RETRACTED ARTICLE: Research on English teaching system based on artificial intelligence and WBIETS wireless network system

Abstract

The English teaching network system is a distant teaching based on the Web. This teaching method can stimulate students’ interests so that students can acquire knowledge voluntarily, and automatic test paper generation is one of the most important modules in the English teaching network system. This article first briefly introduces the architecture of the wireless sensor network and then gives a wireless sensor network teaching experiment system based on a genetic algorithm. The multiple sensor nodes in the system can form a variety of different topologies, the collected data can be sent to the user terminal through the GSM network, and the user can also control the remote sensor node through the GSM network. This paper first describes the automatic test problems, a constrained multi-objective problem, then the design of the genetic algorithm to improve the test paper and puts forward questions based on an encoding method and based on the difficulty and test points of F fitness function for dynamic adjustment of the parameters in the iterative process. Finally, it is verified by experiments that the test paper made by this method satisfies users’ requests for questions, contents, and scores, and at the same time, it also improves the running efficiency of a random optimization algorithm by 7–17 times.

1 Introduction

With the development of our country’s economy and society, the demand for the quality of talent English is becoming more and more prominent. Since the reform and opening up, the scale of English education in China has been expanding. English education has achieved fruitful results and remarkable results and has cultivated more and more comprehensive talents with high language quality [1]. With the development of the society and the adjustment of the economic structure, the demand for language quality of the society has also changed. The contradiction between this change and the objective of English teaching is beginning to show and become increasingly prominent. In today’s society, the requirements for the language quality of talents are more and more focused on the practical application. The college English test form and teaching mode are emphasized on exam-oriented education, too much emphasis on the teaching of English formal stiff, ignoring the essence of language use and the ultimate goal, thus not better adapt to social development, to promote the role of English proficiency [2]. With the rapid popularization of the Internet, English education has provided extensive development space, combining education with modern information technology, forming a network teaching system [3]. The network teaching system is a new teaching way, which is implemented by the computer and Internet technology in the virtual space. Compared with traditional teaching, it has an irreplaceable advantage. The perfect network teaching system can directly replace the traditional teaching process on the Internet.

The specific contributions of this paper include:

• (1) Propose an online teaching system for automatically generating test papers.

• (2) Introduce the architecture of wireless sensor network, then give a teaching experiment system of wireless sensor network based on genetic algorithm.

• (3) A problem-based coding method is proposed, and based on the difficulty and test points of the F fitness function, the parameters in the iterative process are dynamically adjusted.

• (4) The method of this article is applied to English teaching, and the running efficiency of the stochastic optimization algorithm is improved by 7–17 times.

The rest of this paper is organized as follows. Section 2 discusses the related work, followed by the methods in Section 3. The experiment is discussed in Section 4. Section 5 concludes the paper with summary and future research directions.

2 Related work

The English network teaching system has a great advantage over the traditional educational model, which is shown in the following four aspects: First, the network teaching system is designed from enrollment, teaching, learning, self-test, examination, graduation, and statistics, and the whole process covers all aspects of education and training; second is the use of IE landing, not subject to geographical, time constraints at any time to study, self-evaluation, participation in the examination, and other operations; third is the rational management and use of teaching resources. Fourth is that the students’ study is arranged by themselves, which greatly improves the efficiency of learning [4]. The automatic test paper system is one of the most important modules in the English network teaching system. At present, the core algorithms of the automatic test paper system are divided into three categories: random algorithm automatic test paper, backtracking test method automatic test paper, and artificial intelligence algorithm automatic test paper [5]. Artificial intelligence is a new technological science for researching, developing, simulating, extending, and expanding human intelligence theory, methods, technology, and application system [6]. Artificial intelligence is a branch of computer science. It attempts to understand the essence of the intelligent and produce a new way to human intelligence similar to react with intelligent machines; this domain research includes robot, speech recognition, and image recognition, Natural Language Processing, and expert system [7]. The artificial intelligence algorithm will set up some intelligent search questions based on human in the process of automatic test paper generation, so it has certain guarantee for the two indicators of success rate and examination quality [8].

The principle of cluster analysis is to assign data to different clusters or classes by attributes, so as to achieve the process of data collection and analysis. The process of classification is to use groups to group physical or abstract objects into classes with similar attribute characteristics, then conduct directional analysis of different classes. Cluster analysis is applied to many disciplines such as computer science, mathematics, statistics, biology, economics, and so on as the basis of research. As a research branch of statistics, cluster analysis is mainly focused on the cluster analysis based on distance. Clustering is a process of searching a cluster, which is an unsupervised learning state, and the cluster is also in a hidden state. The main difference between clustering and regular classification is that clustering does not need to pre-design labeled classes or data, but automatically classify and mark categories, which is an exemplary study of an observational learning rather than a classification method. In general, clustering will divide objects into several subclasses that are clearly defined; each object belongs to a unique category. But in the clustering of teaching indicators, because of the non-stationary signal characteristics, a fuzzy clustering algorithm model is introduced to achieve effective clustering analysis. Fuzzy clustering can divide the cluster objects into different categories, but the fact of membership is different from the clustering. In order to solve the influence brought by the change of ability evaluation and avoid the smaller number of subset sets and the tight distance of some clustering centers, the time and space constraints are often used to adjust this kind of clustering.

3 Methods

3.1 Genetic algorithm

Genetic algorithm is a kind of randomized search method that evolves from the evolution rule of the biological world (survival of the fittest, genetic mechanism of survival of the fittest). It was first put forward by Professor J.Holland in the USA in 1975. Its main feature is to operate directly on structural objects, without the restriction of derivation and continuity of functions [9]. It has inherent hidden parallelism and better global optimization ability [10]. The method of probability optimization can automatically obtain and guide the optimized search space, adjust the search direction adaptively, and do not need the determined rules [11]. These properties of genetic algorithm have been widely applied in combinatorial optimization, machine learning, signal processing, adaptive control, and artificial life and are also the key technologies in modern intelligent computing [12, 13]. Genetic algorithm is usually implemented in a computer simulation. For an optimization problem, a number of candidate solutions (called individuals) are represented by abstract representations (called chromosomes) to better solutions [14,15,16]. In each generation, the fitness of the whole population is evaluated, and multiple individuals are randomly selected from the current population (based on their fitness), and new life groups are generated through natural selection and mutation, which become the current population in the next iteration of the algorithm [17,18,19].

The English online test system is a unified test for all candidates. It requires that in every test, takers get the same indexes on each test paper. That is, the difference is controlled in a smaller range. The people of setting questions should first set several constraints on the system, such as examination time, each question score, the type, the difficulty coefficient of each question, the knowledge point of each question, the teaching requirement of each question, and so on. According to the direction of the constraint, the genetic algorithm is used to search for the optimization continuously until a test paper near the set condition is produced. Each computer can get a test paper with higher reliability according to the algorithm. Since the papers are all in accordance with the unified constraints, the test paper at the terminal will be the same, for example, the difficulty closed to but not to repeat the test. It also brings unanimous fairness in the examination. Therefore, in many fields of examination, the genetic test paper algorithm is more commonly used. In the traditional genetic algorithm, it is not for the use of the group so it is necessary to improve the genetic algorithm in the application of the paper. First, the mathematical model of this examination paper is constructed. Nine questions will be extracted in the system (2 listening, 2 vocabulary and grammar, 1 reading comprehension, 1 finished fill, 2 tenses, and 1 writing questions). And each question has 4 attributes (topic, number, value, difficulty), so they can construct a 9 × 4 order target matrix A to represent the test paper structure.

$$ A=\left[\begin{array}{cccc}{a}_{11}& {a}_{12}& {a}_{13}& {a}_{14}\\ {}{a}_{21}& {a}_{22}& {a}_{23}& {a}_{21}\\ {}{a}_{31}& {a}_{32}& {a}_{33}& {a}_{34}\\ {}{a}_{41}& {a}_{42}& {a}_{43}& {a}_{44}\\ {}{a}_{51}& {a}_{52}& {a}_{53}& {a}_{54}\\ {}{a}_{61}& {a}_{62}& {a}_{63}& {a}_{64}\\ {}{a}_{71}& {a}_{72}& {a}_{73}& {a}_{74}\\ {}{a}_{81}& {a}_{82}& {a}_{83}& {a}_{84}\\ {}{a}_{91}& {a}_{92}& {a}_{93}& {a}_{94}\end{array}\right] $$

(1)

in which a₁₁~a₉₁ are the title of the question, a₁₂~a₉₂ are the number of the title, a₁₃~a₉₃ are the score of the test, and a₁₄~a₉₄ are the difficulty of the test. The 4 attributes of each problem represent the constraints of 4 aspects. It points out the direction for people to select excellent examination paper. The constraints that the matrix should satisfy the maximum limit are as follows: the total score of the test paper is bound to $\sum \limits_{i=1}^9{a}_{i3}=100$ and number T type scores $\sum \limits_{i=1}^9{a}_{i3}{t}_i$. The system of hearing problems should meet that every question is 5 points, a total of 10 points. Vocabulary and grammar should meet that every question is 5 points, a total of 10 points. Reading comprehension should meet the total score of 20. The finished fill should meet the total score of 20. The tenses should be 5 points per channel, with a total of 10 points. The writing question should meet the total score of 30. The sequence number of the test question is not repeated: that is, in the matrix A, there cannot be a case of the same two or a few elements in the matrix. Otherwise, it shows that the test is repeated. The difficulty of test paper constraint: $\sum \limits_{i=1}^9{a}_{i3}{a}_{i4}/\sum \limits_{i=1}^9{a}_{13}$. The difficulty of setting up the test paper is 3.5 (the highest degree of difficulty is 5). Although the final examination papers are often difficult to meet every constraint we set, there are three hard conditions to be achieved for the system: the first is that the total score of the exam is 100, the second is that the score of each type is the prescribed value, and the third is that there cannot be the same topic. In the case, all of the three conditions can be reached. If a test paper with a 3.5 degree of difficulty can be found, the test paper will be the optimal solution of the genetic algorithm. If you cannot find a test paper that is just 3.5 of the difficulty coefficient, we have to set the difficulty factor with $3.0\le \sum \limits_{i=1}^9{a}_{i3}{a}_{i4}/\sum \limits_{i=1}^9{a}_{13}\le 4.0$ so that the suboptimal solution can be found. The number of each item and the number of each type of question can be entered by an artificial method in the setting of a test paper and the total score is 100 points. Therefore, each test paper meets the total score constraints and the total score constraints for each class of questions at the beginning of random volume extraction. Then judge whether there is the same question or not, which can be avoided in the cross algorithm so the only thing that needs to be controlled is the difficulty of the test paper. In other words, at the initial stage, each test paper is difficult to find a test paper that is difficult to meet the requirements of the setting by genetic algorithm. In this way, the target function is only set for the difficulty of the test paper. The objective function can be defined as:

$$ f=\sum \limits_{j=1}^4{f}_j{w}_j $$

(2)

in which f_j is the absolute value of the actual value of the property of the j volume of the current test paper and the difference between the attribute constraint values of the j paper volume. w_j is the weight of the difference factor. Now only the difficulty coefficient is considered, then the objective function can be simplified to:

$$ f={f}_4{w}_4 $$

(3)

in which ${f}_{4=}\left|\sum \limits_{i=1}^9{a}_{i3}{a}_{i4}/\sum \limits_{i=1}^9{a}_{13}- ND\right|$ is the absolute error between the average difficulty of generating the test paper and the difficulty constraint of the set test. ND is an artificially set test difficulty constraint. w₄ is the weight of the difference factor of the difficulty coefficient. Generally, the fitness function is designed to be inversely proportional to the objective function. In order to prevent the denominator from being zero, the fitness function can be designed as:

$$ {F}_j=1/\left(1+\sum \limits_{j=1}^4{f}_j{w}_j\right) $$

(4)

In the same way, only the difficulty coefficient can be simplified to:

$$ {F}_4=\left(1+{f}_4{w}_4\right) $$

(5)

It can be seen that the better the F value is, the better the quality of the 1 test paper. When F = 0, the best solution is found which fits all the constraints.

3.2 Establishment of a test paper model

The problem of generating test paper is expressed as a five tuple, which corresponds to the attributes of each question, including subjects, contents, knowledge points, difficulties, etc. [20]. It is a set of finite variables, V = {v₁, v₂, ⋯, v_n}; L corresponds to the range of each attribute, which is a finite number set, L = {l₁, l₂, ⋯, l_n}; S corresponds to the constraint of each attribute and is a set of finite rules, S = {s₁, s₂, ⋯, s_n}; R corresponds to the user’s constraint relation, which is a finite rule set, R = {r₁, r₂, ⋯, r_n}; C indicates that the test questions selected from the test library to meet the various constraints are a sequence of questions, C = {c₁, c₂, ⋯, c_n}. The test paper problem solving is from the questions in the selected group of questions C which meet the L value range of each item attribute V with the constraint conditions of S and R and it is a multi-constraint problem solving, the answer is the approximate solution, and the target state is not the only (Veldkamp B P et al. 2017) [21, 22].

The “difficulty coefficient” reflects the difficulty and difficulty of the test, and the “difficulty coefficient” can also be understood as the “degree of ease factor.” The greater the difficulty coefficient, the easier the test question is, and the smaller the difficulty coefficient, the greater the difficulty of the test. The determination of difficulty level is for screening subjects. Usually, test difficulty is conducive to students’ learning, but a certain degree of difficulty can increase the degree of division, which is very important for a comprehensive understanding and mastering of students’ learning situation. The difficulty of the test is as follows:

$$ p=1-d=1-e/g $$

(6)

in which d represents the scoring rate, p indicates the difficulty of the test, e indicates the average score of the question, and g indicates the full score of the test. From formula (1), it shows that the lower the scoring rate is, the more difficult the question is. According to the previous experience, the difficulty of the test is divided into four levels: difficult problems (d < 0.3), more difficult problems (0.3 ≤ d < 0.6), middle questions (0.6 ≤ d < 0.8), and easy questions (0.8 ≤ d ≤ 1). The average difficulty of the test paper is as follows [26-27]:

$$ ND=\sum \limits_{i=1}^m{p}_i{d}_i/\sum \limits_{i=1}^m{p}_i $$

(7)

in which ND indicates the average difficulty of the test paper; m represents the number of questions in the test paper; i indicates the number of the test questions, i = 0, 1, ⋯, m; p_i represents the difficulty of the i test; and d_i indicates the score of the i test. In this paper, the two-end analysis method is used to calculate the score of the test area:

$$ D=\left({R}_H-{R}_L\right)/n $$

(8)

in which D indicates the score of the test area and R_H expresses the qualified number of the high group, and the top 27% of the examinees are selected as the high group. R_L expressed the qualified number of low groups, and 27% were low groups after selecting candidates.

In the course of the test paper, the average difficulty of the test paper is closely related to the difficulty of the test questions. Since the probability of the selected questions does not depend on the results of other questions, the selection of the questions is random, that is, to extract or not be extracted. Therefore, it can be seen that the extraction method belongs to the random problem extraction event and also conforms to the two-distribution function B(n, p) of the discrete random variable. The two distributions are the repeated sub-independent Bernoulli test. There are only two possible results in each test, and the two results are opposite and independent, which are independent of other test results. The probability of event occurring or not remains unchanged in every independent test, and this series of experiments is called n heavy Bernoulli experiment, which is as follows:

$$ {p}_n(k)=\left({}_k^n\right){p}^k{q}^{n-k} $$

(9)

$$ Q= np $$

(10)

in which k represents the level of difficulty, k = 0, 1, 2, ⋯, n, and n is positive integers. p_n(k) represents the probability of a difficulty level of k; Q indicates the average degree of difficulty in the test paper. In the two-distribution function B(n, p), when the n increases, the probability p{x = k} first increases to the maximum and then decreases. This probability distribution is in line with our expectation of the difficulty of the test paper, which is the characteristic of “the big middle, the two heads.” In the application process, the difficulty of the test is set to 4 level, that is, n = 5, and the probability p is obtained according to formula (5). Then, the value of n, p, k is introduced into formula (4), and the percentage of each difficulty test paper in the test paper p_n(k) is obtained. Then, the p_n(k) number multiplied by the total number of questions can get the number of questions that need to be extracted from each level of difficulty.

3.3 Design of the test paper based on the genetic algorithm

Genetic algorithms cannot directly deal with the parameters of the problem space. They must be converted into chromosomes or individuals in a genetic space consisting of a certain structure. This conversion operation is called the encoding and can also be represented (representation). In this paper, binary coding is selected for the problem of the test paper, and 0 indicates that the test question is not selected, and 1 indicates that the question is selected. A test paper is taken in a test paper as a chromosome. Each item is regarded as a gene. The value of the gene is determined according to the item number of the test question, so the chromosome encoding of a test paper is expressed as(G₁, G₂, G₃, ⋯, G_i, ⋯, G_n),i = 1, 2, ⋯, n. n represents the total number of questions in the test paper, and G_i indicates the title number of the test. In the process of coding, the same type of questions will be put together, such as choice questions, fill in questions, simple answer questions, and so on. In order to ensure that the examination points are different, the codes of each fund are not duplicated. If different types of questions may appear the same test sites, the same number of test questions may appear in genetic coding, because they belong to different types of questions and do not affect the correctness of test papers.

In the initial encoding process, according to the examination of the contents of the selected and the number of all kinds of questions, in the unity of the type of questions not repeated, the principle of senior high school entrance examination was randomly selected from a group of questions in the test, the test on the test type, which meet the chromosome fraction, producing initial population. Because genetic algorithm is different from other heuristic algorithms, it mainly searches for the best solution locally, and the global search ability is poor. Therefore, the initial population is directly related to the speed and quality of the final solution.

A fitness value is specified for each solution (chromosome), which is specified according to the actual approaching degree of the problem solving (so as to approximate the answer to the problem). Do not confuse these “solutions” with the question “answer,” and you can understand it as an answer to the characteristics that the system may need to use. The fitness function of this paper is as follows:

$$ {f}_1=\frac{1}{m}\sum \limits_{i=1}^m{d}_i $$

(11)

in which d_i is the item selected in the chromosome test point; m indicates the number of test questions, i = 1, 2, ⋯, m; f₁ represents the fitness function value, 0 ≤ f₁ ≤ 1; and the smaller the f₁’s fitness value is, the more it meets the requirements of the user. Meet the test sets by Ω which is said to contain the test point. When d_i = 1, d_i does not belong to the set Ω. When d_i = 0, d_i belongs to the set Ω. For this index, the fitness function f is set:

$$ {f}_2=\left| DC- NDXS\right| $$

(12)

in which DC represents the difficulty coefficient of the test paper; NDXS indicates the difficulty coefficient specified by the user; 0 ≤ f₂ ≤ 1, the smaller the f₂’s fitness value is, the more it meets the requirements of the user. Therefore, the overall fitness function of the test paper can be set as follows:

$$ {f}_3={f}_1+{f}_2=\frac{1}{m}\sum \limits_{i=1}^m{d}_i+\left| DC- NDXS\right| $$

(13)

In the initial population, the evaluation value of each chromosome is calculated according to the fitness function. The higher the evaluation value is, the greater the probability of the next iteration is. In order to ensure the diversity of the population, the iterative probability of F(i)/ ∑ F(i) is set in this paper. In the genetic algorithm, the cross probability Pc and the mutation probability Pm are the key links and the performance of the direct genetic algorithm. In the iterative process, the greater the cross probability Pc and the mutation probability Pm, the stronger the ability to produce new chromosomes. The smaller the crossover probability Pc and the mutation probability Pm, the algorithm easily converges the individual quickly and may produce the early ripening phenomenon. On this basis, the cross probability Pc and the mutation probability Pm are set up.

$$ Pc=\left\{\begin{array}{l}{P}_{c1}-\frac{\left({P}_{c1}-{P}_{c2}\right)\left({f}^{\prime }-{f}_{\mathrm{avg}}\right)}{f_{\mathrm{max}}-{f}_{\mathrm{avg}}}\kern1.8em {f}^{\prime}\ge {f}_{\mathrm{avg}}\kern0.4em \\ {}{P}_{c1}\kern12.70001em {f}^{\prime }<{f}_{\mathrm{avg}}\kern0.4em \end{array}\right. $$

(14)

$$ Pm=\left\{\begin{array}{l}{P}_{m1}-\frac{\left({P}_{m1}-{P}_{m2}\right)\left(f-{f}_{\mathrm{avg}}\right)}{f_{\mathrm{max}}-{f}_{\mathrm{avg}}}\kern1.8em f\ge {f}_{\mathrm{avg}}\kern0.4em \\ {}{P}_{m1}\kern12.70001em f<{f}_{\mathrm{avg}}\kern0.4em \end{array}\right. $$

(15)

in which f_max represents the maximum value of individual evaluation in all populations. f_avg represents the average value of individual evaluation values in the current population.  f^′ represents the individual evaluation value of cross operation;  f represents the individual evaluation value of the mutation operation. This article sets P_c1 = 0.9, P_c2 = 0.6, P_m1 = 0.1, and P_m2 = 0.001.

4 Experiment

The efficiency of TFLAS algorithm is verified by simulation experiment. The network topology of the experiment can be seen in Fig. 1, and the wired network nodes are respectively S1, S2, S3 while the router link bandwidth is 15 Mbps. The access point between router and wireless network is 15 Mbps in link bandwidth with delay of 2 ms. The transmission control protocol used for various nodes is CP NewReno; the MAC layer of the wireless network adopts IEFE 802.11b protocol. In the experiment, Droptail, LAS, and TFLAS are respectively tested within 60 s and they are compared to each other in time equities and overall network throughout rate of up/down flow.

Fig. 1

Network topology

As in fact, downlink traffic is always higher than uplink traffic, given there are two downstreams and one downstream at a time; nodes S1 and S2 respectively send TCP downstream to D1/D2; and wireless transmission rate is constantly11 Mbps. D3 sends TCP upstream, and at the beginning, wireless transmission rate is constantly 11 Mbps, but later gradually decreases, 5.5 Mbps, 2 Mbps respectively at 20 s/40 s; thus, a scenario of wireless network is created. The experiment summarizes uplink/downlink flow’s average holding-time ratio in the channel and network throughout rate. The average holding-time ratio of uplink/downlink flows in the channel can be seen in Fig. 2 and network throughout rate in Fig. 3.

Fig. 2

Network throughout each flow’s holding time ratio in the channel

Fig. 3

Each up/down flow’s average holding time ratio in the channel

In order to detect the validity of the genetic algorithm in this paper, the English course is taken as an example to compare the method of this paper with the random optimization algorithm. First, determine the amount of various types of test questions in the specific circumstances, as shown in Table 1. The number of various types of questions in the test paper is determined here, as shown in Table 2. The difficulty of setting up a whole test paper is 0.520, and a total of 4 levels of difficulty, that is, n = 5. According to the average difficulty level of formula (5) that is Q = 0.52 × 5 = 2.6, the difficulty distribution of the topic in the test library is shown in Table 3. The problem difficulty distribution is shown in Fig. 4.

Table 1 The distribution of various types of questions

Table 2 Analog test paper structure

Table 3 The difficulty distribution of question bank

Fig. 4

Problem difficulty distribution

(1) Study the influence of various parameters in genetic algorithm on the convergence of the program. First, fix the total iteration number K to 120, and set the number of iterations to 120. When the number of iterations reaches 120 or the best individual’s target evaluation value is F, the algorithm terminates iteration and outputs the final value.

The number of populations was set to 50, Pc = 0, and Pm = 0, which were run randomly, and the results showed that the chromosomes were not produced to meet the requirements of the users. Because when the crossing probability Pc and mutation probability Pm are all 0, only the selected operation is not optimized in the iterative process, and the search space is too limited, so it is difficult to find the global optimal solution.

The set population is 50, and the values of Pm = 0.001 and Pc are set to 0, 0.2, 0.4, 0.6, 0.8, and 1.6 different values, respectively. It can be seen from Table 4 that when the value of parameter Pc is in the range of 0.2~1, the number of successful test papers and the average evolution algebra two indicators all increase first and then decrease with the increase of Pc.

Table 4 The influence of parameter cross probability on the convergence of the algorithm

The set population is 50, Pc = 0.6, Pm is set to 0.001, 0.01, 0.05, and 0.1 4 different values. The results are shown in Table 5. Figure 1 is a random selection of a set of convergent graphs at different Pm values. From Table 5, it shows that the number of the success of the paper and the two indexes of the average evolutionary algebra show a tendency to increase first and then decrease with the increase of Pm. The influence of parameter crossover probability on algorithm convergence is shown in Figs. 5 and 6.

Table 5 The effect of parameter variation probability on the convergence of the algorithm

Fig. 5

Influence of parameter crossover probability on algorithm convergence

Fig. 6

The convergence graph of the algorithm under different values of Pm

(2) Comparison with random algorithms

According to the results of Tables 4 and 5, the value of Pc is set in the range of 0.6~0.9. The value of Pm is set in the range of 0.001~0.1, and the population size is set to 50. Genetic algorithm and random algorithm are used to simulate the test paper, and the time comparison table that meets the user’s requirements is generated. The results are shown in Table 6. Figure 7 is a convergent contrast diagram of random selection of a set of genetic algorithms and random algorithms.

Table 6 Comparison of two kinds of test paper algorithm

Fig. 7

Two algorithm contrastive convergent graphs

It is clearly evident from Table 6 that the use of genetic algorithms in the process of finding the optimal solution is far less than the random algorithm. Due to the global search in genetic algorithm, the range of search is large. According to fitness value, we choose iterated chromosomes and search for the direction of the best solution, which makes it easier to search for the optimal solution. From the results of the whole experiment, we can set up dynamic adjustment parameters to solve the problem of multi-constrained target effectively, that is, intelligent test paper generation of exam questions. Compared with stochastic optimization algorithm, we can quickly find the optimal solution.

5 Results and discussion

As China continues to open to the outside world, progress of science and technology, improvement of international status, and the urgent need for a large number of professionals proficient in foreign languages, with the acceleration of China’s “four modernizations” process, make our country play a more active role in international affairs. Therefore, learning English well is of great practical significance and far-reaching historical significance for the realization of the above goals. In traditional English learning, examination is a very important link. It is a test of students’ learning outcomes. Teachers need to spend a lot of time and energy on problem solving, remarking, and statistical scores. It not only takes time and effort, but also has many disadvantages. With the combination of computer technology and education, examination as an important means to test teaching quality and students’ comprehensive ability also needs to be reformed. An efficient, fast, and scientific English test management system is urgently needed. The use of computer technology for automatic test paper is a key link of English test management system, intelligent test system mainly refers to the computer according to the test parameters of the examiners appointed, and it is a typical multi-constraint problem to extract the test paper which satisfies the constraints of the test paper from the English test question bank. Aiming at the problem of test paper generation in English, this paper designs an improved genetic algorithm to solve it. Through experimental analysis, the design method of this paper satisfies users’ requirements for questions, contents, and scores.