Construction of Evaluation System of Civic Education Work of College Students Based on Decision Tree Algorithm in New Media Era

1)

Principle of Policy Orientation

The purpose of college students’ Civic and Political Education is to improve students’ ideological awareness, moral education level, and political literacy. The Party’s policy reflects the requirements of China’s education sector and its expectations for educational targets, and is the core guidance for the evaluation system of ideological and political education. Since the 19th National Congress, China has frequently issued policies and documents on the work of ideological and political education, and has constantly put forward new requirements and suggestions for the work of ideological and political education in China; these policies and documents not only provide a direction for the development of ideological and political education in China, but also provide a strong reference and theoretical basis for the construction of the evaluation system of ideological and political education in China.

According to the preliminary survey and research and consulting relevant experts, the index system of college students’ civic education is shown in Table 1.

1)

Decision Tree

Decision tree is a tree-shaped data structure. Through the root node extends the various branches, connecting different child nodes, the use of child nodes and the relationship between the parent node and the structure, you can find any one of the nodes or leaf nodes on the entire tree structure.

Decision trees, also known as classification trees, decision tree structure is characterized by a relatively simple, the number of vertices, that is, the root node subordinate to a number of branches, constituting an attribute structure composed of different categories of branches. Each branch represents a set of test outputs. On each branch, different internal nodes as well as leaf nodes are distributed. The internal nodes represent a test, while the leaf nodes represent the branching categories of decisions. After several iterations, the predicted value of the leaf node class that holds the classification is found [19].

1)

For a classification system, suppose X is the set of data samples, x is the number of X’s, C is the attribute category variable taking the value C={C1,C2,C3...Cn}$$C = \left\{ {{C_1},{C_2},{C_3}...{C_n}} \right\}$$, n is the total number of categories, and x_i is the number of samples contained in C_i categories, then the probability that any sample belongs to C_i is: (1)pi=xix$${p_i} = \frac{{{x_i}}}{x}$$

The information entropy of the sample classification is: (2)I(x1,x2...xn)=−∑i=1npilog2pi$$I({x_1},{x_2}...{x_n}) = - \sum\limits_{i = 1}^n {{p_i}} {\log _2}{p_i}$$ 2)

Calculation of information entropy (per attribute)

Let attribute Y has m different values, m different values divide X into {X1,X2,...,Xn}$$\left\{ {{X_1},{X_2},...,{X_n}} \right\}$$, x_ij is the number of samples belonging to C_i in subset X_j, then the information entropy of Y is: (3)E(Y)=∑j=1mx1j+x2j+...+xnjx×I(x1j,x2j,...,xnj)$$E(Y) = \sum\limits_{j = 1}^m {\frac{{{x_{1j}} + {x_{2j}} + ... + {x_{nj}}}}{x}} \times I({x_{1j}},\:{x_{2j}},...,\:{x_{nj}})$$

This subsection focuses on the improvement method to the ID3 algorithm, and the principle of the improvement is described in detail. 1)

Calculation of information entropy

The classical ID3 algorithm is introduced in the previous section, and the arithmetic example is given for deduction. However, in the process of calculation, it is necessary to carry out log operation many times, which is not too demanding in terms of performance for small datasets, but for large datasets, the performance of the operation is particularly important. Therefore, this paper transforms the original log operation, optimizes the algorithm by means of power expansion, and transforms the complex operation into addition, subtraction, multiplication and division with faster operation speed. The specific improvement steps are as follows:

According to the demand can be p, set to: (5)pi=1−si1+si$${p_i} = \frac{{1 - {s_i}}}{{1 + {s_i}}}$$

From equation (5): (6)si=1−pi1+pi$${s_i} = \frac{{1 - {p_i}}}{{1 + {p_i}}}$$

Substituting Eq. (6) into Eq. (3) yields: (7)−∑i=1n1−si1+silog21−si1+si=−∑i=1n1−si1+siln1−si1+siln2$$ - \sum\limits_{i = 1}^n {\frac{{1 - {s_i}}}{{1 + {s_i}}}} {\log _2}\frac{{1 - {s_i}}}{{1 + {s_i}}} = - \sum\limits_{i = 1}^n {\frac{{1 - {s_i}}}{{1 + {s_i}}}} \frac{{\ln \frac{{1 - {s_i}}}{{1 + {s_i}}}}}{{\ln 2}}$$

From the power series expansion, equation (7) can be rewritten as: (8)2ln2∑i=1n1−si1+si(si+13si3+15si5+17si7+...)$$\frac{2}{{\ln 2}}\sum\limits_{i = 1}^n {\frac{{1 - {s_i}}}{{1 + {s_i}}}} \left( {{s_i} + \frac{1}{3}s_i^3 + \frac{1}{5}s_i^5 + \frac{1}{7}s_i^7 + ...} \right)$$

The accuracy of Eq. (8) keeps increasing as the power becomes higher. In order to simplify the calculation process, and because the calculation process is a comparison of the magnitude of information entropy, the first two terms of Eq. (9) are taken here and organized to obtain Eq. (9). (9)2ln2∑i=1n1−si1+si(si+13si3)=23ln2∑i=1nsi(1−si)(3+si2)1+si$$\frac{2}{{\ln 2}}\sum\limits_{i = 1}^n {\frac{{1 - {s_i}}}{{1 + {s_i}}}} \left( {{s_i} + \frac{1}{3}s_i^3} \right) = \frac{2}{{3\ln 2}}\sum\limits_{i = 1}^n {\frac{{{s_i}(1 - {s_i})(3 + s_i^2)}}{{1 + {s_i}}}}$$

Substituting equation (8) into equation (9) yields the following equation (10): (10)83ln2∑i=1npi(1−pi3)(1+pi)3$$\frac{8}{{3\ln 2}}\sum\limits_{i = 1}^n {\frac{{{p_i}(1 - p_i^3)}}{{{{(1 + {p_i})}^3}}}}$$

Substituting equation (5) into equation (10) yields the following equation (11): (11)83xln2∑i=1nxi(x3−xi3)(x+xi)3$$\frac{8}{{3x\ln 2}}\sum\limits_{i = 1}^n {\frac{{{x_i}({x^3} - x_i^3)}}{{{{(x + {x_i})}^3}}}}$$

83xln2$$\frac{8}{{3x\ln 2}}$$ in the above equation is a constant that can be omitted from the calculation of the comparison size, so the final formula for calculating the information entropy is: (12)1x∑i=1nxi(x3−xi3)(x+xi)3$$\frac{1}{x}\sum\limits_{i = 1}^n {\frac{{{x_i}({x^3} - x_i^3)}}{{{{(x + {x_i})}^3}}}}$$ 2)

Create decision tree

When creating a decision tree, it mainly relies on selecting the attribute with the largest information gain as the splitting node, and according to Eq. (12), it can be seen that I(x₁, x₂, ..., x_n) in Eq. (12) is fixed, so the calculation of the attribute information gain mainly depends on the size of the attribute information entropy. The construction process of the decision tree is top-down, i.e., in the early stage of data classification, all the data are concentrated in the root node to start the division, and at the same time, the attribute entropy of the node is calculated, based on which, according to the entropy of the information gain, the data are initially classified. In the process of data classification, it is often carried out according to an agreed statistical way or heuristic rules. If the attributes can not be in the continuation of the split, it has to stop, at this time, it should be merged into the same class.

The method and steps for determining the subjective weight coefficients by the ordinal relationship method are as follows: 1)

Determine the ordinal relationship between evaluation indicators

If the importance of an evaluation indicator in relation to an evaluation objective is greater than or equal to x_j, it is denoted as x_i > x_j. An ordinal relationship is established for a group of evaluation indicators when evaluation indicator x₁, x₂, ⋯, x_m has relationship x1*≥x2*≥⋯xm*$$x_1^* \geq x_2^* \geq \cdots x_m^*$$ in relation to an evaluation objective. Evaluation indicator xj*$$x_j^*$$ refers to the jth evaluation indicator after being ranked according to the sequential relationship.

Where: wz−1*$$w_{z - 1}^*$$ and wz*$$w_z^*$$ are the weights of the z − 1rd and zth indicators respectively, before calculation, weight wz*$$w_z^*$$ is unknown, r_z can be obtained by expert scoring. 3)

Calculate the weight coefficient of each evaluation indicator

Obviously, wz−2*>wz*$$w_{z - 2}^* > w_z^*$$, and because w_z − 1^* > 0, therefore wz−2/wz−1*>wz*/wz−1*$${w_z} - 2/{w_z} - {1^*} > w_z^*/w_{z - 1}^*$$, therefore r_z − 1/r_z. Because: (14)∏i=zmri=wz−1*wz*wz*wz+1*wz+1*wz+2*⋯wm−2*wm−1*wm−1*wm*=wz−1*wz*z≥2$$\prod\limits_{i = z}^m {{r_i}} = \frac{{w_{z - 1}^*}}{{w_z^*}}\frac{{w_z^*}}{{w_{z + 1}^*}}\frac{{w_{z + 1}^*}}{{w_{z + 2}^*}} \cdots \frac{{w_{m - 2}^*}}{{w_{m - 1}^*}}\frac{{w_{m - 1}^*}}{{w_m^*}} = \frac{{w_{z - 1}^*}}{{w_z^*}}\quad z \geq 2$$

Summing over z from 2 to m has: (15)∑z=2m∏i=zmri=∑z=2mwz−1*wm*=1wm*(w1*+w2*+⋯+wm−1*)=1wm*(∑z=1mwz*−wm*)$$\sum\limits_{z = 2}^m {\prod\limits_{i = z}^m {{r_i}} } = \sum\limits_{z = 2}^m {\frac{{w_{z - 1}^*}}{{w_m^*}}} = \frac{1}{{w_m^*}}(w_1^* + w_2^* + \cdots + w_{m - 1}^*) = \frac{1}{{w_m^*}}(\sum\limits_{z = 1}^m {w_z^*} - w_m^*)$$

And because ∑z=1mwz*=1$$\sum\limits_z = {1^m}w_z^* = 1$$, therefore: (16)∑z=2m∏i=zmri=1wm*(1−wm*)=1/wm*−1$$\sum\limits_{z = 2}^m {\prod\limits_{i = z}^m {{r_i}} } = \frac{1}{{w_m^*}}(1 - w_m^*) = 1/w_m^* - 1$$

Then w_m can be solved for: (17)wm=g(1+∑z=2m∏z=imrig)−1$${w_m} = g{(1 + \sum\limits_{z = 2}^m {\prod\limits_{z = i}^m {{r_i}} } g)^{ - 1}}$$

According to the formula: (18)wz−1=rzwz(z=m,m−1,⋯,2)$${w_{z - 1}} = {r_z}\:{w_z}\:(z = m,m - 1, \cdots ,2)$$

The weighting coefficients for each indicator can be derived.

1)

Data standardization

Before using the CRITIC weight method [20], the indexes are firstly standardized, and the commonly used data standardization methods include the method of extremely large value, the method of extremely small value, and the method of eigenvalue assignment, etc. The method of extremely large values is suitable for standardizing the main factor indexes that are proportional to the top plate compressive strength. Extremely large value method is suitable for the standardization of the main factor index which is proportional to the top plate compressive strength, the larger the value of the main factor index, the larger the quantitative value after standardization, and the larger the corresponding top plate compressive strength. On the contrary, the method of very small value applies to the standardization of the main factor index which is inversely proportional to the compressive strength of the roof plate, and the larger the value of the main factor index, the smaller the quantitative value after standardization, and the smaller the corresponding compressive strength of the roof plate.

Extreme value method: (19)standardresulti=xij−min(xj)max(xj)−min(xj)$$s\tan dardresul{t_i} = \frac{{{x_{ij}} - \min ({x_j})}}{{\max ({x_j}) - \min ({x_j})}}$$

Minimal value method: (20)standardresulti=max(xj)−xijmax(xj)−min(xj)$$s\tan dardresul{t_i} = \frac{{\max ({x_j}) - {x_{ij}}}}{{\max ({x_j}) - \min ({x_j})}}$$

Where: x_ij represents the jrd indicator of the ind object to be evaluated; max(x_j) is the maximum value of the jth indicator data; min(x_j) is the minimum value of the jth indicator data. 2)

Calculation of the correlation coefficient and the value of the quantitative indicator of conflictability

The correlation coefficient between the ith indicator and the jth indicator can be calculated by the following formula: (21)rij=∑(xi−xi¯)(xj−xj¯)Σ(xi−xi¯)2(xj−xj¯)2,i≠j$${r_{ij}} = \frac{{\sum {({x_i} - \overline {{x_i}} )} ({x_j} - \overline {{x_j}} )}}{{\sqrt {\Sigma {{({x_i} - \overline {{x_i}} )}^2}{{({x_j} - \overline {{x_j}} )}^2}} }},i \ne j$$

Where: x_i and x_j represent the values of the two variables; x¯i$${\bar x_i}$$ and x¯j$${\bar x_j}$$ represent the mean values of the two variables. When the standard deviation indicator with a scale is used to reflect the discriminatory power of the indicators, due to the different scales and orders of magnitude of the indicators, the standard deviation cannot be directly compared with their discriminatory power, and it should be improved by applying the standard deviation coefficient S_j to measure the discriminatory power of the indicators. (22)Sj=σjx¯jj=1,2......m$${S_j} = \frac{{{\sigma _j}}}{{{{\bar x}_j}}}\quad j = 1,2......m$$

Where: S_j is the standard deviation coefficient of indicator j, σj=∑i=1n(standard_resultij−x¯j)2n−1$${\sigma _j} = \sqrt {\frac{{\sum\limits_{i = 1}^n {{{(s\tan dard\_resul{t_{ij}} - {{\bar x}_j})}^2}} }}{{n - 1}}}$$ is the standard deviation of indicator j, and xj¯$$\overline {{x_j}}$$ is the mean of indicator j.

The value of the quantitative indicator of conflictability of the jth indicator with other indicators is: (23)Aj=∑i=1n(1−rij),i≠j$${A_j} = \sum\limits_{i = 1}^n {(1 - {r_{ij}})} ,\:i \ne j$$

As equation (22) calculates the value of the quantitative indicator of conflict between indicators A_j, the correlation coefficient between indicators i and j may have a negative value, while for positive and negative correlation with the same absolute value, the correlation between the indicators reflected should be the same. Therefore, it is not reasonable to use equation (22) to measure the conflict between indicators. Improvement on this side, using equation (23) to calculate, can effectively avoid this problem. (24)Aj'=∑i=1n(1−|rij|),i≠j$$A_j' = \sum\limits_{i = 1}^n {(1 - |{r_{ij}}|)} , i \ne j$$ 3)

Calculating the information content of indicators

The objective weight of each indicator is measured by the combination of contrast intensity and conflict. Let C_j denote the amount of information contained in the jnd categorized indicator factor, and the formula for calculating C_j is as follows: (25)Cj=Sj×Aj=σjx¯j∑i=1n(1−|rij|),i≠j,j=1,2,⋯n$${C_j} = {S_j} \times {A_j} = \frac{{{\sigma _j}}}{{{{\bar x}_j}}}\sum\limits_{i = 1}^n {(1 - |{r_{ij}}|)} ,\:i \ne j,\:j = 1,2, \cdots n$$

Where: The larger C_j is, the greater the amount of information contained in the jnd evaluation indicator, and the greater the relative importance of that indicator, i.e., the greater the weight.

Game theory is a kind of operation research method to study things with competitiveness, drawing on the idea of game theory, the main and objective weights are regarded as the decision-making subjects in the non-cooperative game, and the two sides look for the balance of interests in the continuous conflict to realize the optimal combination of weights, so as to make the indicator assignment more scientific and reasonable [21]. The specific process is as follows:

Assuming that there is n assignment method, the weight of the indicator calculated by the first assignment method is noted as w₁ = (w₁₁, w₁₂, w₁₃, ⋯w_1m), the weight of the indicator of the second assignment method is w₂ = (w₂₁, w₂₂, w₂₃, ⋯w_2m), and so on the weight of the indicator of the nth assignment method is w_n = (w_n1, w_n2, w_n3, ⋯w_nm). The combination weight is then constructed from the prior combination of w₁, w₂, ⋯w_n, Eq: (27)w=α1w1T+α2w2T+⋯αnwnT$$w = {\alpha _1}w_1^T + {\alpha _2}w_2^T + \cdots {\alpha _n}w_n^T$$

Where: α₁, α₂, ⋯, α_n is the weight combination coefficient of each assignment method respectively.

The Nash equilibrium point is solved according to the principle of game theory, i.e., finding equilibrium between different weights and minimizing the deviation between the combination of weights and each weight, with the objective function and constraints: (28)f=mina1,a2,⋯an∑i=1n|(∑k=1nakwiwkT)−wiwiT|$$f = {\min {_{{a_1},{a_2}, \cdots {a_n}}}}\sum\limits_{i = 1}^n {\left| {(\sum\limits_{k = 1}^n {{a_k}} {w_i}w_k^T) - {w_i}w_i^T} \right|}$$ (29)s.t.∑k=1nak=1$${\text{s}}.{\text{t}}.\sum\limits_{k = 1}^n {{a_k}} = 1$$

By solving the model, the optimal combination of weights for the data under the combined consideration of each assignment method can be obtained. A system of linear equations is established as shown in (30). (30)( w1w1T ⋯ w1wnT ⋮ ⋱ ⋮ wnw1T ⋯ wnwnT)( α1 ⋮ αn)=( w1w1T ⋮ wnwnT)$$\left( {\begin{array}{c} {{w_1}w_1^T}& \cdots &{{w_1}w_n^T} \\ \vdots & \ddots & \vdots \\ {{w_n}w_1^T}& \cdots &{{w_n}w_n^T} \end{array}} \right)\left( {\begin{array}{c} {{\alpha _1}} \\ \vdots \\ {{\alpha _n}} \end{array}} \right) = \left( {\begin{array}{c} {{w_1}w_1^T} \\ \vdots \\ {{w_n}w_n^T} \end{array}} \right)$$

The combination coefficient α₁, α₂, …, α_n is obtained according to Eq. (30) and normalized, i.e., αi*=ai/∑inai$$\alpha _i^* = {a_i}/\sum\limits_i^n {{a_i}}$$ can be used to find the final combination weights as: (31)w*=α1*w1T+α2*w2T+⋯+αn*wnT$${w^*} = \alpha _1^*w_1^T + \alpha _2^*w_2^T + \cdots + \alpha _n^*w_n^T$$

From Eq. (30), it can be seen that the result of the combination weight depends on the linear combination coefficient α₁, α₂, ⋯, α_n, while the combination coefficient α₁, α₂, …, α_n is calculated by Eq. (31), but not every combination coefficient that is sought is positive, if the combination coefficient that is sought: (32)α1,α2,…,αn$${\alpha _1},{\alpha _2}, \ldots ,{\alpha _n}$$

The assumptions of Eq. (32) are not satisfied if there are negative values in them. For this reason this paper uses an improved game theoretic portfolio assignment method.

The objective function f of (32) above is rewritten to ensure that the combination coefficient α₁, α₂, ⋯, α_n is greater than zero. (33)f=mina1,a2,⋯an∑i=1n|(∑k=1nakwiwkr)−wiwir|$$f = {\min {_{{a_1},{a_2}, \cdots {a_n}}}}\sum\limits_{i = 1}^n {\left| {(\sum\limits_{k = 1}^n {{a_k}} {w_i}w_k^r) - {w_i}w_i^r} \right|}$$

The weight combination factor a_k(k = 1, 2⋯n) satisfies the following conditions: (34)∑k=1nak=1$$\sum\limits_{k = 1}^n {{a_k}} = 1$$

Or: (35)∑k=1nak2=1$$\sum\limits_{k = 1}^n {a_k^2} = 1$$

Combining Eq. (34) and constraints (35), the optimization model can be obtained as shown in Eq. (36). (36){ mina1,a2,⋯,anf∑i=1n|(∑k=1nakwiwkT)−wiwiT| s.t.ak>0,k=1,2⋯n∑k=1nak2=1$$\left\{ {\begin{array}{l} {\mathop {\min }\limits_{{a_1},{a_2}, \cdots ,{a_n}} f\sum\limits_{i = 1}^n {\left| {(\sum\limits_{k = 1}^n {{a_k}} {w_i}w_k^T) - {w_i}w_i^T} \right|} } \\ {s.t.\:{a_k} > 0,\:k = 1,2 \cdots n\:\sum\limits_{k = 1}^n {a_k^2} = 1} \end{array}} \right.$$

The following Lagrangian function is established to solve the model: (37)L(ak,γ)=∑i=1n|(∑k=1nakwiwkr)−wiwir|+γ2(∑k=1nak2−1)$$L({a_k},\gamma ) = \sum\limits_{i = 1}^n {\left| {(\sum\limits_{k = 1}^n {{a_k}} {w_i}w_k^r) - {w_i}w_i^r} \right|} + \frac{\gamma }{2}(\sum\limits_{k = 1}^n {a_k^2} - 1)$$

Taking partial derivatives for a_k(k = 1, 2⋯n) and γ, respectively, we have the following partial differential equation based on the conditions for the existence of extreme values: (38)∂L∂ak=∑i=1n(±)−wiwkr+γak=0$$\frac{{\partial L}}{{\partial {a_k}}} = \sum\limits_{i = 1}^n {( \pm )} - {w_i}w_k^r + \gamma \:{a_k} = 0$$ (39)∂L∂γ=∑i=1nak2−1=0$$\frac{{\partial L}}{{\partial \gamma }} = \sum\limits_{i = 1}^n {a_k^2} - 1 = 0$$

From equation (39): (40)ak=∓∑i=1nwiwkTγ$${a_k} = \frac{{ \mp \sum\limits_{i = 1}^n {{w_i}} w_k^T}}{\gamma }$$

Substituting Eq. (38) into Eq. (39) yields: (41)γ=±∑k=1n(∑i=1nwiwkT)2$$\gamma = \pm \sqrt {\sum\limits_{k = 1}^n {{{\left( {\sum\limits_{i = 1}^n {{w_i}} w_k^T} \right)}^2}} }$$

Substituting equation (41) into equation (38) gives: (42)ak=∓∑i=1nwiwkT±∑k=1n(∑i=1nwiwkT)2$${a_k} = \frac{{ \mp \sum\limits_{i = 1}^n {{w_i}} w_k^T}}{{ \pm \sqrt {\sum\limits_{k = 1}^n {{{\left( {\sum\limits_{i = 1}^n {{w_i}} w_k^T} \right)}^2}} } }}$$

Due to the combination factor a_k > 0, equation (42) can be written as: (43)ak=∑i=1nwiwkT∑k=1n(∑i=1nwiwkT)2$${a_k} = \frac{{\sum\limits_{i = 1}^n {{w_i}} w_k^T}}{{\sqrt {\sum\limits_{k = 1}^n {{{\left( {\sum\limits_{i = 1}^n {{w_i}} w_k^T} \right)}^2}} } }}$$

Eq. (43) is the unique solution that satisfies model (42), and finally for a_k(k = 1, 2⋯n), normalization is performed to obtain ak*$$a_k^*$$: (44)ak*=∑i=1nwiwkTγ=∑k=1n∑i=1nwiwkT$$a_k^* = \frac{{\sum\limits_{i = 1}^n {{w_i}} w_k^T}}{{\gamma = \sum\limits_{k = 1}^n {\sum\limits_{i = 1}^n {{w_i}} } w_k^T}}$$

Substituting ak*$$a_k^*$$ into the following equation yields the value of portfolio weights as: (45)w*=∑k=1nak*wkT,(k=1,2⋯n)$${w^*} = \sum\limits_{k = 1}^n {a_k^*} w_k^T\:,\:(k = 1,2 \cdots n)$$

Twenty-seven teachers from each of the five colleges of Computer Science and Engineering, Mechanical Engineering, Electrical and Automation Engineering, Mathematics and Statistics, and Foreign Languages in a university in City A were randomly selected as the subjects of the study. Among them, the age range of the 27 teachers selected from each college should involve young, middle-aged, and old; the title rank should include assistant professor, lecturer, associate professor, and professor; and the degree should involve doctoral, master’s, and bachelor’s degrees. In addition, 44 students were randomly selected as the teaching class for each teacher. Evaluation of decision tree algorithm based Civics teaching using Civics education classroom teaching.

The survey method during research is mostly a questionnaire method, which mainly involves the distribution and retrieval of questionnaires. The research implementation survey was conducted from April to June 2023, and 656 student questionnaires were distributed in this survey. 423 copies were recovered, of which, 389 were valid questionnaires.

In this section of the experiment, DATA is selected as the experimental data to carry out correlation analysis experiments on students’ civic education work. First of all, the data is tested for normality, due to the large amount of data, this section uses the K-S test, and the significance of some indicators p<0.05. It indicates that the evaluation dataset does not satisfy a normal distribution, so the Pearson correlation coefficient was not chosen. Kendall correlation coefficient analysis is suitable for consistency checking of data, such as judges’ scores and data ranking. The Spearman correlation coefficient method, on the other hand, has more relaxed data requirements and is suitable for quantitative data that do not satisfy normal distribution. Therefore, Spearman’s correlation coefficient is used in this section as a method of correlation analysis.

In this section, correlation analysis was carried out using SPSS tool and the Spearman correlation coefficient was selected. The results of the analysis based on the Spearman correlation coefficient are shown in Table 2.

The observation table shows that there is a high degree of correlation between indicator 7 and indicator 10. It is necessary to remove a redundant indicator from these two indicators, this paper utilizes the variable correlation removal method to calculate the correlation mean of each indicator in the above two indicators with all other variables and compare the size, between indicator 7 and indicator 10, the correlation mean of indicator 7 is 0.347, which is larger than the correlation mean of indicator 10, 0.223, therefore, indicator 7 is deleted, and after removing the redundant indicator 7 the experimental data of the data set of students’ civic education work as DATA2.

The subjective weights of Civic Education, Academic Style Construction, Team Building, and Nurturing Effect are calculated, and the subjective weights are finally determined based on the cluster G1 method as shown in Table 4.

On the basis of data normalization, SPSS software was used to calculate the variability σ_j, conflict R_j and informativeness C_j among the indicators, resulting in the objective weight values calculated based on the CRITIC method as shown in Table 5:

In this paper, two kinds of weights have been calculated according to the cluster Gl method and CRITIC method, and according to the game synthesized assignment formula (46), the optimal solution matrix equation can be obtained as: (46)[ Wa1Wa1T Wa1WcancT WcancWa1T WcancWcancT][ λ1 λ2]=[ Wa1Wa1T WcancWcancT]$$\left[ {\begin{array}{c} {{W_{a1}}W_{a1}^T}&{{W_{a1}}W_{canc}^T} \\ {{W_{canc}}W_{a1}^T}&{{W_{canc}}W_{canc}^T} \end{array}} \right]\left[ {\begin{array}{c} {{\lambda _1}} \\ {{\lambda _2}} \end{array}} \right] = \left[ {\begin{array}{c} {{W_{a1}}W_{a1}^T} \\ {{W_{canc}}W_{canc}^T} \end{array}} \right]$$

The final combination weights were found according to formula (46) as shown in Table 6.

According to the results of the combination weights, the highest proportion of the seven evaluation indexes is the one-time employment rate (0.21099), followed by the course learning situation, the team of lecturers, time activities, mental health education, and situation and policy education. It shows that parenting and the construction of academic style have a relatively large impact on the ideological education of college students.