Research on Paths and Countermeasures of Artificial Intelligence to Boost the Development of Russian Language Teachers in Colleges and Universities

It is known from the theory of internal and external causes of material dialectics that the development of anything is the result of the joint action of internal and external causes. The following hypotheses are proposed by selecting the factors influencing the development of Russian language teachers in higher education from the individual and the environment:

This study constructed a hypothetical model of AI-assisted development of Russian-speaking teachers, and the path of the theoretical model is shown in Figure 1, which uses AI, individual factors, and school environment as the independent variables and teacher development as the dependent variable.

Figure 1.

The path of the theoretical model

Russian language teachers in several universities were selected for the questionnaire survey. The questionnaire consists of two main parts: the first part is the personal situation of teachers in universities and colleges, including 4 aspects of gender, age, education, and title. The second part is the main body of the questionnaire, including the evaluation of 4 potential variables: individual teacher factors, school environment factors, artifacts, and teacher development. Each variable was assigned a total of 5 types using the Likert 5-point scale method of strongly agree, agree, generally, disagree, and strongly disagree, respectively, with corresponding values of 5, 4, 3, 2, and 1. Teachers were asked to score each observed variable according to their actual feelings. The questionnaire survey was conducted in the form of network distribution, a total of 100 questionnaires were distributed, 91 questionnaires were recovered, excluding invalid questionnaires such as incomplete completion, and finally 84 valid questionnaires were obtained, with a validity rate of 92.3%. The results of this survey have high reliability and validity and can be analyzed accordingly.

This study uses structural equation modeling to analyze the path of artificial intelligence in promoting the development of Russian language teachers. Structural equation modeling can be highly summarized as a deep integration of “two kinds of variables” and “two kinds of models” as follows: 1)

Two kinds of variables

Variables can be categorized into two distinct groups according to whether they are directly observable or not: observed variables and latent variables. In the broad field of statistics and data analysis, observational variables can be captured directly through measurement and observation, and their values are usually precise and reliable, with negligible errors. In contrast, latent variables can only be traced through the indirect measurement of observed variables. Because of this indirectness, the measurement of latent variables is often accompanied by a certain degree of error, making it necessary to be more careful and meticulous in exploring their true nature.

Structural equations: 3η=Bη+Γξ+ζ

Here, X is set as the observation pointer for the externally derived latent variable (ξ), Y is set as the observation flag for the internally derived latent variable (η), and Λ_y and Λ_x constitute the coefficient loading matrix of the measurement equation.

The structural equation modeling contains the following assumptions: 1)

In the measurement equations, the error terms δ and ε both have a common property that their mean values are zero.

Therefore, the central task in solving the structural equation model is to determine the four key coefficient matrices: Λ_y, Λ_x, B, and Г.

AMOS 23.0 was used to conduct a goodness-of-fit test on the model of AI-assisted Russian language teacher development. The results of the goodness-of-fit test are shown in Table 1, where in terms of the overall fit indicators of the hypothesized model, CMIN/DF = 1.731 < 3, GFI = 0.913 > 0.9, AGFI = 0.938 > 0.9, RMSEA = 0.021 < 0.08, NFI = 0.942 > 0.9, RFI = 0.939 > 0.9, IFI = 0.926 > 0.9, TLI = 0.949 > 0.9, CFI = 0.921 > 0.9, and PGFI = 0.806 > 0.5. All the indicators are in the range of the adapted critical values, which indicates that the model fit is good and acceptable.

Based on the above analysis, the model of AI-assisted teacher development in Russian was run to obtain the standardized path coefficients as shown in Table 2. All six hypotheses proposed in this study were supported at the 5% level, i.e., individual factors, school environment and artificial intelligence all positively and significantly influence teacher development. Meanwhile, the artificial intelligence factor has a positive and significant effect on the individual teacher factor and the school environment factor, and the school environment application also positively affects the individual teacher factor.

The path coefficient of individual teachers’ factors on their development was the largest at 0.571, indicating that individual factors have the most significant effect on the development of Russian teachers in higher education. School environment factors and AI factors also have some influence on the development of Russian language teachers in colleges and universities, with path coefficients of 0.471 and 0.456, in that order.

Cluster analysis is based on the similarity of the data characteristics of the classified samples, according to certain rules to divide the samples into a number of classes, so that the samples in the same class have a high degree of similarity between each other, while the samples in different classes are highly dissimilar to each other. Cluster analysis has been widely used in many fields such as data analysis, pattern recognition and image processing.

The similarity between samples is mainly measured by distance in cluster analysis, and the samples can be clustered using systematic clustering. At the beginning of clustering, each of the n samples is treated as a class, and the distances between the samples and the distances between the classes are calculated. Then the two closest classes are combined into a new class and the distance between the new class and the other classes is calculated. The merging of the two closest classes is repeated until all the samples are merged into one class.

Ward’s systematic clustering method is a clustering method that utilizes the sum-of-differences-squared method for calculating distances; the sum of the squared Euclidean distances from each element of the class to the center of gravity (i.e., the class mean) of the class is referred to as the sum of the squared intraclass departures.

Assuming that classes G_K and G_L are clustered into a new class G_M, the sums of squared intraclass deviations for G_K, G_L and G_M are given by equations (4), (5), and (6), respectively: 4WK=∑xi∈GK(xi−x¯K)′(xi−x¯K) 5WL=∑xi∈GL(xi−x¯L)′(xi−x¯L) 6WM=∑xi∈GH(xi−x¯M)′(xi−x¯M)

When G_K and G_L merge to form the new class G_M, W_M > W_K + W_L, i.e., the sum of squared deviations within the class increases. If G_K and G_K are closer, the value of the increase in the sum of squared deviations should be smaller. Thus the squared distance between G_K and G_L is calculated according to equation (7): 7DKL2=WM−(WK+WL)

The sum of squared deviations method is to apply the idea of analysis of variance in categorization, so that a small sum of squared deviations in the same class indicates a high degree of similarity between the samples. A large sum of squared deviations between different classes indicates a low degree of similarity between the samples. Measuring the similarity between samples by the size of the sum of squared deviations meets the requirements of cluster analysis.

K-means clustering algorithm is a commonly used unsupervised learning algorithm for dividing the sample points in a dataset into a number of predefined clusters. The basic idea of K-means is to assign the data points to the nearest clusters by iterating and updating the centers of the clusters until a stopping condition is reached.

Principle of K-means algorithm, given a dataset X = (x₁, x₂, x₃, x₄ ⋯x_n) containing n sample points and a predefined number of clusters k, firstly, k points are randomly selected as the initial clustering centers C = (C₁, C₂, C₃, ⋯ C_k), and then for each point, the distance of the point to the clustering centers is computed as shown in Eq. (8): 8y¯i=argminidist(xi,Cj)

After completing the calculation, the distance of the point from the center of the clusters is determined and it is classified to the closest center. Finally for each class, the sample mean is calculated and that value is used as the re-selected clustering center as shown in equation (9): 9y¯j=∑iI(y¯i=j)xj∑iI(y¯i=j)

The above steps are looped and the loop terminates when the clustering center no longer changes or the number of iterations is reached.

In the calculation process, the Euclidean distance is usually used as a measure of distance, and the Euclidean distance indicates the geometric distance between two sample points. The formula is shown in equation (10): 10D(x,y)=∑i=1n(xi−yi)2 x(x₁, x₂, …, x_n) and y(y₁, y₂, …, y_n) denote two points respectively, which can represent any point in n-dimensional space.

K-means algorithm has the advantages of simplicity, speed, unsupervised, wide applicability and strong interpretability, but there are many problems in practical applications, such as the selection problem of k values, initializing the clustering center, and simple random selection will lead to biased results. In order to improve the effect of K-means clustering algorithm, the following optimization methods can be considered. 1)

Dynamic adjustment of k value

Dynamic Adjustment of k Values aims to dynamically determine the k values on different datasets according to the characteristics and structure of the data, so as to improve the adaptability of the algorithm and the clustering effect. Common methods for dynamically adjusting k values include the elbow rule, the Gap statistic method, and the contour coefficient method applied in this paper. The principle of the contour coefficient method algorithm is to calculate the contour coefficient of each data point, consider the intra-cluster distance and inter-cluster distance comprehensively, and finally select the k value that maximizes the overall contour coefficient.

In dynamic k value selection, the optimal k value is usually determined by a cluster validity metric or a characterization of the data distribution. As far as the current study is concerned, the commonly used clustering validity metric is the contour coefficient, which measures the degree of similarity of a data point to the data points in the cluster to which it belongs as well as the degree of difference from the data points in other clusters. For each data point, the contour coefficient is defined as shown in equation (11): 11S=b−amax(a,b)

The formula for the contour coefficient is shown in equation (12): 12SC=1n∑i=1n(bi−ai)max(bi,ai)

In Eq. (12), a denotes the average distance between a data point and other data points in the class it belongs to, and b denotes the average distance between a data point and other data points in the nearest cluster. The contour coefficients are taken in the range of [–1,1], and by calculating the contour coefficients at different k values, the k value that maximizes the contour coefficients is selected as the optimal number of clusters.

In the sample of 60 Russian language teachers surveyed for this study, statistics on visits to the platform over a six-month period revealed that the most frequent visits occurred 2758 times, while the lowest was only 54, indicating that there is still very much room for improvement in the engagement of Russian language teachers with the learning management platform.

Figure 2 shows the frequency of the occurrence of the operational behavior of the use of the four pedagogical tools. the comprehensive statistics of the number of times 60 Russian teachers used the different pedagogical tools on the learning management platform, the average number of times the course management tool, the content management tool, the communication and cooperation tool, and the course evaluation tool were used by the Russian teachers was 42.35, 180.78, 67.57, and 183.08. it can be seen that the course management tool had the used the least number of times, followed by the communication and cooperation tool, the content management tool and the course evaluation tool were used relatively frequently, especially the content management tool had a minimum value of the number of times it was used that was greater than zero, indicating that all the Russian language teachers had behaviors such as content retrieval and resource posting occurring when using the platform.

Figure 2.

The use frequency of four teaching tools

Through second-order cluster analysis of the frequency of different types of tool use of the standardized teachers, the 60 teachers were divided into four clusters with different behavioral tendencies. The centroid distribution of the frequency of tool use for the Russian language teachers’ clusters is shown in Figure 3, with (a) to (d) showing the behavioral distribution of the teachers in the first to fourth categories, respectively.

Figure 3.

Frequency distribution of Russian teacher group using tools

The majority of Russian language teachers are categorized into the first category, which consists of 41 teachers, accounting for 68.33% of the teacher sample, and the standardized z-scores of these teachers on the frequency of using course management tools, content management tools, communication and collaboration tools, and course evaluation tools are (-0.51, -0.45, -0.33, and -0.24), respectively, and the frequency of using each type of tool is lower than the. The frequency of use of each type of tool is below average, which indicates that these Russian-speaking teachers have a low level of participation and motivation in the use of all the tools of the platform, and therefore this group is named in this study as the “low participation” group of teachers.

The second group of 10 Russian-speaking teachers, or 16.67% of the sample, showed a high tendency to use course management tools. The frequency of use of the communication and collaboration tool and the course evaluation tool was about the same as the average, while the use of the content management tool (1.39) and the course management tool (1.15) was higher than the average, which suggests that this group of teachers focuses mainly on preparation of course materials and management of the classroom when using the LMS platform. This suggests that teachers in this category focus on preparing course materials and managing the classroom when using the LMS platform, and thus can be called “management-oriented teachers.

The third category of Russian teachers totaled seven, or 11.67% of the sample of teachers. This group of teachers used content management tools (0.34), communication and collaboration tools (2.54), and course management tools (0.62) more often than the average, with communication and collaboration tools being the most prominent, which indicates that this group of teachers emphasizes interaction and group work between teachers and students, and that the modules they use most often are messaging, discussion boards, e-mail, and groups. According to the outstanding characteristics of these teachers, they can be called “interaction-oriented teachers”.

The fourth category of Russian-speaking teachers has the smallest number of teachers, only 2, or 3.33% of the total sample of teachers, and the behavior of this group of teachers in the use of course management tools (-0.55) is low in relation to the mean, suggesting that this group of teachers seldom uses the range of course management tools such as announcements, assignments, rosters, and course reports, and that the use of content management tools (0.24) is below the mean. The most striking feature of this group of teachers is the frequency of use of course evaluation tools. The Z-score for the frequency of use of course evaluation tools is 4.14, which is much higher than the average and the highest of the three groups, indicating that this group of teachers attaches great importance to the evaluation and feedback functions of the platform, and that they make high use of modules such as assignments, surveys and tests, and the grade center, which is why we can refer to the Russian-speaking teachers in this group as “evaluation-oriented teachers”. Therefore, teachers of Russian in this group can be called “evaluation-oriented teachers”.

The analysis of Russian teachers’ behavior is a practical basis for guiding teachers to adjust their teaching programs. The analysis of teachers’ behavior using data mining technology is conducive to the improvement of teaching efficiency and teaching methods, and is of great significance in improving teachers’ teaching effectiveness and promoting teachers’ development under information technology.

Differences in the access behavior of Russian teachers of different genders to the various tools were tested by non-parametric tests, and the results of tool use by teachers of different genders are shown in Figure 4. The most accessed tools by male teachers were content management tools, followed by course evaluation tools, with mean values of 138.54 and 130.58, respectively. The most accessed behaviors by female teachers appeared to be access to content management tools, followed by course evaluation tools with communication and collaboration tools. However, the results of the Mann-Whitney U-test showed that no significant differences existed in the frequency of access to the various types of tools by Russian language teachers of different genders (p > 0.05).

Figure 4.

Tools using for different gender teachers

Table 3 shows the results of the test of differences in the frequency of use of different tools by teachers of different age levels. Russian language teachers under the age of 35 years used the communication and cooperation tools significantly more frequently than the other age groups. The common characteristic of Russian language teachers aged over 46 years is the high frequency of use of content management and course evaluation tools. The results of one-way ANOVA showed that no significant differences exist in the behavior of Russian language teachers of different ages when using the platform (p > 0.05).