TW201013549A - Method of estimation of the examinee's ability on the computerized adaptive testing using adaptive network-based fuzzy inference system - Google Patents

Abstract

A method of estimation of the examinee's ability on the computerized adaptive testing using Adaptive Network-based Fuzzy Inference System simulates and generates a first, a second, and a third training data by a computer system based on Item Response Theory. The first training data is based on a situation the examinee's ability is equal to the examinee's answer. The second training data is based on a situation the examinee's ability is not equal to the examinee's answer. The third training data is based on a situation concerning the standard derivation changes. The method is to train with those training data to establish a model of Adaptive Network-based Fuzzy Inference System to test the examinees and improve the accuracy of the test.

Description

201013549 IX. Description of the invention: [Technical field to which the invention pertains] The present invention relates to a method for measuring a subject's ability in the field of digital learning, in particular, an adaptive network fuzzy inference system for evaluating a computerized fitness test The method of subject ability. [Prior Art] In the aspect of students' ability to test and test students, the traditional method of classifying students is good or bad. In order to classify students with scores under the same test questions, all students must complete all the questions. The method must take a lot of time to identify the students. Therefore, if a computerized fitness test is used to provide a suitable question to the subject, the student's ability can be accurately classified by a number of questions that are less than the traditional test. In digital learning, it is an important study to predict how a subject's ability to provide a suitable test in a computerized fitness test to achieve the best test results. For example, Maximum Likelihood Estimation (MLE) and Bayesian Likelihood Estimation (BLE) are proposed to solve the capability assessment and have good results. In the field of digital learning, the "computerized fitness test" is a measure of abilities. The method of abilities is most often based on the Item ResP〇nse Theory (IRT) in modern tests. And applied by IRT

The digital learning system of the exhibition has become more and more popular in recent years. These are based on the IRT theory and are matched with their respective digital learning systems, which have achieved good results. For example, the familiar measurement system is more commonly used as "adaptive network blur 201013549

System ’ ANFIS), whose architecture is based on fuzzy inference systems (Fuzzy

The Inference System (FIS) is the main subject' combined with the feed-forward neural network's supervised learning method, which enables the system to fully utilize the ability to deal with uncertainty and inaccuracy, as well as self-learning organizational ability and inductive reasoning. Referring to Figure 1, in general, the "Adaptive Network Fuzzy Inference System (ANFIS)" architecture generally includes an input layer, a rule layer, a normalization layer, a result inference layer, and an output layer. Floor. Wherein the input layer (the first layer) maps the input variables into the fuzzy set, the set belonging to X is respectively two sub-sets of 1 A and 2 A; the set belonging to y is 1 B and 2 R忐 other I;隹厶•甘„丄^ ^ often uses a bell-shaped function. The rule layer (the second layer) arranges the fuzzy set and the membership function of each input variable after the input layer determines the fuzzy set between the input variables. Pairing of combinations, the round-out value indicates the triggering of the rule ❹ Strong X The normalized layer (the second layer) is the baby obtained from the upper-layer nodes

The Adaptive Network Fuzzy Inference System (ANFIS) must acquire a set of knowledge, ^) if-then ruies, which can usually be divided into words to produce a set of fuzzy (fiizzy multi-lingual information (linguistic 201013549 inf〇rmation) And numerical information; the knowledge provided by the former for the expert is usually obtained through long-term accumulated experience plus a large number of trial errors, which are used to record the input and output pairs of the system operation _ knowledge. The inference system trains the model with the Hybrid Learning Algorithm as the operation mode of the Adaptive Network Fuzzy Inference System (ANFIS). ❹ However, the aforementioned conventional evaluation system It adopts a fixed topic and does not consider the situation in which the subject's answer value differs from its ability incompatibility and ability assessment criteria, and it may be relatively difficult or relatively simple for students of different levels to answer, so it cannot be effective. Predicting the student's abilities. In addition, the 'traditional assessment system only collects a set of training materials' when the group When the answer value of the training material does not match the ability of the test subject, it is impossible to provide better ability to evaluate the accuracy. [Summary of the Invention] ❿ The present invention provides a computerized fitness test for assessing the computerized fitness test. The method of the subject's ability to provide a better ability to assess the accuracy is the primary objective of the invention. To achieve the foregoing object, the technical means utilized by the present invention and the effects achievable by the technical means include There are: - A variety of adaptability network _ inference system to assess computer aptitude test subject ability method includes a simulation training data step, a learning, force model construction steps and - fitness test ability evaluation steps. The simulation training The data step is based on the test-response theory of the computer system to simulate the generation of 201013549. 2=Training data' wherein the first training situation is the answer value and the test I ability (four) combined 'second _ the situation is the answer value and the subject ability The condition of the non-conformity of 'the second group of training data is to consider the standard difference ^ situation' · the learning ability model construction steps will be the above three sets of training materials Line training 'to complete the construction - the tactile network model system model; the fitness test ability evaluation step, the full-featured network _ inference system for the subject to test, to generate an ability estimate, and then complete It is possible to evaluate the operation, thereby having the effect of improving the accuracy of the evaluation capability. [Embodiment] To make the above and other objects, features and advantages of the present invention more apparent, the following is a summary of the present invention. The preferred embodiment is described in detail with reference to the following drawings. The present invention mainly uses a computer system (not shown) to connect at least a database (not shown) as an execution architecture, and adjusts the network with an adaptive network. Ad The Adaptive-Network-Based Fuzzy Inference System (ANFIS) is used as the basis for the new prediction model of learning ability, supplemented by the Item ReSp〇nse Theory (irt) to adapt the test questions to make use of The degree of discrimination, difficulty, guessing, and ability of the subject before answering the test questions in the computerized fitness test are provided as parameters to provide more conformity to the subject. The questions, then by "adaptability network fuzzy inference system (ANFIS)" deduction amount correction capability of the subject, it can provide better ability to assess accuracy. Referring to FIG. 2, the method for evaluating the ability of the computerized fitness test participant 201013549 using the adaptive network fuzzy inference system generally includes a simulation training data step S1, a wind capability model construction step S2, and a suitability. Test ability evaluation step 〇 • The simulated training data step S1 of the preferred embodiment of the present invention is mainly based on the Item Response Theory (τ), which can quantify the student's ability to between 3 and +3. 'And the three-parameter logarithmic pattern of the test bank is tested, the measurement of the student's ability is related to the dystocia, the degree of discrimination, the guessing degree and the correctness of the question. Therefore, the step S1 of the training data is mainly based on The computer system simulates to generate training materials. Please refer to Figure 3, the detailed steps of the process include: 1. Selecting the topic step S11, the computer system preferably automatically selects five questions to cover the full range of topics for the subject. The person answers, expects to accurately measure the initial ability of the subject, and the subsequent process is based on the last i-th order process on the i-th selection question. The ability estimate is used for selection. The title of the selection is the question of selecting the maximum value of the test message information function /, (θΛ) from the question bank preset in the database. The formula is as follows: !Sense=ϋ * =_1 .72ay2(l- c >.) [cj + + sargmax/,(彡)ζ· = 1,2,··., "where / represents the ί·th flow; "represents the total number of processes; J represents the Estimated tester's ability; /, 〆) representative test questions, · information provided on ability J; /T (valley) represents the derivative of corpse, (valley) value at ability 0; representative ability point 201013549 ^ as); % generation ci on behalf of the test questions / on the test questions / answer _ test questions fresh; _ representative table test questions / Wei Cai 4 represents Lang 7 · _ easy degree; the guess. Mining =====Saki (four) rate side = ^ +(l-cr)

4 is a random chaotic value, (4) 4; z• stands for the first, and the second process. Therefore, when the probability is (8) greater than or equal to the random value in time, the simulation answers R (6〇 is correct, its value is 1, and vice versa, its value is 〇. A calculation ability estimation step S13, subject ability estimation The value valley is calculated by the Maximum Likelihood Estimation (MLE) method, and its formula is as follows: S = aigmaxP(Ul)U2i^-, U„ |0)

Argmax]~J^ (6)Ui Q ψ)1^ /=1 argmaxpc,- +(1_C/)_ ,-1.73⁄4 (4)) ·) Strict ί' = 1,2,.·., " ; Represents the total number of processes; ¢ / stands for the first test test response value 'R value of 1 (for correct response) or 〇 (for incorrect response); 0 for student ability value, _3 <0; In this way, the user's ^, ~..., and the maximum value ❼' is the ability estimate 彡. 201013549 A calculation capability estimation standard deviation step S14, mainly to assess whether the standard deviation is less than the threshold value S141, wherein the certificate (the force is the standard deviation of the capability estimation value j) stops the test if its value is less than the set threshold value, otherwise Continuing with the aforementioned selection topic step S11, the formula is as follows: Scale = 1 > _ _ Qing) =

: The calculation capability estimation candidate value step S15 is to calculate the capability estimation value without the gate $, the confidence interval, and assume the 95% trust product: the range of the 0 confidence interval is ΘΛ±1.96><from The force is the standard deviation. At all points of ^, it is regarded as the candidate point of ability estimation, and the problem is ranked by 0.01. The candidate of the ability estimation selects the estimated value of the estimated calculation ability of the candidate leaf from the question bank according to the test message function, and then calculates the standard. Poor, finally, the point where the standard deviation is the smallest in the ability and the interval, that is, the best ability estimation candidate is more detailed. According to the aforementioned simulation training data step S1, the electric person St collects three sets of training materials. The first group training The data is as follows: the emotional data of the Item Resp Surface Theory (IRT) process is also consistent with the ability of the subject. The first group of training is difficult ^ the last process capacity estimate, this time The process identification degree, the current flow shovel force correction process guessing degree, the current process answer value and the current process value). 〃 quantity (that is, the current process capability estimation value, the last process capability estimation group, the training data is answer Inconsistent with the ability of the subject—11 — 201013549 The first section is the estimated value of the silk ability, the second time of the process, the guess of the process, the current stream __, _ this machine The capability correction amount (that is, the estimated value of the current process capability is estimated). The third group of training data is considered as the target: the monthly condition, by selecting the ability point with the smallest standard deviation as the candidate set for other money. Possible value of the ability, the third group of training data, the best ability estimate, the current process, the current process 0 ❹ Γ i this time recorded _ degree, the nuclear record for the dump and the most

CiEi: The most important record of the money course is the estimated value of the last process. In addition, the simulation environment of the computer system is the subject's ability value _3~^: 'every _ (m as the subject's real ability to simulate the answer test) each ability value will be executed a number of -, each - Process (collecting the three _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ In order to infer the capacity correction amount, the quasi-greenness predicted by the domain capability. The learning ability model construction step S2 of the preferred embodiment of the present invention mainly performs the above three sets of solution data to perform the m's inference ability value suppression. The formula is: x(j) = x(il) + Ax(i~l) M/-1) = /(,〇-I), a(〇s6(〇c(ax〇) where i represents ^current m) represents the U-group training data and the second training data of the third process) or the last -12 - 201013549 best ability estimation candidate (third group training data); α (〇 represents the first ^The number of process title discrimination; () represents the difficulty of the z-th process topic; c(7) represents the z-th process problem guess; ^(7) represents the first, the second process Answer value; Δχ (ί_1) on behalf of, the amount of correction-time process capability or the ability to amend the optimum amount. In this way, the "Adjustment Network Fuzzy Inference System (Anfis)" can be further used to simulate / (), the "adaptive network fuzzy inference system ()" architecture contains: ^

Instead of the following input layer (first layer), this layer node inputs the values x(/-l), α(〇, 6(〇 and each divided into three bell-type membership functions, ><0 Then, it is a two-valve membership function, and the clock type of the input value belongs to the function table 1 + Ni~ejNt - — .dJ^i Οχ = μ]·(Νί) = where % e {χ(ζ. - 1), α(ζ·), δ(7), φ·), 〆,·)} ; = 12, is the number of membership functions of Μ; 4, %, / force is the premise; from ... ... bell-shaped Membership function. The parameter, a regular layer (the second layer), this layer has a total of 3χ3χ3><3 rules, which can be multiplied by the membership function of the input layer, and the formula is as follows: ',,, Starting strength 〇2, ρ=ΤΙμΛΝί) Ν(

Bell-shaped (bell-where M e{x(il)W), 6(〇,c(〇,y(i)}; is the number of membership functions of ';7= 1,2,...,162 Shape) membership function. —13 — 201013549 A normalization layer (third layer), this layer normalizes the results of each node of the rule layer, and the output is between 0 and ,, and its formula is as follows: JP 162 iwp Where % is the launching strength, ^ = 1, 2, .. 462. A result inference layer (fourth layer), this layer node is the adaptive node, multiplying the regular parameter section of the normal ^^ layer section by its equation The formula is as follows: 〇A =^pfp= Wp (qpX(i) + ^α(/) + SpbQ) + (pC(jl) + u^y(i) + ^ } where %,4,\4,^, \ is the conclusion part of the parameter; /? = 1,2,,162. An output layer (fifth layer), this layer node is a fixed node, the result inference layer is summed as the output value, the formula is as follows · 162 〇5 = p—l where 5 is the launching strength, = 1, 2, ..., 162. Overall, when the learning ability model is established, step S2 is completed, and finally, composite learning can be further used. The algorithm (hybrid leamingrule) corrects some of the aforementioned parameters Conclusion Part parameters. Adaptive Testing procedure of Example Assessment of the ability of the present invention the preferred embodiment, the main lines were prepared for the subject, which preferably takes the title optional amount of information covering the whole range of capacity ® subject, the maximum fine estimation approximations (Maximum

Likelihood Estimation, MLE) performs measurements of initial ability. Please refer to FIG. 4' The fitness test capability evaluation step S3 includes a selection question step S31, a simulated subject answer step yang, an inference ability value step S%, a calculation capability estimation standard deviation step S34, and a calculation Can 201013549 force estimate candidate value step S35 and other steps. The selection topic step, the simulation subject answering step S32, the calculation ability estimation standard S34 (S34l), and the calculation capability estimation candidate value step milk execution, are the same as the foregoing selection topic step S11, and the simulated subject answer 2 S12 The calculation capability estimation standard deviation step training and calculation capability estimation: the value steps S15 and the like are the same, and will not be described again. In addition, the inference ability value step S33 is mainly based on the “Adjustment ❹ 模糊 模糊 推 推 推 ( AK AK AK AK AK AK AK AK 魏 魏 魏 魏 魏 , , , , , , , , , , 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏 魏The inferred ability value is used as the ability impairment of the present evaluation ability evaluation S3, and the inference capacity value step S33 is ended when the capacity estimation value is set to the threshold value S341. In this way, the test message function can be selected from the question bank to select the topic with the largest amount of information to give the tester's test message function to make the tester's test message and the maximum value, and the sum of the test message amount and the ability estimation standard. ^ Fang Cheng countdown _, so the standard of the ability of the money can be estimated to produce the most accurate estimate of the ability. Ping Wang takes the method of measuring the test in the field of digital field of the present invention, which does have better ability to evaluate the accuracy below. The establishment of the adaptive network model inference system by Matlab (the change of Nais 'And the subject is 1 person, and the normal distribution is used to simulate the true ability of the subject. Assuming the subject's ability value is (_3 ~ + 3): === two = change as chess The proposed test response "For the mother-to-person test, complete the question bank 5 questions or the ability evaluation standard deviation - 15 ~ 201013549 is less than the door broadcast 0.2. Please refer to Figure 5, which is the sum of the test questions function of the selected 50 questions. Curve, when the ability value is ,, the amount of test information is about 11.5' in the ability value range, its predictive ability will be 'accurate. 'The number of simulated subjects is 100, and the average error is counted. Please refer to it in order. Figures 6, 7, 8, 9, 10, 11 and 12 show the comparison of the Maximum Likelihood Estimation (MLE) and the Bayesian Estimate with the ANFIS of the present invention.

Bayesian Likelihood Estimation, BLE). Among them, ANFIS0 1 is less than MLE, BLE and ANFIS0.5 in the subject's ability (1~+1); MLE will be worse than BLE in the beginning, because BLE tends to be assigned to the previous trend, according to Normal distribution, when the subject's ability value is 0, that is, the middle-capacity candidate has the highest distribution ratio, so BLE will be better than MLE's when the subject's ability value is 1 or 1'. The amount is less than the subject's ability value, 〇. So ANIFS needs to do some number of questions, the evaluation effect will be better than MLE; BLE evaluation ability is slightly worse; in addition, when the subject's ability value is 2 or Because the amount of test questions is small, ANIFS needs to do more questions, and the evaluation effect is better than MLE and BLE. When ANIFS0.5 or ——3, ANFIS evaluates the error value more and it is unstable. Do • More questions to stabilize. Therefore, 'using the adaptive network fuzzy inference system (ANpis) to do the simulation of the subject's ability assessment, can provide a good ^ ability speculation effect when the amount of questions is large, and observe the results of the above simulation experiment, if the test questions ^_ 201013549 large amount As the process increases, the standard deviation will gradually decrease, and the range of best ability estimation candidate values will gradually shrink. The data collected by the process will cover more possibilities, so the effectiveness evaluation will be more stable. As described above, the present invention is mainly based on the adaptive network fuzzy inference system (ANFIS) and is based on the Item Response Theory (IRT), and the test questions can be appropriately selected. ❹ Effectiveness to improve the effectiveness of the assessment, and use the computerized fitness test to identify the degree of difficulty, difficulty, guess and the ability of the subject to answer questions before using the adaptive network fuzzy inference system (ANFIS) to infer the test The ability to correct the ability to assess the subject's ability to answer. More importantly, the subject can be simulated to collect three different assessment scenarios. The three sets of training materials are inferred by various combinations of fuzzy rules to improve the accuracy of the evaluation ability. Although the present invention has been disclosed by the above preferred embodiments, it is not intended to limit the present invention, Various changes and modifications to the above-described embodiments are still within the technical scope of the present invention, and the scope of protection of the present invention is defined by the scope of the patent application, without departing from the spirit and scope of the present invention. —17 — 201013549 [Simple description of the diagram] Figure 1: Schematic diagram of the structure of the Adaptive Network Fuzzy Inference System (ANFIS). Figure 2 is a flow chart showing the steps of the method for assessing the competence of a computerized fitness test subject using an adaptive network fuzzy inference system. Figure 3: Detailed flow diagram of step S1 of the method of the present invention for assessing the ability of a computerized fitness test subject using an adaptive network fuzzy inference system. Figure 4: Detailed flow diagram of step S3 of the method of the present invention for assessing the ability of a computerized fitness test subject using an adaptive network fuzzy inference system. Figure 5: Schematic diagram of the summation of the function of the test message of the simulation experiment using the adaptive network fuzzy inference system to evaluate the competence of the computerized fitness test. Figures 6 to 12: Schematic diagram of the results of a simulation experiment conducted by the present invention using an adaptive network fuzzy inference system to evaluate the ability of a computerized fitness test subject. [Main component symbol description] (none) — 18 —

Claims (1)

201013549 X. Patent Application Scope 1. A method for assessing the ability of computerized fitness to test subjects using an adaptive network fuzzy inference system, the steps of which include: ❹ A simulation training data step, based on the test response theory of a computer system to generate three sets of training data, wherein the first set of training data is the answer value and the subject's ability, and the second set of training data is the answer. If the value does not match the ability of the subject, the third group of training conditions is to consider the standard difference movement; a learning ability model construction step, the above three sets of training materials are _trained to complete the construction-adjustment Sexual network fuzzy inference system model; > appropriate) · Student test ability evaluation (4), Hunan's miscellaneous network fuzzy sounding system for the subject to test, in order to generate an ability estimate, and then complete the ability assessment. The method for assessing the ability of a computerized fitness test subject according to the Xiang etymologic network fuzzy inference system described in the patent specification (4), wherein the simulated training data step includes:, selecting a message amount to cover the full range of test subjects The answering step 'the subject's correctness rate is based on the three-parameter large subject ability estimation value. The most accurate calculation is used to estimate the standard deviation step. The evaluation standard deviation is less than the threshold value of the door 19-19113, if the value is less than the set value. The threshold value is stopped, otherwise the selection step is continued; and a calculation capability estimation candidate step is performed, and the capability estimation value is established as a confidence interval, and all points in the interval are regarded as capability estimation candidate points, each will be The capability estimation candidate points select the title from the question bank according to the test message function, and then calculate the standard deviation, and finally the point where the standard deviation is the smallest in the capacity estimation candidate interval is the ability estimation candidate value for the computer. The system collects three sets of training materials. 3. A method for assessing the competence of a computerized fitness test subject using an adaptive network fuzzy inference system as described in item 2 of the scope of the patent application, wherein the selection step uses the following formula: 1, (9) = iiilL Quef = argmax (§) / = 1,2, · · ·,« where 丨 represents the ith process; “represents the total number of processes; represents the estimated ability of the subject; /, (force represents the test ζ· in ability J The information provided on the above; €, (Valley) represents the derivative of the value in the capacity J; Valley (Valley) represents the probability of the test of the ability of the question 6 in the test questions; represents l is (^) ' 〇, · on behalf of The degree of discrimination of the test questions 矣·; the difficulty level of the test questions 代表• represents the test questions/guess. 4. A method for assessing the ability of a computerized fitness test subject using an adaptive network fuzzy inference system as described in item 2 of the scope of the patent application, wherein - 20 - 201013549 -Ua^eb,) Use the following formula: Pf{9) = Cj -η(1 - e um 〇if pm<Ri where is a random random number, 〇<1<1; / represents the /th flow; when the probability is greater than or equal to chaos In the case of a numerical ruler, the simulation answers π(6>) is correct, its value is 1, and vice versa, its value is 0. 5. A method for assessing the competence of a computerized fitness test subject using an adaptive network fuzzy inference system as described in item 2 of the scope of the patent application, wherein the calculation capability estimation step uses the following formula: 6 = argmaxP^, C /2,...,丨0) η sargmaxH 联)ββ(θ 严/=1 = argm 状) (1 严 strict / = 1, 2,..., " ; « represents the total number of processes; [/, · Represents the first test test value, ¢ /, the value of 1 (for correct response) or 0 (for incorrect response); 0 for student ability value, -3^0S3; household, ^/2,.. .%|6〇 is the maximum value of 0, which is the estimated value of the ability. 6. Method for assessing the competence of computerized fitness test subjects using the adaptive network fuzzy inference system as described in item 2 of the patent application scope. , wherein the calculation capability estimation standard deviation step uses the following formula: —21 — 2201013549 η /=1 SE(0^ — where 5 £: (point) is the standard deviation of the capability estimate J. 7. A method for assessing the competence of a computerized fitness test subject using an adaptive network fuzzy inference system as described in paragraphs 1, 2, 3, 4, 5 or 6 of the patent scope, wherein the first set of training materials Is the last process capability estimate, the current process identification, the current process difficulty, the current process guess, the current process answer value and the current process capability correction (ie, the current process capability estimate - the last process Capacity estimate). 、 The method of assessing the competence of computerized fitness test subjects using the adaptive network fuzzy inference system described in Sections 2, 3, 4, 5 or 6 of the patent scope, wherein the second group of training materials is Estimated value of the last process capability, the process identification degree, the process difficulty level, the current process guess degree, the reading (4) answer value and the cost of the money course correction (that is, the current flow weaving power estimate - on the money estimated value). 9. According to the application for full-time circumstance 2...^ or 6 lysue, the method of assessing the monthly fitness of the computerized fitness test using the adaptive network fuzzy inference system, wherein the second group of training materials is the best in the last process. ^ Estimated value, Benton _ degree, cost of the course, this Si guess ί, the current process answer value and the best ability of this process = wide range best ability estimation candidate value - last process best ability estimate ). 201013549 10's method of assessing the ability of computerized fitness test subjects to evaluate the ability of computerized fitness test subjects based on the patent-adjusted network fuzzy inference system described in the patent paradigm, which applies the inference ability value of the learning ability model construction step JC (7) The formula is as follows: -!) = / (x(i - l), a(〇,0(/),c(〇,j;(〇) ❹ ❿ where f stands for the first, the second flow; the outer one) One of the estimated values of the first process (the first set of training data and the second set of training materials) and the last best ability estimation candidate (the third set of training materials) '·呦 represents the i-th process topic Discrimination degree; 咐 represents the difficulty of the process of the heart process, ¢ (/) represents the guess of the process of the third process; (10) the answer value of the generation; the bee 〗 〖 姊 次 次 次 次 次 次 次 次 次 次 次 次The method of assessing the ability of computerized fitness test subjects according to the benefit of the 1G item, and the method of evaluating the ability of the computerized fitness test, the structure of the system is included in the structure of the human body, _; 12 regularization layer, The result inference layer and the output layer. The adaptive network fuzzy push _ _ _ ability evaluation step package mentioned in the first item —Select:=:, in which the subject is to make a step, a deductive ability value = the step of the difference and the calculation ability includes the candidate value step: select the subject step, simulate the subject to answer the step, select the standard deviation The steps of step and calculation capability estimation candidate are the same as those described in the data step 201013549, and the inference ability value step measures the subject ability correction amount by the adaptive network fuzzy inference system. ❹ —24 —