CN114065031A - Personalized learning path recommendation method based on fuzzy cognitive map - Google Patents

Abstract

The invention relates to a personalized learning path recommendation method based on a fuzzy cognitive map, and belongs to the technical field of education technology. The method mainly uses a fuzzy cognitive diagnosis model and a four-parameter logics model to diagnose the test results of the learner, further carries out fine-grained quantitative analysis on the real cognitive state of the learner, constructs a fuzzy cognitive map of the learner, and generates a dynamic personalized learning path for the learner on the basis. The invention applies the calculation method of energy propagation among construction equipment researched by Vergini and the like to the propagation calculation of the cognitive ability of the learner on the fuzzy cognitive map, so that the dynamic change process of the cognitive level of the learner is visualized and accurate, and the personalized learning resource recommendation of the learner is facilitated.

Description

Personalized learning path recommendation method based on fuzzy cognitive map

Technical Field

The invention relates to a personalized learning path recommendation method based on a fuzzy cognitive map, and belongs to the technical field of education technology.

Background

With the rapid development of artificial intelligence technology and network technology, no matter education in schools or adult education outside schools, "online learning" has become an important learning mode and is highly appreciated.

However, online learning is a main autonomous learning mode, and the learning effect feedback is not timely, the learning is lost, and the like. Most of the traditional education measurement methods analyze the learners from average scores, ranking and the like, do not carry out deep-level and fine-grained diagnosis on the knowledge skills of the learners, and are not beneficial to personalized learning.

On the other hand, the unit test is used as a means for testing the learning condition of the learner, the test result of the learner is usually influenced by the physical condition of the learner during the examination and cannot be normally played, and meanwhile, the phenomenon of guessing the test question accidentally occurs. How to eliminate the factors, obtain the real cognitive level of the students and accurately recommend learning resources is a current research hotspot.

Disclosure of Invention

The invention aims to provide a personalized learning path recommendation method based on a fuzzy cognitive map. The method mainly comprises the steps of diagnosing the test results of the learner by using a fuzzy cognitive diagnosis model and a four-parameter logics model, further carrying out fine-grained quantitative analysis on the real cognitive state of the learner, constructing a fuzzy cognitive map of the learner, and generating a dynamic personalized learning path for the learner on the basis.

The technical scheme of the invention is as follows: a personalized learning path recommendation method based on a fuzzy cognitive map comprises the following specific steps:

step 1: combining the knowledge organization structure of the course classical teaching materials and the course field expert knowledge to determine the knowledge point set C of the course { C ═ C₁,C₂，……，C_NAnd determining the relation between the knowledge points. And constructing a course knowledge graph in a top-down mode by means of the project software.

The course classic textbook is a textbook applied by at least one 3-third of the area where the course is located, and the course field experts are teachers which can accurately describe the logic relationship between the knowledge points of the course for more than 20 years and have more than subordinate high-level job titles.

Obtaining the dependency relationship among the course knowledge points and the 'correlation strength' among the course knowledge points by applying a Delphi method through multiple rounds of opinion solicitation, and further constructing a course weight matrix W_N×N(N is the total number of knowledge points in the course), W_mnIs a knowledge point C when the learner learns_mFor knowledge point C_nWhen m influences the strength<n is the point of knowledge C_mIs a knowledge point C_nThe first repair knowledge of (1).

The multi-turn solicitation is to solicit 10 or more opinions of experts in the field of courses through at least three rounds. The association strength refers to the degree of influence of the mastery condition of one knowledge point on the learning of another knowledge point.

Step 2: with reference to the cognitive classification of the objective of bloom education, the learner's cognitive level is divided into five levels: (1) the people can learn about [0.4-0.6 ], (2) basic understanding [0.6-0.7 ], (3) understanding [0.7-0.8 ], (4) mastering [0.8-0.9 ] and (5) skillful mastering [0.9-1.0 ].

Step 3: according to cognitive level hierarchical division, the learning resource hierarchical division corresponding to the Step2 cognitive level hierarchical division level is as follows: (1) achieve basic understanding [0.6-0.7 ], 2, 0.7-0.8, 3, 0.8-0.9 and 4, 0.9-1.0.

Step 4: and testing the student aiming at a certain learning unit of the course to obtain the testing score of the course unit of the student.

Step 5: computing a learner's knowledge competency level based on the test achievements, assuming J is the learner's set, for each knowledge point C_kE is C, set μ_k:J→[0,1]Is fuzzy set membership function, for each learner J belongs to J, defines J at knowledge point C_kLevel of knowledge ability of_k(j) J for student in fuzzy set (J, mu)_k) Wherein 0 is not more than mu_k(j)≤1，μ_k(j) Calculating according to a Logistic model in a project reaction theory:

i.e. the learner's level of knowledge ability mu_k(j) Depending on the learner's high-order underlying trait θ_jAnd for learner j at knowledge point C_kDifficulty b in_jkAnd degree of distinction a_jk. Wherein, theta_j、b_jk、a_jkCalculated from the learner's test performance using the sirt kit of the RStudio Desktop 1.2.5042 software.

Step 6: calculating the mastery degree of the knowledge points of the learner based on the test results, and under the assumption of the joint type (objective questions), the mastery degree eta of the student j on the objective questions (questions) i_jiComprises the following steps:

wherein q is_ikIndicating whether problem i requires knowledge skills, C_k(q _ik1 denotes a point of knowledge C required for problem i_kCan be solved.

Under the assumption of compensatory (subjective questions), the mastery degree eta of knowledge points of the learner j on the subjective question (question) i_jiComprises the following steps:

wherein q is_ikIndicating whether problem i requires knowledge skills, C_k(q _ik1 denotes a point of knowledge C required for problem i_kCan be solved.

Step 7: and estimating the error and guess parameters by using a four-parameter Logistic model, correcting the phenomena of overestimation and underestimation of the learner ability, and calculating the real answering scores of the learner on the objective questions and the subjective questions by using Bernoulli distribution and Gaussian distribution respectively.

Step 8: suppose that in the unit test, knowledge point C_kThrough the ith₁，i₂，……i_hExamination questionsThe test is carried out, and the scores of the test paper are respectively as follows: s_i1，s_i2，……s_ihBased on the real answering score of the learner j on the Step7, the real answering score is calculated according to the fraction proportion of the test questions in the test paper in a weighted mode, and the learner j on the C is obtained_kTrue cognitive level on (c):

step 9: and constructing a fuzzy cognitive map of the learner on the course test unit based on the knowledge map constructed at Step1 and the weight matrix W and in combination with the cognitive level of the student j measured in the unit test at Step 8.

Step 10: all knowledge points that learner j does not reach "proficiency" in the tested course unit are obtained according to the calculation result of Step8, and an ordered set C' is constructed according to the logical relationship.

Secondly according to the weight matrix W_N×NCalculate learner j at each knowledge point C'_m(m-1, 2,3, …, n) is affected by the disturbance D (i.e. the learner has learned the recommended learning resources).

Through multiple rounds, learning resources are continuously recommended for the learner j, the learning effect is predicted, and a learning path is finally formed, but the learner is considered to acquire the knowledge point when the cognitive level is more than or equal to 0.9.

The Step7 is specifically as follows:

step7.1: estimating the misestimation and guess parameters by using a four-parameter Logistic model to correct the phenomena of overestimation and underestimation of the learner's ability

Wherein, a_i，b_i，c_i，d_iRespectively representing discrimination, difficulty, guess parameter and sleep parameter, c_i，d_iAccording to the test result of the learnerRStudio Desktop 1.2.5042 software sirt toolkit.

Step7.2: respectively calculating the real answering scores R of the learner j on the objective question i and the subjective question i by adopting Bernoulli distribution and Gaussian distribution_ji：

P(R_ji＝1|η_ji,d_i,c_i)＝(1-d_i)η_ji+c_i(1-η_ji)

P(R_ji|η_ji,d_i,c_i)＝N(R_ji|[(1-d_i)η_ji+c_i(1-η_ji)],σ²)

It can be seen that the learner's score on the objective question is mapped to a value in the {0,1} set, and the score on the subjective question is mapped to a value in the [0,1] interval.

The Step10 is specifically as follows:

step10.1: all knowledge points of learner j which do not reach ' proficiency ' in the tested course unit are obtained according to the cognitive level calculated by Step8, and an ordered set C ' ═ C ' is constructed according to the logical relationship of the knowledge points '₁，C′₂，……C′_n}。

Step10.2: according to the weight matrix W between the cognitive level and the knowledge point of the learner j_N×NCalculating and predicting each knowledge point C ' in C ' using the following formula '_m(m-1, 2,3, …, n) cognitive level change after disturbance D (i.e. after learner has learned recommended learning resources):

wherein, C'_m ⁽⁰⁾＝0，C_m ⁽¹⁾Is learner j at knowledge point C'_mInitial cognitive level of D_m ^(k)＝C′_m ^(k)–C′_m ^(k-1)，C′_m ^(k)Represents learner j at C 'after k-th disturbance'_mUpper cognitive level.

Thereby countingCalculating and predicting the cognitive level of the knowledge point after the learner j has interfered for the kth time on the C ', and screening out the knowledge concept C ' which has the greatest influence (namely has the most related knowledge points) on the learning of the unit by the learner j '_x，x∈{1,2,3,…,n}：

Step10.3: if C'_x ^(k+1)0.9 or more, removing the knowledge point C ' from the knowledge point set C ' to be learned '_xAnd updating the set and not performing recommendation learning.

Step10.4: if C'_x ^(k+1)< 0.9, judge knowledge Point C 'calculated by Step9.2'_xThe position in the ordered set of knowledge points C' to be learned.

Step10.4.1: if C'_xIs a leaf node in the ordered set C ' of knowledge points to be learned, namely in C ', C '_xHaving no look ahead knowledge, then according to learner at C'_xAnd (4) recommending learning resources corresponding to the cognitive level for the cognitive level. After learning, learner j is at C'_xThe upper cognitive level is preset as an average of an upper limit and a lower limit of the recommended learning resource hierarchy.

Step10.4.2: if C'_xIs a non-leaf node in the ordered set C ' of knowledge points to be learned, namely in C ', C '_xIf the pre-repair knowledge exists, finding out learning knowledge points C 'according to the weight matrix W'_xInfluencing the most advanced prior knowledge point C ″)_x. Then, according to the learner at C ″)_xThe upper level of cognition is that it recommends learning resources corresponding to the level of cognition. After learning, learner j is at C ″)_xThe upper cognitive level is preset as an average of an upper limit and a lower limit of the recommended learning resource hierarchy.

Step10.5: repeating Step10.2-Step10.4 until the learner learns all knowledge points in C', thereby forming a learning resource path.

The invention has the beneficial effects that: the calculation method of energy propagation among the building devices researched by Vergini and the like is applied to the propagation calculation of the cognitive ability of the learner in the fuzzy cognitive map, so that the dynamic change process of the cognitive level of the learner is visualized and accurate, and the personalized learning resource recommendation of the learner is facilitated.

The invention and the commonly used FCM (fuzzy Cognitive maps) function KL_i(t+1)＝(KL_i(t)±w_ji*p_j*KL_i(t)/100, carrying out t test, wherein the result shows that the two have no significant difference, which shows that the method can be used for calculating the cognitive competence of the cognitive map of the learner, but the method better visualizes the change propagation state of the cognitive level of the learner in the learning process.

Drawings

FIG. 1 is a flow chart of the present invention;

FIG. 2 is a "advanced math" course knowledge graph in an embodiment of the present invention;

FIG. 3 is a fuzzy cognitive map of a learner in the "advanced math" lesson in an embodiment of the present invention.

Detailed Description

The invention is further described with reference to the following drawings and detailed description.

On the basis of a fuzzy cognitive diagnosis model, the cognitive level of a learner in objective questions and subjective questions is diagnosed from unit test results of advanced math courses by combining a four-parameter logistic model, and a learner fuzzy cognitive map is generated by combining a course knowledge map. And (4) combining the calculation of the change of the cognitive level of the learner after the interference (recommending learning resources), recommending an individualized learning path most suitable for the learner to learn. The present invention is further explained by taking the cognitive diagnosis analysis of the grade-one students of certain colleges and universities in the "advanced mathematics" course as an example.

Step1, course knowledge graph construction: the knowledge organization structure of the classic teaching materials of the ' advanced mathematics ' course (advanced mathematics ' written by college university (seventh edition), published in 2014) and the knowledge of the course experts are combined to determine the knowledge point set C ═ C of the course₁,C₂,……C_NDetermining the relation between the knowledge points; with the help of the project software, a "higher mathematics" course knowledge graph is constructed in a top-down manner, as shown in fig. 2. Obtaining dependency relationship between course knowledge points and their relation by applying Delphi method through multiple rounds of opinion solicitationThe influence strength between course weight matrixes W_N×N(N is the total number of knowledge points in the course), and as shown in Table 1, is a weight matrix between some knowledge points of higher mathematics (seventh edition of the book) written at college university, W_mnRepresents a knowledge point C_mIs a knowledge point C_nAnd the prior correction of knowledge, and knowledge point C_nTo knowledge point C_mThe influence strength of (a);

	function(s)	Extreme limit	Function derivation	Limit operation	Integral of uncertainty	Fixed integral
							Function(s)	0	0.84	0.86	0.85	0.85	0.85
Extreme limit	0.37	0	0.67	0.90	0.46	0.47
							Function derivation	0.39	0.59	0	0.39	0.90	0.90
Limit operation	0.42	0.56	0.26	0	0.45	0.46
							Integral of uncertainty	0.39	0.31	0.60	0.29	0	0.92
Fixed integral	0.39	0.31	0.60	0.29	0.65	0

Table 1: weight matrix

Step2, cognitive level hierarchy partitioning: with reference to the cognitive classification of the objective of bloom education, the learner's cognitive level is divided into five levels: (1) the identification [0.4-0.6 ], (2) the basic understanding [0.6-0.7 ], (3) the understanding [0.7-0.8 ], (4) the mastery [0.8-0.9 ], (5) the mastery [0.9-1.0 ];

step3, learning resource hierarchy division: according to cognitive level hierarchical division, the learning resource hierarchical division corresponding to the Step2 cognitive level hierarchical division level is as follows: (1) achieving basic understanding [0.6-0.7), (2) achieving understanding [0.7-0.8), (3) achieving mastery [0.8-0.9), (4) achieving skillful mastery [0.9-1.0 ];

the relationship between Step2 and Step3 is shown in table 2.

Cognitive level hierarchy	Recommended learning resource hierarchy
		Memory [0.4-0.6)	Basic understanding [0.6-0.7)
Basic understanding [0.6-0.7)	It is understood that [0.7-0.8),
		it is understood that [0.7-0.8),	master [0.8-0.9)
Master [0.8-0.9)	Is well mastered at 0.9-1.0]
		Is well mastered at 0.9-1.0](knowledge acquisition)

Table 2: relationship between cognitive level hierarchy and recommended learning resource hierarchy

Step4, unit testing of course: aiming at a certain learning unit of the course, organizing a teacher question and group paper, and testing by students to obtain course unit test scores of the students;

step5, knowledge competency level calculation of learner: suppose J is a set of learners, for each knowledge point C_kE is C, set μ_k:J→[0,1]Is a fuzzy set membership function. For each student J ∈ J, defining the knowledge point C of the student J_kLevel of knowledge ability of_k(j) J for student in fuzzy set (J, mu)_k) Wherein 0 is not more than mu_k(j)≤1。μ_k(j) Calculating according to a Logistic model in a project reaction theory, wherein the calculation result is as follows:

i.e. the learner's level of knowledge ability mu_k(j) Depending on the learner's high-order underlying trait θ_jAnd for learner j at knowledge point C_kDifficulty b in_jkAnd degree of distinction a_jkWherein theta_j、b_jk、a_jkBased on the learner's test results, calculated using the sirt kit of the R software.

Table 3 shows the results of the calculation of knowledge ability levels of some students on the objective problem of "function" and the subjective problem of "limit operation";

table 3: level of knowledge ability of students

Step6, calculating the mastery degree of the knowledge points of the learner: under the assumption of a connected (objective) type, student j isDegree of mastery η on objective question i_jiCalculated using the following formula, the calculation results are shown in Table 3, wherein q is_ikIndicating whether problem i requires knowledge skill C_k(q _ik1 indicates that the test question i needs knowledge point k to be solved).

Under the assumption of compensatory (subjective questions), the mastery degree eta of knowledge points of the student j on the subjective question i_jiCalculated using the following formula.

Table 4 shows the results of the calculation of the knowledge point mastery degree of some students on the objective function questions and the subjective limit calculation questions;

	degree of mastering objective question' function	Degree of mastery of subjective question "ultimate operation
			stu1	0.59	0.99
stu2	0.59	0.90
			stu3	0.49	0.31
stu4	0.56	0.92
			stu5	0.59	0.88

Table 4: degree of mastery of learner

Step7, the true score of the learner's response is calculated as follows:

step7.1: and estimating the mismatching and guessing parameters by using a four-parameter Logistic model, and correcting the phenomena of overestimation and underestimation of the student ability. The four parameter Logistic model is:

wherein, a_i，b_i，c_i，d_iRespectively representing discrimination, difficulty, guess parameter and sleep parameter, c_i，d_iCalculated according to the test results of the students by using the sirt toolkit of the R software.

Step7.2: respectively calculating the real answering scores R of the learners j on the objective questions i and the subjective questions i by adopting Bernoulli distribution and Gaussian distribution_ji：

P(R_ji＝1|η_ji,d_i,c_i)＝(1-d_i)η_ji+c_i(1-η_ji)

P(R_ji|η_ji,d_i,c_i)＝N(R_ji|[(1-d_i)η_ji+c_i(1-η_ji)],σ²)

It can be seen that the scores of the students on the objective questions are mapped to values in the {0,1} set, and the scores on the subjective questions are mapped to values in the [0,1] interval.

Table 5 shows the results of the calculation of the actual scores of some students on the objective function questions and the subjective limit calculation questions;

table 5: student true score

Step8, cognitive level calculation of learners: suppose that in the unit test, knowledge point C_kThrough the ith₁，i₂，……i_hThe test is carried out on the test questions, and the scores of the test questions in the test paper are respectively as follows: s_i1，s_i2，……s_ih. Therefore, the real answering score of the learner j on the test paper calculated based on Step7 is calculated according to the proportion of the scores of the test questions in the test paper in a weighted mode, namely the learner j on the C is calculated according to the following formula_kTrue cognitive level (see values on nodes in fig. 3):

step9, constructing a fuzzy cognitive map: constructing a fuzzy cognitive map of the learner on the course testing unit based on the knowledge map constructed at Step1 and the weight matrix W and combined with the cognitive level of the cognitive point measured by the student j in the unit test, which is calculated at Step8, as shown in 3;

step10, learning resource path generation steps are as follows:

step10.1: all knowledge points of learner j which do not reach ' proficiency ' in the tested course unit are obtained according to the cognitive level calculated by Step8, and an ordered set C ' ═ C ' is constructed according to the logical relationship of the knowledge points '₁，C′₂，……C′_n}；

Step10.2: according to the weight matrix W between the cognitive level and the knowledge point of the learner j_N×NCalculating and predicting each knowledge point C ' in C ' using the following formula '_m(m-1, 2,3, …, n) cognitive level changes after disturbance D (i.e. after learner has learned recommended learning resources)

Wherein, C'_m ⁽⁰⁾＝0，C_m ⁽¹⁾Is learner j at knowledge point C'_mInitial cognitive level of D_m ^(k)＝C′_m ^(k)–C′_m ^(k-1)，C′_m ^(k)Represents learner j at C 'after k-th disturbance'_m(iii) cognitive level of;

the cognitive level of the knowledge point after the learner j has the kth interference on the knowledge point C ' is calculated and predicted, and the knowledge concept C ' having the greatest influence (namely the most related knowledge points) on learning of the unit by the learner j is screened out '_x，x∈{1,2,3,…,n}；

Step10.3: if C'_x ^(k+1)0.9 or more, removing the knowledge point C ' from the knowledge point set C ' to be learned '_xUpdating the set and not performing recommendation learning any more;

step10.4: if C'_x ^(k+1)< 0.9, judge knowledge Point C 'calculated by Step9.2'_xA position in the ordered set of knowledge points C' to be learned;

step10.4.1: if C'_xIs a leaf node in the ordered set C ' of knowledge points to be learned (i.e., in C ', C '_xNo prior repair knowledge has been available), then according to learner at C'_xAnd (4) recommending learning resources corresponding to the cognitive level for the cognitive level (see Step2 and Step 3). After learning, learner j is at C'_xThe upper cognition level is preset as an average of the upper limit and the lower limit (see Step3) of the recommended learning resource hierarchy.

Step10.4.2: if C'_xIs a non-leaf node in the ordered set C ' of knowledge points to be learned (i.e., in C ', C '_xAnd repair prior knowledge), finding out learning knowledge points C 'according to the weight matrix W'_xInfluencing the most advanced prior knowledge point C ″)_x. Then, according to the learner at C ″)_xThe upper cognitive level is that which recommends learning resources corresponding to the cognitive level (see Step2 and Step 3). After learning, learner j is at C ″)_xThe upper cognition level is preset as an average of the upper limit and the lower limit (see Step3) of the recommended learning resource hierarchy.

Step10.5: repeating Step10.2-Step10.4 until the learner learns all knowledge points in C', thereby forming a learning resource path.

While the present invention has been described in detail with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, and various changes can be made without departing from the spirit and scope of the present invention.

Claims (3)

1. A personalized learning path recommendation method based on a fuzzy cognitive map is characterized by comprising the following steps:

step 1: combining the knowledge organization structure of the course classical teaching materials and the course field expert knowledge to determine the knowledge point set C of the course { C ═ C₁,C₂，……，C_NDetermining the relation between the knowledge points; constructing a course knowledge graph in a top-down mode by means of the software;

obtaining the dependency relationship among the course knowledge points and the 'correlation strength' among the course knowledge points by applying a Delphi method through multiple rounds of opinion solicitation, and further constructing a course weight matrix W_N×N(N is the total number of knowledge points in the course), W_mnIs a knowledge point C when the learner learns_mFor knowledge point C_nWhen m influences the strength<n is the point of knowledge C_mIs a knowledge point C_nThe first repair knowledge;

step 2: with reference to the cognitive classification of the objective of bloom education, the learner's cognitive level is divided into five levels: (1) the identification [0.4-0.6 ], (2) the basic understanding [0.6-0.7 ], (3) the understanding [0.7-0.8 ], (4) the mastery [0.8-0.9 ], (5) the mastery [0.9-1.0 ];

step 3: according to cognitive level hierarchical division, the learning resource hierarchical division corresponding to the Step2 cognitive level hierarchical division level is as follows: (1) achieving basic understanding [0.6-0.7), (2) achieving understanding [0.7-0.8), (3) achieving mastery [0.8-0.9), (4) achieving skillful mastery [0.9-1.0 ];

step 4: aiming at a certain learning unit of a course, testing a student to obtain the testing score of the course unit of the student;

step 5: computing a learner's knowledge competency level based on the test achievements, assuming J is the learner's set, for each knowledge point C_kE is C, set μ_k:J→[0,1]Is fuzzy set membership function, for each learner J belongs to J, defines J at knowledge point C_kLevel of knowledge ability of_k(j) J for student in fuzzy set (J, mu)_k) Wherein 0 is not more than mu_k(j)≤1，μ_k(j) Calculating according to a Logistic model in a project reaction theory:

i.e. the learner's level of knowledge ability mu_k(j) Depending on the learner's high-order underlying trait θ_jAnd for learner j at knowledge point C_kDifficulty b in_jkAnd degree of distinction a_jk；

Step 6: calculating the mastery degree of the knowledge points of the learner based on the test results, and under the assumption of a binding type, the mastery degree eta of the student j on the objective question i_jiComprises the following steps:

wherein q is_ikIndicating whether problem i requires knowledge skills, C_k(q_ik1 denotes a point of knowledge C required for problem i_kCan solve；

Under the compensatory assumption, the mastery degree eta of the knowledge point of the learner j on the subjective question i_jiComprises the following steps:

wherein q is_ikIndicating whether problem i requires knowledge skills, C_k(q_ik1 denotes a point of knowledge C required for problem i_kCan be solved;

step 7: estimating the error and guess parameters by using a four-parameter Logistic model, correcting the phenomena of overestimation and underestimation of the learner ability, and respectively calculating the real answering scores of the learner on objective questions and subjective questions by adopting Bernoulli distribution and Gaussian distribution;

step 8: suppose that in the unit test, knowledge point C_kThrough the ith₁，i₂，……i_hThe test is carried out on the test questions, and the scores of the test questions in the test paper are respectively as follows: s_i1，s_i2，……s_ihBased on the real answering score of the learner j on the Step7, the real answering score is calculated according to the fraction proportion of the test questions in the test paper in a weighted mode, and the learner j on the C is obtained_kTrue cognitive level on (c):

∑R_jts_t

t∈{i₁，i₂，.......，i_h}

step 9: based on the knowledge map and the weight matrix W constructed at Step1, combining the cognitive level of the cognitive point of student j in the unit test calculated at Step8, constructing a fuzzy cognitive map of the learner on the course test unit;

step 10: obtaining all knowledge points which are not reached by the learner j in the tested course unit according to the calculation result of Step8, and constructing an ordered set C' according to the logical relationship;

secondly according to the weight matrix W_N×NCalculate learner j at each knowledge point C'_m(m-1, 2,3, …, n) cognitive level changes following interference D;

through multiple rounds, learning resources are continuously recommended for the learner j, the learning effect is predicted, and a learning path is finally formed, but the learner is considered to acquire the knowledge point when the cognitive level is more than or equal to 0.9.

2. The method for recommending a personalized learning path based on a fuzzy cognitive map as claimed in claim 1, wherein Step7 is specifically:

step7.1: estimating the misestimation and guess parameters by using a four-parameter Logistic model to correct the phenomena of overestimation and underestimation of the learner's ability

Wherein, a_i，b_i，c_i，d_iRespectively representing discrimination, difficulty, guess parameter and sleep parameter, c_i，d_iAccording to the test result of the learner, the learner is calculated by using a sirt kit of RStudio Desktop 1.2.5042 software;

step7.2: respectively calculating the real answering scores R of the learner j on the objective question i and the subjective question i by adopting Bernoulli distribution and Gaussian distribution_ji：

P(R_ji＝1|η_ji,d_i,c_i)＝(1-d_i)η_ji+c_i(1-η_ji)

P(R_ji|η_ji,d_i,c_i)＝N(R_ji|[(1-d_i)η_ji+c_i(1-η_ji)],σ²)

It can be seen that the learner's score on the objective question is mapped to a value in the {0,1} set, and the score on the subjective question is mapped to a value in the [0,1] interval.

3. The method for recommending a personalized learning path based on a fuzzy cognitive map as claimed in claim 1, wherein Step10 is specifically:

step10.1: all knowledge points of learner j which do not reach ' proficiency ' in the tested course unit are obtained according to the cognitive level calculated by Step8, and an ordered set C ' ═ C ' is constructed according to the logical relationship of the knowledge points '₁，C′₂，……C′_n}；

Step10.2: according to the weight matrix W between the cognitive level and the knowledge point of the learner j_N×NCalculating and predicting each knowledge point C ' in C ' using the following formula '_m(m ═ 1,2,3, …, n) cognitive level changes following interference D:

wherein the content of the first and second substances,

C_m ⁽¹⁾is learner j at knowledge point C'_mThe initial level of cognition of the user,

represents learner j at C 'after k-th disturbance'_m(iii) cognitive level of;

calculating and predicting the cognitive level of the knowledge point after the learner j has the k-th interference on the knowledge point C ', and screening out the knowledge concept C ' which has the greatest influence on the learning of the learner j to the unit '_x，x∈{1,2,3,…,n}：

Step10.3: if it is

Removing knowledge points C ' from the knowledge point set C ' to be learned '_xUpdating the set and not performing recommendation learning any more;

step10.4: if it is

Judging the calculated knowledge point C 'of Step9.2'_xA position in the ordered set of knowledge points C' to be learned;

step10.4.1: if C'_xIs a leaf node in the knowledge point ordered set C 'to be learned according to the learner being in C'_xC, recommending learning resources corresponding to the cognitive level, and learning learner j is at C'_xThe upper cognition level is preset as an average value of an upper limit and a lower limit of a recommended learning resource level;

step10.4.2: if C'_xFinding out learning knowledge points C ' according to the weight matrix W for non-leaf nodes in the knowledge point ordered set C ' to be learned '_xInfluencing the most advanced prior knowledge point C ″)_xThen, according to the learner at C ″)_xThe cognitive level above is that the learner j recommends learning resources corresponding to the cognitive level, and after learning, the learner j is in C ″_xThe cognitive level of the upper level is preset as an average value of an upper limit and a lower limit of a recommended learning resource level;

step10.5: repeating Step10.2-Step10.4 until the learner learns all knowledge points in C', thereby forming a learning resource path.

Images