Research on optimal allocation of accounting teaching resources in cloud computing environment based on genetic algorithm framework design

With the development of information technology, the social demand for accounting talents has changed greatly. The comprehensive quality and professional practice skills of accounting talents are more and more demanding. How to meet the social demand for accounting talents is the key issue considered by every university with accounting majors [1–2]. How to integrate the internal and external resources of the school, realize the optimal allocation of internal and external resources, build a platform for the internship and practical training of accounting majors, provide students with practical opportunities, and enhance the practical application ability is an urgent problem to be solved for the development of accounting majors [3–5]. The efficient and reasonable allocation of teaching resources needs to be considered not only in the scheduling process before the beginning of the semester, but also for the temporary increase of teaching plans after the beginning of the school year, such as training courses for working people, remedial classes or re-training classes, which are even more necessary to optimize the allocation of resources due to the fact that the teaching resources are more limited at this time [6–8]. In essence, the scheduling of these additional teaching classes is actually a scheduling problem, which has been shown to be NP-complete. For this type of complete polynomial nondeterministic problems, although it is possible to check the problem in polynomial time and then use the exhaustive method to obtain the solution of the problem, the computational time in this way increases exponentially with the complexity of the problem, so this solution algorithm does not have practical application [9–12]. In order to meet the requirements of computational performance in practical applications, the NP-complete problem usually adopts the strategy of finding approximate solutions, and the genetic algorithm in this kind of algorithm searches for the approximate optimal solution by simulating the natural evolution process, which has mature theory as the basis and is more widely used [13–15]. In this regard, the study of the optimal allocation of accounting teaching resources based on the framework of the genetic algorithm helps to solve the shortage of accounting teaching resources in colleges and universities, and creates a useful attempt to create a “teaching community” with complementary advantages, mutual benefits and win-win cooperation, which greatly alleviates the situation of insufficient teaching resources and enables students to enjoy more high-quality educational resources, and strongly guarantees the quality of teaching. The shortage of teaching resources in accounting is a beneficial attempt to create a “teaching community”, which can greatly alleviate the situation of insufficient teaching resources, enable students to enjoy more quality educational resources, and strongly guarantee the quality of teaching [16].

This paper briefly introduces the paper forming process using the intelligent paper forming algorithm. Taking the high-quality accounting question paper needed for the actual examination as the starting point, a mathematical model of the paper-grouping problem is established. Then, according to the constraints that need to be considered when organizing accounting paper resources, the chromosome coding method and fitness function are designed, and the crossover and mutation methods in the algorithm are adjusted. Then the operations of the improved genetic algorithm are mapped to the mathematical model of the grouping problem. Finally, the improved genetic algorithm is verified through experimental analysis in solving the problem of optimal allocation of accounting teaching resources in a cloud computing environment. It has been verified through experiments that the improved genetic algorithm is capable of producing higher quality test papers when solving the accounting paper-forming problem.

2

Optimization of the problem of grouping accounting resources in a cloud environment

2.1

Study on the organization of volumes

The task of grouping is to group the necessary test papers from the massive test bank with high speed and quality, and the excellent grouping algorithm is the foundation for realizing its function.

According to the conditions of high-quality test papers, the intelligent grouping algorithm completes the grouping process as shown in Figure 1.

In this paper, the quality of test questions will be evaluated in terms of six aspects: the type of questions to which the accounting resource test questions belong, the number of questions, the difficulty of the questions, the knowledge of the questions, the differentiation of the questions, and the exposure of the questions.

The specific test question attributes will be briefly described below. 1)

Types of test questions

According to the standards of modern educational examination, combined with the actual teaching, this paper mainly adopts several types of test questions: multiple-choice questions, fill-in-the-blank questions, short-answer questions and calculation questions.

2)

The number of test questions

In general, in order to meet the time requirements of the actual test paper, the objective questions of the test paper should account for the number of test questions accounted for some significant, subjective questions accounted for the number of test questions proportionally less.

3)

Difficulty of the paper

The difficulty of the test paper reflects the degree of difficulty of the students in solving the problem. When the test questions are entered into the test bank, the difficulty coefficient K_i should be entered for each test question based on the scores of the students in the last three years, which is calculated by the following formula: 1 $K_{i} = 1 - \frac{q_{i}}{p_{i}}$ where q_i denotes the average score obtained by all students for the i h trial and p_i denotes the actual score for the i th trial.

4)

Knowledge points of test questions

Test question knowledge points reflect the scope of knowledge of the examined test questions, is able to detect the examination of the scope of the test paper group paper to meet the specific requirements of the important indicators. In the test questions entered into the test bank, should indicate that the test question belongs to the knowledge points in the specific test bank of knowledge points can be expressed in numerical values.

5)

Test question differentiation

The differentiation of the test questions reflects the differences in the scores of different students on the test questions, and can accurately assess the extent to which students have mastered the knowledge examined. When the test questions are entered into the test bank, the differentiation coefficient Q_i of each test question should be entered according to the students’ scores in the past three years, and the specific formula is as follows: 2 $Q_{i} = \frac{\bar{p_{H}} - \bar{p_{L}}}{p_{i}}$ where $\bar{p_{H}}$ denotes the average score obtained for the i th test question in the higher score band (e.g., top 20%), $\bar{p_{L}}$ denotes the average score obtained for the i th test question in the lower score band (bottom 20%), and p_i denotes the actual score for the i th test question.

6)

Test question exposure

The exposure of test questions accurately reflects the frequency of their use. When test questions are entered into the test bank, the exposure of each test question should be entered according to the probability of the test question appearing in the examination paper within three years, and if the test question is entered for the first time and there is no historical data, the exposure of the test question is set to zero.

2.2

Mathematical modeling of grouping problems

This paper takes the actual examination demand as the starting point, establishes the mathematical model of the group paper problem, first of all, according to the actual situation of the examination paper to make scientific and reasonable assumptions on the mathematical model of the group paper problem, secondly, the constraints of the group paper problem are set up specifically, and finally sets up the objective function needed for the group paper problem.

Intelligent paper organization problem is a problem of seeking optimal or near-optimal solutions under multiple constraints. Therefore, the mathematical model of the paper grouping problem Z involved in this paper can be defined as Z = {T, S, D, SP, DP, SE}, such a kind of six-tuple problem, based on which the mathematical model matrix of the paper is established, in which the rows indicate the number of questions and the columns indicate the attributes of each question, and the specific formulas are as follows:

Where m is the total number of genes (question ID) in one paper chromosome, K_i indicates the difficulty coefficient of the i th test gene, and P_i is the actual score of the i th test gene.

Knowledge point coverage constraint SP, which specifies the content and scope of the knowledge points contained in the group of question papers on the content you want to examine the specifics, which will help teachers to accurately understand the students’ chapter knowledge mastery. In this paper, taking into account the actual needs, in order to more scientifically represent the knowledge points examined, the percentage of knowledge points scored is used to represent the knowledge point coverage calculation. Assuming that the actual number of knowledge points examined in a paper totaled n, the specific formula for a certain knowledge point coverage SP_i is as follows: 3 $S P_{i} = \frac{S D_{i}}{\sum_{i = 1}^{n} S D_{i}}$ 4 $\sum_{i = 1}^{n} S P_{i} = 1$ where SP_i is the percentage of the total marks of the paper accounted for by the marks of the i th knowledge point, and SD_i denotes the marks of the i th knowledge point.

The differentiation constraint DP, through the test paper differentiation, can help teachers to grasp the students’ learning situation in a real and effective way. In this paper, the specific formula for test paper differentiation DP is as follows: 5 $D P = \frac{\sum_{i = 1}^{m} Q_{i} \cdot P_{i}}{\sum_{i = 1}^{m} P_{i}}$

Where Q_i refers to the coefficient of differentiation for the i th question in a paper, P_i is the actual mark for the i th question on the paper, and DP refers to the overall differentiation of a paper. The question differentiation factor should be updated in real time after each examination.

The exposure constraint SE, which describes the probability of certain test questions being repeated, indicates the exposure of the paper by the frequency of the examination questions being repeated in the group of papers. This constraint should be considered based on the actual nature of the subject, e.g., the papers required for screening type subject exams may have a lower exposure than the papers required for examination type subjects. In this paper, the specific formula for the exposure SE is given below: 6 $S E = \frac{\sum_{i = 1}^{m} a_{i}}{m}$

Where a_i ndicates whether the i th question in the group paper is repeated or not, 1 if it is repeated and 0 if it is not repeated.

The fitness function [17] reflects the adaptation of the individuals selected by the genetic algorithm to the problem under study. Since the improved genetic algorithm satisfies some constraints at the beginning of the initialized population design, the remaining indicators influence the fitness function in four main ways: the difficulty constraint of the test paper, the percentage of knowledge scores constraint, the differentiation of the test paper, and the overall exposure constraint of the test paper. Set the weights of these four influencing aspects as T_i(i ≤ 4), and take the absolute value of the difference between the expected value and the actual value of the constraints of different groups of volumes, and the difference result is expressed by ω_i(i ≤ 4). The objective function g in this paper is calculated as follows: 7 $g = \sum_{i = 1}^{4} T_{i} \cdot ω_{i}$ 8 $\sum_{i = 1}^{4} T_{i} = 1$

From the formula of the adapted function in this paper, it can be seen that if the result of the calculation of the g value of the objective function of group winding is smaller, it indicates that the result of individual evolution is more in line with the expectation. Therefore, in the actual process of grouping, in order to obtain the optimal individual solution of grouping, the g value of the objective function should be as small as possible. At the same time, in order to make the fitness value positively reflect the individual evolutionary advantages and disadvantages, and to ensure that the time complexity of the calculation of the fitness function is small, this paper is based on the objective function g of the grouping problem, and designs the formula for its relationship with the fitness function f as follows: 9 $f = \frac{1}{1 + g}$ where f is the fitness function of the grouped problem and g is the objective function of the grouped problem.

3

Research on Intelligent Accounting Grouping Model

3.1

Brief description of genetic algorithm

Genetic algorithm [18] mainly includes initial population, fitness function, chromosome coding, genetic operators and so on. Genetic algorithms have gained wide use in production scheduling problems, function optimization, machine learning and image processing, etc. intelligent grouping is the study of genetic algorithms as an optimization tool.

Simple Genetic Algorithm (SGA) can be defined using an octet as shown in equation (10): 10 $S G A = {C, E, P_{0}, M, Φ, Γ, Ψ, T}$

The significance of each element in the formula is as follows: symbol C : coding method of chromosome; symbol E : fitness function; symbol P_o : initial population; symbol M : population size; symbol Φ : selection operator; symbol Γ : crossover operator; symbol Ψ : variation operator; symbol T : termination condition of the genetic operation.

By analyzing the principle of genetic algorithm, the basic flow of the algorithm is obtained. 1)

Initial population: set the iteration counter to randomly generate the initial population.

2)

Calculate the fitness value: calculate the fitness value of individuals in the population.

3)

Selection operation: selection is based on the principle of survival of the fittest to select the more adapted individuals from the population to form the offspring. The main methods are: roulette selection, tournament selection, random traversal sampling and so on.

4)

Crossover: In individuals selected for reproduction of the next generation, the homozygotes of two different individuals are interchanged to produce a new individual. The main methods of crossover are: single-point crossover, multi-point crossover, etc.

5)

Mutation operation: Mutation is based on the principle of genetic variation in biological inheritance, according to the selection probability of roulette to randomly select an individual as the parent generation, and randomly select the gene position with the probability of mutation of certain individuals of certain gene positions to perform anisotropic transformation.

6)

Judgment of termination conditions: the computation is terminated when the fitness value of the optimal individual reaches a given threshold or meets the maximum number of iterations 29.

3.2

Application of Improved Genetic Algorithms to Grouping Problems

3.2.1

Chromosome coding design and population initialization

The use of genetic algorithms for the group paper, first of all, to map the ID sequence of the questions in the question bank and its composition of strings, composed of strings representing an individual that is a set of questions in the combination of questions in the paper, that is, the chromosome coding process corrected for the grouping of the question paper.

For the group paper problem, if we use binary encoding to map the ID of the questions in the question bank as a combination of characters such as “01”, then it will lead to the encoding scale is too large, making the algorithm take up more resources, and the amount of computation also rises. In this paper, there are four different types of questions in the question bank, i.e., multiple choice, fill-in-the-blank, short answer, and computational questions. Therefore, real number segments are used to encode the different types of questions separately, the advantage of which is that the encoding between the types of questions is not interfered with, so that it is easy to perform crossover and mutation operations at a later stage.

In the beginning of the algorithm search needs to be carried out in the initialization of the population operation, this paper to the random function of the way of the exercises in the exercise library for the selection and composition of the exercises collection, in the process of selection if there is a ID duplicate test questions, then re-selected from the exercise library, until the number of individuals in the population is, so as to ensure that the differences in the population, the specific operation process shown in Figure 2.

3.2.2

Adaptation function design

The mathematical model for grouping papers is designed in Chapter 2 along with its constraints. Since the algorithm satisfies some constraints i.e., the number of questions and score constraints of the test questions contained in a set of test papers before the population initialization operation, when designing the fitness function for the grouping problem, the size of the fitness value of an individual test paper is only related to the difficulty coefficient of the test paper D, the knowledge coverage R and the differentiation of the test paper TD then the composition of the fitness function is as shown in Eq. (11): 11 $F = 1 - (1 - R) \times α_{1} - | D - D E | \times α_{2} - | T D - T D E | \times α_{3}$

In the formula, α₁, α₂ and α₃ are the weights of the knowledge point coverage, difficulty coefficient and differentiation of the test paper, respectively, while DE denotes the difficulty coefficient of the expected test paper and TDE denotes the differentiation of the expected test paper. The value of α₁ is assigned 0.5, the value of α₂ is assigned 0.3, and the value of α₃ is assigned 0.2. Observation shows that the greater the knowledge point coverage R, the closer the values of paper difficulty coefficient DE and paper differentiation TD are to their respective expected values, the greater the value of the fitness function.

3.2.3

Genetic operator design

Since the selection operation and the population individual optimization operation in this algorithm have already been described [19], this section only focuses on the design of the crossover and mutation operations for the group volume problem in detail. 1)

Adaptive crossover operation

Since the chromosome encoding method of group paper problems is the encoding method of real segments, this paper adopts the method of two-point crossover based on the same question type. The specific idea is: firstly, select the parent individual A and the parent individual B from the population, secondly, set the crossover point on the same type of question of two individuals, such as the parent individual A set three multiple choice questions, then the parent individual B also set three in the same type of question, and finally, carry out the crossover operation to exchange the genes between the crossover points of the two parent individuals, if there is a duplication of the question ID after the exchange, then the crossover is unsuccessful. If there is a duplicate question ID after the exchange, then the crossover is not successful, and the test questions with duplicate IDs need to be replaced, and the replaced test questions should be guaranteed to be of the same question type and have the same knowledge points as the previous test questions.

Before performing the crossover, the r ∈ (0,1) is set using a random function, and this paper uses adaptive dynamic adjustment to make changes to the crossover probability P_c. If r < P_c, the crossover operation is performed to exchange part of the gene segments to form a new offspring chromosome individual, otherwise the crossover operation is not performed.

2)

Adaptive mutation operation

According to the real number segment coding method, this paper adopts the way of segment mutation based on the same question type to perform the mutation operation. The specific idea is: before executing the mutation, generate random number r ∈(0,1) in advance, and at the same time, utilize the adaptive way to dynamically calculate the crossover probability P_m, which has been described in Chapter 3. If r < P_m, the mutation operation is performed, and the test ID under a certain question type is changed so as to form a new individual: otherwise, the mutation operation is not performed. The specific mutation operation is as follows:

Assuming that a chromosome individual (a test paper) contains M segments of genes, that is, there are M types of questions, and these segments contain N types of genes, that is, there are N test questions. First, set up two sets, the first is the gene segment set R = [0,M – 1],then R_i(i ≤ M – 1) is denoted as the i th gene segment in the set of individual papers, i.e., the i th question type.

The second is the set of test questions within the gene segment K = [0, N – 1], where K_j(j ≤ N – 1) is denoted as the j gene (ID of the test question) in gene segment i (question type i).

Next, an integer k is arbitrarily chosen from set R according to a random function, and if k ≤ M – 1, the mutation will be performed within the R_k gene segment, followed by a randomly chosen integer q from set K, and if q ≤ N – 1, then the mutation will be performed on the gene at position q of the k gene segment.

A final judgment is made, and if the mutated test ID is duplicated with other test IDs within the gene segment, then a new test question is selected that is consistent with the knowledge of the original question and is not duplicated with any other IDs within the segment, otherwise no further selection is required.

3.3

Solution flow of the intelligent grouping algorithm

The flow of the intelligent paper grouping algorithm based on the improved genetic algorithm is shown in Figure 3, and the specific execution steps are as follows: 1)

Enter the specific parameters of the intelligent paper grouping algorithm, such as the question type constraints, the total number of questions and the difficulty coefficient of the test paper.

2)

Take the ID of the test questions as the gene and encode real segments according to the question types, thus forming a gene segment, i.e., an individual test paper.

3)

Population initialization, generating a population of size N, while setting the initial iteration number G to zero.

4)

Calculate the fitness value of the test paper represented by each individual in the population based on the fitness function.

5)

A roulette wheel is used to perform selection operations until a new population of number N is generated.

6)

Adaptive crossover operations on individuals from the parents of a new population using two-point crossover based on the same question.

7)

Adaptive mutation operations are performed on individuals in the population by adopting a segmented mutation approach based on the same question types.

8)

Perform population optimization operations to calculate the probability of acceptance of individuals in the test paper, thus deciding whether the poorer individuals in the population will enter the next generation or not.

9)

Determine whether the number of iterations G reaches the maximum number of iterations gen, if it does, the algorithm ends and outputs the optimal solution result, otherwise G = G + 1 and jump to step 4) to start a new round of iterations.

4

Validation of the effectiveness of the improved algorithm

In this subsection, the improved adaptive algorithm will be used to conduct grouping experiments and compare the grouping results with those of GA and AGA, so as to compare and analyze the grouping results of different accounting resource grouping algorithms from multiple perspectives, and verify the effectiveness and superiority of the improved algorithms from multiple aspects.

4.1

Parameterization

In order to ensure the fairness of the algorithm validation, the validation experiments will be carried out in the same equipment and environment, and the parameter information of the system is shown in Table 1, the parameters include hardware parameters and software parameters. To avoid the influence of other factors, other processes running on the device will be cleared before the start of each validation experiment.

Table 1.

System parameter information table

Classification	Entry	Concretely
Hardware parameter	Operating system	Windows10
	Running memory	10G
	processor	i7-7500U
Software parameter	Running language	Python3
Software parameter	Running tool	Visual Studio Code

In this validation, 4 test papers will be used as the constraints for grouping papers, and the grouping effect of the algorithm will be observed through several comparisons, and the information of the test papers is shown in Table 2. The data corresponding to the difficulty level field in Table 2 is the number of test questions. Because the algorithm uses real-number encoding, there is no need to specify the type of test questions. The other fixed parameters of the algorithm are specified as follows: maximum number of iterations 400, population size 200, and fitness expectation 0.98. Traditional genetic algorithms need to specify the crossover probability Pc the variance probability Pm, and according to the existing arguments, specifying the crossover probability Pc as 0.6 and the variance probability Pm as 0.001.

Table 2.

Examination paper information sheet

Serial number	Simplicity	Simpler	Medium	Harder	Difficulty	Expectation difficulty
1	6	5	10	4	5	48.8
2	6	5	10	4	5	49.5
3	6	15	18	3	3	48.8
4	6	15	18	3	3	49.5

4.2

Comparison of results

The group paper results of the GA algorithm are shown in Table 3. From the multi-group grouping results, the average grouping time of the GA algorithm is 250ms, and the actual difficulty of the results deviates somewhat from the expected difficulty, with an average accuracy rate of 87.8%.

Table 3.

GA group volume information

Serial number	Iteration number	Group time/ms	Expectation difficulty	Actual difficulty	Accuracy rate
1	300	235	48.8	42.5	87.1%
2	300	257	49.5	43.4	87.7%
3	300	256	48.8	44.1	90.4%
4	300	252	49.5	42.6	86.1%

The grouping results of the AGA algorithm are shown in Table 4, the average grouping time of the algorithm is 260ms, which is slower than GA, but the average accuracy reaches 90.35%, which is 2.55% higher than GA.

Table 4.

AGA group volume information

Serial number	Iteration number	Group time/ms	Expectation difficulty	Actual difficulty	Accuracy rate
1	300	254	48.8	44.5	91.2%
2	300	258	49.5	44.4	89.7%
3	300	265	48.8	44.1	90.4%
4	300	263	49.5	44.6	90.1%

The grouping results of the improved algorithm of this paper are shown in Table 5, the average grouping time of the algorithm is 265ms, which is slower than AGA, and the average accuracy is 94.05%, which is 3.7% higher than AGA. From the grouping results of the three algorithms, GA, AGA, and this paper’s algorithm, it can be seen that the average accuracy of GA is 87.8%, AGA is 90.35%, and this paper is 94.2%. The average accuracy of the improved algorithm of this paper is 6.25% higher than GA and 3.7% higher than AGA. Although the improved algorithm in this paper is the slowest in terms of average grouping time, the grouping time of 265ms is within the acceptable range and does not have a great impact in real grouping scenarios. Through the above comparative analysis, it can be seen that the grouping results of the improved algorithm in this paper have been improved compared to GA and AGA, which are closer to the expectation.

Table 5.

This algorithm group volume information

Serial number	Iteration number	Group time/ms	Expectation difficulty	Actual difficulty	Accuracy rate
1	300	260	48.8	45.8	93.9%
2	300	268	49.5	46.4	93.7%
3	300	264	48.8	46.9	96.1%
4	300	268	49.5	45.8	92.5%

Figure 4 shows the change of population mean fitness of GA and AGA group volumes.GA algorithm has a better performance in the early stage, the average fitness increases rapidly, but in the late stage of the performance is worse, the convergence results gradually appear deviation, and the final population mean fitness value of GA algorithm is 0.47.We can see that the AGA algorithm does not converge to a certain value in the early stage of the algorithm, but jumps out of the algorithm after converging to a certain value. Convergence, although in the early part of the algorithm, is not as fast as the GA algorithm, but in the continuous evolution process, the algorithm gradually converges to the global optimum.

Figure 5 shows the average fitness change of AGA and this paper’s algorithm, from the figure we can see that the two initial trend is similar, first rapid convergence and then jump out of the convergence to find the global optimum, and gradually converge in the late stage, but this paper’s convergence is much better, and the final results are more accurate. The final population average fitness value of the AGA algorithm is 0.63, and the final population average fitness value of the algorithm of this paper is 0.75. It can be seen that the average fitness value of the final population result of this paper is 59.57% higher than GA and 19.05% higher than AGA.

The probabilities of crossover and mutation are denoted as Pc and Pm, respectively. The adaptive probabilities of crossover and mutation of the AGA algorithm are shown in Fig. 6.The adaptive crossover mutation probability of the AGA algorithm decreases linearly with the increase in the value of individual fitness, i.e., in the case where the value of individual fitness is greater than the average fitness value, the crossover mutation probability decreases by the same proportion regardless of how close the value of individual fitness approaches the maximum fitness value.

The adaptive crossover and mutation probability of this paper’s algorithm is shown in Fig. 7. This paper’s algorithm distinguishes itself from the AGA algorithm in that it does not decrease significantly at the beginning stage when the individual fitness value is close to the maximum fitness value, but maintains a higher crossover mutation probability to increase the diversity of the population, and decreases rapidly when it is about to approach the maximum fitness value to reduce the individual crossover mutation probability and accelerate the convergence of the algorithm. In summary, the improved algorithm in this paper has better performance in exam grouping scenarios compared to GA and AGA, which provides an improved solution idea for exam grouping scenarios of accounting resources in a cloud environment. At the same time, the improved algorithm can also achieve higher quality grouping through difficulty constraints and knowledge point coverage constraints on top of the basic constraints such as the basic number constraints and total score constraints. Therefore, the improved algorithm does not need to care about the types and subjects of the test questions, and it is suitable for accounting paper grouping scenarios that include constraints such as difficulty and knowledge point coverage, and it still performs well in scenarios with a large amount of data.

5

Evaluation of the quality of accounting teaching resource allocation in the cloud computing environment

5.1

Panel rationalization

In this study, the rationality of the panel is examined by three indicators: question difficulty, expected ability of respondents, and test validity. Question difficulty refers to the average difficulty of the questions for each primary routing path of the panel formed by the algorithm in this paper. The expected ability of the test taker refers to the ability value of the test taker after doing a round of testing, which is calculated by the true value of the ability of the test taker and the question difficulty parameter. Test validity refers to the validity of the empirical study of the paper’s grouping method, and is obtained by correlation of the mean of the testee’s two scores with the last accounting level score (standardized). Table 6 shows the results of the group paper quality analysis index data, twice in this paper grouping method under the test test question usage, TOR, selecting questions consume less time than the traditional test, especially selecting questions consume time, this paper grouping method only consumes 0.05s to complete the accounting paper grouping, while the traditional test grouping to spend about a week, and the panel of reasonable index values are also good, the overall validity reaches the standard. The correlation coefficients are all less than 0.05, indicating that the test reliability of the algorithm presented in this paper also meets the requirements.

Table 6.

Group volume quality analysis index

Test category			Two test mean	Last traditional test
Test quantity			11.56	1:1:1
Content expectation ratio			21	4:5:1
Test overlap			0.25	1
The topic is time-consuming (mean)			0.05s	Over a week
Panel rationality	Problem difficulty	Mean	1.02	1.25
	Problem difficulty	SD	0.93	13.56
	Expected capacity	Mean	1.01	125
	Expected capacity	SD	0.95	7.58
	Correlation coefficient		0.001	0.001

5.2

Assessment of students’ accounting competencies

ABS reflects the degree of absolute deviation between the estimated value of the parameter and the true value, the smaller the value indicates that the grouping model designed based on this paper’s algorithm is more accurate in estimating the respondent’s competence. RMSE reflects the degree of root-mean-square deviation between the estimated value of the competence and the true value of this paper’s algorithm, the smaller the value indicates that the grouping model designed based on this paper’s algorithm is more accurate in estimating the return to truth of the respondent’s competence. RMSD is used to measure the ability of the ability parameter estimates of the intelligent accounting grouping model to repair the true value, which reflects the robustness of the ability value estimates of the intelligent accounting grouping model. Table 7 shows the results of the analysis of the accuracy metrics of the ability estimation of the intelligent grouping model, which ranges from -5 to +5logits for the ability estimation of the accounting level of the respondents by the intelligent accounting grouping model. The average value of the test results of the two accounting level adaptive tests is 1.08logits, which is relatively close to each other, in line with the expectation of taking short intervals to conduct the two tests, and on the other hand, it also reflects that the system’s estimation of the ability of the respondents tends to be more stable. In addition, there is a large gap between the highest value and the lowest value of the two English reading comprehension adaptive test results, indicating that the system designed by the intelligent accounting grouping model has a high degree of differentiation of the English reading comprehension ability level of the testees. In conclusion, it is proved through experiments that the ABS, RMSE, and RMSD of the intelligent accounting grouping model test are all within a reasonable range, and its RMSD is only 0.144, which indicates that the accuracy, return to truth, and robustness of the parameter estimation of the intelligent accounting grouping model are all better, and the quality is stable, which is in line with the design expectations.

Table 7.

Model ability estimation accuracy analysis index

Test category	Two test mean	Last traditional test
Minimum grade	-0.25	51
Average performance	1.08	78
Top score	2.26	100
ABS	0.265	/
RMSE	0.158	/
RMSD	0.144	/

6

Conclusion

This paper takes the intelligent grouping problem as the core of its research, realizes intelligent grouping through the improved genetic algorithm, and further improves the allocation of accounting teaching resources in a cloud computing environment.

Comparing the grouping results of GA, AGA, and the algorithm in this paper, the average accuracy of the improved algorithm in this paper is 6.25% and 3.7% higher than that of GA and AGA, respectively. The average grouping time of the improved algorithm is the slowest, but the grouping time of 265ms will not have a significant impact in the actual grouping scenario. Therefore, the grouping results of the improved algorithm in this paper are overall improved relative to GA and AGA.

The improved algorithm in this paper has better and more accurate convergence results in the iterative process. Its average adaptation value is 59.57% and 19.05% higher than GA and AGA respectively. It indicates that it can be better adapted to the accounting grouping issue.

The amount of test questions, the proportion of content expectations, the test overlap rate, and the time consumed in selecting questions under the two methods of this paper’s paper-grouping method are less than that of the traditional test, and the values of all the indexes of its panel reasonably perform well with the correlation coefficients less than 0.05, which indicates that the test reliability of the paper-grouping of this paper’s improved algorithm is up to the standard.

Besides, the three indicators of ABS, RMSE and RMSD tested by the intelligent accounting group paper model are all within reasonable range, respectively 0.265, 0.158 and 0.144, with low values, which indicates that the accuracy, return to truth, and robustness estimated by the intelligent accounting group paper model are relatively excellent, reaching the expected goal of the design, and can effectively test out the level of students’ accounting.

Idioma:: Inglés

Calendario de la edición:: 1 veces al año
Temas de la revista:: Ciencias de la vida, Ciencias de la vida, otros, Matemáticas, Matemáticas aplicadas, Matemáticas generales, Física, Física, otros

RSS Feed de revista

Research on optimal allocation of accounting teaching resources in cloud computing environment based on genetic algorithm framework design

Qianbao Zheng

He Huang

Publicado en línea: 19 mar 2025

Recibido: 15 nov 2024

Aceptado: 18 feb 2025

DOI: https://doi.org/10.2478/amns-2025-0452

Palabras claveAccounting teaching resources, Genetic algorithm, Fitness function, Intelligent grouping algorithm

© 2025 Qianbao Zheng et al., published by Sciendo

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

Palabras clave
Accounting teaching resources, Genetic algorithm, Fitness function, Intelligent grouping algorithm