CN102184345A - Test-paper generation method based on genetic algorithm - Google Patents

Abstract

The invention discloses a test-paper generation method based on a genetic algorithm. The method has functions of automatically extracting test questions and generating test paper. By using the test-paper generation method based on the genetic algorithm, the test paper is generated, so the electronization and automation of test-paper generating operation can be realized, burdens on a teacher are reduced, the teaching efficiency of the teacher is improved, and teaching quality is evaluated truly and objectively; and by the test-paper generation method based on the genetic algorithm, indexes such as test type, difficulty, discrimination and knowledge point distribution can be balanced relatively, user requirements can be met to the largest extent, and teaching management is more scientized and standardized.

Description

Group volume method based on genetic algorithm

Technical field

The present invention relates to paper and generate method automatically, in particular, the present invention relates to group volume method based on genetic algorithm.

Background technology

In education sector, examination is the important step in the whole teaching process, and it is that the student is gained knowledge and a kind of evaluation of ability, also is a kind of educational measurement book section of weighing teachers ' teaching effect quality.In examination in the past, there are many problems in group volume mode.Traditional group volume mode mainly relies on the teacher that the direction that stresses of knowledge point is collected and chosen examination question finishing the establishment of paper, required content, standards of grading etc. since etc. the interference of various human factors, lacks general comparability, the reliability of taking an examination and validity are lower.Utilize the computer set volume, can not only save the quality time in classroom, increase work efficiency, and can eliminate the influence of the people's subjective will that makes the test, make the standardization more of taking an examination, the actual effect of more objective, true, comprehensive reflection teaching helps promoting the raising of quality of instruction.

Along with deepening constantly of the content of courses and improving constantly to the teaching requirement, and the enhancing of student knowledge being grasped the examination needs of global level, this just improves gradually to the requirement of paper quality, and everybody can not satisfy the needs of multi-angle high-quality papers such as knowledge point broad covered area, item difficulty is moderate, examination question topic type is evenly distributed for the group volume by making the test in the past.Therefore, need a kind of science, effective, correct algorithm to assist the work of making the test of finishing.

Summary of the invention

Purpose of the present invention based on genetic algorithm, satisfies the constraint condition that the people proposed that makes the test exactly for addressing the above problem as much as possible, and optimum paper is provided.Genetic algorithm is relevant with the standard that the people carries that makes the test, and at this this paper four standards will be set: topic type-examination question number distributes, difficulty-examination question number distributes, on average difficulty, knowledge point-examination question number distribute.

For achieving the above object, the present invention adopts following technical scheme:

The invention has the beneficial effects as follows:

1, have the function of taking out topic group volume automatically, utilize the group volume method based on genetic algorithm to generate paper automatically, electronization and robotization that can the work of realization group volume alleviate teacher's burden, improve teachers ' teaching efficient, truly the objective evaluation quality of instruction;

2, the group volume method based on genetic algorithm can reach relative equilibrium between every indexs such as topic type, difficulty, discrimination and knowledge point distribution, can at utmost satisfy customer requirements, promotes the scientific of teaching management and standardization.

Description of drawings

Fig. 1 is the group volume method flow diagram based on genetic algorithm;

Fig. 2 is fitness value calculation method step figure;

Fig. 3 selects block diagram for paper;

Fig. 4 is paper intersection block diagram;

Fig. 5 is paper variation block diagram.

Embodiment

Below in conjunction with drawings and Examples the present invention is further specified:

The group volume will be picked out satisfactory exercise question exactly and form paper from thousands of roads examination question.Because each examination question all can have a plurality of attributes, finding the solution of Here it is so-called multi-objective problem, because multi-objective problem might not be separated, therefore even traveled through examination questions all in the test item bank, attempted all possible array mode, also also one find a cover to satisfy the paper that user's soft quota requires surely, therefore we should be not stuck with condition in the process of group volume, but should carry out the iteration of limited number of times in line with the principle of optimization aim as far as possible, the paper that obtains is got final product near user's demand as far as possible.

Whether force to meet the demands when strategy index and group volume as shown in table 1

Table 1

Sequence number	Index name	Whether force to satisfy during the group volume
			1	The topic type	Be
2	Outline	Not
			3	Average difficulty	Not
4	The difficulty interval	Not
			5	The examination time	Not
6	The anchor topic	Not

Group volume method based on genetic algorithm may further comprise the steps:

A. generate initial paper set, from test item bank, select the examination question that meets the knowledge point requirement, press chapters and sections, the new table of topic type classification composition;

B. calculate the fitness value of every paper in the paper set;

C. check that whether the paper that is generated satisfies the various constraint conditions that group volume algorithm requires, and then is provided with successfully sign if exist, and turns to step H;

D. select paper to duplicate according to the fitness value of each paper;

E. select two papers at random, carry out interlace operation, examination question corresponding in two papers is exchanged at the per pass examination question in the paper;

F. select a paper at random, carry out the generation that mutation operation (new individual the every of gene strand is made a variation by probability P m, is negate) is finished paper set of new generation concerning the two-value gene strand at per pass examination question in the paper;

G. turn to step B;

H. check successfully sign, if the success of then expression group of success volume, otherwise, the failure of group volume.

The generation of initial paper set comprises following steps in the described steps A:

A1. the policy information that distributes of the topic type that is provided with according to the people that makes the test-examination question number selected topic group volume at random, the examination question of paper was organized and was satisfied topic type distribution requirement fully this moment;

A2. organize the paper set of some.

The fitness value calculation method comprises following steps among the described step B:

Select follow-on individuality according to survival of the fittest principle.When selecting, be selection principle with the fitness, the fitness criterion has embodied the survival of the fittest, the natural law that uncomfortable person eliminates.Choosing of fitness value comprises:

B1. the function distribution value of difficulty in computation-examination question number:

$D*Q\left(x\right)=\frac{\left|\right|\stackrel{\‾}{Q\left(x\right)}-\stackrel{\‾}{Q}\left|\right|}{M_0}=\frac{\left|\right|(q_1-q_1,q_2-q_2,...,q_{n_Q}-q_{n_Q})\left|\right|}{M_0}$

$=\frac{\sqrt{{\mathrm{\Σ}}_{i=1}^{n_Q}{(q_1-q_1)}^2}}{M_0}$

Formula is explained:

The difficulty interval	0.1-0.2	0.2-0.5	0.5 more than	The vector title
					The user sets	2	5	3	Q
Exercise question quantity in the paper	1	6	4	Q’

Each quantitative value in " exercise question quantity in the paper " is that the property calculation according to each examination question is come out in current paper.So the function distribution function of difficulty-examination question number gets final product for the distance of relatively vectorial (Q, Q '), M0 is total number of examination question in the formula.

The function distribution value of B2. average difficulty

$D*T\left(x\right)=\frac{\left|\right|T_0-T_0\left(X\right)\left|\right|}{T_0}=|1-\frac{\left(\frac{{\mathrm{\Σ}}_{i=1}{d}_i*f_i}{{\mathrm{\Σ}}_{i=1}^n{f}_i}\right)}{T_0}|$

Formula is explained: the average difficulty that T0 sets for the user, the average difficulty of T0 ' for coming out according to the difficulty and the fractional computation of paper in the present paper is so the function of average difficulty is distributed as the comparison between T0 and the T0 '.

B3. the function distribution value of calculation knowledge point-examination question number

$D*K\left(x\right)=\frac{\left|\right|\stackrel{\‾}{K\left(X\right)}-\stackrel{\‾}{K}\left|\right|}{M_0}=\frac{\left|\right|(k_1-k_1,k_2-k_2,...,k_{n_k}-k_{n_k})\left|\right|}{M_0}=\frac{\sqrt{{\mathrm{\Σ}}_{i=1}^{n_k}{(k_1-k_1)}^2}}{M_0}$

Formula is explained: the knowledge point distribute with the interval distributional class of difficulty seemingly, finally for the user sets vectorial K, with present each knowledge of paper the distance value of vectorial K ' of distributed quantity composition.

B4. calculating target function

f(X)＝w ₀D*Q(X)+w ₁D*K(X)+w _dD*T(X)

Objective function is made up of 3 parts, and the function that this 3 part is respectively difficulty-examination question number distributes, average difficulty function distributes, the function of knowledge point-examination question number distributes, and each part all has weights Wi separately, W1+W2+W3=1.

B5. calculate fitness function

Fitness function transforms by objective function and comes, and uses fitness function to judge that high more this paper of proof of the fitness of paper is more near target in genetic algorithm.

Paper is selected may further comprise the steps among the described step D:

D1. calculate the fitness value sum sum of all papers;

D2. calculate the selection probability of each paper, P=Fi/sum (Fi is an i paper fitness value);

D3. find out paper and concentrate the selection probability max of all paper maximums;

D4. select the random number A of a 0-max at random;

D5. as if A＜P, this paper is selected

Intersect in the described step e and may further comprise the steps:

E1. set intersection constant Pc (generally more suitable in 0.4-0.6);

E2. produce the random number R 1 of 0-1;

E3. if R1＜=Pc rolls up the examination questions exchange for two.

Variation may further comprise the steps among the described F:

F1. set variation constant Pm (more suitable in the general 0.001-0.01 scope);

F2. between 0-1, produce random number R 2;

F3. as if R2＜=Pm, from the examination question set, get the examination question of identical topic type at random, replace this examination question in this paper.

Claims (6)

1. based on the group volume method of genetic algorithm, it is characterized in that the performing step of this group volume method is as follows:

A. from test item bank, select the examination question that meets the knowledge point requirement, form by chapters and sections, the classification of topic type and newly show and generate initial paper set;

B. calculate the fitness value of every paper in the paper set;

C. check whether the paper that is generated satisfies the various constraint conditions that group volume algorithm requires, if exist successfully sign is set then, if the success of then expression group of success volume, otherwise the failure of group volume turns to step D to carry out;

D. select paper to duplicate according to the fitness value of each paper;

E. select two papers at random, carry out interlace operation, examination question corresponding in two papers is exchanged at the per pass examination question in the paper;

F. select a paper at random, carry out the generation that mutation operation is finished paper set of new generation at per pass examination question in the paper;

G. turn to step B.

2. the group volume method based on genetic algorithm as claimed in claim 1 is characterized in that, the generation of initial paper set comprises following steps in the described steps A:

A1. the policy information that distributes of the topic type that is provided with according to the people that makes the test-examination question number selected topic group volume at random, the examination question of paper was organized and was satisfied topic type distribution requirement fully this moment;

A2. organize the paper set of some.

3. the group volume method based on genetic algorithm as claimed in claim 1 is characterized in that choosing of fitness value comprises among the described step B:

B1. the function distribution value of difficulty in computation-examination question number:

$D*Q\left(x\right)=\frac{\left|\right|\stackrel{\‾}{Q\left(x\right)}-\stackrel{\‾}{Q}\left|\right|}{M_0}=\frac{\left|\right|(q_1-q_1,q_2-q_2,...,q_{n_Q}-q_{n_Q})\left|\right|}{M_0}$

$=\frac{\sqrt{{\mathrm{\Σ}}_{i=1}^{n_Q}{(q_1-q_1)}^2}}{M_0}$

Each quantitative value in " exercise question quantity in the paper " is that the property calculation according to each examination question is come out in current paper, gets final product so the function of difficulty-examination question number is distributed as the distance of comparison vector (Q, Q '), and M0 is total number of examination question in the formula;

The function distribution value of B2. average difficulty

$D*T\left(x\right)=\frac{\left|\right|T_0-T_0\left(X\right)\left|\right|}{T_0}=|1-\frac{\left(\frac{{\mathrm{\Σ}}_{i=1}{d}_i*f_i}{{\mathrm{\Σ}}_{i=1}^n{f}_i}\right)}{T_0}|$

Wherein, n is a paper exercise question sum, f _iBe the score value of i problem, d _iIt is the difficulty value of i problem.The average difficulty that T0 sets for the user, the average difficulty of T0 ' for coming out according to the difficulty and the fractional computation of paper in the present paper, the function of average difficulty is distributed as the comparison between T0 and the T0 ';

B3. the function distribution value of calculation knowledge point-examination question number

$D*K\left(x\right)=\frac{\left|\right|\stackrel{\‾}{K\left(X\right)}-\stackrel{\‾}{K}\left|\right|}{M_0}=\frac{\left|\right|(k_1-k_1,k_2-k_2,...,k_{n_k}-k_{n_k})\left|\right|}{M_0}$

$=\frac{\sqrt{{\mathrm{\Σ}}_{i=1}^{n_k}{(k_1-k_1)}^2}}{M_0}$

The knowledge point distribute with the interval distributional class of difficulty seemingly, finally for the user sets vectorial K, with present each knowledge of paper the distance value of vectorial K ' of distributed quantity composition;

B4. calculating target function

Objective function f (X)=w ₀D*Q (X)+w ₁D*K (X)+w _dD*T (X)

Wherein, w ₁, w ₂..., w _gBe the weight of each index, and $\underset{i=1}{\overset{g}{\mathrm{\Σ}}}{w}_i=1$

Objective function is made up of 3 parts, and the function that this 3 part is respectively difficulty-examination question number distributes, average difficulty function distributes, the function of knowledge point-examination question number distributes, and each part all has weights Wi separately, W1+W2+W3=1;

B5. calculate fitness function

Fitness function transforms by objective function and comes, and uses fitness function to judge that high more this paper of proof of the fitness of paper is more near target in genetic algorithm;

Fitness function:

Obviously exist multiple mode to select C _Mn, C _MnBe an input value, or the maximal value of g (x) in the evolutionary process, or the maximal value of g (x).

4. the group volume method based on genetic algorithm as claimed in claim 1 is characterized in that, paper is selected may further comprise the steps among the described step D:

D1. calculate the fitness value sum sum of all papers;

D2. calculate the selection probability P of each paper, P=Fi/sum, wherein Fi is an i paper fitness value;

D3. find out paper and concentrate the selection probability max of all paper maximums;

D4. select the random number A of a 0-max at random;

D5. as if A＜P, then this paper is selected.

5. the group volume method based on genetic algorithm as claimed in claim 1 is characterized in that cross-over principle is as follows in the described step e:

E1. set intersection constant Pc;

E2. produce the random number R 1 of 0-1;

E3. if R1＜=Pc rolls up the examination questions exchange for two.

6. the group volume method based on genetic algorithm as claimed in claim 1 is characterized in that, the variation principle is as follows in the described step F:

F1. set variation constant Pm;

F2. between 0-1, produce random number R 2;

F3. as if R2＜=Pm, from the examination question set, get the examination question of identical topic type at random, replace this examination question in this paper.

Images