A study on dynamic data modelling in the evaluation of the effectiveness of ideological education in colleges and universities based on big data

The smaller the value of the negative indicator indicates a more desirable situation, so it is normalized to: (2) rij=maxuj−uijmaxuj−minuj

Where, fij=1+rij∑i=1m(1+rij) can thus be based on Eq: (4) ai=1−hin−∑i=1nhi, And ∑i=1nai=1

The entropy weight of each secondary evaluation index is obtained as A=(a₁,a₂,…a₁₆)

The entropy weight of the first level indicator was further found to be W=(W₁,W₂,W₃,W₄,W₅).

According to Equation (5), the combined weights of all primary and secondary indicators can be found as F=(f₁,f₂,f₃,f₄,f₅), B=(B₁,B₂,B₃,B₄,B₅).

The weights of the index system for evaluating the effectiveness of civic education in colleges and universities calculated using the above method are shown in Table 2. Among the level 1 indicators, B, “educational content,” has the highest weight of 0.3220, and the weights of B2, “national sentiment” (0.0837) and B6, “scientific attitude” (0.0862) are higher under the level 1 indicator. Under this level, B2 “national sentiment” (0.0837) and B6 “scientific attitude” (0.0862) have higher weights

Comprehensively comparing the characteristics and typical functions of representative neural networks, this study chooses to use the evaluation model based on the Adaboost-RBF neural network to assess the effect of ideological education in colleges and universities.

The structure of RBNN is similar to a multilayer forward network [23], which is a three-layer forward neural network with a single hidden layer. The change from the input layer space to the hidden layer space is not linear, but the change from the hidden layer space to the output layer space is linear. The transformation function of the hidden layer neurons is RBF, which is a locally distributed, non-negative, nonlinear function with radially symmetric decay of the center.RBF. RBNN can effectively solve the limitations of traditional BP neural network gradient descent learning algorithms, such as slow convergence speed and easy falling into the local minima, which makes RBNN have good generalization ability.RBNN uses RBF as the “base” of the hidden unit, maps the input data from the low-dimensional space to the high-dimensional space, makes the problems that are linearly indivisible in the low-dimensional space to achieve linear separability in the high-dimensional space, and then solves the complex pattern classification problems. Problems that are linearly indivisible in the low-dimensional space can be linearly differentiated in the high-dimensional space, thus solving complex pattern classification problems.

The most commonly used RBF is Gaussian function, and its expression is as follows: (6) G(χp−ci)=exp(−‖ χp−ci ‖22σ2)

where ǁχ_p–c_iǁ is the Euclidean paradigm and σ is the variance of the Gaussian function. At this point, the linearly weighted expression of the RBFNN for the output of the hidden layer neurons is as follows: (7) { yj=∑i=1mwijexp(−‖ χp−ci ‖22σ2)j=1,2,3,⋯,l

Where χ_p=(χ^p₁,χ^p₂,⋯,χ^p_n)^· is the pth input sample, p=1,2,⋯,P, a total of P input samples, c_i is the center of the implicit layer node, w_ij is the connection weight from the implicit layer to the output layer, i=1,2,⋯,m, a total of m nodes in the implicit layer, j=1,2,⋯,l, a total of l output nodes, and y_j is the output of the jth output node of the network that corresponds to the input sample.

There are two ways to select the center of the hidden layer nodes, one is selected from the sample input, and the other is self-organized selection using various dynamic clustering algorithms, and the commonly used clustering method is K-means, which has the advantage of being able to determine the expansion constants of each hidden node based on the distance between the centers of the clusters.

Adaboost is a state-of-the-art integrated learning algorithm [24], the core idea of which is to use the same training set to train different weak classifiers and then combine these different weak classifiers to form a classifier with stronger performance. The main idea of the Adaboost algorithm is realized by changing the distribution of data. After classifying by a weak classifier, the new weight ω and the weight coefficient α of the weak classifier are calculated, the weight α represents the weight of this classifier in the final composition of the strong classifier, and finally, all the weak classifiers are summed up by the summation operation. The use of the Adaboost algorithm minimizes the impact of extraneous factors on the data and enhances training on critical data. The theory of statistical learning also suggests that Adaboost-based learners are not prone to overfitting and achieve good performance on test sets.

The modeling process of the classification evaluation model based on the Adaboost-RBF neural network is shown in Figure 1. The model uses the RBNN model as the base classifier, repeatedly trains the RBNN classification sample output, and then applies the Adaboost algorithm to weigh multiple RBNN base classifiers to form a strong classifier, and finally applies the model to the evaluation of the effect of civic education in colleges and universities.

Figure 1.

The Adaboost-RBNN classification evaluation model flow chart

Baoji College of Arts and Sciences is a provincial general undergraduate institution of higher education that specializes in teacher education and has multiple disciplines such as literature, history, philosophy, science, engineering, management, law, education, economics, etc. The college now has several undergraduate programs, such as philosophy, civics and politics, politics and law, education, science, and technology.The college now offers a number of undergraduate majors, such as Philosophy, Civics and Politics, Teaching and Research Department, Politics and Law, Education Science and Technology. The university has always insisted on nurturing people as the root, emphasized and strengthened the construction of university culture and ideological and political education, and carried forward the good teaching style of “revering strictness, emphasizing guidance, seeking refinement and exploring new ideas” and the strong academic style of “learning diligently, practicing thinking, being truthful and practical” to guide and motivate students to grow up and achieve success. To be successful. In recent years, the university has actively constructed a long-term mechanism for improving the quality of education and education, carried out the evaluation and construction work of Civic and Political Education, and reformed the curriculum system, educational content and educational methods. Therefore, this paper chooses to evaluate the effect of Civic and Political Education using this college as the object. The data related to Civic and political education (such as daily examination results and assessment data) are collected from the teaching affairs system in Baoji College of Arts and Sciences and are inputted into the Civic and Political Education Effectiveness Evaluation System for Colleges and Universities on the Adaboost-RBNN neural network model according to the items.

The 634 sample data collected from five majors (Philosophy, Biology, Educational Science and Technology, Fine Arts, and Computer Science) were submitted to the Adaboost-RBNN network model for training to approximate the complex mapping relationship between the assessment metrics and the various assessment results, and the network converged after 4269 steps of iteration. The network’s output is a real number, so the quantitative numerical results must be converted into qualitative assessment ratings. Let the numerical result of the ith sample output be y_i, if y_i < 0.59 it is fail, if 0.6 ≤ y_i < 0.69 it is pass, if 0.70 ≤ y_i < 0.79 it is moderate, if 0.80 ≤ y_i < 0.89 it is good, if 0.90 ≤ y_i < 1.00 it is excellent. According to the scheme designed above, the training results of some samples are shown in Table 3. For the trained network, if in the future, we wish to evaluate the effect of Civic and Political education for students of a certain major, we only need to obtain the corresponding data from the teaching system and submit the resulting sample data directly to the Adaboost-RBNN network model, so that we can directly obtain the results of the Civic and Political education evaluation of the major.

The initial and final goal of designing neural network training is application, and whether the trained network can be applied lies in whether it has a good ability to generalize and promote. This is the key criterion to measure the advantages and disadvantages of neural network evaluation methods. If a network with good performance in the training stage cannot generalize or if it can generalize but is not ideal, the model is considered a failure and unusable.The generalization promotion capacity of the validated trained Adaboost-RBNN network model is examined in the following way. The trained Adaboost-RBNN network model was tested using 149 samples from chemistry students who were not used in the training phase, and the evaluation results of some samples are shown in Table 4. Out of all 149 test samples, the maximum test error is 0.0069. Of these, 144 were judged correctly, 1 was judged excellent, and 4 were judged moderate. Thus, the Adaboost-RBNN network model has a correct evaluation rate of 96.64%. The validation results show that the assessment model based on the BP neural network and in the Civic education effect proposed in this paper has strong generalization and promotion ability, and it is a feasible and reasonable assessment model, thus providing a new way to solve the problem of comprehensive and dynamic assessment of the effect of Civic education in colleges and universities.

This paper analyzes the dynamic assessment of the effectiveness of Civic and Political Education in Baoji College of Arts and Sciences, which began on October 9, 2023, and ended on January 14, 2024, with a total of 15 weeks. The results of the dynamic assessment of the effectiveness of Civic and Political Education in this college during 15 weeks are shown in Fig. 3, and the total scores of A “Educational Objectives”, D “Student Activities,” and E “Educational Effect” are 10, 20 and 15, respectively. The total assessment scores of A “Educational Objectives”, D “Student Activities,” and E “Educational Effectiveness” are 10, 20 and 15, respectively, while the total assessment scores of B “Educational Content” and C “Teacher Education” are 30. From the assessment results of the first week, Baoji College of Arts and Sciences has a slightly lower overall assessment of the effectiveness of ideological education. By comparing the average of the assessment scores among the five dimensions, it is found that the best assessment score of the first-level index of teacher education is 20.36, which is a “medium” level. The assessment score of educational objectives is only 3.85, which is “poor”, indicating that Baoji College of Arts and Sciences does not pay attention to the formulation of educational objectives in Civic and Political Education. In the 7th week, the assessment score of “educational objectives” in the effectiveness of Civic and Political Education decreased from 5.19 in the 6th week to 5.06, and the assessment scores of other level 1 indicators were also unsatisfactory. Therefore, from the 7th week onwards, the teachers of the university intervened and adjusted the strategy of ideological education in a timely manner according to the results analyzed by the dynamic evaluation model. In the 8th week, the assessment scores of “educational objectives”, “teacher education,” and “educational effect” improved greatly, from 5.06, 24.88 and 10.32 to 6.06, respectively. 10.32 to 6.14, 26.57, and 11.62 respectively.This shows that the dynamic assessment model has provided a greater impetus to improve the effectiveness of the school’s ideological education.

Figure 3.

The overall assessment of the effect of ideological and political education

This section analyzes in detail each level of indicators in the Civic and Political Education Effectiveness Assessment Indicator System in order to further explore the impact of the application of the dynamic assessment model on each indicator.

Figure 4.

Second level indicator dynamic evaluation results