Knowledge tracking method based on forgetful attention mechanism, knowledge tracking model, computer system and storage medium
By introducing a forgetting factor with relative position encoding to correct similarity weights, and by fusing question and answer sequence embeddings, the knowledge tracking model is optimized, thus solving the prediction bias problem caused by the forgetting effect and improving prediction accuracy and interpretability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-05
- Publication Date
- 2026-04-10
AI Technical Summary
Existing knowledge tracing methods have failed to effectively model the forgetting effect of students, resulting in prediction bias and insufficient prediction accuracy.
We introduce a forgetting factor based on relative position encoding, correct similarity weights, and integrate question and answer sequence embeddings to optimize the prediction model.
It improves the predictive accuracy and relevance of the knowledge tracing model, and enhances its ability to explain the student learning process.
Smart Images
Figure CN121638508B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of knowledge tracing, and particularly relates to a knowledge tracing method based on a forgetting attention mechanism, a knowledge tracing model, a computer system and a storage medium. BACKGROUND
[0002] Knowledge tracing (KT) is one of the core tasks in intelligent education systems, and its goal is to predict the answering performance of a student on future exercises based on the student's historical answering records.
[0003] Traditional knowledge tracing methods mainly include two categories: one is a traditional method based on a probabilistic graphical model, and the other is a model method based on deep learning. However, both have their own limitations: the traditional probabilistic graphical model has certain interpretability, for example, the Rasch model can depict the answering accuracy through the difference between the student ability parameter and the question difficulty parameter, but its prediction performance is limited and it is difficult to capture complex learning processes; while the deep learning model can automatically learn the dynamic changes of the student's knowledge state through the neural network, and the prediction accuracy is significantly improved, but it often has the problem of insufficient interpretability.
[0004] On the other hand, the inventors found in the process of conceiving and implementing the present application that in real learning scenarios, the knowledge memory of a student will gradually decline over time, which is called the “forgetting effect”, but the current mainstream knowledge tracing methods using attention mechanisms generally ignore this effect: on the one hand, traditional knowledge tracing models only rely on historical sequence similarity for prediction when modeling, and fail to effectively model the forgetting process, thereby causing prediction bias; on the other hand, existing models usually only rely on historical learning states when predicting the next question answering performance of a student, and lack explicit introduction of “target exercise” information, thereby causing the model to be difficult to distinguish between different questions, reducing the prediction specificity and accuracy.
[0005] Therefore, the present application proposes a new knowledge tracing method to overcome the prediction bias caused by the “forgetting effect”. SUMMARY
[0006] The main purpose of the present application is to provide a knowledge tracing method based on a forgetting attention mechanism, which aims to solve the problem of how to overcome the knowledge tracing prediction bias caused by the “forgetting effect”.
[0007] To achieve the above purpose, the present application provides a knowledge tracing method based on a forgetting attention mechanism, which comprises:
[0008] obtaining a question sequence embedding representation and an answering sequence embedding representation obtained after embedding representation of student learning record data is obtained;
[0009] The question sequence embedding representation is represented as a corresponding similarity weight matrix, and a forgetting factor generated based on relative position coding is introduced to correct the similarity weight;
[0010] The answer sequence embedding representation is weighted and summed based on the corrected similarity weight to obtain a fusion embedding representation;
[0011] According to the fusion embedding representation and the question sequence embedding representation, an optimization representation is determined to predict the correctness of the student in the next answering based on the optimization representation.
[0012] Optionally, the forgetting factor generated based on relative position coding is introduced to correct the similarity weight, specifically including:
[0013] The sequence length of the question sequence embedding representation is obtained, and a relative position embedding matrix is constructed according to the sequence length ;
[0014] The relative position embedding matrix is subjected to learnable embedding and linear transformation to obtain a forgetting rate :
[0015]
[0016] In the formula, is a learnable parameter, is a Sigmoid function, is a relative position index, is the position index of the question in the sequence at the current time, is the position index of the question in the sequence at the historical time;
[0017] The similarity weight is corrected according to the forgetting rate to obtain the corrected similarity weight :
[0018]
[0019] In the formula, is a normalized exponential function.
[0020] Optionally, the question sequence embedding representation includes a knowledge point embedding representation and a question difficulty embedding representation, and the answer sequence embedding representation includes a knowledge point embedding representation and an answer correctness embedding representation.
[0021] Optionally, the expression of the question sequence embedding representation is:
[0022]
[0023] wherein, represents a latent representation of a knowledge point , wherein, is a one-hot encoding of the knowledge point , and is a learnable mapping matrix; represents a difficulty vector of a question ; represents an interaction vector of a question and a knowledge point , used to depict the difference of different questions under the same knowledge point, represents a feature combination, represents an element-wise multiplication.
[0024] Optionally, the answer sequence embedding representation is expressed as:
[0025]
[0026] wherein, represents a latent representation of a knowledge point , wherein, is a one-hot encoding of the knowledge point , and is a learnable mapping matrix; represents an answer vector, wherein is a one-hot encoding of the correctness of the answer, is a learnable mapping matrix; represents a feature combination, represents an element-wise multiplication.
[0027] Optionally, the determining an optimization representation according to the fusion embedding representation and the question sequence embedding representation specifically comprises:
[0028] the fusion embedding representation and the question sequence embedding representation , sequentially passing through a full connection layer, a residual connection and a normalization processing, to obtain the optimization representation :
[0029]
[0030] wherein, represents a normalization function, represents a residual connection function.
[0031] Optionally, the predicting an answer result of the student in the next question based on the optimization representation specifically comprises:
[0032] when a response prediction instruction is responded to, embedding a current input question sequence into the optimization representation generated at a historical moment The stitching is performed to obtain a predicted vector :
[0033]
[0034] The student's accuracy in the next question is predicted by a fully connected layer and a sigmoid activation function :
[0035]
[0036] wherein, is a learnable parameter, is a sigmoid function.
[0037] In addition, to achieve the above object, the present application also provides a knowledge tracking model, comprising:
[0038] An input and embedding representation module is configured to obtain a question sequence embedding representation and an answer sequence embedding representation obtained by embedding representation of student learning record data;
[0039] A forgetting attention modeling and feature optimization module is configured to calculate a similarity weight matrix corresponding to the question sequence embedding representation, introduce a forgetting factor generated based on relative position coding to correct the similarity weight, and perform weighted summation on the answer sequence embedding representation based on the corrected similarity weight to obtain a fusion embedding representation; and determine an optimized representation according to the fusion embedding representation and the question sequence embedding representation.
[0040] A prediction and output module is configured to predict the student's accuracy in the next question based on the optimized representation.
[0041] In addition, to achieve the above object, the present application also provides a computer system, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the computer program is executed by the processor to implement the steps of the knowledge tracking method based on the forgetting attention mechanism according to any one of the above.
[0042] In addition, to achieve the above object, the present application also provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the knowledge tracking method based on the forgetting attention mechanism according to any one of the above.
[0043] The present application has at least the following beneficial effects:
[0044] By introducing a forgetting factor based on relative position coding to simulate the law of student knowledge decay over time, a more realistic learning scenario is constructed for the answer sequence embedding representation. The answer sequence embedding representation is fused with the question sequence embedding representation to make the prediction more targeted and accurate. BRIEF DESCRIPTION OF DRAWINGS
[0045] Figure 1 A flowchart of the knowledge tracking method based on the forgetting attention mechanism involved in the embodiments of the present application;
[0046] Figure 2 A schematic diagram of the architecture of the knowledge tracking model involved in the embodiments of the present application;
[0047] Figure 3 A schematic diagram of the hardware operating environment of the computer system involved in the embodiments of the present application.
[0048] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION
[0049] In order to better understand the above technical solutions, the exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.
[0050] First embodiment
[0051] Referring to Figure 1 , the present embodiment provides a knowledge tracking method based on a forgetting attention mechanism, the method comprising the following steps:
[0052] Step S10, obtaining the question sequence embedding representation and the answer sequence embedding representation obtained by embedding the student learning record data;
[0053] In the present embodiment, the student learning record data is embedded in advance to obtain the question sequence embedding representation and the answer sequence embedding representation. The execution subject performs the action of obtaining the two embedding representations when performing knowledge tracking.
[0054] In some optional embodiments, the student learning record data is embedded based on the Rasch model.
[0055] Further and optionally, the student learning record data includes three types of data: knowledge points, exercise difficulty, and answer correctness. An exercise embedding representation is obtained from the knowledge point embedding representation and the exercise difficulty embedding representation, and an answer sequence embedding representation is obtained from the knowledge point embedding representation and the answer correctness embedding representation.
[0056] Further and optionally, the expression of the question sequence embedding representation is:
[0057]
[0058] In the formula, represents a potential representation of a knowledge point , wherein, is one-hot encoding of the knowledge point , and is a learnable mapping matrix; represents a difficulty vector of a question ; represents an interaction variation vector of a question and a knowledge point , used to depict the difference of different questions under the same knowledge point, represents feature combination, represents element-wise multiplication.
[0059] Further and optionally, the expression of the answer sequence embedding representation is:
[0060]
[0061] In the formula, represents a potential representation of a knowledge point , wherein, is one-hot encoding of the knowledge point , and is a learnable mapping matrix; represents an answer vector, wherein is one-hot encoding of the answer correctness, is a learnable mapping matrix; represents feature combination, represents element-wise multiplication.
[0062] It should be noted that the sequence embedding representation structure obtained by the above method. This structure maintains the idea of "question difficulty affecting answer performance" in the Rasch model, while enhancing scalability in vector form.
[0063] In step S20, a similarity weight matrix corresponding to the question sequence embedding representation is calculated, and a forgetting factor generated based on relative position coding is introduced to correct the similarity weight.
[0064] After obtaining the two embedding representations, the question sequence embedding representation is selected, the similarity weight corresponding to the single question embedding representation in the question sequence embedding representation is calculated, and the similarity weights are constructed into a similarity weight matrix according to the corresponding order of the question embedding representations in the sequence.
[0065] Exemplarily, the similarity weight is , and the calculation expression is:
[0066]
[0067] , wherein represents the similarity between the question sequence embedding representation at the moment and another question sequence embedding representation , and is a scaling factor.
[0068] In this embodiment, considering that in a real learning scenario, the knowledge memory of a student will gradually decline over time, and the model under this system usually only relies on historical sequence similarity for prediction, failing to effectively model the forgetting process, resulting in prediction deviation. Therefore, a forgetting factor generated based on relative position coding is introduced in this step to simulate the law of student knowledge decline over time, aiming to improve the prediction accuracy and explanation ability of the knowledge tracking model.
[0069] Further and optionally, the forgetting factor generated based on the relative position coding is introduced to correct the similarity weight, and specifically includes:
[0070] Step S21, obtaining the sequence length of the question sequence embedding representation, and constructing a relative position embedding matrix according to the sequence length.
[0071] Step S22, performing learnable embedding and linear transformation on the relative position embedding matrix to obtain a forgetting rate .
[0072]
[0073] , wherein is a learnable parameter, is a Sigmoid function, is a relative position index, is the position index of the current moment question in the sequence, is the position index of the historical moment question in the sequence.
[0074] Step S23, correcting the similarity weight according to the forgetting rate to obtain a corrected similarity weight :
[0075]
[0076] wherein, is a normalized exponential function.
[0077] Step S30, weighting and summing the answer sequence embedding representations based on the revised similarity weights to obtain a fusion embedding representation;
[0078] After obtaining the revised similarity weights, the answer sequence embedding representations are weighted and summed based on the revised similarity weights to obtain a fusion embedding representation.
[0079] Exemplarily, the mathematical expression of the fusion embedding representation is:
[0080]
[0081] in sequence form , and T is the number of fusion embedding representations.
[0082] Step S40, determining an optimization representation according to the fusion embedding representation and the question sequence embedding representation, to predict the correctness of the student in the next answering based on the optimization representation.
[0083] After obtaining the fusion embedding representation, the feature integration and optimization representation are performed with the previously obtained question sequence embedding representation.
[0084] Further and optionally, the fusion embedding representation and the question sequence embedding representation are sequentially subjected to a full connection layer, a residual connection and a normalization process to obtain the optimization representation :
[0085]
[0086] wherein, represents a normalization function, represents a residual connection function.
[0087] After obtaining the optimization representation, the optimization representation is used to predict the correctness of the student in the next answering.
[0088] Further and optionally, the optimization representation is used for prediction, not instantaneously, but called when needed:
[0089] When receiving a prediction instruction input by a user, in response to the prediction instruction, the embedding of the target exercise is explicitly input , and is combined with the optimization representation generated at the historical time The splicing is performed to obtain a predicted vector :
[0090]
[0091] The correct rate of the student in the next answering is predicted through a full connection layer and a Sigmoid activation function :
[0092]
[0093] In the formula, is a learnable parameter, is a Sigmoid function.
[0094] In the technical scheme provided in the embodiment, by introducing a forgetting factor based on relative position coding to simulate the law of student knowledge decay over time, a more realistic learning scenario is constructed for answer sequence embedding representation, the answer sequence embedding representation is fused with the question sequence embedding representation, and the prediction is more targeted and accurate.
[0095] Verification embodiment
[0096] To verify the effectiveness of the knowledge tracking method based on the forgetting attention mechanism provided in the embodiments of the present application, the knowledge tracking model (referred to as Ours) built based on the knowledge tracking method based on the forgetting attention mechanism in the present verification embodiment is used to perform experiments on multiple public data sets, and the hyperparameter search is used to find the optimal hyperparameter configuration to improve the model performance and generalization ability.
[0097] In order to ensure the fairness of comparison, the hyperparameter search is performed on all models (including DKT, DKT+, SAKT and SimpleKT) compared in the present verification embodiment, and the experiments are performed under the optimal configuration of each model, so as to ensure that the comparison result has reference value.
[0098] The experimental results are shown in the following Tables 1 and 2:
[0099] Table 1. AUC comparison result
[0100]
[0101] Table 2. ACC comparison result
[0102]
[0103] In Table 1 and Table 2, Question Level represents an evaluation mode taking a question as a basic prediction object; KC Level represents an evaluation mode taking a knowledge point as a basic prediction object. All-in-One represents that when a question is associated with multiple knowledge points, all related knowledge points are synchronously predicted to avoid the label leakage problem introduced by one-by-one prediction.
[0104] AS2009, AL2005, BD2006, NIPS34, Statics2011, AS2015 respectively represent commonly used knowledge tracking public datasets, wherein AS2009 is an ASSISTments2009 dataset, AL2005 is an Algebra2005 dataset, BD2006 is a Bridge2006 dataset, NIPS34 is a Task3&4 dataset in NeurIPS 2020 Education Challenge, and all of them are widely used for comparative evaluation of knowledge tracking models. POJ is a public dataset for code clone detection task, and is used to evaluate the performance of the model in retrieving semantically similar codes.
[0105] The results show that the AUC (Area Under Curve) and ACC (Accuracy) indicators of the method proposed in the application are better than those of the existing comparative models on most datasets, indicating that the prediction performance and generalization ability are further improved. Compared with the traditional knowledge tracking model based on attention mechanism, the application achieves stable improvement on different sizes and types of datasets, indicating that the proposed mechanism can effectively enhance the expression ability of the model.
[0106] The forgetting attention mechanism introduced in the application improves the utilization of historical information by the model to some extent, so that the prediction result is more in line with the actual learning process of the student. The experimental results show good performance, verifying the effectiveness and applicability.
[0107] As an implementation scheme, Figure 2 The schematic diagram of the architecture of the hardware running environment of the computer system involved in the embodiment scheme of the application.
[0108] As Figure 2As shown, the computer system can include: a processor 1001, such as a CPU, a memory 1005, a user interface 1003, a network interface 1004, and a communication bus 1002. The communication bus 1002 is used to realize the connection communication between these components. The user interface 1003 can include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 can also include a standard wired interface, a wireless interface. The network interface 1004 can optionally include a standard wired interface, a wireless interface (such as a WI-FI interface). The memory 1005 can be a high-speed RAM memory, or a stable memory (non-volatile memory) such as a disk memory. The memory 1005 can also be an independent storage device from the aforementioned processor 1001.
[0109] Those skilled in the art can understand that Figure 2 The computer system architecture shown in the embodiment does not constitute a limitation on the computer system, and can include more or fewer components than shown, or combine certain components, or different component arrangements.
[0110] As Figure 2 As shown, the memory 1005 as a storage medium can include an operating system, a network communication module, a user interface module, and a computer program. The operating system is a program that manages and controls the hardware and software resources of the computer system, and the running of the computer program and other software or programs.
[0111] In Figure 2 In the computer system shown, the user interface 1003 is mainly used to connect the terminal and communicate data with the terminal; the network interface 1004 is mainly used for the background server and communicates data with the background server; and the processor 1001 can be used to call the computer program stored in the memory 1005.
[0112] In the embodiment, the computer system includes: a memory 1005, a processor 1001, and a computer program stored on the memory and executable on the processor, wherein:
[0113] When the processor 1001 calls the computer program stored in the memory 1005, the following operations are performed:
[0114] Obtain the question sequence embedding representation and the answer sequence embedding representation obtained by embedding the student learning record data;
[0115] Calculate the similarity weight matrix corresponding to the question sequence embedding representation, and introduce a forgetting factor generated based on relative position coding to correct the similarity weight;
[0116] weighting and summing the answer sequence embedding representation based on the modified similarity weight to obtain a fusion embedding representation;
[0117] determining an optimization representation based on the fusion embedding representation and the question sequence embedding representation, and predicting the accuracy of the student in the next answering based on the optimization representation.
[0118] In addition, with reference to Figure 3 , the embodiment also proposes a knowledge tracking model, which comprises:
[0119] a question sequence embedding representation and an answer sequence embedding representation obtained after embedding representation of student learning record data are acquired by an input and embedding representation module 100;
[0120] a forgetting attention modeling and feature optimization module 200, configured to calculate a similarity weight matrix corresponding to the question sequence embedding representation, and introduce a forgetting factor generated based on relative position coding to modify the similarity weight; weighting and summing the answer sequence embedding representation based on the modified similarity weight to obtain a fusion embedding representation; and determining an optimization representation based on the fusion embedding representation and the question sequence embedding representation;
[0121] a prediction and output module 300, configured to predict the accuracy of the student in the next answering based on the optimization representation.
[0122] Further and optionally, the knowledge tracking model can further perform the following operations:
[0123] obtaining a sequence length of the question sequence embedding representation, and constructing a relative position embedding matrix based on the sequence length ;
[0124] performing learnable embedding and linear transformation on the relative position embedding matrix to obtain a forgetting ratio :
[0125]
[0126] wherein, is a learnable parameter, is a Sigmoid function, is a relative position index, is a position index of the question in the sequence at the current time, is a position index of the question in the sequence at the historical time;
[0127] the forgetting ratio is used to modify the similarity weight to obtain a modified similarity weight :
[0128]
[0129] wherein, is a normalized exponential function.
[0130] Further and optionally, the knowledge tracing model can further perform the following operations:
[0131] The question sequence embedding representation comprises a knowledge point embedding representation and a question difficulty embedding representation, and the answer sequence embedding representation comprises a knowledge point embedding representation and an answer correctness embedding representation.
[0132] Further and optionally, the knowledge tracing model can further perform the following operations:
[0133] The expression of the question sequence embedding representation is:
[0134]
[0135] wherein, represents a latent representation of a knowledge point , wherein, is a one-hot encoding of the knowledge point , and is a learnable mapping matrix; represents a difficulty vector of a question ; represents an interaction variation vector of a question and a knowledge point , used to depict the difference of different questions under the same knowledge point, represents feature combination, represents element-wise multiplication.
[0136] Further and optionally, the knowledge tracing model can further perform the following operations:
[0137] The expression of the answer sequence embedding representation is:
[0138]
[0139] wherein, represents a latent representation of a knowledge point , wherein, is a one-hot encoding of the knowledge point , and is a learnable mapping matrix; represents an answer vector, wherein is a one-hot encoding of answer correctness, is a learnable mapping matrix; represents feature combination, represents element-wise multiplication.
[0140] Further and optionally, the knowledge tracking model can further perform the following operations:
[0141] concatenate the fusion embedding representation and the question sequence embedding representation , sequentially pass through a fully connected layer, a residual connection and normalization processing, to obtain the optimized representation :
[0142]
[0143] wherein, denotes a normalization function, denotes a residual connection function.
[0144] Further and optionally, the knowledge tracking model can further perform the following operations:
[0145] In response to the prediction instruction, concatenate the current input question sequence embedding representation with the optimized representation generated at the historical time, to obtain a prediction vector :
[0146]
[0147] pass through a fully connected layer and a Sigmoid activation function to predict the accuracy of the student in the next question :
[0148]
[0149] wherein, is a learnable parameter, is a Sigmoid function.
[0150] In addition, it can be understood by those skilled in the art that all or part of the processes in the method of implementing the above embodiments can be completed by a computer program instructing related hardware. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer readable storage medium. The program instructions are executed by at least one processor in a computer system to implement the process steps of the above-mentioned embodiments of the method.
[0151] Therefore, the present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement each step of the knowledge tracking method based on the forgetting attention mechanism as described in the above embodiments.
[0152] The computer readable storage medium can be a U disk, a mobile hard disk, a read-only memory (ROM), a magnetic disk or an optical disk, and the like.
[0153] It should be noted that the storage medium provided by the embodiments of the present application is a storage medium used to implement the method of the embodiments of the present application. Therefore, based on the method introduced in the embodiments of the present application, the specific structure and modification of the storage medium can be understood by those skilled in the art, and therefore will not be described here. Any storage medium used by the method of the embodiments of the present application belongs to the scope of protection of the present application.
[0154] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the present application can adopt a completely hardware embodiment, a completely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes.
[0155] The present application is described with reference to flowcharts and / or block diagrams according to the method, device (system) and computer program product of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in one or more flows and / or blocks.
[0156] These computer program instructions can also be stored in a computer readable storage medium that can guide the computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer readable storage medium produce a product including instruction apparatus, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in one or more flows and / or blocks.
[0157] These computer program instructions can also be loaded into a computer or other programmable data processing device, so that a series of operation steps are performed on the computer or other programmable device to produce a computer implemented process, so that the instructions executed on the computer or other programmable device provide a means for implementing the functions specified in the flowcharts and / or block diagrams.Figure 1 one or more processes and / or functions specified in one or more blocks. Figure 1 one or more blocks.
[0158] It is noted that in the claims the word "comprising" does not exclude not having other parts than those specified in the claim. The word "a" or "an" preceding the citation of a generic term does not exclude a plurality of those generic terms. This application can be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the unitary claim, the features of the different embodiments can be combined with each other if not incompatible. The use of the word "a" or "an" does not exclude the plural, and the use of "comprising" does not exclude other steps or elements not listed. The word "comprising" other opening terms such as "consisting of" should be construed as specifying steps or components which are in the embodiments "only". Recitation of ranges of values herein are merely intended to serve as a shorthand method of referring individually to each separate value falling within the range. Unless otherwise indicated herein, each individual value is incorporated into the specification as if it were individually recited herein. All methods described herein can be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The use of any and all examples, or exemplary language (e.g., "such as") provided herein, is intended merely to better illuminate the application and does not pose a limitation on the scope of the application unless otherwise claimed. No language is intended to indicate that the application will not also cover any
[0159] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the preferred embodiments by those skilled in the art once they learn of the basic inventive concepts. Therefore, the appended claims are intended to cover all such variations and modifications as fall within the scope of the application.
[0160] Obviously, many modifications and variations of the present application are possible in light of the above teachings. It is, therefore, to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.
Claims
1. A knowledge tracking method based on the forgetting attention mechanism, characterized in that, The method includes the following steps: Obtain the question sequence embedding representation and the answer sequence embedding representation after embedding the student learning record data; Calculate the similarity weight matrix corresponding to the embedding representation of the question sequence, and introduce a forgetting factor generated based on relative position encoding to correct the similarity weight; The answer sequence embedding representation is weighted and summed based on the corrected similarity weights to obtain the fused embedding representation; Based on the fusion embedding representation and the question sequence embedding representation, an optimized representation is determined to predict the student's accuracy rate in the next question based on the optimized representation; The introduction of a forgetting factor based on relative position encoding to correct the similarity weights specifically includes: Obtain the sequence length of the question sequence embedding representation, and construct a relative position embedding matrix based on the sequence length. ; Embedding matrix of relative positions By performing learnable embeddings and linear transformations, the forgetting rate can be obtained. : ； In the formula, For learnable parameters, For the Sigmoid function, Relative position index, This represents the index of the question's position in the sequence at the current moment. This represents the position index of the question at a historical moment within the sequence. According to the forgetting rate Similarity weights The similarity weights are then adjusted to obtain the corrected weights. : ； In the formula, It is a normalized exponential function.
2. The knowledge tracking method based on the forgetting attention mechanism as described in claim 1, characterized in that, The question sequence embedding representation includes knowledge point embedding representation and question difficulty embedding representation, and the answer sequence embedding representation includes knowledge point embedding representation and correct / incorrect answer embedding representation.
3. The knowledge tracking method based on the forgetting attention mechanism as described in claim 2, characterized in that, The expression representing the embedding of the question sequence is: ； In the formula, Representing knowledge points The potential representation, in which, For knowledge points One-hot encoding, It is a learnable mapping matrix; Indicates the topic The difficulty vector; Indicates the question and knowledge point The interactive variation vector is used to characterize the differences between different questions on the same knowledge point. Represents feature combinations, This indicates element-wise multiplication.
4. The knowledge tracking method based on the forgetting attention mechanism as described in claim 2 or 3, characterized in that, The embedded representation of the response sequence is: ； In the formula, Representing knowledge points The potential representation, in which, For knowledge points One-hot encoding, It is a learnable mapping matrix; Let represent the response vector, where One-hot encoding for correct or incorrect answers. It is a learnable mapping matrix; Represents feature combinations, This indicates element-wise multiplication.
5. The knowledge tracking method based on the forgetting attention mechanism as described in claim 1, characterized in that, The step of determining the optimized representation based on the fused embedding representation and the question sequence embedding representation specifically includes: The fusion embedding representation and the embedded representation of the question sequence The optimized representation is obtained by sequentially passing through a fully connected layer, residual connections, and normalization. : ； In the formula, Represents the normalization function. This represents the residual connection function.
6. The knowledge tracking method based on the forgetting attention mechanism as described in claim 1 or 5, characterized in that, The prediction of students' answers in the next round based on the optimized representation specifically includes: When responding to a prediction instruction, the current input sequence of questions is embedded in the representation. The optimized representation generated at historical moments By concatenating the vectors, the prediction vector is obtained. : ； Predicting students' accuracy in the next test using a fully connected layer and a Sigmoid activation function. : ； In the formula, For learnable parameters, This is the Sigmoid function.
7. A knowledge tracking model for implementing the knowledge tracking method based on the forgetting attention mechanism as described in claim 1, characterized in that, The knowledge tracing model includes: The input and embedding representation module is used to obtain the question sequence embedding representation and the answer sequence embedding representation after embedding the student learning record data; The forgetting attention modeling and feature optimization module is used to calculate the similarity weight matrix corresponding to the question sequence embedding representation, and to introduce a forgetting factor generated based on relative position encoding to correct the similarity weight; the answer sequence embedding representation is weighted and summed based on the corrected similarity weight to obtain the fused embedding representation; and the optimized representation is determined based on the fused embedding representation and the question sequence embedding representation. The prediction and output module is used to predict the student's accuracy rate in the next round of questions based on the optimized representation.
8. A computer system, characterized in that, The computer system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the knowledge tracking method based on the forgetting attention mechanism as described in any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the knowledge tracking method based on the forgetting attention mechanism as described in any one of claims 1 to 6.