Research on an early warning model of effectiveness evaluation in ideological and political teaching based on big data

“Social computing” has two meanings [10]: one is to serve social work as a tool of “social software”, which means to apply the latest information computing technology to social activities; the second one is to embed humanities and social sciences into computer or computing technology so as to realise in-depth research in social and humanities fields.

Although the concepts of social computing and big data are contemporary words, in recent years, with the rise of big data, both of them are based on computing technology and gradually blend with each other and complement each other. It can even be redefined as follows: scientific computing theories and technologies such as system theory, cloud computing, artificial intelligence and data analysis are used as methods and means in social science theory, which effectively solves the multifarious problems in the field of humanities and social sciences. Social computing in the era of big data is based on massive data, by continuously recording, mining and analysing personal thought and behaviour data information, it is possible for scientific, quantitative analysis and accurate research and calculation of humanities and social sciences, interpersonal interaction and social complex adaptive system.

The assumption and possibility of applying big data prediction function in the field of IAP education are based on social computing theory. The guidance of social computing ideas provides a new opportunity for big data technicians and researchers of IAP education to explore deeply, that is, using big data prediction technology and cloud computing technology to explore the IAP education in the field of humanities and social sciences.

As shown in Figure 1, at present, there are two explanations of educational informatization in academic circles: “technology theory” and “process theory”.

Fig. 1

Technology theory takes information as a scientific and technological means and applies modern information technologies such as computer, Internet, mobile terminal, big data and cloud computing to the process of education, thus promoting the comprehensive reform of education and enhancing its effectiveness. This theory holds that information technology assumes a significant part in the advancement of education, which is an indispensable tool and means to realise educational informatization; so, it implies the attributes of “technology” and “education” [11].

Process theory holds that educational informatization is a process of improving the effectiveness of teaching methods by modern information technology so as to cultivate the information literacy of educational objects and realise the modernisation of education. Specifically, it refers to taking the educational thought as the guidance in all categories; by using information technology, we can effectively develop and make full use of information resources at all levels and train high-tech talents for the society, thus accelerating the process of educational modernisation.

The method of IAP education is accompanied by educational practice, which is attached to the different stages of the process of IAP education and the constant renewal of its activity forms; so, there are various concrete forms, such as indoctrination education method, practice exercise method, example demonstration method, self-education method, psychological guidance method, etc. [12, 13]. The method of IAP education follows the principle of creativity, that is to say, educators should constantly study new situations, creatively use traditional educational methods or summarise and explore new educational methods. The most important innovative way is to innovate IAP education methods by absorbing and applying research results of related disciplines or to use modern scientific and technological achievements to realise the modernisation of educational means.

With the advent of the era of big data, big data thinking and technology not only participate in the whole process of IAP education research and innovate the practical activities of IAP education but also promote its traditional process, that is, the process of making, implementing and evaluating educational programmes can be realised by new ways of thinking. Meanwhile, under the background of big data prediction function, the IAP education modes such as intelligent calculation, precise customisation and immediate prevention have been born, which also provide new educational methods for educators.

It is essentially a user learning behaviour collection system based on the computer system, which is used to collect user learning behaviour [14]. As shown in Figure 2, it has the functions of interactive, analogue and management platform.

Fig. 2

With the high-level development of quality-oriented education, educational data mining is gradually becoming a mainstream research trend. In recent years, the rapid development has attracted many educational scholars and research institutions to invest in this research. At present, education data mining mainly includes the following processes and steps:

The dynamic early warning system of IAP education based on the improved SVM algorithm mainly realises the data collection and potential law mining of the courses, effect evaluation, ideological dynamic control, early warning feedback and other links of IAP education for students in higher vocational colleges with obvious potential ideological dynamic differences [18] so as to realise the complete life cycle information management and control of IAP education in higher vocational colleges. This is shown in Figure 3.

Fig. 3

As shown in Figure 4, corresponding to the dynamic early warning model of IAP education based on the improved SVM algorithm, the internal workflow flow order is as follows:

Step 1: Enter the important speech made by General Secretary Xi Jinping under the new situation and relevant national policies and regulations as the general basis for higher vocational colleges to formulate individualised IAP education programmes;

Step 2: Carry out standardised inspection and small-scale experiments on the individualised IAP education plan formulated by higher vocational colleges;

Step 3: Decide whether to popularise according to the experimental results. If the experimental results support popularisation, start the IAP education effect analysis submodule and evaluate the personalised IAP education scheme formulated by higher vocational colleges from multi-dimensional effect evaluation; otherwise, return to the first step;

Step 4: Start the dynamic early warning submodule of IAP education, find out the students with dangerous ideological trends in advance and give early warning tips and take timely and active intervention measures to ensure that the IAP education in higher vocational colleges plays an effective role.

Fig. 4

To make comprehensive fuzzy evaluation on the quantitative data of users' invisible early warning, it is integrated into the decision analysis system mechanism, as shown in Figure 5.

Fig. 5

Fuzzy neural network is introduced for decision evaluation and adaptive training of fuzzy rules. Structurally, the user's invisible interest input layer, membership function fuzzification layer [19], fuzzy rule adaptive training layer, output variable fuzziness division layer and user's invisible interest decision analysis layer are formed. Functionally, the BP neural network is used as the pattern memory. The self-learning ability is used to continuously update the optimisation weight coefficient, which makes the fuzzy system have generalisation ability, realises the comprehensive evaluation of personalised service and feeds back the results to the improved support vector machine algorithm, thus realising a benign closed loop.

Support vector machine (SVM) algorithm essentially belongs to an efficient, limited and generalised classifier with supervised and extensible classifiers, which can overcome linear and nonlinear obstacles. In order to realise nonlinear multi-core data mining and clustering effect, improve the memory consumption ratio, balance generalisation ability and learning ability, improve the interpretability of data sets and strengthen the adaptability of kernel functions and so on, introducing cold and hot data separation factor and random gradient descent factor to improve the SVM algorithm [20]. The specific steps are as follows.

It can be seen from Figure 3 that the dynamic early warning model of IAP education oriented to multidimensional application is transformed into an unlimited experience loss minimisation problem with penalty factors, and the objective function is defined as Eq. (1). (1) f(ω)=minωλ2||ω||2+1m∑l(ω,(x,y)) f(\omega ) = \mathop {\min }\limits_\omega {\lambda \over 2}{\left| {\left| \omega \right|} \right|^2} + {1 \over m}\sum l(\omega ,(x,y)) (ω, (x, y)) in Eq. (1) can be calculated using Eq. (2). (2) l(ω,(x,y))=m{0,1−y<ω,x≫ω} l(\omega ,(x,y)) = {\bf m}\left\{ {0,1 - y < \omega ,x \gg \omega } \right\}

Using random gradient descent to solve the objective function, in each iteration period, randomly select training samples and map the corresponding objective function gradient and select the gradient step length on the opposite side of the iteration direction to ensure that the running time of the algorithm is satisfied O(nλg) O\left( {{n \over {{\lambda _g}}}} \right) , among them, n is the sum of dimensions in the constraint space [21] of ω and x. To solve the duality problem of nonlinear correspondence, the following mapping transformation is carried out as shown in Eq. (3). (3) ∑αiyixi→(≠0 is an integer) \sum {\alpha _i}{y_i}{x_i} \to \left( { \ne 0\;{\rm{is}}\;{\rm{an}}\;{\rm{integer}}} \right)

After the mapping of Eq. (3), the problem of minimising unlimited experience loss with penalty factors under the multidimensional constraint given in Eq. (1) is transformed into the problem of solving extreme value under the single constraint. Further, smooth loss functions are used to replace hinge loss to further transform the problem into a smooth and unconstrained optimisation problem under the hyperplane. The specific solution process is as follows:

A training sample i_t is randomly selected in the hyperplane constraint space, among them, i characterises intrinsic property of the sample, t characterises the external activity (iteration times) of the sample, and these are integrated into Eq. (1), such as Eq. (4). (4) f(ω,it)=λ2||ω||2+l(ω,(xii,yit)) f(\omega ,{i_t}) = {\lambda \over 2}{\left| {\left| \omega \right|} \right|^2} + l\left( {\omega ,\left( {{x_{{i_i}}},{y_{{i_t}}}} \right)} \right)

Sub-gradients are used to solve Eq. (4), such as Eq. (5). (5) ∇t=λωt−I[yit{ωt,xii}<1]yitxii {\nabla _t} = \lambda {\omega _t} - I\left[ {{y_{{i_t}}}\left\{ {{\omega _t},{x_{{i_i}}}} \right\} < 1} \right]{y_{{i_t}}}{x_{{i_i}}}

In Eq. (5), I [y_it {ω_t, x_ii} < 1] is the indicator function, with a range of two values; if it is true, it is 1; otherwise, it is 0. Based on Eq. (5), input the data set of user interest points S, regularization factor λ, and sample external activity (number of iterations) T, the iteration of one cycle can be expressed as Eq. (6). (6) ωt+1≤ωt−βt∇t {\omega _{t + 1}} \le {\omega _t} - {\beta _t}{\nabla _t}

In Eq. (6), βt=1λt {\beta _t} = {1 \over {{\lambda _t}}} is an adaptive step factor, which is negatively correlated with the number of iterations, and Eq. (7) can be obtained by bringing Eq. (5) into Eq. (6). (7) ωt+1≤ωt−βλωt−I[yit{ωt,xii}<1]yitxii {\omega _{t + 1}} \le {\omega _t} - {\beta _\lambda }{\omega _t} - I\left[ {{y_{{i_t}}}\left\{ {{\omega _t},{x_{{i_i}}}} \right\} < 1} \right]{y_{{i_t}}}{x_{{i_i}}}

Further, simplify the Eq. (7) and backward deduce the available Eq. (8). (8) ωt+1≤(1−1t)ωt+βtI[yit{ωt,xii}<1]yitxii {\omega _{t + 1}} \le \left( {1 - {1 \over t}} \right){\omega _t} + {\beta _t}I\left[ {{y_{{i_t}}}\left\{ {{\omega _t},{x_{{i_i}}}} \right\} < 1} \right]{y_{{i_t}}}{x_{{i_i}}}

Based on Eq. (8), if the indicator function is true, it is converted to Eq. (9). (9) ωt+1≤(1−1t)ωt+βtyitxii {\omega _{t + 1}} \le \left( {1 - {1 \over t}} \right){\omega _t} + {\beta _t}{y_{{i_t}}}{x_{{i_i}}}

Based on Eq. (8), if the indicator function is false, it is converted to Eq. (10). (10) ωt+1≤(1−1t)ωt {\omega _{t + 1}} \le \left( {1 - {1 \over t}} \right){\omega _t}

It can be seen from Eq. (8) that the cold and hot data can be separated by setting the comprehensive bias term. Furthermore, by introducing the online learning mechanism, the predictor with low generalisation error can be obtained, which balances the generalisation ability and learning ability.

The construction of the early warning index system is the basis of whether its evaluation method is effective or not. The basic way is to decompose the target layer by layer according to certain standards to form an index system. Through interviews and investigations with educational scholars and experts from industrial enterprises in higher vocational colleges, the early warning evaluation index system of IAP teaching effectiveness as shown in Table 1 is adopted.

The early warning evaluation value of higher vocational education quality is directly related to the determination of early warning limits. In view of its fuzziness and subjectivity, this paper uses the fuzzy comprehensive evaluation method [22], which combines fuzzy mathematics with analytic hierarchy process to calculate.

In early warning, the early warning limit is determined by the range of early warning value, and different early warning signals need to be determined according to the early warning limit so as to send a warning to the manager with an intuitive image. So, the rationality of early warning limit determination is an important factor to determine the accuracy of early warning. This paper uses the theory of statistics 3σ to determine the early warning limit [23] (14) σi=∑j=1N(x¯i−xij')2m−1 {\sigma _i} = \sqrt {{{\sum\nolimits_{j = 1}^N {{\left( {{{\overline x }_i} - x_{ij}^\prime} \right)}^2}} \over {m - 1}}} where σ_i is the standard deviation of the i-th index. Then, the comprehensive standard deviation σ of the warning sign early warning evaluation value is calculated using Eq. (15). (15) σ=∑i=1m(x¯−xi')211 \sigma = \sqrt {{{\sum\nolimits_{i = 1}^m {{\left( {\overline x - x_i^\prime} \right)}^2}} \over {11}}}

According to the principle of normal distribution in statistics, when the early warning evaluation value of each index is within the interval of (x̅−σ,x̅+σ], it shows that most of the early warnings are in a loose state, which needs to be analysed according to the specific situation and belongs to the benign warning situation. If the early warning evaluation value is within the interval of (x̅−2σ, x̅−σ], it shows that a small number of warning signs are in a state of tension, which belongs to mild warning situation. If the early warning evaluation value is within the interval of (x̅−3σ, x̅−2σ], it means that more than half of the warning signs are in a state of tension, which is a moderate warning situation. If the early warning evaluation value is within the interval of (−∞,x̅−3σ, it means that most or even all warning signs are in a state of tension, which is a serious warning situation.

After dividing the warning interval, it is necessary to determine the warning level and output the corresponding warning status by signal light method. In the design of warning signal system, ★ indicates the first-level warning, ★★ indicates the second-level warning, ★★★ indicates the third-level warning and ★★★★ indicates the crisis warning. The more serious the warning situation, the fewer stars, as shown in Table 2.