CN115510286A - Multi-relation cognitive diagnosis method based on graph convolution network - Google Patents

Multi-relation cognitive diagnosis method based on graph convolution network Download PDF

Info

Publication number
CN115510286A
CN115510286A CN202211389531.5A CN202211389531A CN115510286A CN 115510286 A CN115510286 A CN 115510286A CN 202211389531 A CN202211389531 A CN 202211389531A CN 115510286 A CN115510286 A CN 115510286A
Authority
CN
China
Prior art keywords
student
node
formula
ith
students
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
Application number
CN202211389531.5A
Other languages
Chinese (zh)
Inventor
张琨
吴开放
杨永辉
吴乐
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hefei University of Technology
Original Assignee
Hefei University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hefei University of Technology filed Critical Hefei University of Technology
Priority to CN202211389531.5A priority Critical patent/CN115510286A/en
Publication of CN115510286A publication Critical patent/CN115510286A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/901—Indexing; Data structures therefor; Storage structures
    • G06F16/9024—Graphs; Linked lists
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/02—Knowledge representation; Symbolic representation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/20—Education
    • G06Q50/205—Education administration or guidance

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Educational Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Computing Systems (AREA)
  • Tourism & Hospitality (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Strategic Management (AREA)
  • Health & Medical Sciences (AREA)
  • Educational Administration (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Primary Health Care (AREA)
  • General Business, Economics & Management (AREA)
  • Molecular Biology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a multi-relation cognitive diagnosis method based on a graph convolution network, which comprises the following steps: 1. constructing heterogeneous data: historical answer recording matrixes of students, incidence matrixes of exercises and knowledge points and interaction matrixes of students and knowledge points; 2. carrying out characteristic propagation on different answer results of the students by utilizing graph convolution respectively; the problem of data sparsity caused by wrong division of student answer records is relieved through graph comparison learning; 3. modeling the interaction relation between the student and the knowledge points and the inclusion relation between the exercises and the knowledge points by using an attention mechanism; 4. and fusing different characteristics obtained by the two modules, and predicting the performance of the student through a neurocognitive diagnosis model. The invention fully excavates the influence of different answer results of students on the abilities of the students and models the mastering conditions of the students on different knowledge points from two angles, thereby realizing more accurate student performance prediction and the mastering degree of the students on specific knowledge points.

Description

Multi-relation cognitive diagnosis method based on graph convolution network
Technical Field
The invention relates to cognitive diagnosis in the field of intelligent education, in particular to a multi-relation cognitive diagnosis method based on a graph convolution network.
Background
In the field of education, with the wide application of intelligent education systems, how to model knowledge mastery conditions of students and to recommend personalized exercises become a focus of attention of people.
The cognitive diagnosis model is used for modeling the mastering conditions of students at different knowledge points according to the historical answer conditions of the students, and then predicting the accuracy of the students on the problem which is not done by the students by using the commonly used diagnosis model. Early work utilized manually designed functions to model the linear relationship of students to problems, as traditional project response theory models single student abilities. The complex relation of interaction between students and exercises cannot be fully mined by the representation of the students and the exercises obtained by self-training of the neural network in the existing work, the hierarchical relation graph between the students, the exercises and the concepts is modeled, the influence of different answer results of the students on the mastering of knowledge points is ignored, in addition, the number of the exercises is far more than that of the knowledge points, one exercise usually comprises a plurality of knowledge points, the representation of the students and the exercises is relatively rough only through the interactive modeling of the students and the exercises, and the process of direct interaction between the students and the knowledge points is omitted in the prior cognitive diagnosis work. How to use the historical answer records of students and the association relationship between exercises and knowledge points to obtain the mastery degree of the students on different knowledge points and the student answer prediction becomes a problem worthy of research.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a multi-relation cognitive diagnosis method based on a graph-volume network, so that the influence of different answer results of students on the abilities of the students can be fully exploited, and the mastering conditions of the students on different knowledge points can be modeled from two angles, thereby realizing more accurate student performance prediction and the mastering degree of the students on specific knowledge points.
The invention adopts the following technical scheme for solving the technical problems:
the invention relates to a multi-relation cognitive diagnosis method based on a graph convolution network, which is characterized by comprising the following steps of:
step 1, obtaining historical answer records of students and constructing heterogeneous data:
denote student set by S, and S = { S = 1 ,...,s a ,...,s b ,...,s M }，s a Denotes the a student, s b Represents the b-th student, M represents the total number of students, a is more than or equal to 1, and b is more than or equal to M; denote the problem set by E, and E = { E = { 1 ,...,e i ,...,e j ,...,e N }，e i Represents the ith exercise, e j Representing the jth problem, N representing the total number of the problems, i is more than or equal to 1, and j is more than or equal to N; the set of knowledge points is represented by C, and C = { C 1 ,...,c f ,...,c g ,...,c K }，c f Represents the f-th knowledge point, c g Representing the g-th knowledge point, K represents the total number of the knowledge points, f is more than or equal to 1, and g is more than or equal to K;
let r be ai Represents the a-th student s a Answer ith exercise e i As a result, the historical answer matrix of the student is R = { R = { R } ai } M×N (ii) a And dividing the historical answer matrix R of the student into an answer pair exercise matrix R + Sum-and-error problem matrix R - (ii) a Let p be af Represents the a-th student s a The exercise contains the f-th knowledge point c f Then the interaction matrix of the student and the knowledge point is P = { P af } M×K (ii) a Let q be ig Show the ith track problem e i Containing the g-th knowledge point c g If the correlation matrix of the problem and knowledge point is Q = { Q = ig } N×K ；
Step 2, initializing student characterization vectors, exercise characterization vectors and knowledge point characterization vectors;
step 2.1, initializing student set S by using an Xavier method to obtain student representation T = { T = (zero-to-zero) after obtaining student representation S 1 ,...t a ,...t M Where t is a Represents the a-th student s a D-dimensional vector of (1);
step 2.2, initializing the problem set E by using an Xavier method to obtain a problem representation X = { X = (X) } 1 ,...x i ,...x N In which x i Represents the ith exercise e i D-dimensional vector of (1);
step 2.3, initializing the knowledge point set C by using an Xavier method to obtain a knowledge point representation O = { O = (O) } 1 ,...o f ,...o K In which o f Represents the f-th knowledge point c f D-dimensional vector of (1);
step 3, constructing a multi-relationship cognitive diagnosis model based on the graph convolution network, comprising the following steps: the system comprises a relational graph-based exercise level learning module, an attention mechanism-based knowledge point level learning module and a neuro-cognitive diagnosis module;
step 3.1, the processing of the problem level learning module based on the relational graph is as follows:
step 3.1.1, combining the right and wrong two relations in the student historical answer records to construct two student-exercise bipartite graphs to obtain a graph G + And graph G - Wherein, diagram G + Graph G representing the composition of the student's joint-border with the question for which the answer is correct - A graph formed by connecting students and exercises with wrong answers;
step 3.1.2, for FIG. G + Calculating the feature vector of the ith student node after the ith update by using the formula (1)
Figure BDA0003931398850000021
And feature vectors of the first problem node
Figure BDA0003931398850000022
Figure BDA0003931398850000023
In the formula (1), the acid-base catalyst,
Figure BDA0003931398850000024
represents the a-th student s a A set of questions that are correctly answered,
Figure BDA0003931398850000025
indicating the correct ith answer question e i A set of students;
Figure BDA0003931398850000026
and
Figure BDA0003931398850000027
respectively representing the feature vector of the ith student node and the feature vector of the ith exercise node after the ith-1 time of updating; when l =1, let
Figure BDA0003931398850000031
Step 3.1.3, for FIG. G + Obtaining the a-th student node characterization of the problem level using equation (2)
Figure BDA0003931398850000032
And ith problem node representation of problem level
Figure BDA00039313988500000311
Figure BDA0003931398850000033
Step 3.1.4, for FIG. G - Obtaining the characterization of the student node at the process of step 3.1.2 and step 3.1.3
Figure BDA0003931398850000034
And characterization of problem nodes
Figure BDA0003931398850000035
Step 3.1.4, the a-th student node representation of the exercise layer is obtained by using the formula (3)
Figure BDA0003931398850000036
And ith problem node characterization
Figure BDA0003931398850000037
Figure BDA0003931398850000038
In formula (3), R' represents a relation set containing right and wrong answers;
step 3.1.5, compare FIG. G + Student nodes or problem nodes in (1) as anchor point samples, in graph G - Taking the student node or the exercise node corresponding to the anchor point sample as a positive sample, and taking the graph G as a positive sample + Taking other student nodes or exercise nodes except the anchor point sample as negative samples;
step 3.1.6, constructing a contrast loss function L of the student node by using the formula (4) s ：
Figure BDA0003931398850000039
In the formula (4), T + Shows diagram G + Student node characterization of (1), T - Shows diagram G - The student node representation in (1); s' represents other student node sets except the a-th student node; tau is a temperature parameter, sim represents cosine similarity;
step 3.1.7, constructing contrast loss function L of problem node by using formula (5) e ：
Figure BDA00039313988500000310
In the formula (5), X + Shows diagram G + Middle problem node characterization vector, X - Shows diagram G - The problem node in (1) represents a vector; e' represents other exercise node sets except the ith exercise node;
step 3.2, the attention mechanism-based knowledge point level learning module performs processing:
step 3.2.1, obtaining an interaction matrix P of the students and the knowledge points from the historical answer records R and the association matrix Q of the exercises and the knowledge points of the students, and obtaining the weight alpha of the mastery degree of the a-th student node to the f-th knowledge point by using the formula (6) af So as to obtain the a-th student node characterization vector of the knowledge level by using the formula (7)
Figure BDA0003931398850000041
Figure BDA0003931398850000042
Figure BDA0003931398850000043
In the formula (6), the reaction mixture is,
Figure BDA0003931398850000044
representing a set of knowledge points having interaction with an a-th student node;
step 3.2.2 obtaining the associated weight beta of the ith problem node and the g knowledge point by using the formula (8) ig So as to obtain the ith exercise node characterization vector of the knowledge level by using the formula (9)
Figure BDA0003931398850000045
Figure BDA0003931398850000046
Figure BDA0003931398850000047
In the formula (8), the reaction mixture is,
Figure BDA0003931398850000048
representing a knowledge point set contained in the ith problem node;
step 3.3, processing of the neurocognitive diagnosis module:
step 3.3.1, respectively obtaining the final characterization vectors of the a-th student node by using the formula (10)
Figure BDA0003931398850000049
And the final characterization vector of the ith problem node
Figure BDA00039313988500000410
Figure BDA00039313988500000411
In the formula (10), [;]represents a tandem operation, σ (-) represents a sigmoid activation function; FC s A fully connected layer representing a student representation fusion; FC e Representing the problem to characterize the fused fully connected layer;
step 3.3.2, predicting the score of the ith student node for correctly answering the ith exercise node by using the formula (11)
Figure BDA00039313988500000412
Figure BDA00039313988500000413
In the formula (11), F (-) represents a multilayer perceptron of two fully-connected layers, and the weight value is a non-negative value; k represents a one-hot coded vector of the ith problem and knowledge point mapping;
Figure BDA0003931398850000051
is the ith problem node characterization vector
Figure BDA0003931398850000052
The scalar value after dimensionality reduction is obtained by the formula (12):
Figure BDA0003931398850000053
in formula (12), FC 1 Represents a fully connected layer;
step 4, constructing a cross entropy loss function L by using the formula (13) se ：
Figure BDA0003931398850000054
In the formula (13), (a, i) represents a sample pair of students and exercises collected in the training process, wherein a belongs to S, and i belongs to E; y is ai Representing the real score of the ith student on the ith problem;
and 5, constructing a total Loss function Loss by using the formula (14):
Loss＝L se +λ(L s +L e ) (14)
in the formula (14), λ is a weight for the super parameter to balance different loss values;
and 6, training the cognitive diagnosis model by using an Adam optimizer, minimizing a total Loss function Loss to update model parameters until convergence, so that the trained cognitive diagnosis model is used for predicting answers of students, and the mastering degree of the corresponding students on different knowledge points is obtained according to the obtained final characterization vectors of the student nodes.
The electronic device comprises a memory and a processor, and is characterized in that the memory is used for storing programs for supporting the processor to execute the multi-relation cognitive diagnosis method, and the processor is configured to execute the programs stored in the memory.
The invention relates to a computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, performs the steps of the multi-relationship cognitive diagnostic method.
Compared with the prior art, the invention has the beneficial effects that:
1. the invention divides the relation between the student behavior and the exercise into the correct relation and the wrong relation aiming at different answering behaviors of the student, adopts the mode of the graph to carry out representation learning on the interactive relation between the student and the exercise, and carries out propagation on the fused features through the graph volume layer to learn more accurate representation of the student, the exercise and the knowledge points, thereby better diagnosing the mastering condition of the student on the knowledge points.
2. In order to better perform cognitive diagnosis, the invention develops a multi-relationship cognitive diagnosis method based on a graph convolution network, can learn student characteristics and problem characteristics from two aspects of problems and knowledge points, fully considers the interactive relationship among multi-source data, and is more beneficial to measuring the abilities of students at different knowledge points. Meanwhile, the accuracy of student score prediction is effectively improved.
3. The invention designs a graph comparison learning method based on relational data, positive and negative sample pairs are constructed from two angles of right answer and wrong answer exercises, the problem of data sparsity is effectively relieved, and therefore the performance of a model is effectively improved.
Drawings
FIG. 1 is a flow chart of a multi-relationship cognitive diagnosis model based on a graph convolution network.
Detailed Description
In the embodiment, the influence difference of the correctness of the answer of the student on the mastery degree of the knowledge point is considered, and the feature fusion is performed on the learning of the exercise level based on the self-adaptive graph and volume network and the learning of the knowledge point level based on the attention mechanism to obtain more accurate student capability representation and exercise characteristic representation, so that the mastery condition of the student on the knowledge point is better diagnosed. As shown in fig. 1, the method comprises the following steps:
step 1, obtaining historical answer records of students and constructing heterogeneous data:
denote student set by S, and S = { S = 1 ,...,s a ,...,s b ,...,s M }，s a Denotes the a student a, s b Representing the b-th student, M represents the total number of students, a is more than or equal to 1, and b is more than or equal to M; denote the problem set by E, and E = { E = { 1 ,...,e i ,...,e j ,...,e N }，e i Represents the ith exercise, e j Representing the jth problem, N representing the total number of the problems, i is more than or equal to 1, and j is more than or equal to N; the set of knowledge points is denoted by C, and C = { C 1 ,...,c f ,...,c g ,...,c K }，c f Represents the f-th knowledge point, c g Representing the g-th knowledge point, K represents the total number of the knowledge points, f is more than or equal to 1, and g is more than or equal to K;
let r be ai Represents the a-th student s a Answer ith exercise e i As a result, the historical answer matrix of the student is R = { R = { R } ai } M×N (ii) a And dividing the historical answer matrix R of the student into an answer pair exercise matrix R + Sum-and-error problem matrix R - (ii) a Let p be af Represents the a-th student s a The exercise contains the f-th knowledge point c f Then the interaction matrix of the student and the knowledge point is P = { P af } M×K (ii) a Let q be ig Show the ith track problem e i Containing the g-th knowledge point c g If the correlation matrix of the problem and knowledge point is Q = { Q = ig } N×K ；
Step 2, initializing student characterization vectors, exercise characterization vectors and knowledge point characterization vectors;
step 2.1, initializing student set S by using an Xavier method to obtain student representation T = { T = (zero-to-zero) after obtaining student representation S 1 ,...t a ,...t M Where t is a Represents the a-th student s a D-dimensional vector of (1);
step 2.2, initializing the problem set E by using an Xavier method to obtain a problem representation X = { X = (X) } 1 ,...x i ,...x N In which x i Represents the ith exercise e i D-dimensional vector of (1);
step 2.3, initializing the knowledge point set C by using an Xavier method to obtain a knowledge point representation O = { O = (O) } 1 ,...o f ,...o K In which o is f Represents the f-th knowledge point c f D-dimensional vector of (a);
step 3, constructing a multi-relation cognitive diagnosis model based on a graph volume network, comprising the following steps: the system comprises a relational graph-based exercise level learning module, an attention mechanism-based knowledge point level learning module and a neuro-cognitive diagnosis module;
step 3.1, the processing of the problem level learning module based on the relational graph is as follows:
step 3.1.1, combining the right and wrong two relations in the student historical answer records to construct two student-exercise bipartite graphs to obtain a graph G + And graph G - Wherein, diagram G + Graph G representing the composition of the student's joint-border with the question for which the answer is correct - A graph formed by connecting students and exercises with wrong answers;
step 3.1.2, for FIG. G + Calculating the feature vector of the ith student node after the ith update by using the formula (1)
Figure BDA0003931398850000071
And feature vectors of the first problem node
Figure BDA0003931398850000072
Figure BDA0003931398850000073
In the formula (1), the reaction mixture is,
Figure BDA0003931398850000074
represents the a-th student s a A set of questions to be answered correctly,
Figure BDA0003931398850000075
indicating the correct ith answer question e i A set of students;
Figure BDA0003931398850000076
and
Figure BDA0003931398850000077
are respectively provided withRepresenting the feature vector of the ith student node and the feature vector of the ith exercise node after the (l-1) th update; when l =1, let
Figure BDA0003931398850000078
Step 3.1.3, for FIG. G + Obtaining the a-th student node characterization of the problem level using equation (2)
Figure BDA0003931398850000079
And ith problem node representation of problem level
Figure BDA00039313988500000710
Figure BDA00039313988500000711
Step 3.1.4, for FIG. G - Obtaining the characterization of the student node at the process of step 3.1.2 and step 3.1.3
Figure BDA00039313988500000712
And characterization of problem nodes
Figure BDA00039313988500000713
Step 3.1.4, the a-th student node representation of the exercise layer is obtained by using the formula (3)
Figure BDA00039313988500000714
And ith problem node characterization
Figure BDA00039313988500000715
Figure BDA0003931398850000081
In the formula (3), R' represents a relationship set, and comprises two relationships of the answer pair and the answer mistake;
step 3.1.5, compare FIG. G + The student node or exercise node in (1) is used as an anchor point sample in (G) - Taking the student node or the exercise node corresponding to the anchor point sample as a positive sample, and taking the graph G as a positive sample + Taking other student nodes or exercise nodes except the anchor point sample as negative samples;
step 3.1.6, constructing a contrast loss function L of the student node by using the formula (4) s ：
Figure BDA0003931398850000082
In the formula (4), T + Shows diagram G + Student node characterization of (1), T - Shows diagram G - The student node representation in (1); s' represents other student node sets except the a-th student node; tau is a temperature parameter, sim represents cosine similarity;
step 3.1.7, constructing contrast loss function L of problem node by using formula (5) e ：
Figure BDA0003931398850000083
In the formula (5), X + Shows diagram G + Middle problem node characterization vector, X - Shows diagram G - The problem node in (1) represents a vector; e' represents other exercise node sets except the ith exercise node;
step 3.2, the attention mechanism-based knowledge point level learning module performs processing:
step 3.2.1, obtaining an interaction matrix P of the students and the knowledge points from the historical answer records R and the association matrix Q of the exercises and the knowledge points of the students, and obtaining the weight alpha of the mastery degree of the a-th student node to the f-th knowledge point by using the formula (6) af So as to obtain the a-th student node characterization vector of the knowledge level by using the formula (7)
Figure BDA0003931398850000084
Figure BDA0003931398850000085
Figure BDA0003931398850000086
In the formula (6), the reaction mixture is,
Figure BDA0003931398850000087
representing a set of knowledge points having interaction with an a-th student node;
step 3.2.2 obtaining the associated weight beta of the ith problem node and the g knowledge point by using the formula (8) ig So as to obtain the ith problem node characterization vector of the knowledge level by using the formula (9)
Figure BDA0003931398850000091
Figure BDA0003931398850000092
Figure BDA0003931398850000093
In the formula (8), the reaction mixture is,
Figure BDA0003931398850000094
representing a knowledge point set contained in the ith problem node;
and 3.3, processing by a neurocognitive diagnosis module:
step 3.3.1, respectively obtaining the final characterization vectors of the a-th student node by using the formula (10)
Figure BDA0003931398850000095
And the final characterization vector of the ith problem node
Figure BDA0003931398850000096
Figure BDA0003931398850000097
In the formula (10), [;]represents a tandem operation, σ (-) represents a sigmoid activation function; FC s A fully connected layer representing a student representation fusion; FC e Representing the problem to characterize the fused fully connected layer;
step 3.3.2, predicting the score of the ith student node for correctly answering the ith exercise node by using the formula (11)
Figure BDA0003931398850000098
Figure BDA0003931398850000099
In the formula (11), F (-) represents a multilayer perceptron of two fully-connected layers, and the weight value is a non-negative value; k represents a one-hot coded vector of the ith problem and knowledge point mapping;
Figure BDA00039313988500000910
is the ith problem node characterization vector
Figure BDA00039313988500000911
The scalar value after dimensionality reduction is obtained by the formula (12):
Figure BDA00039313988500000912
in formula (12), FC 1 Represents a fully connected layer;
step 4, constructing a cross entropy loss function L by using the formula (13) se ：
Figure BDA00039313988500000913
In the formula (13), (a, i) represents that the training has been performedThe method comprises the steps of collecting sample pairs of students and exercises, wherein a belongs to S, and i belongs to E; y is ai Representing the real score of the ith student on the ith problem;
and 5, constructing a total Loss function Loss by using the formula (14):
Loss＝L se +λ(L s +L e ) (14)
in the formula (14), λ is a weight for the super parameter to balance different loss values;
and 6, training the cognitive diagnosis model by using an Adam optimizer, minimizing a total Loss function Loss to update model parameters until convergence, so that the trained cognitive diagnosis model is used for predicting answers of students, and the mastering degree of the corresponding students on different knowledge points is obtained according to the obtained final characterization vectors of the student nodes.
In this embodiment, an electronic device includes a memory for storing a program that supports a processor to execute the above-described multi-relationship cognitive diagnostic method, and a processor configured to execute the program stored in the memory.
In this embodiment, a computer-readable storage medium stores a computer program, and the computer program is executed by a processor to execute the steps of the multi-relationship cognitive diagnosis method.
Example (b):
to verify the effectiveness of the present method, the present invention employs two public datasets commonly used in intelligent educational research: ASSISTMENTs, EDNet. Screening not less than 15 students in answer records of each data set, removing exercises which do not contain knowledge point information, and randomly selecting 1000 students for research on the large-scale data set EDNet.
For the student achievement prediction task, the ACC, the AUC and the RMSE are adopted as evaluation standards. The invention selects 5 methods for effect comparison, namely IRT, MIRT, DINA, neurolCD and RCD. In particular, table 1 shows the experimental results in the above data set, and it can be observed that the method proposed by the present invention is superior to the comparative method in all three indexes of ACC, AUC and RMSE.
TABLE 1 student Performance prediction results on datasets for the methods of the invention and comparative methods
Figure BDA0003931398850000101
The invention carries out ablation experiments to explore the influence of each component, wherein MRCD-GRAPH represents that only the characterization learning of the problem level is carried out; MRCD-KC represents that only the representation learning of the knowledge point level is carried out, MRCD-GRAPH-KC (T) represents that the wrong division of answer records is not carried out; MRCD-GRAPH-KC (RGCN) represents the use of R-GCN to model learning student-problem interaction diagrams.
TABLE 2 student performance prediction results of ablation experiments according to the method of the invention
Figure BDA0003931398850000111
From table 2, it can be observed that the learning module at the problem level plays the most important role. This module is very important in the MRCD model because it focuses on the high-order interactions between students and problems and uses a graph-contrast learning framework to alleviate sparsity problems. Furthermore, when the correct answer and the wrong answer behavior are considered to be the same, the model performance is also degraded, which indicates that it is necessary to deal with the wrong relation of students to problems.
The number of layers of the hyper-reference convolutional network in the module is verified through experiments, the experimental results are shown in the following table 3, the model performance becomes better along with the increase of GCN layers, and information in learning high-order interaction is helpful for student achievement prediction.
TABLE 3 comparison of the Performance of the method of the invention for different propagation depths D
Figure BDA0003931398850000112
In conclusion, the method provided by the invention is remarkably superior to a plurality of comparative methods, thereby proving the feasibility of the method provided by the invention.

Claims (3)

1. A multi-relation cognitive diagnosis method based on a graph convolution network is characterized by comprising the following steps:
step 1, obtaining historical answer records of students and constructing heterogeneous data:
denote student set by S, and S = { S = 1 ,...,s a ,...,s b ,...,s M }，s a Denotes the a student a, s b Representing the b-th student, M represents the total number of students, a is more than or equal to 1, and b is more than or equal to M; denote the problem set by E, and E = { E = { 1 ,...,e i ,...,e j ,...,e N }，e i Represents the ith exercise, e j Representing the jth problem, N representing the total number of the problems, i is more than or equal to 1, and j is more than or equal to N; denote the knowledge point set by C, and C = { C 1 ,...,c f ,...,c g ,...,c K }，c f Represents the f-th knowledge point, c g Representing the g-th knowledge point, K represents the total number of the knowledge points, f is more than or equal to 1, and g is more than or equal to K;
let r be ai Represents the a-th student s a Answer ith exercise e i As a result, the historical answer matrix of the student is R = { R = { R } ai } M×N (ii) a And dividing the historical answer matrix R of the student into an answer pair exercise matrix R + Sum-and-error problem matrix R - (ii) a Let p be af Represents the a-th student s a The exercise contains the f-th knowledge point c f Then the interaction matrix of the student and the knowledge point is P = { P af } M×K (ii) a Let q be ig Show the ith track problem e i Containing the g-th knowledge point c g If the correlation matrix of exercise and knowledge points is Q = { Q = ig } N×K ；
Step 2, initializing student characterization vectors, exercise characterization vectors and knowledge point characterization vectors;
step 2.1, initializing student set S by using an Xavier method to obtain student representation T = { T = (zero-to-zero) after obtaining student representation S 1 ,...t a ,...t M Where t is a Represents the a-th student s a D-dimensional vector of (1);
step 2.2, initializing the problem set E by using an Xavier method to obtain a problem representation X = { X = (X) } 1 ,...x i ,...x N In which x i Represents the ith exercise e i D-dimensional vector of (1);
step 2.3, initializing the knowledge point set C by using an Xavier method to obtain a knowledge point representation O = { O = (O) } 1 ,...o f ,...o K In which o f Represents the f-th knowledge point c f D-dimensional vector of (1);
step 3, constructing a multi-relationship cognitive diagnosis model based on the graph convolution network, comprising the following steps: the system comprises a relational graph-based exercise level learning module, an attention mechanism-based knowledge point level learning module and a neuro-cognitive diagnosis module;
step 3.1, the processing of the problem level learning module based on the relational graph is as follows:
step 3.1.1, combining the right and wrong two relations in the student historical answer records to construct two student-exercise bipartite graphs to obtain a graph G + And graph G - Wherein, diagram G + Graph G representing the composition of the student's joint-border with the question for which the answer is correct - A graph formed by connecting students and exercises with wrong answers;
step 3.1.2, for FIG. G + Calculating the feature vector of the ith student node after the ith update by using the formula (1)
Figure FDA0003931398840000021
And feature vectors of the first problem node
Figure FDA0003931398840000022
Figure FDA0003931398840000023
In the formula (1), the reaction mixture is,
Figure FDA0003931398840000024
denotes the firsta students s a A set of questions that are correctly answered,
Figure FDA0003931398840000025
indicating the correct ith answer question e i A set of students;
Figure FDA0003931398840000026
and
Figure FDA0003931398840000027
respectively representing the feature vector of the ith student node and the feature vector of the ith exercise node after the ith-1 time of updating; when l =1, let
Figure FDA0003931398840000028
Step 3.1.3, for FIG. G + Obtaining the a-th student node characterization of the problem level using equation (2)
Figure FDA0003931398840000029
And ith problem node representation of problem level
Figure FDA00039313988400000210
Figure FDA00039313988400000211
Step 3.1.4, for FIG. G - Obtaining the characterization of the student node at the process of step 3.1.2 and step 3.1.3
Figure FDA00039313988400000212
And characterization of problem nodes
Figure FDA00039313988400000213
Step 3.1.4, obtaining the a-th student section of the exercise layer by using the formula (3)Point characterization
Figure FDA00039313988400000214
And ith problem node characterization
Figure FDA00039313988400000215
Figure FDA00039313988400000216
In formula (3), R' represents a relation set containing right and wrong answers;
step 3.1.5, compare FIG. G + Student nodes or problem nodes in (1) as anchor point samples, in graph G - Taking the student node or the exercise node corresponding to the anchor point sample as a positive sample, and drawing G + Taking other student nodes or exercise nodes except the anchor point sample as negative samples;
step 3.1.6, constructing a contrast loss function L of the student node by using the formula (4) s ：
Figure FDA0003931398840000031
In the formula (4), T + Shows diagram G + Student node characterization of (1), T - Shows diagram G - The student node representation in (1); s' represents other student node sets except the a-th student node; tau is a temperature parameter, sim represents cosine similarity;
step 3.1.7, constructing contrast loss function L of problem node by using formula (5) e ：
Figure FDA0003931398840000032
In the formula (5), X + Shows diagram G + Middle problem node characterization vector, X - Shows diagram G - Exercise section of (1)Point characterization vectors; e' represents other exercise node sets except the ith exercise node;
step 3.2, the attention mechanism-based knowledge point level learning module performs processing:
step 3.2.1, obtaining an interaction matrix P of the students and the knowledge points from the historical answer records R and the association matrix Q of the exercises and the knowledge points of the students, and obtaining the weight alpha of the mastery degree of the a-th student node to the f-th knowledge point by using the formula (6) af So as to obtain the a-th student node characterization vector of the knowledge level by using the formula (7)
Figure FDA0003931398840000033
Figure FDA0003931398840000034
Figure FDA0003931398840000035
In the formula (6), the reaction mixture is,
Figure FDA0003931398840000036
representing a set of knowledge points having interaction with an a-th student node;
step 3.2.2 obtaining the associated weight beta of the ith problem node and the g knowledge point by using the formula (8) ig So as to obtain the ith problem node characterization vector of the knowledge level by using the formula (9)
Figure FDA0003931398840000037
Figure FDA0003931398840000038
Figure FDA0003931398840000039
In the formula (8), the reaction mixture is,
Figure FDA00039313988400000310
representing a knowledge point set contained in the ith exercise node;
and 3.3, processing by a neurocognitive diagnosis module:
step 3.3.1, respectively obtaining the final characterization vectors of the a-th student node by using the formula (10)
Figure FDA0003931398840000041
And the final characterization vector of the ith problem node
Figure FDA0003931398840000042
Figure FDA0003931398840000043
In the formula (10), [;]represents a tandem operation, σ (-) represents a sigmoid activation function; FC s A fully connected layer representing a student representation fusion; FC e Representing the problem to characterize the fused fully connected layer;
step 3.3.2, predicting the score of the ith student node for correctly answering the ith exercise node by using the formula (11)
Figure FDA0003931398840000044
Figure FDA0003931398840000045
In the formula (11), F (-) represents a multilayer perceptron of two fully-connected layers, and the weight value is a non-negative value; k represents a one-hot coded vector of the ith problem and knowledge point mapping;
Figure FDA0003931398840000046
is the ith problem node characterization vector
Figure FDA0003931398840000047
The scalar value after dimensionality reduction is obtained by the formula (12):
Figure FDA0003931398840000048
in formula (12), FC 1 Represents a fully connected layer;
step 4, constructing a cross entropy loss function L by using the formula (13) se ：
Figure FDA0003931398840000049
In the formula (13), (a, i) represents a sample pair of students and exercises collected in the training process, wherein a belongs to S, and i belongs to E; y is ai Representing the real score of the ith student on the ith problem;
and 5, constructing a total Loss function Loss by using the formula (14):
Loss＝L se +λ(L s +L e ) (14)
in the formula (14), λ is a weight for the super parameter to balance different loss values;
and 6, training the cognitive diagnosis model by using an Adam optimizer, minimizing a total Loss function Loss to update model parameters until convergence, so that the trained cognitive diagnosis model is used for predicting answers of students, and the mastering degree of the corresponding students on different knowledge points is obtained according to the obtained final characterization vectors of the student nodes.
2. An electronic device comprising a memory and a processor, wherein the memory is configured to store a program that enables the processor to perform the multi-relationship cognitive diagnostic method of claim 1, and the processor is configured to execute the program stored in the memory.
3. A computer-readable storage medium, having a computer program stored thereon, where the computer program is executed by a processor to perform the steps of the multi-relationship cognitive diagnostic method of claim 1.
CN202211389531.5A 2022-11-08 2022-11-08 Multi-relation cognitive diagnosis method based on graph convolution network Pending CN115510286A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211389531.5A CN115510286A (en) 2022-11-08 2022-11-08 Multi-relation cognitive diagnosis method based on graph convolution network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211389531.5A CN115510286A (en) 2022-11-08 2022-11-08 Multi-relation cognitive diagnosis method based on graph convolution network

Publications (1)

Publication Number Publication Date
CN115510286A true CN115510286A (en) 2022-12-23

Family

ID=84513966

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211389531.5A Pending CN115510286A (en) 2022-11-08 2022-11-08 Multi-relation cognitive diagnosis method based on graph convolution network

Country Status (1)

Country Link
CN (1) CN115510286A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116166998A (en) * 2023-04-25 2023-05-26 合肥师范学院 Student performance prediction method combining global and local features
CN116306863A (en) * 2023-01-06 2023-06-23 山东财经大学 Collaborative knowledge tracking modeling method and system based on contrast learning
CN117994092A (en) * 2023-11-30 2024-05-07 安徽大学 Cognitive diagnosis method based on entity internal relation and learning curve theory

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116306863A (en) * 2023-01-06 2023-06-23 山东财经大学 Collaborative knowledge tracking modeling method and system based on contrast learning
CN116166998A (en) * 2023-04-25 2023-05-26 合肥师范学院 Student performance prediction method combining global and local features
CN117994092A (en) * 2023-11-30 2024-05-07 安徽大学 Cognitive diagnosis method based on entity internal relation and learning curve theory

Similar Documents

Publication Publication Date Title
CN113033808B (en) Deep embedded knowledge tracking method based on problem difficulty and student capability
CN112116092B (en) Interpretable knowledge level tracking method, system and storage medium
CN115510286A (en) Multi-relation cognitive diagnosis method based on graph convolution network
CN110941723A (en) Method, system and storage medium for constructing knowledge graph
CN112529155B (en) Dynamic knowledge mastering modeling method, modeling system, storage medium and processing terminal
CN113283488B (en) Learning behavior-based cognitive diagnosis method and system
CN114429212A (en) Intelligent learning knowledge ability tracking method, electronic device and storage medium
CN114385801A (en) Knowledge tracking method and system based on hierarchical refinement LSTM network
CN115544158A (en) Multi-knowledge-point dynamic knowledge tracking method applied to intelligent education system
CN115310520A (en) Multi-feature-fused depth knowledge tracking method and exercise recommendation method
CN114676903A (en) Online prediction method and system based on time perception and cognitive diagnosis
CN118193920A (en) Knowledge tracking method of personalized forgetting mechanism based on concept driving
CN111898803B (en) Problem prediction method, system, equipment and storage medium
CN117807422A (en) Higher-order cognitive diagnosis method based on multilayer attention network
CN117011098A (en) Prediction method for learning ability of students based on MKVMN model
CN116611513A (en) Cognitive diagnosis method based on intra-layer similarity relationship
CN114707775B (en) Knowledge tracking model training method, tracking method, device, equipment and medium
CN116402134A (en) Knowledge tracking method and system based on behavior perception
CN114925610A (en) Learner knowledge structure and level modeling method, system, equipment and terminal
CN114971972A (en) Deep knowledge tracking method integrating forgetting factor and exercise difficulty and intelligent terminal
CN115205072A (en) Cognitive diagnosis method for long-period evaluation
Kang et al. Personalized exercise recommendation via implicit skills
CN112001536A (en) High-precision finding method for minimal sample of mathematical capability point defect of primary and secondary schools based on machine learning
CN113989080B (en) Learner representation method and system based on deep knowledge-project joint tracking
CN118035568B (en) Intelligent recommendation method for education problems with knowledge mastering degree perception

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination