Deep knowledge tracking method fusing multiple features and learning ability

Through in-depth knowledge tracking methods that integrate multiple features and learning abilities, we build an embedding layer, learning layer and forgetting layer, deal with knowledge points-exercise characteristics and answering ability characteristics, optimize the model to predict the answering situation, and solve the problem of existing models that are difficult to simulate human learning. The process and the problem of high reliability achieve more accurate learning performance prediction.

CN118982444AInactive Publication Date: 2024-11-19LANZHOU UNIVERSITY OF TECHNOLOGY +1

Patent Information

Application Number
CN202411013382.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-26
Publication Date
2024-11-19
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing knowledge tracking model is difficult to truly simulate the human learning process, and it is highly reliable in the real learning environment, so it is impossible to effectively consider factors such as the learning environment, learning ability, memory and attention.

Method used

Using a deep knowledge tracking method that integrates multiple features and learning abilities, we use multiple feature vectors related to students' learning abilities to build a knowledge tracking model, including an embedding layer, a learning layer, a forgetting layer and a prediction layer, and use long and short-term memory. Neural network processes knowledge points - exercise characteristics and learners' answering ability characteristics, and optimizes the model to predict students' answering situation.

Benefits of technology

The reliability of the knowledge tracking model is improved, so that it can predict learners' learning performance more accurately and simulate the complexity of the human learning process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982444A_ABST
    Figure CN118982444A_ABST
Patent Text Reader

Abstract

The invention relates to a deep knowledge tracking method fusing multiple features and learning ability, and the method comprises the steps: determining a plurality of feature vectors related to the learning ability of a student, and constructing a knowledge tracking model according to all feature vectors; inputting the current knowledge state of the student into a knowledge tracking model, and enabling the knowledge tracking model to output an answer condition prediction result of the student; and obtaining an actual answer condition result of the student, and constructing a target function according to the answer condition prediction result and the actual answer condition result to optimize the knowledge tracking model. According to the technical scheme, multiple features and the learning ability of the learner are fused, so that the knowledge tracking model can accurately predict the learning performance of the learner, and the knowledge tracking model has better reliability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of knowledge tracking technology, and in particular to a deep knowledge tracking method integrating multiple features and learning capabilities. Background Art

[0002] With the continuous advancement of information technology, people's learning methods have also changed. Online education has gradually become an important way in the field of education. It not only breaks through the time and place restrictions of traditional education, but also gives learners more autonomy and choice. Through online education, learners can choose courses according to their needs and interests, arrange learning time and place more flexibly, and thus better improve their learning results. However, with the rise and development of online education, how to use massive user learning data combined with artificial intelligence technology to promote the development of education and meet the personalized needs of learners is a problem that scholars need to explore for a long time. Among them, knowledge tracking is one of the current research hotspots. Knowledge tracking is to deeply explore the learner's learning trajectory, analyze the learner's interaction sequence with the exercises during the learning process, accurately track the learner's knowledge mastery status, and predict the probability of the learner answering the question correctly at the next moment.

[0003] As a time series task in educational data mining, knowledge tracing, combined with the powerful feature extraction and modeling capabilities of deep learning models, has a unique advantage in processing sequential tasks. Therefore, the knowledge tracing model based on deep learning has been widely used in the field of education. However,

[0004] Most existing knowledge tracking models only use students' answer information to build knowledge tracking models for simple learning scenarios. However, in a real learning environment, human learning and cognitive processes are affected by factors such as the learning environment, one's own learning ability, memory, attention, and the difficulty of knowledge points. Existing knowledge tracking models are difficult to truly simulate the human learning process and have low reliability. Summary of the invention

[0005] In order to at least to some extent overcome the problem that the knowledge tracking model in the related art is difficult to truly simulate the human learning process and has low reliability, the present application provides a deep knowledge tracking method that integrates multiple features and learning capabilities.

[0006] The scheme of this application is as follows:

[0007] A deep knowledge tracking method that integrates multiple features and learning capabilities, including:

[0008] Determine multiple eigenvectors related to students' learning abilities, and construct a knowledge tracking model based on all eigenvectors;

[0009] Inputting the student's current knowledge status into the knowledge tracking model, so that the knowledge tracking model outputs the predicted result of the student's answering situation;

[0010] The actual results of the students' answers are obtained, and an objective function is constructed according to the predicted results of the answers and the actual results of the answers to optimize the knowledge tracking model.

[0011] Preferably, constructing a knowledge tracking model includes:

[0012] Construct an embedding layer, obtain the student's learning process data in the embedding layer, and determine the basic learning unit vector corresponding to the learning process data;

[0013] Construct a learning layer, and obtain the relevant knowledge state vector at the previous moment based on the knowledge state output vector at the previous moment and the semantic representation vector of the knowledge points contained in the current exercise;

[0014] In the learning layer, a learning vector is obtained according to the basic learning unit vector at the current moment, the basic learning unit vector at the previous moment, the time interval vector between two learnings, and the relevant knowledge state vector at the previous moment;

[0015] The learning vector is subjected to a first activation function to obtain an initial learning gain vector;

[0016] The learning vector is subjected to a second activation function to obtain a learning gate vector;

[0017] According to the initial learning gain vector and the learning gate vector, an actual learning gain vector is obtained;

[0018] Multiply the actual learning gain vector by the exercise knowledge point vector to obtain the relevant learning gain vector;

[0019] Construct a forgetting layer, in which a forgetting erasure vector is obtained according to the knowledge state vector at the previous moment, the actual learning gain vector and the time interval vector between two learnings through the action of the second activation function;

[0020] In the forgetting layer, according to the knowledge state vector at the previous moment, the actual learning gain vector and the time interval vector between two learnings, a forgetting update vector is obtained through the action of the first activation function;

[0021] Obtaining a forgetting vector according to the forgetting erasure vector and the forgetting update vector;

[0022] Obtaining the student's current knowledge state according to the forgetting vector, the knowledge state vector at the previous moment and the related learning gain vector;

[0023] Construct a prediction layer so that the prediction layer can output the predicted results of students' answers based on their current knowledge status.

[0024] Preferably, the basic learning unit vector includes: a knowledge point-exercise feature vector and a learner's answering ability feature vector;

[0025] The knowledge point-exercise feature vector includes: a knowledge point number feature vector, an exercise number feature vector and a knowledge point difficulty feature vector;

[0026] The learner's ability to answer questions feature vector includes: the answer result vector of the knowledge point corresponding to the question, the time used to answer the question corresponding to the knowledge point, and the number of answer prompts;

[0027] The knowledge point-exercise feature vector and the learner's question-answering ability feature vector are input into a long short-term memory neural network for processing to obtain the basic learning unit vector.

[0028] Preferably, the knowledge point-exercise feature vector and the learner's question-answering ability feature vector are input into a long short-term memory neural network for processing, including:

[0029]

[0030] Among them, eu t represents the knowledge point-exercise feature vector; e t represents the knowledge point number feature vector; c t Represents the characteristic vector of exercise numbers; df t Indicates the knowledge point difficulty feature vector; pu t represents the learner's ability to answer questions; a t Indicates the answer result vector of the knowledge point corresponding to the question; at t Indicates the time used to answer the corresponding knowledge point question; ht t Indicates the number of prompts for answering questions; l t Represents the basic learning unit vector; W1 T 、W2 T denote the first weight and the second weight respectively, b1 and b2 denote the first bias term and the second bias term respectively;

[0031] in, N represents the total number of times a question has been answered, P i Indicates the score for answering a question correctly.

[0032] Preferably, the relevant knowledge state vector at the previous moment is obtained according to the knowledge state output vector at the previous moment and the semantic representation vector of the knowledge point included in the current exercise, including:

[0033]

[0034] in, Represents the relevant knowledge state vector at the previous moment; represents the semantic representation vector of the knowledge points contained in the current exercise, h t-1 Represents the knowledge state vector at the previous moment;

[0035] In the learning layer, the learning vector is obtained according to the basic learning unit vector at the current moment, the basic learning unit vector at the previous moment, the time interval vector between two learnings, and the relevant knowledge state vector at the previous moment, including:

[0036]

[0037] Among them, L t Represents the learning vector.

[0038] Preferably, determining the relevant learning gain vector of the exercise knowledge point according to the learning vector includes:

[0039] The learning vector is subjected to the first activation function to obtain an initial learning gain vector:

[0040]

[0041] Among them, lg t represents the initial learning gain vector; tanh represents the first activation function; W3 T represents the third weight; b3 represents the third bias term;

[0042] The learning vector is subjected to the second activation function to obtain the learning gate vector:

[0043]

[0044] in, represents the learning gate vector; σ represents the second activation function; W4 T represents the fourth weight; b4 represents the fourth bias term;

[0045] According to the initial learning gain vector and the learning gate vector, the actual learning gain vector is obtained:

[0046]

[0047] Among them, LG t represents the actual learning gain vector;

[0048] Multiply the actual learning gain vector by the exercise knowledge point vector to get the relevant learning gain vector:

[0049]

[0050] Among them, RLG t Represents the relevant learning gain vector.

[0051] Preferably, in the forgetting layer, according to the knowledge state vector at the previous moment, the actual learning gain vector and the time interval vector between two learnings, a forgetting erasure vector is obtained through the action of a second activation function, including:

[0052]

[0053] In the forgetting layer, according to the knowledge state vector at the previous moment, the actual learning gain vector and the time interval vector between two learnings, after the first activation function, the forgetting update vector is obtained, including:

[0054]

[0055] Obtaining a forgetting vector according to the forgetting erasure vector and the forgetting update vector includes:

[0056] F t =(1-Fe t )(1+Fa t );

[0057] Among them, Fe t represents the forgotten erasure vector; Fa t represents the forgetting update vector; F t represents the forgetting vector; W5 T represents the fifth weight; b5 represents the fifth bias term; W6 T represents the sixth weight; b6 represents the sixth bias term;

[0058] According to the forgetting vector, the knowledge state vector at the previous moment and the related learning gain vector, the current knowledge state of the student is obtained, including:

[0059] h t =RLG t +F t ·h t-1 ;

[0060] Among them, h t Indicates the student's current state of knowledge.

[0061] Preferably, the prediction layer outputs the prediction result of the student's answer according to the student's current knowledge status, including:

[0062]

[0063] Among them, p t+1 Indicates the answer to Exercise e t+1 The predicted results of students’ answers, p t+1 The output range is (0, 1), p t+1 >0.5 indicates predicted problem et+1 Answer correctly, otherwise it means predicting exercise e t+1 Wrong answer; W7 T represents the seventh weight; b7 represents the seventh bias term.

[0064] Preferably, constructing an objective function according to the predicted answer result and the actual answer result includes:

[0065]

[0066] Among them, p t Indicates the answer to Exercise e t The predicted results of students’ answers; a t Indicates the answer to Exercise e t The actual results of the students’ answers; θ represents all parameters; λ(θ) represents the regularization hyperparameter.

[0067] The technical solution provided by this application may include the following beneficial effects: The deep knowledge tracking method that integrates multiple features and learning ability in this application includes: determining multiple feature vectors related to students' learning ability, and constructing a knowledge tracking model based on all feature vectors; inputting students' current knowledge status into the knowledge tracking model, so that the knowledge tracking model outputs the predicted results of students' answering; obtaining students' actual answering results, and constructing an objective function based on the predicted results and actual results of answering to optimize the knowledge tracking model. The technical solution in this application integrates multiple features and learners' learning ability, so that the knowledge tracking model can more accurately predict learners' learning performance, and the knowledge tracking model has better reliability.

[0068] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0070] Figure 1 It is a flowchart of a deep knowledge tracking method integrating multiple features and learning capabilities provided by an embodiment of the present application;

[0071] Figure 2 It is a schematic diagram of the structure of a knowledge tracking model in a deep knowledge tracking method that integrates multiple features and learning capabilities provided by an embodiment of the present application. DETAILED DESCRIPTION

[0072] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0073] Figure 1 This is a flowchart of a deep knowledge tracking method that integrates multiple features and learning capabilities provided by an embodiment of the present application, referring to Figure 1 , a deep knowledge tracking method that integrates multiple features and learning capabilities, including:

[0074] S1. Determine multiple feature vectors related to students’ learning abilities and construct a knowledge tracking model based on all feature vectors;

[0075] S2, inputting the student's current knowledge status into the knowledge tracking model, so that the knowledge tracking model outputs the prediction result of the student's answering situation;

[0076] S3. Obtain the actual results of students' answers, and build an objective function based on the predicted results and the actual results to optimize the knowledge tracking model.

[0077] It should be noted that, refer to Figure 2 ,The knowledge tracking model includes: embedding layer, learning layer, forgetting layer and prediction layer.

[0078] Based on this, a knowledge tracking model is constructed, including:

[0079] Construct an embedding layer, obtain the student's learning process data in the embedding layer, and determine the basic learning unit vector corresponding to the learning process data;

[0080] Construct a learning layer, and obtain the relevant knowledge state vector at the previous moment based on the knowledge state output vector at the previous moment and the semantic representation vector of the knowledge points contained in the current exercise;

[0081] In the learning layer, a learning vector is obtained according to the basic learning unit vector at the current moment, the basic learning unit vector at the previous moment, the time interval vector between two learnings, and the relevant knowledge state vector at the previous moment;

[0082] The learning vector is subjected to the first activation function to obtain an initial learning gain vector;

[0083] The learning vector is subjected to the second activation function to obtain the learning gate vector;

[0084] According to the initial learning gain vector and the learning gate vector, an actual learning gain vector is obtained;

[0085] Multiply the actual learning gain vector by the exercise knowledge point vector to obtain the relevant learning gain vector;

[0086] Construct a forgetting layer, in which a forgetting erasure vector is obtained according to the knowledge state vector at the previous moment, the actual learning gain vector and the time interval vector between two learnings through the action of the second activation function;

[0087] In the forgetting layer, according to the knowledge state vector at the previous moment, the actual learning gain vector and the time interval vector between two learnings, a forgetting update vector is obtained through the action of the first activation function;

[0088] Obtaining a forgetting vector according to the forgetting erasure vector and the forgetting update vector;

[0089] According to the forgetting vector, the knowledge state vector of the previous moment and the related learning gain vector, the student's current knowledge state is obtained;

[0090] Construct a prediction layer so that the prediction layer can output the predicted results of students' answers based on their current knowledge status.

[0091] Specifically, the basic learning unit vector includes: knowledge point-exercise feature vector and learner answering ability feature vector;

[0092] The knowledge point-exercise feature vector includes: knowledge point number feature vector, exercise number feature vector and knowledge point difficulty feature vector;

[0093] The learner's ability to answer questions feature vector includes: the answer result vector of the knowledge point corresponding to the question, the time used to answer the question corresponding to the knowledge point, and the number of answer prompts;

[0094] The knowledge point-exercise feature vector and the learner's answering ability feature vector are input into the long short-term memory neural network for processing to obtain the basic learning unit vector.

[0095] The knowledge point-exercise feature vector and the learner's answering ability feature vector are input into the long short-term memory neural network for processing, including:

[0096]

[0097] Among them, eu t represents the knowledge point-exercise feature vector; e t represents the knowledge point number feature vector; c t Represents the characteristic vector of exercise numbers; df t Indicates the knowledge point difficulty feature vector; pu trepresents the learner's ability to answer questions; a t Indicates the answer result vector of the knowledge point corresponding to the question; at t Indicates the time used to answer the corresponding knowledge point question; ht t Indicates the number of prompts for answering questions; l t Represents the basic learning unit vector; W1 T 、W2 T denote the first weight and the second weight respectively, b1 and b2 denote the first bias term and the second bias term respectively;

[0098] in, N represents the total number of times a question has been answered, P i Indicates the score for answering a question correctly.

[0099] Specifically, according to the knowledge state output vector at the previous moment and the semantic representation vector of the knowledge point contained in the current exercise, the relevant knowledge state vector at the previous moment is obtained, including:

[0100]

[0101] in, Represents the relevant knowledge state vector at the previous moment; represents the semantic representation vector of the knowledge points contained in the current exercise, h t-1 Represents the knowledge state vector at the previous moment;

[0102] In the learning layer, the learning vector is obtained according to the basic learning unit vector at the current moment, the basic learning unit vector at the previous moment, the time interval vector between two learnings, and the relevant knowledge state vector at the previous moment, including:

[0103]

[0104] Among them, L t Represents the learning vector.

[0105] It should be noted that With h t-1 It is a different concept. According to h t-1 and The obtained relevant knowledge state vector refers to the correlation with the semantic representation vector of the knowledge point contained in the current exercise.

[0106] Specifically, the relevant learning gain vector of the exercise knowledge point is determined according to the learning vector, including:

[0107] The learning vector is passed through the first activation function to obtain the initial learning gain vector:

[0108]

[0109] Among them, lg t represents the initial learning gain vector; tanh represents the first activation function; W3 T represents the third weight; b3 represents the third bias term;

[0110] The learning vector is passed through the second activation function to obtain the learning gate vector:

[0111]

[0112] in, represents the learning gate vector; σ represents the second activation function; W4 T represents the fourth weight; b4 represents the fourth bias term;

[0113] According to the initial learning gain vector and the learning gate vector, the actual learning gain vector is obtained:

[0114]

[0115] Among them, LG t represents the actual learning gain vector;

[0116] Multiply the actual learning gain vector by the exercise knowledge point vector to get the relevant learning gain vector:

[0117]

[0118] Among them, RLG t Represents the relevant learning gain vector.

[0119] It should be noted that the main function of the learning gate vector is to control students' learning absorption ability.

[0120] RLG t Represents the relevant learning gain vector, which also refers to the correlation with the semantic representation vector of the knowledge point contained in the current exercise.

[0121] Specifically, in the forgetting layer, according to the knowledge state vector at the previous moment, the actual learning gain vector and the time interval vector between two learnings, after the second activation function is applied, a forgetting erasure vector is obtained, including:

[0122]

[0123] In the forgetting layer, according to the knowledge state vector at the previous moment, the actual learning gain vector and the time interval vector between two learnings, after the first activation function, the forgetting update vector is obtained, including:

[0124]

[0125] Obtaining a forgetting vector according to the forgetting erasure vector and the forgetting update vector includes:

[0126] F t =(1-Fe t )(1+Fa t );

[0127] Among them, Fe t represents the forgotten erasure vector; Fa t represents the forgetting update vector; F t represents the forgetting vector; W5 T represents the fifth weight; b5 represents the fifth bias term; W6 T represents the sixth weight; b6 represents the sixth bias term;

[0128] According to the forgetting vector, the knowledge state vector of the previous moment and the related learning gain vector, the student’s current knowledge state is obtained, including:

[0129] h t =RLG t +F t ·h t-1 ;

[0130] Among them, h t Indicates the student's current state of knowledge.

[0131] It should be noted that according to the Ebbinghaus forgetting curve theory, the number of remembered knowledge points decays exponentially over time. Although forgetting is inevitable, memory can be consolidated and prevented through regular review and repeated learning. In order to model complex forgetting defects, the knowledge tracking model in this solution designs a forgetting layer.

[0132] By multiplying the forgetting vector with the knowledge state vector at the previous moment, the impact of forgetting can be eliminated.

[0133] Specifically, the prediction layer outputs the predicted results of the students' answers according to their current knowledge status, including:

[0134]

[0135] Among them, p t+1 Indicates the answer to Exercise e t+1 The predicted results of students’ answers, p t+1 The output range is (0, 1), p t+1 >0.5 indicates predicted problem e t+1 Answer correctly, otherwise it means predicting exercise e t+1 Wrong answer; W7 T represents the seventh weight; b7 represents the seventh bias term.

[0136] It should be noted that in this embodiment, the output threshold of the prediction layer is set to 0.5, p t+1 The output range is (0, 1), p t+1 >0.5 indicates predicted problem e t+1 Answer correctly, otherwise it means predicting exercise e t+1 Wrong answer.

[0137] Specifically, the objective function is constructed based on the predicted answer results and the actual answer results, including:

[0138]

[0139] Among them, p t Indicates the answer to Exercise e t The predicted results of students’ answers; a t Indicates the answer to Exercise e t The actual results of the students’ answers; θ represents all parameters; λ(θ) represents the regularization hyperparameter.

[0140] In practice, the Adam optimizer is used to minimize the objective function on mini-batch processing.

[0141] This technical solution integrates multiple features and learners' learning abilities, so that the knowledge tracking model can more accurately predict learners' learning performance, and the knowledge tracking model has better reliability.

[0142] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0143] It should be noted that, in the description of this application, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of this application, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0144] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0145] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0146] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0147] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0148] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0149] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0150] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A deep knowledge tracking method integrating multiple features and learning capabilities, characterized in that: include: Determine multiple eigenvectors related to students' learning abilities, and construct a knowledge tracking model based on all eigenvectors; Inputting the student's current knowledge status into the knowledge tracking model, so that the knowledge tracking model outputs the predicted result of the student's answering situation; The actual results of the students' answers are obtained, and an objective function is constructed according to the predicted results of the answers and the actual results of the answers to optimize the knowledge tracking model.

2. The method according to claim 1, characterized in that: Build a knowledge tracking model, including: Construct an embedding layer, obtain the student's learning process data in the embedding layer, and determine the basic learning unit vector corresponding to the learning process data; Construct a learning layer, and obtain the relevant knowledge state vector at the previous moment based on the knowledge state output vector at the previous moment and the semantic representation vector of the knowledge points contained in the current exercise; In the learning layer, a learning vector is obtained according to the basic learning unit vector at the current moment, the basic learning unit vector at the previous moment, the time interval vector between two learnings, and the relevant knowledge state vector at the previous moment; The learning vector is subjected to a first activation function to obtain an initial learning gain vector; The learning vector is subjected to a second activation function to obtain a learning gate vector; According to the initial learning gain vector and the learning gate vector, an actual learning gain vector is obtained; Multiply the actual learning gain vector by the exercise knowledge point vector to obtain the relevant learning gain vector; Construct a forgetting layer, in which a forgetting erasure vector is obtained according to the knowledge state vector at the previous moment, the actual learning gain vector and the time interval vector between two learnings through the action of the second activation function; In the forgetting layer, according to the knowledge state vector at the previous moment, the actual learning gain vector and the time interval vector between two learnings, a forgetting update vector is obtained through the action of the first activation function; Obtaining a forgetting vector according to the forgetting erasure vector and the forgetting update vector; Obtaining the student's current knowledge state according to the forgetting vector, the knowledge state vector at the previous moment and the related learning gain vector; Construct a prediction layer so that the prediction layer can output the predicted results of students' answers based on their current knowledge status.

3. The method according to claim 2, characterized in that The basic learning unit vector includes: a knowledge point-exercise feature vector and a learner's answering ability feature vector; The knowledge point-exercise feature vector includes: a knowledge point number feature vector, an exercise number feature vector and a knowledge point difficulty feature vector; The learner's ability to answer questions feature vector includes: the answer result vector of the knowledge point corresponding to the question, the time used to answer the question corresponding to the knowledge point, and the number of answer prompts; The knowledge point-exercise feature vector and the learner's question-answering ability feature vector are input into a long short-term memory neural network for processing to obtain the basic learning unit vector.

4. The method according to claim 3, characterized in that Inputting the knowledge point-exercise feature vector and the learner's answering ability feature vector into a long short-term memory neural network for processing, including: Among them, eu t represents the knowledge point-exercise feature vector; e t represents the knowledge point number feature vector; c t Represents the characteristic vector of exercise numbers; df t Indicates the knowledge point difficulty feature vector; pu t represents the learner's ability to answer questions; a t Indicates the answer result vector of the knowledge point corresponding to the question; at t Indicates the time used to answer the corresponding knowledge point question; ht t Indicates the number of prompts for answering questions; l t Represents the basic learning unit vector; W1 T 、W2 T denote the first weight and the second weight respectively, b1 and b2 denote the first bias term and the second bias term respectively; in, N represents the total number of times a question has been answered, P i Indicates the score for answering a question correctly.

5. The method according to claim 4, characterized in that According to the knowledge state output vector at the previous moment and the semantic representation vector of the knowledge points contained in the current exercise, the relevant knowledge state vector at the previous moment is obtained, including: in, Represents the relevant knowledge state vector at the previous moment; represents the semantic representation vector of the knowledge points contained in the current exercise, h t-1 Represents the knowledge state vector at the previous moment; In the learning layer, the learning vector is obtained according to the basic learning unit vector at the current moment, the basic learning unit vector at the previous moment, the time interval vector between two learnings, and the relevant knowledge state vector at the previous moment, including: Among them, L t Represents the learning vector.

6. The method according to claim 5, characterized in that Determining a relevant learning gain vector of the exercise knowledge point according to the learning vector includes: The learning vector is subjected to the first activation function to obtain an initial learning gain vector: Among them, lg t represents the initial learning gain vector; tanh represents the first activation function; W3 T represents the third weight; b3 represents the third bias term; The learning vector is subjected to the second activation function to obtain the learning gate vector: in, represents the learning gate vector; σ represents the second activation function; W4 T represents the fourth weight; b4 represents the fourth bias term; According to the initial learning gain vector and the learning gate vector, the actual learning gain vector is obtained: Among them, LG t represents the actual learning gain vector; Multiply the actual learning gain vector by the exercise knowledge point vector to get the relevant learning gain vector: Among them, RLG t Represents the relevant learning gain vector.

7. The method according to claim 6, characterized in that In the forgetting layer, according to the knowledge state vector at the previous moment, the actual learning gain vector and the time interval vector between two learnings, after the second activation function, the forgetting erasure vector is obtained, including: In the forgetting layer, according to the knowledge state vector at the previous moment, the actual learning gain vector and the time interval vector between two learnings, after the first activation function, the forgetting update vector is obtained, including: Obtaining a forgetting vector according to the forgetting erasure vector and the forgetting update vector includes: <h2 style=";text-align:left;direction:ltr">F<h2 style=";text-align:left;direction:ltr"> t <h2 style=";text-align:left;direction:ltr"> (1-Fe<h2 style=";text-align:left;direction:ltr"> t <h2 style=";text-align:left;direction:ltr"> (1+Fa)<h2 style=";text-align:left;direction:ltr"> t <h2 style=";text-align:left;direction:ltr"> ); Among them, Fe t represents the forgotten erasure vector; Fa t represents the forgetting update vector; F t Represents the forgetting vector; W5 T represents the fifth weight; b5 represents the fifth bias term; W6 T represents the sixth weight; b6 represents the sixth bias term; According to the forgetting vector, the knowledge state vector at the previous moment and the related learning gain vector, the current knowledge state of the student is obtained, including: h t =RLG t +F t ·h t-1 ; Among them, h t Indicates the student's current state of knowledge.

8. The method according to claim 7, characterized in that The prediction layer outputs the predicted results of the students' answers based on their current knowledge status, including: Among them, p t+1 Indicates the answer to Exercise e t+1 The predicted results of students’ answers, p t+1 The output range is (0, 1), p t+1 >0.5 indicates predicted problem e t+1 Answer correctly, otherwise it means predicting exercise e t+1 Wrong answer; W7 T represents the seventh weight; b7 represents the seventh bias term.

9. The method according to claim 1, characterized in that: Constructing an objective function according to the predicted answer result and the actual answer result, including: Among them, p t Indicates the answer to Exercise e t The predicted results of students’ answers; a t Indicates the answer to Exercise e t The actual results of the students’ answers; θ represents all parameters; λ(θ) represents the regularization hyperparameter.

Citation Information

Patent Citations

  • Deep embedded knowledge tracking method based on exercise difficulty and student ability

    CN113033808A

  • Knowledge tracking method integrating learning process and difficulty features of question knowledge points

    CN114781710A

  • Student knowledge state analysis method and system based on real factor constraint deep knowledge tracking

    CN117314696A

  • Exercise recommendation method and device, electronic equipment and storage medium

    CN118035428A

  • A unified platform for domain adaptable human behaviour inference

    EP3561815A1

Cited By

  • Student ability portraying method and device, electronic equipment and storage medium

    CN115983119A

  • Double-state knowledge tracking method fusing dynamic domain learning ability

    CN122088640A