CN113282723A - Deep knowledge tracking pre-training method based on graph neural network - Google Patents
Deep knowledge tracking pre-training method based on graph neural network Download PDFInfo
- Publication number
- CN113282723A CN113282723A CN202110557176.7A CN202110557176A CN113282723A CN 113282723 A CN113282723 A CN 113282723A CN 202110557176 A CN202110557176 A CN 202110557176A CN 113282723 A CN113282723 A CN 113282723A
- Authority
- CN
- China
- Prior art keywords
- low
- knowledge
- knowledge point
- dimensional vector
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 36
- 238000012549 training Methods 0.000 title claims abstract description 14
- 239000013598 vector Substances 0.000 claims abstract description 48
- 230000003993 interaction Effects 0.000 claims abstract description 18
- 238000012512 characterization method Methods 0.000 claims description 13
- 238000013507 mapping Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 6
- 230000003213 activating effect Effects 0.000 claims description 2
- 238000011478 gradient descent method Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 7
- 230000000694 effects Effects 0.000 description 6
- 230000015654 memory Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 239000002131 composite material Substances 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
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
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3346—Query execution using probabilistic model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Animal Behavior & Ethology (AREA)
- Probability & Statistics with Applications (AREA)
- Human Computer Interaction (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
A knowledge tracking pre-training method based on a graph neural network improves the accuracy of the existing deep knowledge tracking method, and relates to the field of intelligent education method exploration. The problem low-dimensional representation is finally obtained by constructing a bipartite graph of the problem and knowledge point relation, extracting the corresponding relation, the problem similarity and the knowledge point similarity of the problem and the knowledge points, deploying a graph neural network to obtain the low-dimensional vector representation, simultaneously considering various characteristics of the problem, such as problem difficulty, and learning the interaction of the multi-field characteristics by using a multiplicative neural network. The representation integrates the characteristics of abundant problems, knowledge point relations, problem difficulty and the like, and can be used as the input of the existing deep knowledge tracking method. The knowledge tracking pre-training method can be combined with any depth knowledge tracking method for use, and practice proves that the accuracy of the existing depth knowledge tracking method can be greatly improved, and meanwhile, the problem low-dimensional representation with higher interpretability is obtained.
Description
Technical Field
The invention relates to a knowledge tracking task in the field of intelligent education, in particular to a knowledge tracking pre-training method based on a graph neural network.
Background
Knowledge tracking means that an evaluation function is established by using historical learning data of students and characteristics of learning contents to predict the average probability of each question answered by the students later. Generally, the current intelligent education system records the learning data of a plurality of students, and the students can learn a certain knowledge point in the forms of making questions, watching a web course and the like. Aiming at each knowledge point, the intelligent education system can design a plurality of corresponding questions to help students master the knowledge points. Knowledge tracking is to track the learning rule of students on teaching contents by using the existing historical answer sequences of students and further predict the average answer probability of the students on relevant questions. Knowledge tracking can help teachers to know the characteristics of teaching contents and the learning rules of students, and then more reasonable teaching arrangement is designed.
Analyzing recent patent technologies about knowledge tracking:
1. the Chinese patent application with application number 201911250785.7, namely 'knowledge tracking data processing method, system and storage medium based on graph convolution', provides a method for modeling a student learning sequence by using a convolutional neural network, and the method only aims at knowledge points, ignores the difference of problems under each knowledge point and also ignores the relation between the problems and the knowledge points;
2. the Chinese patent with application number of 201911115390.6 discloses a knowledge tracking system and method based on a hierarchical memory network, which simulates the modes of human long-term memory and short-term memory based on the knowledge tracking system of the hierarchical memory network, establishes a deep network and performs classified attenuation storage on input knowledge information. Similarly, the method ignores the difference of different problems contained in the knowledge points and also ignores the relationship between the problems and the knowledge points.
(II) analyzing the recent research of the knowledge tracking method based on deep learning:
deep Knowledge tracking, published by Piech et al at the Advances In Neural Information Processing Systems conference (2015, 28, 505, 513), first uses a Deep Neural network to solve the Knowledge tracking task and uses a recurrent Neural network pair to capture the sequence dependency of the student's historical answer sequence. The defects are as follows: the method only analyzes the learning rule of the knowledge points. The method neglects the difference of different problems under the same knowledge point, and is not suitable for the scene that the number of the knowledge points is large, but the interaction between students and the knowledge points is sparse.
Dynamic key-value memory network for knowledge tracking, published by Zhang Jiani et al at the proceedings of International Conference on World Wide Web Conference (26 th 2017), which uses different memory slots to record the mastery states of different knowledge points and automatically learns the relationship between questions and knowledge points based on the attention mechanism. The disadvantage is that the relationship between the problem and the knowledge point is easily obtained from the true scene, and the learning of the relationship based on the attention mechanism has a certain deviation. Meanwhile, the method ignores the similarity of knowledge points and the similarity of problems.
The following conclusions can be drawn from the domestic and foreign relevant patent analysis and relevant research: the current deep knowledge tracking method directly focuses on the learning rule of knowledge points. In practice, however, each knowledge point contains different problems, and direct modeling of knowledge points ignores the unique information of these problems, compromising the accuracy of the method. Meanwhile, the number of problems is usually far larger than the number of knowledge points, and the interaction of the problems is more sparse (generally, one student does not repeatedly do the same problem for many times, but learns the same knowledge point for many times). This makes it difficult for current deep knowledge tracking methods to be applied directly to problem-level prediction.
Therefore, those skilled in the art are dedicated to develop a knowledge tracking pre-training method, a low-dimensional characterization of a problem is pre-trained on the basis of a complex relationship between the problem and a knowledge point and a unique feature of each problem, and the obtained problem low-dimensional characterization can be used as an input of any deep knowledge tracking method, so that the fine-grained average answer-to-pair probability of each problem is predicted, and the prediction accuracy of the knowledge tracking method based on deep learning is improved.
Disclosure of Invention
In view of the above-mentioned drawbacks of the prior art, the present invention is directed to solving the technical problem that the prior deep knowledge tracking method ignores the difference of the problems and has limited expression results.
In order to achieve the purpose, the invention provides a deep knowledge tracking pre-training method based on a graph neural network, which comprises the following steps:
step 1, constructing a relation bipartite graph G of problems and knowledge points;
step 2, extracting three relations, namely a corresponding relation between the question and the knowledge point, a question similarity relation and a knowledge point similarity relation;
step 3, obtaining low-dimensional vector representations of problem nodes and knowledge point nodes by using a neural network;
step 4, obtaining low-dimensional vector representation of the auxiliary information of the problem;
step 5, fusing problem node low-dimensional vector representation, corresponding knowledge point node low-dimensional vector representation and problem auxiliary information by using a multiplicative neural network to obtain problem low-dimensional representation;
preferably, the low-dimensional vector characterization learning of the problem node and the knowledge point node in the step 3 and the parameter learning of the multiplicative neural network in the step 5 are performed simultaneously to form a complete end-to-end knowledge tracking pre-training method.
Further, in step 1, the bipartite graph G includes two types of nodes, namely a problem node and a knowledge point node, and edges only exist between the different types of nodes; if problem node qiAnd knowledge point nodes sjThere is an edge between them, indicating that the knowledge point contains the problem.
Further, in step 2, the corresponding relationship between the problem and the knowledge point refers to an edge in the graph G, which is a first-order similarity of the nodes in the graph G;
the problem similarity relation and the knowledge point similarity relation are determined according to the existence of common neighbors of the nodes in the graph G, and are second-order similarity of the nodes in the graph G;
for two problem nodes qiAnd q isjIf the neighbor node sets of the nodes are overlapped, namely if a knowledge point contains the two problems at the same time, the nodes have a similar relation, otherwise, the nodes do not have the similar relation; similarly, the similarity relationship between two knowledge points can be extracted, that is, if a problem exists and two knowledge points simultaneously contain the problem, the two knowledge points have the similarity relationship, otherwise, the similarity relationship does not exist.
Preferably, in step 3, the neural network is a graph neural network.
Further, step 3 further comprises:
step 3.1, randomly initializing low-dimensional vector representations of problem nodes and knowledge point nodes in the graph G;
step 3.2, representing the low-dimensional vectors of the problem and the knowledge point nodes, performing inner product on every two low-dimensional vectors, and constraining by using the corresponding relation of the problem and the knowledge point through activating function mapping;
3.3, performing inner product on the low-dimensional vector representations of the problem nodes pairwise, mapping by an activation function, and using problem similarity for constraint;
step 3.4, performing inner product on the low-dimensional vector representations of the knowledge point nodes pairwise, mapping by an activation function, and using the similarity of the knowledge points for constraint;
further, step 4 further comprises: vectorizing auxiliary information of a problem, preferably, converting discrete value features into one-hot codes, normalizing continuous value features, splicing all coded features, and mapping dimensions by using a fully-connected neural network to obtain low-dimensional vector representation of the auxiliary information of required dimensions.
Further, in step 5, for a problem i, we have its corresponding problem node q in graph GiLow dimensional vector characterization ofNode s of knowledge point corresponding to qiiLow dimensional vector characterization of(if the question i corresponds to a plurality of knowledge points, we take the average value of the low-dimensional vector representations of the knowledge point nodes), and the low-dimensional vector representation of the auxiliary information of the question iThen, a multiplicative neural network is used for feature interaction, and the specific steps are as follows:
Step 5.2, toPerforming pairwise interaction by using inner product operation to obtain a secondary characteristic interaction information matrix
Step 5.3, mixingAndspliced and mapped intoIs the low dimensional characterization of problem i we need;
step 5.4, useAs input to a fully connected neural network, and using an activation function to predict an average answer probability for each question, where the average answer probability reflects the difficulty of the question, using the true question average answer probability to constrain, learning multiplicationParameters of the neural network.
Preferably, the network parameters in steps 3 and 5 can be trained end to end by a gradient descent method.
Further, the problem low-dimensional representation obtained by pre-training can be used as the input of the existing deep knowledge tracking method, and whether the student can answer the question or not is predicted at each time step.
The applied graph neural network regularly restricts the low-dimensional vector representation of the problem nodes and the knowledge point nodes in the vector space. That is, when there is a corresponding relationship between the problem and the knowledge point, the low-dimensional vector representations of the problem node and the knowledge point node are close to each other in the vector space. Similarly, similar knowledge points or problems, corresponding to low-dimensional vector representations, are also close in vector space. Although we have problem node representations and their corresponding knowledge point node representations, we wish to obtain that the final problem low-dimensional representation can include such relationship information, together with auxiliary information for the problem, so that it is easier to distinguish between different problems. Therefore, the multiplication neural network is used by the people, so that the three types of characteristics are fully interacted, and the required problem is represented in a low-dimensional mode.
The invention has the following technical effects:
1. the problem low-dimensional representation obtained by pre-training in the invention is widely and conveniently applied, and can be directly used as the input of the knowledge tracking methods on the premise of not changing the structure of the existing deep knowledge tracking method.
2. The invention fully utilizes the complex relation between the problems and the knowledge points, and solves the problem that the prior deep knowledge tracking method has limited performance when the number of the problems is large and the interaction between students and the learned problems is very sparse.
3. Practice proves that the pre-trained problem low-dimensional representation can effectively improve the accuracy of the existing deep knowledge tracking method. And the visualization effect of the problem low-dimensional representation also shows that the learned problem representation is more explanatory.
The conception, the specific structure and the technical effects of the present invention will be further described with reference to the accompanying drawings to fully understand the objects, the features and the effects of the present invention.
Drawings
FIG. 1 is a flow chart of one embodiment of the present invention.
Detailed Description
The preferred embodiments of the present application will be described below with reference to the accompanying drawings for clarity and understanding of the technical contents thereof. The present application may be embodied in many different forms of embodiments and the scope of the present application is not limited to only the embodiments set forth herein.
The conception, the specific structure and the technical effects of the present invention will be further described below to fully understand the objects, the features and the effects of the present invention, but the present invention is not limited thereto.
The embodiment of the invention provides a deep knowledge tracking pre-training method based on a graph neural network, which is applied to an intelligent education system environment for carrying out a knowledge tracking task, wherein the environment comprises a plurality of knowledge points, each knowledge point comprises a plurality of problems, and each problem corresponds to one knowledge point or a plurality of knowledge points. Each question has corresponding auxiliary information, such as average answering time of each question, question type (choice question, filling in blank question, etc.), question difficulty, etc. The student interacts with the intelligent education system, answers a question at each moment, and the system records the label of the question and whether to answer the question. The method comprises the following specific steps:
step 1, defining a problem and knowledge point relation bipartite graph G, wherein the G comprises two types of nodes which are respectively a problem node and a knowledge point node, and edges only exist between the different types of nodes. If problem node qiAnd knowledge point nodes sjThere is an edge between them, indicating that the knowledge point contains the problem.
Step 2, corresponding relation between the problem and the knowledge point can be directly obtained from the graph G, and meanwhile, the two problem nodes q are subjected toiAnd q isjIf their neighbor node sets have coincidence (i.e. there is a knowledge point containing both problems), they are similar, otherwise they are dissimilar. In the same way, the method for preparing the composite material,the similarity of two knowledge points can be extracted.
And 3, randomly initializing low-dimensional vector representation of the problem nodes and the knowledge point nodes, and obtaining a relation matrix, a problem similarity matrix and a knowledge point similarity matrix of the problems and the knowledge points by utilizing vector inner product operation and activation function mapping. And (4) constraining by using the three relations extracted in the step (2) and learning the low-dimensional vector representation of the problem nodes and the knowledge point nodes.
And 4, vectorizing auxiliary information of the problem, specifically converting discrete value features into one-hot codes, and normalizing continuous value features. And then splicing all the coded features, and performing dimensionality mapping by using a fully-connected neural network to obtain the low-dimensional vector representation of the auxiliary information with the required dimensionality.
Step 5, for a problem i, we have its corresponding problem node q in the graph GiLow dimensional vector characterization ofNode s of knowledge point corresponding to qiiLow dimensional vector characterization of(if the question i corresponds to a plurality of knowledge points, we take the average value of the low-dimensional vector representations of the knowledge point nodes), and the low-dimensional vector representation of the auxiliary information of the question iThen, the multiplicative neural network is used for feature interaction, and the specific steps are as follows
Step 5.2, toPerforming pairwise interaction by using inner product operation to obtain a secondary characteristic interaction information matrix
Step 5.3, mixingAndspliced and mapped intoIs the low dimensional characterization of problem i that we need.
Step 5.4, useThe average answer pair probability of each question is predicted as an input to the fully connected neural network and using the activation function. (where the average answer probability reflects the difficulty of the question). And (4) using the real question average answer pair probability to constrain and learning the parameters of the multiplicative neural network.
And 6, performing low-dimensional vector representation learning of the problem nodes and the knowledge point nodes in the step 3 and parameter learning of the multiplicative neural network at the same time to form a complete end-to-end knowledge tracking pre-training method.
Further, the problem low-dimensional representation obtained by pre-training can be used as the input of the existing deep knowledge tracking method, and whether the student can answer the question or not is predicted at each time step.
In the knowledge tracking task of the intelligent education system, the method can enable the existing deep knowledge tracking method to make full use of the complex relation between the question and the knowledge point, and can fully consider the learning degree of the knowledge point corresponding to the question, the wrong answer condition of the relevant question and the difference of different questions when the answer probability of the question is predicted. The existing deep knowledge tracking method is used for directly predicting the answer probability of knowledge points, but in the actual situation, the mastering degree of students on the knowledge points cannot be directly obtained. The feedback that the intelligent education system can take is that the student answers a right or wrong question, which greatly limits the performance of the existing deep knowledge tracking method. If the existing deep knowledge tracking method is used for directly predicting whether the problem is answered or not, the dilemma that the number of the problems is huge and the interaction between students and the problems is sparse (the number of times that the students usually do the same problem repeatedly is limited) is faced, and the sparse data has great damage to the performance of the deep neural network. The problem representation is fully pre-trained by utilizing the complex relation between the problems and the knowledge points and the auxiliary information of each problem, so that the problem interaction sparseness limit of students and the problems is effectively solved, the learning conditions of the related problems and the knowledge points are considered, and the method is more practical. Meanwhile, the accuracy of the conventional depth knowledge tracking method is greatly improved.
The foregoing detailed description of the preferred embodiments of the present application. It should be understood that numerous modifications and variations can be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the concepts of the present application should be within the scope of protection defined by the claims.
Claims (10)
1. A deep knowledge tracking pre-training method based on a graph neural network is characterized by comprising the following steps:
step 1, constructing a relation bipartite graph G of problems and knowledge points;
step 2, extracting three relations, namely a corresponding relation between the question and the knowledge point, a question similarity relation and a knowledge point similarity relation;
step 3, obtaining the low-dimensional vector representation of the problem node and the low-dimensional vector representation of the knowledge point node in the graph G by using a neural network;
step 4, obtaining low-dimensional vector representation of the auxiliary information of the problem;
and 5, fusing the low-dimensional vector representation of the problem node, the corresponding low-dimensional vector representation of the knowledge point node and the auxiliary information of the problem by using a multiplicative neural network to obtain the low-dimensional representation of the problem.
2. The method according to claim 1, wherein in the step 1, the graph G includes two types of nodes, namely a problem node and a knowledge point node, and edges exist only between the different types of nodes; if problem node qiAnd knowledge point nodes sjIf there is an edge between them, it indicates that the knowledge point contains the problem.
3. The method according to claim 2, wherein in the step 2, the corresponding relationship between the question and the knowledge point refers to an edge in the graph G, and is a first-order similarity of nodes in the graph G;
the problem similarity relation and the knowledge point similarity relation are determined according to the existence of common neighbors of the nodes in the graph G, and are second-order similarity of the nodes in the graph G;
for two problem nodes qiAnd q isjIf the neighbor node sets of the two problems are overlapped, namely if one knowledge point contains the two problems, the two problems have a similar relation, otherwise, the two problems do not have the similar relation; similarly, if a problem exists and is simultaneously contained by two knowledge points, the two knowledge points have a similar relationship, otherwise, the similar relationship does not exist.
4. The method of claim 1, wherein in step 3, the neural network is a graph neural network.
5. The method of claim 3, wherein step 3 further comprises:
step 3.1, randomly initializing low-dimensional vector representation of the problem node in the graph G and low-dimensional vector representation of the knowledge point node;
step 3.2, performing inner product on the low-dimensional vector representation of the problem node and the low-dimensional vector representation of the knowledge point node in pairs, and constraining by using the corresponding relation of the problem and the knowledge point through activating function mapping;
3.3, performing inner product on the low-dimensional vector representations of the problem nodes pairwise, mapping by an activation function, and using the problem similarity relation for constraint;
and 3.4, performing inner product on the low-dimensional vector representations of the knowledge point nodes pairwise, mapping by an activation function, and using the similarity relation of the knowledge points for constraint.
6. The method of claim 1, wherein step 4 further comprises: vectorizing the auxiliary information of the problem, preferably converting discrete value features into one-hot codes, and normalizing continuous value features; and then splicing all the coded features, and performing dimensionality mapping by using a fully-connected neural network to obtain the low-dimensional vector representation of the auxiliary information.
7. The method of claim 5, wherein in step 5, for a question i, there is its corresponding question node q in the graph GiLow dimensional vector characterization ofqiCorresponding knowledge point node siLow dimensional vector characterization ofAnd low-dimensional vector characterization of auxiliary information for problem iThen, a multiplicative neural network is used for feature interaction, and the specific steps are as follows:
Step 5.2, toPerforming pairwise interaction by using inner product operation to obtain a secondary characteristic interaction information matrix
step 5.4, useAnd as the input of the fully-connected neural network, predicting the average answer probability of each question by using an activation function, wherein the average answer probability reflects the difficulty of the question, and using the real question average answer probability to constrain and learn the parameters of the multiplicative neural network.
9. The method of claim 1, wherein in steps 3 and 5, the network parameters are trained end-to-end by a gradient descent method.
10. The method of claim 1, wherein the low dimensional representation of the question obtained in step 5 is used as an input to an existing deep knowledge tracking method to predict at each time step whether the student will answer the question.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110557176.7A CN113282723A (en) | 2021-05-21 | 2021-05-21 | Deep knowledge tracking pre-training method based on graph neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110557176.7A CN113282723A (en) | 2021-05-21 | 2021-05-21 | Deep knowledge tracking pre-training method based on graph neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113282723A true CN113282723A (en) | 2021-08-20 |
Family
ID=77280755
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110557176.7A Pending CN113282723A (en) | 2021-05-21 | 2021-05-21 | Deep knowledge tracking pre-training method based on graph neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113282723A (en) |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107273490A (en) * | 2017-06-14 | 2017-10-20 | 北京工业大学 | A kind of combination mistake topic recommendation method of knowledge based collection of illustrative plates |
CN109639469A (en) * | 2018-11-30 | 2019-04-16 | 中国科学技术大学 | A kind of sparse net with attributes characterizing method of combination learning and system |
CN109829057A (en) * | 2019-01-11 | 2019-05-31 | 中山大学 | A kind of knowledge mapping Entity Semantics spatial embedding method based on figure second order similitude |
CN110362723A (en) * | 2019-05-31 | 2019-10-22 | 平安国际智慧城市科技股份有限公司 | A kind of topic character representation method, apparatus and storage medium |
US20190333400A1 (en) * | 2018-04-27 | 2019-10-31 | Adobe Inc. | Personalized e-learning using a deep-learning-based knowledge tracing and hint-taking propensity model |
CN111159419A (en) * | 2019-12-09 | 2020-05-15 | 浙江师范大学 | Knowledge tracking data processing method, system and storage medium based on graph convolution |
CN111538868A (en) * | 2020-04-28 | 2020-08-14 | 中国科学技术大学 | Knowledge tracking method and exercise recommendation method |
CN111813921A (en) * | 2020-08-20 | 2020-10-23 | 浙江学海教育科技有限公司 | Topic recommendation method, electronic device and computer-readable storage medium |
CN112001536A (en) * | 2020-08-12 | 2020-11-27 | 武汉青忆辰科技有限公司 | High-precision finding method for minimal sample of mathematical capability point defect of primary and secondary schools based on machine learning |
CN112085168A (en) * | 2020-09-11 | 2020-12-15 | 浙江工商大学 | Knowledge tracking method and system based on dynamic key value gating circulation network |
CN112257966A (en) * | 2020-12-18 | 2021-01-22 | 北京世纪好未来教育科技有限公司 | Model processing method and device, electronic equipment and storage medium |
CN112671716A (en) * | 2020-12-03 | 2021-04-16 | 中国电子科技网络信息安全有限公司 | Vulnerability knowledge mining method and system based on map |
-
2021
- 2021-05-21 CN CN202110557176.7A patent/CN113282723A/en active Pending
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107273490A (en) * | 2017-06-14 | 2017-10-20 | 北京工业大学 | A kind of combination mistake topic recommendation method of knowledge based collection of illustrative plates |
US20190333400A1 (en) * | 2018-04-27 | 2019-10-31 | Adobe Inc. | Personalized e-learning using a deep-learning-based knowledge tracing and hint-taking propensity model |
CN109639469A (en) * | 2018-11-30 | 2019-04-16 | 中国科学技术大学 | A kind of sparse net with attributes characterizing method of combination learning and system |
CN109829057A (en) * | 2019-01-11 | 2019-05-31 | 中山大学 | A kind of knowledge mapping Entity Semantics spatial embedding method based on figure second order similitude |
CN110362723A (en) * | 2019-05-31 | 2019-10-22 | 平安国际智慧城市科技股份有限公司 | A kind of topic character representation method, apparatus and storage medium |
CN111159419A (en) * | 2019-12-09 | 2020-05-15 | 浙江师范大学 | Knowledge tracking data processing method, system and storage medium based on graph convolution |
CN111538868A (en) * | 2020-04-28 | 2020-08-14 | 中国科学技术大学 | Knowledge tracking method and exercise recommendation method |
CN112001536A (en) * | 2020-08-12 | 2020-11-27 | 武汉青忆辰科技有限公司 | High-precision finding method for minimal sample of mathematical capability point defect of primary and secondary schools based on machine learning |
CN111813921A (en) * | 2020-08-20 | 2020-10-23 | 浙江学海教育科技有限公司 | Topic recommendation method, electronic device and computer-readable storage medium |
CN112085168A (en) * | 2020-09-11 | 2020-12-15 | 浙江工商大学 | Knowledge tracking method and system based on dynamic key value gating circulation network |
CN112671716A (en) * | 2020-12-03 | 2021-04-16 | 中国电子科技网络信息安全有限公司 | Vulnerability knowledge mining method and system based on map |
CN112257966A (en) * | 2020-12-18 | 2021-01-22 | 北京世纪好未来教育科技有限公司 | Model processing method and device, electronic equipment and storage medium |
Non-Patent Citations (8)
Title |
---|
余传明等: "基于深度学习的领域知识对齐模型研究:知识网络视角", 《情报学报》 * |
傅国绩: "基于事件的异质信息网络表征学习", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 * |
刘恒宇等: "知识追踪综述", 《华东师范大学学报(自然科学版)》 * |
周晓旭等: "网络顶点表示学习方法", 《华东师范大学学报(自然科学版)》 * |
曲良: "基于MetaGNN的异质信息网络表征学习", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 * |
田满鑫等: "一种基于实体时间敏感度的知识表示方法", 《软件工程》 * |
郭崇慧等: "一种基于集成学习的试题多知识点标注方法", 《运筹与管理》 * |
马骁睿等: "一种结合深度知识追踪的个性化习题推荐方法", 《小型微型计算机系统》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110264091B (en) | Student Cognitive Diagnosis Method | |
Chang et al. | A Bayes net toolkit for student modeling in intelligent tutoring systems | |
CN111753054B (en) | Machine reading inference method based on graph neural network | |
CN111274800A (en) | Inference type reading understanding method based on relational graph convolution network | |
CN113344053B (en) | Knowledge tracking method based on examination question different composition representation and learner embedding | |
CN112116092A (en) | Interpretable knowledge level tracking method, system and storage medium | |
CN110377707B (en) | Cognitive diagnosis method based on depth item reaction theory | |
Chaplot et al. | Learning cognitive models using neural networks | |
CN114969298A (en) | Video question-answering method based on cross-modal heterogeneous graph neural network | |
CN111553166A (en) | Scene cognition calculation-based online learner dynamic model prediction method | |
CN115329096A (en) | Interactive knowledge tracking method based on graph neural network | |
CN115546196A (en) | Knowledge distillation-based lightweight remote sensing image change detection method | |
CN113591988A (en) | Knowledge cognitive structure analysis method, system, computer equipment, medium and terminal | |
CN114328943A (en) | Question answering method, device, equipment and storage medium based on knowledge graph | |
CN109934350B (en) | Method, device and platform for realizing one-question multi-solution of mathematical questions | |
CN113282723A (en) | Deep knowledge tracking pre-training method based on graph neural network | |
CN114117033B (en) | Knowledge tracking method and system | |
CN112256858B (en) | Double-convolution knowledge tracking method and system fusing question mode and answer result | |
CN116431821A (en) | Knowledge graph completion method and question-answering system based on common sense perception | |
CN113360669B (en) | Knowledge tracking method based on gating graph convolution time sequence neural network | |
CN114911930A (en) | Global and local complementary bidirectional attention video question-answering method and system | |
Liu et al. | A modeling method based on bayesian networks in intelligent tutoring system | |
CN114880443B (en) | Problem generation method, device, computer equipment and storage medium | |
CN116680502B (en) | Intelligent solving method, system, equipment and storage medium for mathematics application questions | |
CN116502713B (en) | Knowledge tracking method for enhancing topic similarity embedding based on weighted element path |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210820 |
|
RJ01 | Rejection of invention patent application after publication |