\mathop \sum }}\,{{({{x}_{ij}}-\overline{{{x}_{ij}}})}^{2}}\;/\;(m-1)}\]]

3)

Calculation of indicator conflictability

The conflict of indicators is based on the correlation between the indicators, the stronger the correlation, the weaker the conflict of indicators, the weight is reduced accordingly, such as a strong positive correlation between the two indicators, indicating that the conflict between the two indicators is low, indicating that these two indicators in the evaluation process to reflect the amount of information has a greater degree of similarity, so with which corresponds to the smaller the weight. The quantification of conflict is shown below.

(12)

r_{k j} = \frac{\sum_{i = 1}^{m} (x_{i k} - \bar{x_{k}}) (x_{i j} - \bar{x_{j}})}{\sqrt{\sum_{i = 1}^{m} {(x_{i k} - \bar{x_{k}})}^{2}} \sqrt{\sum_{i = 1}^{m} {(x_{i j} - \bar{x_{j}})}^{2}}} k \neq j

\[{{r}_{kj}}=\frac{\underset{i=1}{\overset{m}{\mathop \sum }}\,({{x}_{ik}}-\overline{{{x}_{k}}})({{x}_{ij}}-\overline{{{x}_{j}}})}{\sqrt{\underset{i=1}{\overset{m}{\mathop \sum }}\,{{({{x}_{ik}}-\overline{{{x}_{k}}})}^{2}}}\sqrt{\underset{i=1}{\overset{m}{\mathop \sum }}\,{{({{x}_{ij}}-\overline{{{x}_{j}}})}^{2}}}}\;k\ne j\]]

In Eq:

x_ik,x_ij - indicates the value of the i nd measurement point with respect to the k rd and j th indicators, respectively.

r_kj - the correlation coefficient of the k th and j th indicators.

4)

Calculation of the information load of the indicators

The information carrying capacity of the j th indicator is set to C_j, and the calculation formula is shown below.

(13)

C_{j} = θ_{j} \sum_{k = 1}^{n} (1 - r_{k j})

\[{{C}_{j}}={{\theta }_{j}}\underset{k=1}{\overset{n}{\mathop \sum }}\,(1-{{r}_{kj}})\]]

5)

Calculate the objective weights of the indicators

The final objective weight y_j is obtained by using the indicator information quantity calculation as follows.

(14)

v_{j} = \frac{C_{j}}{\sum_{j = 1}^{n} C_{j}}

\[{{v}_{j}}=\frac{{{C}_{j}}}{\underset{j=1}{\overset{n}{\mathop \sum }}\,{{C}_{j}}}\]]

3.1.3

Analysis and Determination of Composite Weights

This paper adopts the multiplicative synthetic normalization method to combine the subjective and objective weights in the form of equal preference, and obtains the comprehensive weight w_j of each index in the evaluation index system through normalization, as shown in the formula below.

(15)

w_{j} = \frac{u_{j} v_{j}}{\sum_{j = 1}^{n} u_{j} v_{j}}

\[{{w}_{j}}=\frac{{{u}_{j}}{{v}_{j}}}{\underset{j=1}{\overset{n}{\mathop \sum }}\,{{u}_{j}}{{v}_{j}}}\]]

3.2

Cloud Model Evaluation Methods

The cloud model evaluation method can effectively convert qualitative concepts and quantitative information, taking into account the arbitrariness and ambiguity of vocational higher education’s evaluation process. Based on the evaluation index system for vocational college education, the paper quantifies each index, determines subjective and objective weights, and performs the combination of weights. The cloud model theory is introduced to establish the evaluation model of vocational college education based on cloud model; finally, with the help of MATLAB software, through the inverse cloud generator, the comprehensive cloud of indicators is inputted, the comprehensive cloud diagram is outputted, and compared with the standard cloud diagram, the preliminary construction safety risk level is determined, and in order to make the evaluation results more accurate, the similarity between the indicator cloud and the standard risk level is calculated, and the maximum value of the similarity degree is selected as the comprehensive risk evaluation grade, and finally arrive at the vocational system college education grade.

The basic algorithm of cloud is composed of forward cloud generator and reverse cloud generator. Forward cloud generator is the process of changing three cloud digital features into cloud drops, i.e., completing the conversion of qualitative concepts to qualitative data. The inverse cloud generator is the inverse operation of the forward cloud generator, i.e., it is the process of transforming the cloud droplets into cloud digital features, which realizes the conversion from quantitative data to qualitative concepts.

1)

Forward cloud generator

The main calculation steps of forward cloud generator are as follows:

(1) Input the cloud digital features (E_x,E_n,H_e) and cloud drop numbers into MATLAB software and program them.

(2) Generate normal random numbers E_n ~ N(E_n,H_e) and X ~ N(E_x,E_n′).

(3) Calculate: (16) $y_{i} = \exp [- \frac{{(x_{i} - E_{x})}^{2}}{2 {(E_{x}^{'})}^{2}}]$ \[{{y}_{i}}=\exp [-\frac{{{({{x}_{i}}-{{E}_{x}})}^{2}}}{2{{(E_{x}^{'})}^{2}}}]\]]

(4) Let (x_i,y_i) be a cloud droplet, X denote the quantitative value from which the qualitative concept is transformed by means of a forward generator, and Y denote the measure.

(5) Repeat the above steps, to be repeated until the number of cloud droplets that meet the conditions of the study can be generated.

2)

Inverse cloud generator

The reverse cloud generator refers to the conversion of the sample from a quantitative process to a qualitative process.

Input: cloud droplet X and its quantitative value u, i.e. (x_i,y_i).

Output: features reflecting the qualitative concept of cloud droplets (E_x,E_n,H_e).

Calculate the sample mean: (17) $\bar{x} = \frac{1}{n} \sum_{i = 1}^{n} x_{i}$ \[\bar{x}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathop \sum }}\,{{x}_{i}}\]]

The center distance of the sample: (18) $A = \frac{1}{n} \sum_{i = 1}^{n} | x_{i} - \bar{x} |$ \[A=\frac{1}{n}\underset{i=1}{\overset{n}{\mathop \sum }}\,|\;{{x}_{i}}-\bar{x}|\]]

Variance: (19) $S^{2} = \frac{1}{n - 1} \sum_{i = 1}^{n} {(x_{i} - \bar{x})}^{2}$ \[{{S}^{2}}=\frac{1}{n-1}\underset{i=1}{\overset{n}{\mathop \sum }}\,{{({{x}_{i}}-\bar{x})}^{2}}\]]

Calculate expectations: (20) $E_{x} = \bar{x}$ \[{{E}_{x}}=\bar{x}\]]

Calculate the entropy: (21) $E_{n} = \sqrt{\frac{π}{2}} \times \frac{1}{n} \sum_{i = 1}^{n} | x_{i} - \bar{x} |$ \[{{E}_{n}}=\sqrt{\frac{\pi }{2}}\times \frac{1}{n}\underset{i=1}{\overset{n}{\mathop \sum }}\,|{{x}_{i}}-\bar{x}|\]]

Calculating superentropy: (22) $H_{e} = \sqrt{(S^{2} - E_{n}^{2})}$ \[{{H}_{e}}=\sqrt{({{S}^{2}}-E_{n}^{2})}\]]

3.3

Steps for evaluating higher education based on cloud modeling

3.3.1

Determination of portfolio weights

1)

Determine subjective weights

After constructing the evaluation ANP structure, determine the degree of dominance and use the ANP method to calculate the limit supermatrix to obtain the subjective weights.

2)

Determine objective weights

Using the improved CRITIC method, the weight indicators are assigned, and the weight vector is obtained as ω₂ = (ω₂₁,ω₂₂,ω₂₃,...,ω_2n)^T.

3)

Determine the comprehensive weights

On the basis of subjective and objective weights use the multiplication synthesis normalization method for weight combination.

3.3.2

Determination of evaluation sets and criteria clouds

Evaluation set V refers to the composition of the results of educational evaluations conducted by relevant experts. The set of rating results is represented by V, i.e. V = {V₁,V₂,V₃,…V_m}. Where n represents the number of rating levels, and n should be taken as an appropriate value, otherwise it will lead to strong or weak ambiguity of the evaluation results due to subjective factors. Therefore, the thesis combines the characteristics of higher education and divides the education scale into four levels, which are: “very poor”, “poor”, “better”, “good “.

Based on the evaluation set, the domain U is sliced into M subintervals, and the standard cloud numerical characteristics for each risk interval [x_min,x_max] are derived according to Eq. E_x,E_n,H_e from which the standard cloud model is obtained as a comparison reference.

3.3.3

Determination of the evaluation indicator cloud

The raw data are obtained by distributing questionnaires to experts and scoring the security level. Input the raw data into the inverse cloud generator and calculate using the following formula to find the evaluation cloud of vocational system college education evaluation index.

(23)

E_{x j} = \frac{\sum_{p - 1}^{t} x_{p j}}{t}

\[{{E}_{xj}}=\frac{\underset{p-1}{\overset{t}{\mathop \sum }}\,{{x}_{pj}}}{t}\]]

(24)

E_{n j} = \sqrt{\frac{π}{2}} \sum_{p - 1}^{t} \frac{| x_{p j} - E_{x j} |}{t}

\[{{E}_{nj}}=\sqrt{\frac{\pi }{2}}\underset{p-1}{\overset{t}{\mathop \sum }}\,\frac{|{{x}_{pj}}-{{E}_{xj}}|}{t}\]]

(25)

H_{e j} = \sqrt{| S_{j}^{2} - E_{n j}^{2} |}

\[{{H}_{ej}}=\sqrt{|S_{j}^{2}-E_{nj}^{2}|}\]]

In order to ensure that the obtained data are real and objective, it is necessary to adjust the expert scoring data several times. First, the standard cloud parameters obtained from the first scoring results of the experts are input into the forward cloud generator to obtain the evaluation cloud diagram. Observe the dispersion of the cloud map to determine whether the expert scoring data is feasible; if the dispersion is too strong, it indicates that the scoring difference is large and needs to be adjusted. Coordinate with the experts to modify the scoring results until they are more centralized in the cloud diagram and meet the requirements of the cloud model input data.

3.3.4

Identification of a consolidated indicator cloud

With the above obtained evaluation cloud and comprehensive weights, the comprehensive evaluation cloud can be obtained by utilizing the following formula calculation.

(26)

E_{x}^{*} = \sum_{j - 1}^{n} ω_{j}^{*} E_{x j}

\[E_{x}^{*}=\underset{j-1}{\overset{n}{\mathop \sum }}\,\omega _{j}^{*}{{E}_{xj}}\]]

(27)

E_{n}^{*} = \sqrt{\sum_{j - 1}^{n} ω_{j}^{*} E_{n j}^{2}}

\[E_{n}^{*}=\sqrt{\underset{j-1}{\overset{n}{\mathop \sum }}\,\omega _{j}^{*}E_{nj}^{2}}\]]

(28)

H_{e}^{*} = \sum_{j - 1}^{n} ω_{j}^{*} H_{e j}

\[H_{e}^{*}=\underset{j-1}{\overset{n}{\mathop \sum }}\,\omega _{j}^{*}{{H}_{ej}}\]]

The comprehensive cloud parameters are obtained, and the cloud parameters are input into the MATLAB software, which can get the cloud map of the whole target, which can comprehensively reflect the results of the evaluation indexes on the whole.

3.3.5

Determination of evaluation results

For the determination of the evaluation results there are two methods of similarity comparison and cloud comparison, in order to make the evaluation results more real and reasonable thesis synthesize the above two methods to determine the evaluation level of vocational system of higher education.

1)

Similarity comparison

Input two evaluation clouds in the cloud forward generator, respectively, for the comparison of the integrated cloud and the standard cloud, calculate the similarity between the two, with the help of MATLAB software to complete, the specific process is:

(1) Generate random numbers $E_{x}^{'} ~ N (E_{n}, H_{e}^{2})$ $E_{x}^{\prime }\tilde{\ }N({{E}_{n}},H_{e}^{2})$] and $χ_{i} ~ N (E_{x}, E_{n}^{' 2})$ ${{\chi }_{i}}\tilde{\ }N({{E}_{x}},E_{n}^{\prime 2})$] using software.

(2) Calculate the degree of certainty μ′ of x_i: (29) $μ^{'} = e^{\frac{{(x_{i} - E_{x}^{2})}^{2}}{2 {(E_{n})}^{2}}}$ \[{{\mu }^{'}}={{e}^{\frac{{{({{x}_{i}}-E_{x}^{2})}^{2}}}{2{{({{E}_{n}})}^{2}}}}}\]]

(3) Find the overall similarity μ; (30) $μ = \frac{1}{n} \sum_{i = 1}^{n} σ$ \[\mu =\frac{1}{n}\underset{i=1}{\overset{n}{\mathop \sum }}\,\sigma \]]

The similarity value μ is used to compare the two clouds, the larger the value the closer it is to the standard cloud.

2)

Comparison of cloud diagrams

With the help of MATLAB software, the comprehensive cloud of each index is compared with the standard cloud in the same coordinate system to get the comparison cloud map of evaluation indexes. By observing the location of the comprehensive cloud in the comparison cloud diagram, the nearest standard cloud level is the specific evaluation level of college education, i.e., the evaluation result.

Therefore, the thesis synthesizes the advantages of the above two methods, combines the two, and determines the evaluation grade of vocational college education more accurately through the method of calculating the similarity of cloud diagram comparison pairs, which makes the evaluation results more accurate.

4

Cloud modeling-based assessment of educational levels in vocationally oriented colleges and universities

The weights obtained by the three methods are summarized in Table 2. The multiplicative synthetic normalized combined assignment method is to optimize the weights obtained by ANP method and CRITIC method for countermeasures, in the multiplicative synthetic normalized combined assignment model, the subjective weights react to the experience and preference of decision-makers and experts, and the objective weights reflect the objective performance of the evaluation object of the objective presentation of the current status of in-depth learning, and the two weights will be optimized to the smallest deviation in the weights. After the multiplicative synthetic normalization assignment, the importance of the learning process of the first-level indicator is still in the first place, and in the second-level indicator, the learning link and learning participation are 0.2148 and 0.1317 respectively, which indicates that the deep learning pays more attention to the learning link and learning participation, which is in line with the goal of the deep learning with a certain degree of scientificity, and the evaluation index has reliability.

Table 2.

Index of various methods of various methods

Primary indicator	ANP	CRITIC	Composite weight	Secondary indicator	ANP	CRITIC	Composite weight
Learning design	0.1055	0.1043	0.1180	Learning goal	0.0142	0.0255	0.0173
				Learning strategy	0.0344	0.0762	0.0478
				Learning method	0.0451	0.0730	0.0529
Learning environment	0.2542	0.3366	0.2235	Learning support	0.0314	0.0306	0.0313
Learning environment	0.2542	0.3366	0.2235	Learning resources	0.2208	0.1443	0.1922
Learning process	0.4784	0.3833	0.4790	Learning link	0.0357	0.0545	0.0428
				Learning content	0.2380	0.1754	0.2148
				Learning participation	0.0814	0.1015	0.0897
				Learning assessment	0.1278	0.1380	0.1317
Learning effect	0.1619	0.1758	0.1795	Learning will	0.0455	0.0731	0.0555
				Learning ability	0.1011	0.1083	0.1036
				Learning quality	0.0150	0.0295	0.0204

School A was selected and its educational status was evaluated using the cloud model, and the educational assessment grade was divided into four grades, i.e., S=(S1,S2,S3,S4)={extremely poor, poor, better, good}, corresponding to the assessment value range of [0,100], in which [0,25] indicates that vocational colleges and universities are extremely poor in terms of educational effect, (25,50] indicates that vocational colleges and universities are poor in terms of educational effect, (50,75 ] indicates that the vocational system of higher education has a good effect, and (75,100] indicates that the vocational system of higher education has a good effect. Calculated evaluation standard cloud model, the standard cloud model is a measure in the process of vocational college education assessment, the final results of vocational college education assessment can be compared and analyzed with it, in order to determine the vocational college education assessment level.

According to the characteristic parameters of the standard cloud model obtained from the calculation, the parameters of the standard cloud model and the cloud droplets are used as inputs, and according to the principle of the cloud generator, the calculation is carried out through the MATLAB software, which generates the standard cloud diagram of the educational assessment level of vocational colleges and universities in turn, and the cloud diagram is shown in Fig. 2, which shows that the standard clouds corresponding to Levels I to IV are shown in the figure in the order from the left to the right.

Using MATLAB, the standard cloud and the cloud diagrams of the indicator layer evaluation indexes due to V1-V3, V4-V5, V6-V9, and V10-V12 are plotted into the same coordinate system respectively, and the comparison diagrams of the generated indicator layer factor evaluation cloud and the standard cloud are shown in Figures 3-Figure 6.

The V3 assessment result map is between “Extreme” and “Poor”, which is closer to “Extreme”, so the evaluation result of the V3 level of learning method is “Extremely Poor”. The evaluation results of learning strategy V2 are between “good” and “poor”, which is closer to “better”, so the evaluation result of learning strategy V2 is “good”. In the same way, the level of learning objective V1 is assessed as “poor”.

The evaluation results of Learning Support V4 are between “good” and “poor”, which is closer to “poor”, so the evaluation result of Learning Support V4 is “poor”. In the same way, the assessment result of learning resource V5 is closer to “poor”, and the assessment result of V5 is “poor”.

The evaluation results of learning session V6 are between “good” and “good”, which is closer to “good”, so the evaluation result of learning session V6 is “good”. In the same way, the evaluation results of learning content V7 are closer to “excellent”, and the evaluation results of learning participation V8 and learning assessment V9 are both “poor”.

The cloud map of the assessment results of V10 is between “very poor” and “poor”, which is closer to “poor”, so the grade assessment result caused by V10 is “poor”. In the same way, the assessment results of learning ability V11 and learning quality V12 were closer to “good”, and the assessment results of V11 and V12 were both “good”.

MATLAB was used to plot the four index cloud maps of the standard cloud and the target layer into the same coordinate system, and the comparison charts of the target layer cloud and the standard cloud were shown in Figure 7-Figure 10. The educational evaluation level of Learning Design U1 is between “very poor” and “poor”, which is closer to the “poor” level, so the evaluation result of learning design U1 is “poor”. In the same way, the U2 assessment result of the learning environment was “poor”, the assessment result of the learning process U3 was “poor”, and the assessment result of the learning effect U4 was “good”. As shown in Figure 11, it can be seen that the comprehensive evaluation cloud map is closer to the “poor” level, and the entropy value En and superentropy He of the calculated comprehensive cloud are relatively small, indicating that the educational evaluation model in this paper has good stability and high credibility.

In order to further analyze the results of the school’s educational evaluation, the similarity between the target layer index and the standard clouds of each grade was calculated, and the calculation results were shown in Table 3, and the maximum similarity between U1 and the “poor” grade cloud was 0.478, so the calculation result of the cloud similarity of U1 was “poor”, which was consistent with the results obtained by the cloud map method. In the same way, the similarity results of U2, U3 and U4 were “poor”, “poor” and “good”, respectively, which were consistent with the results of the cloud map, and the maximum similarity between the comprehensive cloud and the “poor” grade cloud was 0.547, which was consistent with the cloud map. Therefore, according to the above-mentioned cloud evaluation model, cloud map and similarity calculation, the evaluation result of the school’s education grade is “poor”, and further teaching rectification is needed.

Table 3.

Cloud similarity calculation results

Target level indicator	Grade				Evaluation result
Target level indicator	Bad	Worse	Better	Good	Evaluation result
U1	0.035	0.478	0.000	0.000	Worse
U2	0.000	0.215	0.046	0.000	Worse
U3	0.000	0.409	0.017	0.000	Worse
U4	0.000	0.048	0.215	0.000	Better
Integrated cloud	0.002	0.547	0.008	0.000	Worse

5

Conclusion

Due to the complexity of the indicators and the ambiguity of the educational assessment process, the single assignment method causes a single weight value and error with the actual situation, and the combination assignment increases the accuracy of the weight value and minimizes the weight deviation. After the combination of assignment, the importance of the learning process of the first-level indicator is still in the first place, and the weights of the two indicators of the second-level indicator, learning link and learning participation, are 0.2148 and 0.1317, respectively, which indicates that the deep learning attaches importance to the process of the learning link and learning participation, and the result is in line with the goal of the deep learning, which indicates that the vocational system of higher education evaluation indicators constructed in this paper are reliable.

The educational grade assessment model of higher education based on the combined empowerment-cloud model was used to evaluate and validate the current educational status of School A. During the evaluation process, the educational evaluation grades of learning design, learning environment, and learning process were found to be “poor”, and the educational evaluation results of learning effect U4 were found to be “Better”. The final evaluation of the school is “poor”, and school A should strengthen the construction of professional teaching team, further create a good teaching environment, and better utilize the teaching evaluation index system to promote the quality of teaching and the level of teaching management.

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

Integration of deep learning and vocational college education evaluation index system model construction

Xiaona Xie

Wenliang Zhang

Published Online: Mar 21, 2025

Received: Oct 21, 2024

Accepted: Feb 14, 2025

DOI: https://doi.org/10.2478/amns-2025-0608

KeywordsDeep learning, ANP-CRITIC, Cloud model, Evaluation index system

© 2025 Xiaona Xie et al., published by Sciendo

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

Keywords
Deep learning, ANP-CRITIC, Cloud model, Evaluation index system