CN1588308A - Method for realizing automatically compiling test paper from item pool using improved genetic calculation - Google Patents
Method for realizing automatically compiling test paper from item pool using improved genetic calculation Download PDFInfo
- Publication number
- CN1588308A CN1588308A CN 200410062414 CN200410062414A CN1588308A CN 1588308 A CN1588308 A CN 1588308A CN 200410062414 CN200410062414 CN 200410062414 CN 200410062414 A CN200410062414 A CN 200410062414A CN 1588308 A CN1588308 A CN 1588308A
- Authority
- CN
- China
- Prior art keywords
- paper
- examination question
- value
- tactic
- test paper
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The automatic test paper composing process adopting improved genetic algorithm test question library includes mainly the following steps: making test paper composing policy, selecting test questions, composing test paper automatically, and analysis and evaluation. The present invention improves traditional simple genetic algorithm, maintains the features of simple genetic algorithm, including built-in parallelism, overall optimization and fast convergence speed, and avoids the demerits of early blind search, premature convergence, etc. By means of the improved genetic algorithm, it is possible to compose test paper fast in computer, the test paper has its parameters meeting the requirement of the test paper composing policy, and the test questions are highly random, scientific and reasonable and meet the requirement of network test. The present invention has excellent application foreground.
Description
Technical field
The present invention relates to a kind of test item bank automatic volume group implementation method of improving genetic algorithm that adopts, exactly, relate to a kind ofly in Web education, the method for automatic composition examination paper is provided for the network test system that has a standardized electronics test item bank; And this automatic volume group method is based on the electronics test item bank of standard, self-defining tactic of generating test paper and carries out belonging to the digital information processing field on the basis of improved genetic algorithm at the existing problem of present existing automatic volume group.
Background technology
In the building-up of question bank based on the modern network education skill, the test item bank automatic volume group is the information processing technologies such as data search, optimization of utilization computing machine, selects examination question from test item bank automatically, and test question information is handled, and forms paper at last.Specifically, the automatic volume group problem is exactly the automatic volume group strategy that computing machine is formulated according to man-machine interaction, convert thereof into to selecting attribute (difficulty, discrimination, the suggestion score value etc.) constraint condition of examination question, utilize the information processing capability of computing machine then, according to scientific and reasonable search and disposal route, automatically search out one group of examination question and form paper from test item bank, this part paper should satisfy the indexs such as difficulty, discrimination, reliability and time efficiency that educational measurement is learned.
At present, the main difficult problem of automatic volume group is how to guarantee that the paper that generates can farthest satisfy user's different needs, and has randomness, science, rationality.Especially under the network interactive environment, the user has relatively high expectations to group volume speed, and can search in theory global optimum algorithm may with the sacrifice time as cost, often can not produce a desired effect.Therefore, how selecting efficient a, science, strong algorithm is the key of automatic volume group.
Existing test item bank automatic volume group method has random approach, backtracking method, genetic algorithm or the like.Wherein random approach is simple in structure, and is very fast for the extraction travelling speed of single track topic; But the efficient of this method is not high, and main problem is not only to require the examination question amount of test item bank big, and it is good also will to distribute.Random approach group volume needs the time of search longer usually, and this is insufferable for online exam.Backtracking method is to belong to depth-first algorithm with good conditionsi, and for the simple paper of group volume index, it is higher that group is rolled into power.But, find that when practical application this algorithm will take a large amount of internal memories, program structure is relatively complicated, and chooses examination question shortage randomness, and the group volume time is long.
Genetic algorithm is a kind of computer random optimized Algorithm of superseded selection mode and genetic mechanism of the simulating nature circle survival of the fittest.The survival processes of biotic population is generally followed Darwinian survival of the fittest genetic evolution criterion: the individuality in the colony according to it to the adaptive faculty of environment and selected or superseded by the Nature.The result of evolutionary process is reflected on the individual configurations, and the chromosome that promptly comprises some genes is formed in the structure, and corresponding phenotype and genotypic contact have embodied individual external characteristic and the logical relation between internal mechanism.Chromosome also adapts to large natural environment by the intersection between the individuality, variation.Biological stain body or the chromosome represented with computer mode are exactly a string binary numeral, also cry individual sometimes.Its adaptive faculty is that the some numerical value in corresponding each chromosome is weighed; Chromosomal selection or to eliminate then be according to problem maximizing or the minimum value faced.In order to embody chromosomal adaptive faculty, introduced function one fitness function that to measure each involved chromosome of problem.Fitness function has embodied the survival of the fittest principle in the natural evolution, can determine chromosomal excellent, bad degree by fitness function.For optimization problem, fitness function is exactly an objective function.
The genetic manipulation of genetic algorithm mainly contains three kinds: select (selection), intersection (crossover), variation (mutation).Selection operation is named replicate run again, and it is eliminated or by heredity the next generation according to measured excellent, the bad degree decision of the fitness function value of individuality; The plain mode of interlace operation be with selected two individualities that go out respectively as father and mother's individuality, and both numerical value of partial digits carried out corresponding exchange; The plain mode of mutation operation is the numeral on certain numerical digit that changes in the digital string, is appreciated that certain gene in the chromosome for a change.
In genetic algorithm, not all selecteed chromosome all will carry out interlace operation and/or mutation operation, but carry out with certain probability: the probability P c that takes place that intersects usually compares with the probability P m that variation takes place, the former is than the big several number magnitude of the latter, for example: the crossover probability span is: [0.6,0.95], variation probability span is: [0.001,0.01].The chromosome of population sum is population scale, and it has tangible influence to efficiency of algorithm, and scale is too for a short time to be unfavorable for evolution, and scale too senior general cause the evolution time long.Different problems is had suitable separately population scale, and population scale is 30 to 100 usually.
Genetic algorithm has the characteristics of inherent concurrency, global optimizing and fast convergence rate simultaneously, and these characteristics all are suitable for handling the problem of test item bank automatic volume group.The example that adopts simple generic algorithm to carry out the test item bank automatic volume group below in conjunction with an example describes this algorithm in detail:
At first determine the coding method of computing machine: suppose total total n problem in the exam pool, computing machine is represented with the binary digit string of a n position, form such as F
1F
2F
3... Fn.Wherein Fi has two kinds of values, when i inscribes when selected, and Fi=1, when the i topic is not selected, Fi=0.Like this, a paper can be expressed as the binary digit string of the n position of shape such as 00101...0010.
Then carry out the initialization of colony, concrete grammar is that computer random generates m the binary digit string as above-mentioned form, and this m individual (being chromosome) forms genetic group.In this colony, the numerical digit length of the binary digit string that each is individual is all identical, and promptly each chromosomal gene dosage is identical.
Then colony is begun to carry out genetic manipulation, the operation that colony is selected circularly, intersects, makes a variation just up to satisfying the stop condition of setting, perhaps reaches till the cycle index of maximum.
Wherein selection operation is to use fitness function to calculate each individual fitness value, therefrom select the higher individuality of fitness value (promptly satisfying the individuality of group volume constraint condition preferably) again, be genetic to the next generation after duplicating in proportion, leave out the low individuality of fitness value at last, but will keep the size of population of colony constant.
Interlace operation is a back part that two individualities (being binary digit string) is exchanged each other these two numeric strings according to crossover probability Pc from the beginning of a certain position.With all numeric strings in the colony all randomly in twos the combination after, to numeric string, between 0～1, produce a random number at each, if this number is greater than crossover probability Pc, numeric string is selected an exchange spot at random then for this, the numeric string after this exchange spot is carried out interlace operation; Otherwise maintain the original state.
Mutation operation is that certain selected at random in binary digit string one-bit digital is reversed according to variation probability P m: 1 becomes 0,0 becomes 1.Lick change to some extent in the journey specific to group, promptly when changing a certain position, the another one of Xuan Zeing also reverses at random, shows as a problem of replacing in the current paper.
At last select the highest individuality of fitness value in cycling reaches the colony of stipulated number, just select wherein to satisfy best relatively an examination question combination of tactic of generating test paper, this combination is exactly last paper.
At present, when carrying out the test item bank automatic volume group, use the many of traditional genetic algorithm, because this method can find satisfied result quickly, but also have some problems: can produce early stage blind search, premature convergence, converge on disadvantages such as locally optimal solution, local search ability are not strong, these defective effects the further expansion of this method use.Therefore, how to solve automatic volume group be easy to realize and organize the high contradiction of volume method efficient, group is rolled into the contradiction that power is high and the group volume time is short and paper better satisfies predetermined tactic of generating test paper and the good contradiction of paper examination question randomness has become the research and development problem that the insider is badly in need of solving.
Summary of the invention
In view of this, the purpose of this invention is to provide a kind of test item bank automatic volume group implementation method of improving genetic algorithm that adopts, this method is improved traditional genetic algorithm, can be suitable for the paper of online exam for the automatic generation of the network test system that has the standardization test storehouse more, simultaneously, various contradictions and problem that prior art exists have been solved preferably.
The object of the present invention is achieved like this: a kind of test item bank automatic volume group implementation method of improving genetic algorithm that adopts, and it is characterized in that: this method comprises following operation steps:
(1) work out tactic of generating test paper: the examination question distributed intelligence that provides according to test item bank produces human-computer interaction interface, works out the strategy of a cover automatic volume group by this interface user; Described tactic of generating test paper is exactly standard and the constraint condition to the attribute proposition of all examination questions in the target paper;
(2) choose examination question: choose examination question according to the examination question screening conditions that described tactic of generating test paper forms from test item bank, the property parameters of these selected examination questions will be used to the genetic algorithm automatic volume group;
(3) automatic volume group: form fitness function according to described tactic of generating test paper, then according to each part paper, be the fitness function value of chromosome or individuality and the genetic screening that the genetic algorithm after the improvement is carried out examination question colony, form paper until selecting one group of the highest examination question of fitness function value.
This method also comprises following operation steps:
(4) assay: whether the tactic of generating test paper with reference to working out, satisfy tactic of generating test paper to the paper that has generated and require to provide evaluation, and analyze its examination question distribution and other property parameters; Reference when group was rolled up after described analysis and evaluation were used for.
Work out the automatic volume group strategy in the described step (1) and comprise following operation steps at least:
(11) work out the examination question filtercondition: the following filtercondition including, but not limited to this that candidate's exercise question is set: maximum access times, minimum interval, exercise question purposes, the exercise question audit of candidate's exercise question time shutter apart from the time that makes the test identify and the exercise question correlativity;
(12) work out the tactic of generating test paper attribute: the test paper examination time span of volume face total points and suggestion is set, and with the interval significance level of described these two attributes of integer representation in whole tactic of generating test paper that be [0,10], big more this attribute that shows of numerical value is important more;
(13) work out the examination question distribution occasion: according to the following examination question distribution occasion in the DATA DISTRIBUTION information setting target paper of adding up in the database: exercise question type, knowledge point, exercise question difficulty, the degree of awareness, and the significance level of relevant every attribute including, but not limited to this; And with the significance level of the every attribute in the described examination question distribution occasion of integer representation that is [0,10] of an interval, big more this attribute that shows of numerical value is important more.
When working out the every attribute in the examination question distribution occasion in the described step (13), with the examination question number as chorologic unit, or with the examination question mark as chorologic unit.
The screening conditions that form according to tactic of generating test paper in the described step (2) are chosen examination question and are meant from test item bank: carry out selecting for the first time the operation of examination question earlier from test item bank according to the examination question filtercondition, then, again in the examination question that screens for the first time, property parameters according to the examination question distribution occasion carries out the programmed screening operation, makes the examination question that screens should satisfy every screening conditions simultaneously; These selected examination questions will be used to the genetic algorithm automatic volume group then.
Described step (3) automatic volume group comprises following operation steps at least:
(31) initialization colony: at first design the chromosome coding method in the genetic algorithm: promptly represent paper chromosome with the binary digit string of a n position, promptly individual; This individuality is to choose examination question and a paper that forms from candidate's test item bank of total total n road examination question, and wherein each binary digital two kinds of value 1 or 0 represents that respectively whether its pairing certain examination question choose; According to the property parameters of examination question distribution occasion in the tactic of generating test paper this binary digit string is divided then, promptly rearrange this binary digit string, make binary digit string in each functional block represent all candidates' the examination question set of satisfying certain distribution occasion attribute according to functional block;
(32) calculate fitness value: the constraint condition in the tactic of generating test paper is mapped to fitness function, and, according to value resequences again with each the individual functional value in this fitness function calculating colony; The height of functional value is represented the good and bad degree of this individuality, promptly chromosomal good and bad degree and the degree that meets tactic of generating test paper;
(33) carry out selection operation: select and duplicate to be used to enter follow-on individuality according to ideal adaptation degree functional value, be that the high individuality of fitness function value will be replicated a plurality of next generations of entering, the individuality that the fitness function value is low will be eliminated, and the quantity of colony of future generation should equate with the quantity of previous generation colony;
(34) carry out interlace operation: the combination of in population, all individualities being selected at random earlier in twos, then in earlier stage in search, single-point intersection according to traditional genetic algorithm is carried out overall interlace operation to two individualities in every pair of combination, just on the whole binary digit string of these two individualities, select earlier a point of crossing corresponding to each other at random, again these two individual corresponding all later binary digits of point of crossing are all exchanged, to keep colony's diversity; And in the search later stage, adopt the crossover operator of belt restraining to carry out conditional local genetic manipulation, promptly at first on the whole binary digit string of these two individualities, select a functional block at random, select a point of crossing in this functional block internal random again, at last these two individual corresponding point of crossing are all exchanged to all binary digits at functional block end later on, to improve the local search ability of genetic algorithm, improve simultaneously and find the solution quality and improve speed of convergence;
(35) carry out mutation operation: in search in earlier stage, in whole chromosome or individual scope, select two numerical digits to make a variation at random, to improve colony's diversity as change point with different numerical value; And in the search later stage, at first in whole chromosome or individual scope, select a functional block at random, select two numerical digits to carry out the variation of belt restraining in this functional block internal random then, improve local search ability, and accelerate the speed of convergence in later stage as change point with different numerical value;
(36) judge whether to stop heredity: judge whether to occur satisfying the tactic of generating test paper requirement fully or have the individuality of high fitness value, perhaps whether finished the maximal value that preestablishes genetic evolution algebraically, it is the right maximum times of hereditary recycle ratio, if, then population is evolved and stops, no longer the better paper of search enters step (37); Otherwise, change step (32) over to and continue genetic manipulation;
(37) the highest individuality of output fitness: choosing the individuality that fitness value is the highest in the population is optimized individual, carries out Gray code again, forms paper.
Fitness function in the described step (32) is to take all factors into consideration the volume face total points that is provided with in the step (12) and the test paper examination time span of suggestion, and every attribute of the examination question distribution occasion that is provided with in the step (13) and a definite computing formula:
The fitness function value
A in the formula
1Be the actual examination duration of the corresponding paper of this individuality, A is the suggestion test time length of working out in the tactic of generating test paper; I
ASignificance level value for examination duration in the tactic of generating test paper; B
1Be the total points of the corresponding paper of this individuality, B is the volume face total points requirement of working out in the tactic of generating test paper, I
BSignificance level value for volume face total points in the tactic of generating test paper; All the other relevant characters in the formula are then corresponding with every properties of distributions and significance level value thereof in the corresponding paper distribution occasion of this individuality; Fitness function value F
nNumerical value high more, the degree that the paper of representing this individuality correspondence satisfies tactic of generating test paper is high more.
Selection number of times in the selection operation that carries out in the described step (33) depends on fitness ratio formula: the selecteed probability of individual i
N is a number of groups in the formula, f
iFitness function value for the individual i in this colony; The selecteed Probability p of individual i
SiBe used for reflecting the ideal adaptation degree value summation shared ratio of the fitness value of this individuality i in whole colony, individual fitness value is big more, and its selecteed probability is just high more.
Take the search early stage of different genetic manipulations and the division boundary in search later stage to depend on whether computing machine finds most individual situations that occur homologous genes in same position in population in monitoring population evolutionary process in described step (34) and (35); The standard of dividing boundary is: whether the fluctuation range of the fitness value of the individuality over half above population quantity is less than 10% of these ideal adaptation degree mean values.
In the described step (35), it is whole numeric string length value divided by the merchant's who obtains after 3 integer that the maximum times of change point position operation is selected the early stage in search, and it is the integer of the length value of numeric string in the described functional block divided by the merchant who obtains after 3 that the search later stage is selected the maximum times of change point position operation; Do not select the numerical digit that can make a variation if surpass described maximum times yet, then abandon this operation.
The inventive method is improved traditional simple generic algorithm, not only kept the inherent concurrency that traditional genetic algorithm has, the characteristics of global optimizing and fast convergence rate, and can avoid traditional genetic algorithm can produce early stage blind search preferably, premature convergence, converge on locally optimal solution, defectives such as local search ability is not strong, use this improved genetic method, computing machine can be formed paper fast, and the paper parameters all can satisfy the tactic of generating test paper requirement preferably, the examination question randomness of the paper that generates of automatic volume group repeatedly, scientific and rational index height, can satisfy the requirement of online exam, have good application prospects.
Description of drawings
Fig. 1 adopts the process flow diagram of the test item bank automatic volume group implementation method of improving genetic algorithm for the present invention;
Fig. 2 is the detail flowchart of automatic volume group operation in the test item bank automatic volume group implementation method of the present invention;
Fig. 3 reorganizes an embodiment synoptic diagram of examination question binary digit string according to functional block (being the examination question type) in the operation of initialization colony for the present invention.
Fig. 4 (A), (B) are respectively an embodiment synoptic diagram of the interlace operation that interlace operation that traditional genetic algorithm carries out and the present invention only carry out in functional block on whole chromosome.
Fig. 5 (A), (B) are respectively an embodiment synoptic diagram of the two point mutation operation that two point mutation operation that traditional genetic algorithm carries out and the present invention only carry out in functional block on whole chromosome.
Embodiment
For making the purpose, technical solutions and advantages of the present invention clearer, the present invention is described in further detail below in conjunction with accompanying drawing.
Referring to Fig. 1, the present invention is a kind of test item bank automatic volume group implementation method of improving genetic algorithm that adopts, and mainly comprises following operation steps:
(1) work out tactic of generating test paper: the examination question distributed intelligence that provides according to test item bank produces human-computer interaction interface, works out the strategy of a cover automatic volume group by this interface user; Described tactic of generating test paper is exactly standard and the constraint condition to the attribute proposition of all examination questions in the target paper.
Tactic of generating test paper comprises works out examination question filtercondition, tactic of generating test paper attribute and examination question distribution occasion respectively.Wherein working out the examination question filtercondition is that following content is set: the maximum access times of candidate's exercise question, the minimum interval when candidate's exercise question time shutter, distance made the test, exercise question purposes, exercise question audit sign and exercise question correlativity.Working out the tactic of generating test paper attribute is the suggestion test time length and the volume face total points of working out paper, and the significance level of these two attributes.Work out the examination question distribution occasion and comprise and be provided with that target paper exercise question type distribution, knowledge point distribute, the exercise question difficulty distributes, the degree of awareness distributes, and the significance level of every attribute; Here adopt the significance level of integer representation association attributes in tactic of generating test paper of an interval for [0,10], big more this attribute that shows of numerical value is important more.When working out the every attribute of examination question distribution occasion, can be with the examination question number as chorologic unit, the number of for example working out individual event multiple-choice question in the examination question type distribution is 3 roads; Can it be 10 minutes as the exercise question mark of working out chapter 1 first segment in the distribution of knowledge point also with the examination question mark as chorologic unit.
(2) choose examination question: choose examination question according to the examination question screening conditions that tactic of generating test paper forms from test item bank, the property parameters of these selected examination questions will be used to the genetic algorithm automatic volume group.This step is to form screening conditions according to the knowledge point properties of distributions in examination question filtercondition of working out in the abovementioned steps and the examination question distribution occasion, wherein the examination question filtercondition is a first step screening conditions, the maximum access times of for example working out candidate's exercise question in the tactic of generating test paper are 3, and computing machine only can be chosen access times and is used for automatic volume group less than 3 examination question so; The knowledge point properties of distributions is the second step filtercondition, and for example working out the 5th chapter in the tactic of generating test paper is not examination content, and computing machine just can not chosen the examination question of the 5th chapter as automatic volume group so.In a word, screen the examination question that is used for automatic volume group at last and must satisfy every screening conditions simultaneously.
(3) automatic volume group: form fitness function according to described tactic of generating test paper, then according to each part paper, be the fitness function value of chromosome or individuality and the genetic screening that the genetic algorithm after the improvement is carried out examination question colony, form paper until selecting one group of the highest examination question of fitness function value.
Referring to Fig. 2,, specifically introduce its included operations below in conjunction with accompanying drawing because this step is key operation of the present invention.
(31) initialization colony: the at first chromosomal coding method in the needs of the automatic volume group design genetic algorithm according to the present invention.Suppose the always total n road of candidate's examination question, just represent paper chromosome with the binary digit string of a n position, promptly individual; This individuality is to choose examination question and a paper that forms from candidate's test item bank of total total n road examination question, its form such as F
1F
2F
3... Fi...Fn, Fi represents certain examination question in candidate's test item bank in the formula, each examination question has two kinds of values: when this i inscribes when selected, Fi=1, when this i topic is not selected, Fi=0.Like this, a paper can be expressed as the binary digit string of the n position of shape such as 00101...0010, claims paper chromosome again, or individual.It below also is the coding method that paper adopted of traditional genetic algorithm.But, the present invention also will improve this numerical coding string: adopt functional block that this binary digit string is divided, just this binary digit string is divided according to the property parameters (for example examination question type distribution attribute) of examination question distribution occasion in the tactic of generating test paper, promptly rearrange tissue, make binary digit string in each functional block represent all candidates' the examination question set of satisfying certain distribution occasion attribute by the examination question type.For example shown in Figure 3, wherein the examination question in 010100 all is a multiple-choice question, belongs to functional block 1; Thereafter 100011 in examination question all be the topic of filling a vacancy, belong to functional block 2.
The initialized conventional method of colony is the binary coding string that generates m above-mentioned form at random, forms genetic group by each and every one body of this m.In the present invention, just be absorbed in roaming state at random for fear of the search beginning, colony's initialization operation is improved: in adopting traditional chromosome individuality that generates at random, being distributed with selectively to generate and having the initial population of better individuality according to dissimilar exercise questions in the paper again.For example working out the topic of filling a vacancy in the examination question type distribution condition of tactic of generating test paper needs to select 10 roads, and the value that 10 digit order numbers are set in the functional block of multiple-choice question so at random is 1, and other position is 0.
(32) calculate fitness value: the constraint condition in the tactic of generating test paper is mapped to fitness function, and, according to value resequences again with each the individual functional value in this fitness function calculating colony; The height of functional value is represented the good and bad degree of this individuality, promptly chromosomal good and bad degree and the degree that meets tactic of generating test paper.
Fitness function is to take all factors into consideration the volume face total points of setting and the test paper examination time span of suggestion, and every attribute of examination question distribution occasion and definite computing formula; Be a formula of recommending below:
The fitness function value
A in the formula
1Be the actual examination duration of the corresponding paper of this individuality, A is the suggestion test time length of working out in the tactic of generating test paper; I
ASignificance level value for examination duration in the tactic of generating test paper; B
1Be the total points of the corresponding paper of this individuality, B is the volume face total points requirement of working out in the tactic of generating test paper, I
bSignificance level value for volume face total points in the tactic of generating test paper; All the other relevant characters in the formula all are that every properties of distributions and the significance level thereof in the corresponding paper distribution occasion with this individuality is worth corresponding quantized value similarly; Fitness function value F
nNumerical value high more, the degree that the paper of representing this individuality correspondence satisfies tactic of generating test paper is high more.
(33) carry out selection operation: because fitness function value F
nDetermined chromosomal excellent, bad degree, in order to embody the survival of the fittest principle in the nature evolution, this step is selected and duplicates to be used to enter follow-on individuality according to ideal adaptation degree functional value, be about to a plurality of next generations of entering of individual replicate of fitness function value height (being that performance is excellent), the individuality of fitness function value low (being that performance is bad) will be eliminated, and the quantity of colony of future generation should equate with the quantity of previous generation colony.
Selection number of times in the selection operation depends on fitness ratio formula: the selecteed probability of individual i
N is a number of groups in the formula, f
iFitness function value for the individual i in this colony; The selecteed probability P of individual i
SiThe fitness value shared ratio in the ideal adaptation degree value summation of whole colony that reflects this individuality i, individual fitness value is big more, and its selecteed probability is just high more.
(34) carry out interlace operation: the combination of in population, all individualities being selected at random earlier in twos, then in earlier stage in search, single-point intersection according to traditional genetic algorithm is carried out overall interlace operation to two individualities in every pair of combination, just on the whole binary digit string of these two individualities, select earlier a point of crossing corresponding to each other at random, again these two individual corresponding all later binary digits of point of crossing are all exchanged, to keep colony's diversity; And in the search later stage, adopt the crossover operator of belt restraining to carry out conditional local genetic manipulation, promptly at first on the whole binary digit string of these two individualities, select a functional block at random, select a point of crossing in this functional block internal random again, at last these two individual corresponding point of crossing are all exchanged to all binary digits at functional block end later on, to improve the local search ability of genetic algorithm, improve simultaneously and find the solution quality and improve speed of convergence;
In genetic process, computing machine will be monitored evolution situation of population, mainly monitors two aspects: the one, whether certain individual amount too much in the population, promptly should individuality other individualities are more outstanding relatively, if but duplicate in proportion too much, also can be unfavorable for next step evolution of population.The 2nd, whether occur most individual in the population at the identical gene of same position appearance.The gene here can be understood as several bit digital continuous on whole number of individuals word string, and this phenomenon represents that this gene is better, but not necessarily best.At this moment the evolution in can carrying out among a small circle at this gene is optimized this gene more.If COMPUTER DETECTION is second kind of situation above population occurs, show that population evolved to reasonable degree.Reach better degree in order to make evolution be unlikely stagnation, the present invention is divided into two periods with swap operation in the whole evolutionary process and mutation operation: early stage is adopted different operations respectively with the search later stage in search.Specifically, search point of crossing selection in earlier stage is to select at random on the whole numeric string of every pair of individuality; It at first is to select a functional block at random earlier in every pair of individuality that the point of crossing in later stage is selected, and selects a numerical digit as the point of crossing then on this functional block at random.And the division boundary in search early stage and search later stage depends on whether computing machine finds most individual situations that occur homologous genes in same position in population in monitoring population evolutionary process; The standard of dividing boundary is: this is that the source of the overwhelming majority appropriateness values of these individualities of proof is necessary conditions of same gene less than the 10%-of these ideal adaptation degree mean values to surpass the fluctuation range of fitness value of population quantity individuality over half.
Referring to Fig. 4, figure is the new individual C and the D of the interlace operation generation carried out after the point of crossing of adopting traditional genetic algorithm to select at random on whole chromosome to individual A and B shown in (A).Be that individual A and B are adopted the improved genetic algorithm of the present invention shown in the figure (B): select earlier a functional block at random, select a point of crossing in this functional block internal random then, again this point of crossing after to the new individual E and the F that carry out the interlace operation generation between the functional block end.
(35) carry out mutation operation: in search in earlier stage, in whole chromosome or individual scope, select two numerical digits to make a variation at random, to improve colony's diversity as change point with different numerical value; And in the search later stage, at first in whole chromosome or individual scope, select a functional block at random, select two numerical digits to carry out the variation of belt restraining in this functional block internal random then, improve local search ability, and accelerate the speed of convergence in later stage as change point with different numerical value.
It is to select earlier two numerical digits on whole numeric string at random that search change point is in earlier stage selected, (that is: one is 1 if the numeral of these two numerical digits is different, another is 0) then meet the demands, otherwise select two numerical digits again at random, if selected also not choose for m time, then abandoned current mutation operation; The selection of search later stage change point is to select earlier a functional block at random, on this functional block, select two numerical digits then at random, if these two locational digital differences then meet the demands, otherwise on this functional block, select two numerical digits again at random, m ' is inferior also not to be chosen if selected, and then abandons current mutation operation.
Referring to Fig. 5, figure carries out the new individual G that produces behind the mutation operation after the different change point 1,2 of two numerical value adopting traditional genetic algorithm to select at random to individual A on whole chromosome.Be that individual A is adopted the improved genetic algorithm of the present invention shown in the figure (B): select earlier a functional block at random, select two different change points of numerical value to carry out the new individual H that mutation operation produces in this functional block internal random then.
When carrying out mutation operation, the maximum times (being aforementioned m and m ') of the selection change point position operation that search early stage and search later stage carry out is respectively whole numeric string length value divided by the length value of numeric string in the merchant's who obtains after 3 integer and the functional block integer divided by the merchant who obtains after 3.Do not select the numerical digit that can make a variation if surpass this maximum times yet, then abandon this operation.
(36) judge whether to stop heredity: judge whether to occur satisfying the tactic of generating test paper requirement fully or have the individuality of high fitness value, perhaps whether finished the maximal value that preestablishes genetic evolution algebraically, it is the right maximum times of hereditary recycle ratio, if, then population is evolved and stops, no longer the better paper of search enters step (37); Otherwise, change step (32) over to, continue genetic manipulation.
(37) the highest individuality of output fitness: choosing the individuality that fitness value is the highest in the population is optimized individual, carries out Gray code again, forms paper.
Because the characteristics of genetic algorithm, the paper of using the automatic composition of the improved genetic algorithm of the present invention also not necessarily can satisfy the requirement of tactic of generating test paper fully, therefore after forming paper automatically, computing machine also can be estimated and be analyzed the paper that has generated with reference to the tactic of generating test paper of previous formulation.This also can be described as the 4th operation steps of the present invention:
(4) assay:, whether the paper that has generated is satisfied tactic of generating test paper require to provide evaluation and analyze its examination question distribution and other property parameters with reference to the previous tactic of generating test paper of working out; Reference when group was rolled up after these analyses and evaluation all can be used for.
The present invention has carried out test and has implemented, from test item bank, choose satisfactory different examination question automatic volume group according to described improved genetic algorithm, the effect of test is successful, repeatedly examination question randomness, science and the rational index of the paper of automatic volume group generation are than higher, can satisfy the requirement of online exam, realize the goal of the invention of expection, had good application prospects.
Claims (10)
1, a kind of test item bank automatic volume group implementation method of improving genetic algorithm that adopts, it is characterized in that: this method comprises following operation steps:
(1) work out tactic of generating test paper: the examination question distributed intelligence that provides according to test item bank produces human-computer interaction interface, works out the strategy of a cover automatic volume group by this interface user; Described tactic of generating test paper is exactly standard and the constraint condition to the attribute proposition of all examination questions in the target paper;
(2) choose examination question: choose examination question composition paper according to the examination question screening conditions that described tactic of generating test paper forms from test item bank, the property parameters of the selected examination question on the described paper will be used to the genetic algorithm automatic volume group;
(3) automatic volume group: form fitness function according to described tactic of generating test paper, then according to each part paper, be the fitness function value of chromosome or individuality and the genetic screening that the genetic algorithm after the improvement is carried out examination question colony, form paper until selecting one group of the highest examination question of fitness function value.
2, employing according to claim 1 improves the test item bank automatic volume group implementation method of genetic algorithm, and it is characterized in that: this method also comprises following operation steps:
(4) assay: whether the tactic of generating test paper with reference to working out, satisfy tactic of generating test paper to the paper that has generated and require to provide evaluation, and analyze its examination question distribution and other property parameters; Reference when group was rolled up after described analysis and evaluation were used for.
3, employing according to claim 1 improves the test item bank automatic volume group implementation method of genetic algorithm, it is characterized in that: work out the automatic volume group strategy in the described step (1) and comprise following operation steps at least:
(11) work out the examination question filtercondition: the following filtercondition including, but not limited to this that candidate's exercise question is set: maximum access times, minimum interval, exercise question purposes, the exercise question audit of candidate's exercise question time shutter apart from the time that makes the test identify and the exercise question correlativity;
(12) work out the tactic of generating test paper attribute: the test paper examination time span of volume face total points and suggestion is set, and with the interval significance level of described these two attributes of integer representation in whole tactic of generating test paper that be [0,10], big more this attribute that shows of numerical value is important more;
(13) work out the examination question distribution occasion: according to the following examination question distribution occasion in the DATA DISTRIBUTION information setting target paper of adding up in the database: exercise question type, knowledge point, exercise question difficulty, the degree of awareness, and the significance level of relevant every attribute including, but not limited to this; And with the significance level of the every attribute in the described examination question distribution occasion of integer representation that is [0,10] of an interval, big more this attribute that shows of numerical value is important more.
4, employing according to claim 1 improves the test item bank automatic volume group implementation method of genetic algorithm, it is characterized in that: when working out the every attribute in the examination question distribution occasion in the described step (13), with the examination question number as chorologic unit, or with the examination question mark as chorologic unit.
5, employing according to claim 1 improves the test item bank automatic volume group implementation method of genetic algorithm, it is characterized in that: the screening conditions that form according to tactic of generating test paper in the described step (2) are chosen examination question and are meant from test item bank: carry out selecting for the first time the operation of examination question earlier from test item bank according to the examination question filtercondition, then, again in the examination question that screens for the first time, property parameters according to the examination question distribution occasion carries out the programmed screening operation, makes the examination question that screens should satisfy every screening conditions simultaneously; These selected examination questions will be used to the genetic algorithm automatic volume group then.
6, employing according to claim 1 improves the test item bank automatic volume group implementation method of genetic algorithm, and it is characterized in that: described step (3) automatic volume group comprises following operation steps at least:
(31) initialization colony: at first design the chromosome coding method in the genetic algorithm: promptly represent paper chromosome with the binary digit string of a n position, promptly individual; This individuality is to choose examination question and a paper that forms from candidate's test item bank of total total n road examination question, and wherein each binary digital two kinds of value 1 or 0 represents that respectively whether its pairing certain examination question choose; According to the property parameters of examination question distribution occasion in the tactic of generating test paper this binary digit string is divided then, promptly rearrange this binary digit string, make binary digit string in each functional block represent all candidates' the examination question set of satisfying certain distribution occasion attribute according to functional block;
(32) calculate fitness value: the constraint condition in the tactic of generating test paper is mapped to fitness function, and, according to value resequences again with each the individual functional value in this fitness function calculating colony; The height of functional value is represented the good and bad degree of this individuality, promptly chromosomal good and bad degree and the degree that meets tactic of generating test paper;
(33) carry out selection operation: select and duplicate to be used to enter follow-on individuality according to ideal adaptation degree functional value, be that the high individuality of fitness function value will be replicated a plurality of next generations of entering, the individuality that the fitness function value is low will be eliminated, and the quantity of colony of future generation should equate with the quantity of previous generation colony;
(34) carry out interlace operation: the combination of in population, all individualities being selected at random earlier in twos, then in earlier stage in search, single-point intersection according to traditional genetic algorithm is carried out overall interlace operation to two individualities in every pair of combination, just on the whole binary digit string of these two individualities, select earlier a point of crossing corresponding to each other at random, again these two individual corresponding all later binary digits of point of crossing are all exchanged, to keep colony's diversity; And in the search later stage, adopt the crossover operator of belt restraining to carry out conditional local genetic manipulation, promptly at first on the whole binary digit string of these two individualities, select a functional block at random, select a point of crossing in this functional block internal random again, at last these two individual corresponding point of crossing are all exchanged to all binary digits at functional block end later on, to improve the local search ability of genetic algorithm, improve simultaneously and find the solution quality and improve speed of convergence;
(35) carry out mutation operation: in search in earlier stage, in whole chromosome or individual scope, select two numerical digits to make a variation at random, to improve colony's diversity as change point with different numerical value; And in the search later stage, at first in whole chromosome or individual scope, select a functional block at random, select two numerical digits to carry out the variation of belt restraining in this functional block internal random then, improve local search ability, and accelerate the speed of convergence in later stage as change point with different numerical value;
(36) judge whether to stop heredity: judge whether to occur satisfying the tactic of generating test paper requirement fully or have the individuality of high fitness value, perhaps whether finished the maximal value that preestablishes genetic evolution algebraically, it is the right maximum times of hereditary recycle ratio, if, then population is evolved and stops, no longer the better paper of search enters step (37); Otherwise, change step (32) over to and continue genetic manipulation;
(37) the highest individuality of output fitness: choosing the individuality that fitness value is the highest in the population is optimized individual, carries out Gray code again, forms paper.
7, improve the test item bank automatic volume group implementation method of genetic algorithms according to claim 3 or 6 described employings, it is characterized in that: the fitness function in the described step (32) is to take all factors into consideration the volume face total points that is provided with in the step (12) and the test paper examination time span of suggestion, and every attribute of the examination question distribution occasion that is provided with in the step (13) and a definite computing formula:
The fitness function value
A in the formula
1Be the actual examination duration of the corresponding paper of this individuality, A is the suggestion test time length of working out in the tactic of generating test paper; I
ASignificance level value for examination duration in the tactic of generating test paper; B
1Be the total points of the corresponding paper of this individuality, B is the volume face total points requirement of working out in the tactic of generating test paper, I
BSignificance level value for volume face total points in the tactic of generating test paper; All the other relevant characters in the formula are then corresponding with every properties of distributions and significance level value thereof in the corresponding paper distribution occasion of this individuality; Fitness function value F
nNumerical value high more, the degree that the paper of representing this individuality correspondence satisfies tactic of generating test paper is high more.
8, according to the test item bank automatic volume group implementation method of claim 3 or 6 described employings improvement genetic algorithms, it is characterized in that: the selection number of times in the selection operation that carries out in the described step (33) depends on fitness ratio formula: the selecteed probability of individual i
N is a number of groups in the formula, f
lFitness function value for the individual i in this colony; The selecteed probability P of individual i
SiBe used for reflecting the ideal adaptation degree value summation shared ratio of the fitness value of this individuality i in whole colony, individual fitness value is big more, and its selecteed probability is just high more.
9, employing according to claim 6 improves the test item bank automatic volume group implementation method of genetic algorithm, it is characterized in that: take the search early stage of different genetic manipulations and the division boundary in search later stage to depend on whether computing machine finds in population that the situations of homologous genes appear in most individualities in same position in monitoring population evolutionary process in described step (34) and (35); The standard of dividing boundary is: whether the fluctuation range of the fitness value of the individuality over half above population quantity is less than 10% of these ideal adaptation degree mean values.
10, employing according to claim 6 improves the test item bank automatic volume group implementation method of genetic algorithm, it is characterized in that: in the described step (35), it is whole numeric string length value divided by the merchant's who obtains after 3 integer that the maximum times of change point position operation is selected the early stage in search, and it is the integer of the length value of numeric string in the described functional block divided by the merchant who obtains after 3 that the search later stage is selected the maximum times of change point position operation; Do not select the numerical digit that can make a variation if surpass described maximum times yet, then abandon this operation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200410062414 CN1588308A (en) | 2004-07-02 | 2004-07-02 | Method for realizing automatically compiling test paper from item pool using improved genetic calculation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200410062414 CN1588308A (en) | 2004-07-02 | 2004-07-02 | Method for realizing automatically compiling test paper from item pool using improved genetic calculation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN1588308A true CN1588308A (en) | 2005-03-02 |
Family
ID=34603694
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200410062414 Pending CN1588308A (en) | 2004-07-02 | 2004-07-02 | Method for realizing automatically compiling test paper from item pool using improved genetic calculation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1588308A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102184345A (en) * | 2011-06-27 | 2011-09-14 | 山东地纬计算机软件有限公司 | Test-paper generation method based on genetic algorithm |
| CN103136453A (en) * | 2013-03-25 | 2013-06-05 | 华北电力大学（保定） | Automatic test paper formation method and automatic scoring method of document manipulation subjects |
| CN103136714A (en) * | 2011-12-05 | 2013-06-05 | 北大方正集团有限公司 | Determination method of test question record and device of test question record |
| CN103810150A (en) * | 2012-11-09 | 2014-05-21 | 中国银联股份有限公司 | Automatic relation nestable questionnaire generating method and device |
| CN104021261A (en) * | 2013-02-28 | 2014-09-03 | 国际商业机器公司 | Method and device of processing data in the medical field |
| CN104504627A (en) * | 2014-12-03 | 2015-04-08 | 中建材国际贸易有限公司 | Test paper automatic composing method utilizing genetic algorithm |
| CN106570186A (en) * | 2016-11-11 | 2017-04-19 | 网易（杭州）网络有限公司 | Electronic test paper generation method and apparatus |
| CN106710344A (en) * | 2017-02-13 | 2017-05-24 | 盐城工学院 | Computer application examination system |
| CN106897767A (en) * | 2017-03-03 | 2017-06-27 | 盐城工学院 | Automatic volume group method and device |
| CN109118402A (en) * | 2018-08-31 | 2019-01-01 | 中国人民解放军军事科学院评估论证研究中心 | A kind of army's cadre's on-line study examination assessment system and method |
| CN109241516A (en) * | 2018-10-30 | 2019-01-18 | 辽宁科技大学 | A kind of intelligent Auto-generating Test Paper method based on improved adaptive GA-IAGA |
| CN110599839A (en) * | 2019-10-23 | 2019-12-20 | 济南盈佳科技有限责任公司 | Online examination method and system based on intelligent paper grouping and text analysis review |
| CN111369063A (en) * | 2020-03-06 | 2020-07-03 | 北京易真学思教育科技有限公司 | Test paper model training method, test paper combining method and related device |
| CN113052315A (en) * | 2019-12-26 | 2021-06-29 | 北京易真学思教育科技有限公司 | Volume model training, volume method and device and computer storage medium |
-
2004
- 2004-07-02 CN CN 200410062414 patent/CN1588308A/en active Pending
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102184345A (en) * | 2011-06-27 | 2011-09-14 | 山东地纬计算机软件有限公司 | Test-paper generation method based on genetic algorithm |
| CN103136714A (en) * | 2011-12-05 | 2013-06-05 | 北大方正集团有限公司 | Determination method of test question record and device of test question record |
| CN103810150A (en) * | 2012-11-09 | 2014-05-21 | 中国银联股份有限公司 | Automatic relation nestable questionnaire generating method and device |
| US11004564B2 (en) | 2013-02-28 | 2021-05-11 | International Business Machines Corporation | Method and apparatus for processing medical data |
| US10068668B2 (en) | 2013-02-28 | 2018-09-04 | International Business Machines Corporation | Method and apparatus for processing medical data |
| CN104021261A (en) * | 2013-02-28 | 2014-09-03 | 国际商业机器公司 | Method and device of processing data in the medical field |
| CN103136453A (en) * | 2013-03-25 | 2013-06-05 | 华北电力大学（保定） | Automatic test paper formation method and automatic scoring method of document manipulation subjects |
| CN103136453B (en) * | 2013-03-25 | 2015-09-16 | 华北电力大学（保定） | The automatic volume group method of document function topic and automatic marking method |
| CN104504627A (en) * | 2014-12-03 | 2015-04-08 | 中建材国际贸易有限公司 | Test paper automatic composing method utilizing genetic algorithm |
| CN106570186A (en) * | 2016-11-11 | 2017-04-19 | 网易（杭州）网络有限公司 | Electronic test paper generation method and apparatus |
| CN106710344A (en) * | 2017-02-13 | 2017-05-24 | 盐城工学院 | Computer application examination system |
| CN106897767A (en) * | 2017-03-03 | 2017-06-27 | 盐城工学院 | Automatic volume group method and device |
| CN109118402A (en) * | 2018-08-31 | 2019-01-01 | 中国人民解放军军事科学院评估论证研究中心 | A kind of army's cadre's on-line study examination assessment system and method |
| CN109118402B (en) * | 2018-08-31 | 2021-12-07 | 中国人民解放军军事科学院评估论证研究中心 | Army cadre online learning, examination and evaluation system and method |
| CN109241516A (en) * | 2018-10-30 | 2019-01-18 | 辽宁科技大学 | A kind of intelligent Auto-generating Test Paper method based on improved adaptive GA-IAGA |
| CN110599839A (en) * | 2019-10-23 | 2019-12-20 | 济南盈佳科技有限责任公司 | Online examination method and system based on intelligent paper grouping and text analysis review |
| CN113052315A (en) * | 2019-12-26 | 2021-06-29 | 北京易真学思教育科技有限公司 | Volume model training, volume method and device and computer storage medium |
| CN113052315B (en) * | 2019-12-26 | 2022-05-17 | 北京易真学思教育科技有限公司 | Volume model training, volume method and device and computer storage medium |
| CN111369063A (en) * | 2020-03-06 | 2020-07-03 | 北京易真学思教育科技有限公司 | Test paper model training method, test paper combining method and related device |
| CN111369063B (en) * | 2020-03-06 | 2022-08-09 | 北京易真学思教育科技有限公司 | Test paper model training method, test paper combining method and related device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN1588308A (en) | Method for realizing automatically compiling test paper from item pool using improved genetic calculation | |
| CN112488315B (en) | Batch scheduling optimization method based on deep reinforcement learning and genetic algorithm | |
| CN105825270A (en) | Automatic test paper generation method for test questions in question bank through genetic algorithm | |
| CN106599995A (en) | Method for predicting ship navigation meteorological factors according to marine meteorological information | |
| WO2021088436A1 (en) | Project scheduling method and apparatus for multi-language collaborative development | |
| CN108510050A (en) | It is a kind of based on shuffling the feature selection approach to leapfrog | |
| CN107122890A (en) | Personal credit score optimizing evaluation method | |
| CN111967721A (en) | Comprehensive energy system greening level evaluation method and system | |
| CN110390360A (en) | Method and apparatus are determined based on the rectangular block stock layout sequence of genetic algorithm | |
| CN103440525A (en) | Urban lake and reservoir water bloom emergency treatment multiple-target multiple-layer decision-making method based on Vague value similarity measurement improved algorithm | |
| CN103164631B (en) | A kind of intelligent coordinate expression gene analyser | |
| Xu et al. | Strategic diagnosis of China’s modern coal-to-chemical industry using an integrated SWOT-MCDM framework | |
| Fei et al. | LTPConstraint: a transfer learning based end-to-end method for RNA secondary structure prediction | |
| Guanqi et al. | Evolutionary parallel local search for function optimization | |
| CN114217580B (en) | Functional fiber production scheduling method based on improved differential evolution algorithm | |
| Lu et al. | Research on the allocation of water resources engineering projects based on multi-objective optimization | |
| CN1460966A (en) | Computer performance index quantification method based on fuzzy logic | |
| CN114742173A (en) | Transformer fault diagnosis method and system based on neural network | |
| CN115600913A (en) | Main data identification method for intelligent mine | |
| Wang et al. | Air quality data analysis and forecasting platform based on big data | |
| Wang | Evaluation of the Practical Effects of Environmental Measures in the Conservation of Architectural Heritage in Yan’an Based on Recurrent Neural Networks | |
| CN110738398A (en) | trial risk analysis method based on fuzzy number overall utility | |
| Jia et al. | Optimization of mine ventilation network feature graph | |
| Li | A construction of online teaching quality evaluation model based on big data mining | |
| CA2494799A1 (en) | Method for clustering decision trees in data classifiers |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C12 | Rejection of a patent application after its publication | ||
| RJ01 | Rejection of invention patent application after publication |