Knowledge tracking method based on hierarchical data and oriented to cognitive behavior modeling

Abstract

The invention relates to the technical field of knowledge tracking, in particular to a knowledge tracking method based on hierarchical data and oriented to cognitive behavior modeling, which comprises the following steps: acquiring a learning sequence, carrying out data analysis and data cleaning on the acquired learning sequence to extract learning features, constructing an interactive embedding vector based on the extracted learning features, and carrying out data analysis and data cleaning on the interactive embedding vector. According to the method, knowledge mastering and forgetting rules of a learner are systematically simulated, a multi-dimensional memory enhanced cognitive behavior model is constructed, cognitive state updating is carried out on the learner from multiple dimensions, and the updated cognitive state is constrained and optimized through a cognitive state fine adjustment mechanism so as to obtain a final cognitive state. And prediction is carried out based on the obtained final cognitive state to output a prediction result, and a loss function is constructed, so that the knowledge representation degradation problem, the knowledge state unreasonable fluctuation problem and the non-physiological forgetting phenomenon caused by continuous state updating can be avoided, and the significant deviation between the non-physiological forgetting phenomenon and the human cognitive law can be avoided.

Description

Technical Field

[0001] This application belongs to the field of knowledge tracing technology, specifically relating to a knowledge tracing method based on hierarchical data and cognitive behavior modeling. Background Technology

[0002] Driven by the digital wave, smart education, leveraging technologies such as artificial intelligence and big data, is reshaping traditional education models and becoming a crucial engine for educational modernization. Knowledge tracing (KT), a powerful tool in smart education, models learners' dynamic cognitive states by collecting and analyzing behavioral data during the learning process. Based on these cognitive states, it predicts learners' performance in future learning, providing key support for personalized education and playing an indispensable role in innovative educational applications. With breakthroughs in deep learning technology, deep knowledge tracing has demonstrated significant advantages in predicting learners' future performance and assessing knowledge mastery, becoming an important research direction for the intelligent development of education and attracting widespread attention from academia and the education community.

[0003] Existing traditional deep knowledge tracing methods neglect the hierarchical structural features of learning behavior data (such as the difficulty level of test questions and the distribution of answering time), and in the process of cognitive behavior modeling, it is difficult to avoid the problem of knowledge representation degradation caused by continuous state updates. The non-physiological forgetting phenomenon caused by this deviates significantly from the laws of human cognition. Summary of the Invention

[0004] To address the technical problems existing in the background art, the present invention provides a knowledge tracking method based on hierarchical data and cognitive behavior modeling. This method enables hierarchical processing of learning features and helps to avoid [problems].

[0005] In a first aspect, the knowledge tracing method based on hierarchical data and cognitive behavior modeling provided by the present invention includes the following steps: Step 1: Obtain the learning sequence and perform data analysis and cleaning on the obtained learning sequence to extract learning features, including: question difficulty features, response time features, question features, concept features, and response features; Step 2: Construct interactive embedding vectors based on the extracted learning features; Step 3: Systematically simulate the learner's knowledge acquisition and forgetting patterns, construct a multidimensional memory enhancement cognitive behavior model, and update the learner's cognitive state from multiple dimensions through the constructed multidimensional memory enhancement cognitive behavior model and the learner's interactive embedding; Step 4: Constrain and optimize the updated cognitive state through a cognitive state fine-tuning mechanism to obtain the final cognitive state; Step 5: Make predictions based on the obtained final cognitive state to output the prediction results, and construct the loss function.

[0006] Furthermore, step two includes: The characteristics of question difficulty and response time are processed hierarchically to obtain hierarchical representations of question difficulty and response time. The problem features, hierarchical representation of problem difficulty, hierarchical representation of response time, conceptual features, and response features are embedded to obtain problem feature embedding vectors, hierarchical embedding vectors of problem difficulty, hierarchical embedding vectors of response time, conceptual feature embedding vectors, and response feature embedding vectors. Enhance the question embedding representation based on the obtained question feature embedding vector, question difficulty hierarchical embedding vector, and response time hierarchical embedding vector; The concept feature embedding vectors are aggregated to obtain multi-concept aggregated embedding vectors; We integrate enhanced question embedding representation, multi-concept aggregation embedding representation, and response embedding representation to construct an interactive embedding vector.

[0007] Furthermore, the method is characterized in that the hierarchical processing of the question difficulty features and response time features to obtain hierarchical representations of question difficulty and response time includes: The difficulty features of problems are hierarchically divided based on error rate, and each problem is assigned an appropriate difficulty label to obtain a hierarchical representation of problem difficulty. The specific process is as follows: ; in, It is the level of difficulty of the problem. Is it the correct answer to the question? Number of times, It is to answer questions Total number of times; The response time characteristics are hierarchically divided according to the response time, and a label is assigned to each time stage. The specific process is as follows: ; in, It is the current moment. t Time allotted for answering questions It is a scaling factor that hierarchically represents the response time. It refers to the current moment t A hierarchical representation of the response time.

[0008] Furthermore, the embedding of question features, hierarchical representation of question difficulty, hierarchical representation of response time, conceptual features, and response features to obtain question feature embedding vectors, hierarchical embedding vectors of question difficulty, hierarchical embedding vectors of response time, conceptual feature embedding vectors, and response feature embedding vectors includes: The problem features are embedded to obtain the problem feature embedding matrix. ,in, Through questions ID Query To obtain problem characteristics t Problem feature embedding vector at time step ,in, Represents the total number of questions. d Represents dimensions; The hierarchical representation of problem difficulty is embedded to obtain the hierarchical representation embedding matrix of problem difficulty. ,in, Then, the problem difficulty is hierarchically represented by the embedding matrix. Perform a query to obtain a hierarchical representation of the problem's difficulty level. t Hierarchical embedding vector of problem difficulty at each time step ,in, It is the level of difficulty of the problem. d Represents dimensions; The hierarchical representation of response time is embedded to obtain the hierarchical representation of response time embedding matrix. ,in, Then, the hierarchical representation of the response time is embedded in the matrix. Perform a query to obtain a hierarchical representation of the response time. t Hierarchical embedding vector of response time corresponding to each moment , Z The level representing the time taken to respond; Conceptual features are embedded to obtain a conceptual feature embedding matrix. ,in, Through conceptual features ID Embedding matrix of concept features Perform a query to obtain concept features. t Multiple concept feature embedding vectors corresponding to time step ,in, The total number of representative concepts, d Represents dimension, 1、2 and max Indicates the sequence number of multiple conceptual features corresponding to a problem; The response features are embedded to obtain the response feature embedding matrix. ,in, , for response feature embedding matrix Perform a query to obtain the characteristics of the problem response. t Problem response feature embedding vector at time step .

[0009] Furthermore, the aggregation of concept feature embedding vectors to obtain multi-concept aggregated embedding vectors includes: By using a multilayer perceptron network, the conceptual features of multiple concepts under the same problem are embedded into vectors and mapped to dimensions. d The multi-concept aggregated embedding vector is obtained, and the specific formula is as follows: ; ； in, This indicates a splicing operation. This represents the maximum number of concepts a problem can contain. This represents the concatenation and embedding of feature vectors from multiple concepts. , It is a weight matrix. 、 It is a bias term. Represents the feature dimension; This represents a multi-concept aggregation embedding vector.

[0010] Furthermore, step three includes: Based on the forgetting mechanism, forgetting effects are constructed targeting the problem cognition level, the concept cognition level, and the global cognition level, respectively. The specific formulas are shown below: ; ; ; in, This indicates the forgetting effect at the cognitive level of the problem; This indicates the forgetting effect at the conceptual cognitive level. The forgetting effect at the global cognitive level Indicates the historical state of problem cognition. Indicates the historical state of conceptual cognition. Indicates the overall historical state of cognition. This indicates the time interval between two interactions that represents the same question. This indicates the embedding of the same concept between two interactions. Embed the time interval for each time step. , and This represents the weight matrix that undergoes a linear transformation on the vector concatenated with historical states and time intervals. , and Indicates the bias term. represent Sigmoid Activation function; Based on the forgetting effect at the problem cognition level, concept cognition level, and global cognition level, the cognitive state at each level is updated respectively.

[0011] Furthermore, the updating of the cognitive state at each level based on the forgetting effect at the problem cognition level, concept cognition level, and global cognition level includes: A: Update the cognitive state at the problem-solving level. The specific update rules are as follows: ； ； in, This indicates the memory state when the problem last occurred. After the forgetting effect The filtered result For the current moment t Interactive embedding vectors, It is the result after the cognitive state has been updated. Represents the learnable parameter matrix. It is a bias term. It is an activation function; B: Update the conceptual cognitive state at the conceptual cognition level, with the following specific rules: ； ； Since the problem involves multiple concepts, among which This represents the result of the aggregation of memory states of each concept corresponding to the current problem after the forgetting effect, i.e., the... The memory state of the concept when it last appeared. After the forgetting effect After filtering, one of the concept memory states is obtained. This refers to the updated cognitive state corresponding to the current problem. C: Update the global cognitive state at the global cognitive level, with the following specific rules: ； ； in, The global cognitive state at the previous moment After the forgetting effect The filtered result Represents the learnable parameter matrix. It is a bias term; This is the result of updating the global cognitive state of the current problem.

[0012] Furthermore, step four includes: Aggregate the updated multi-dimensional cognitive states; The aggregated cognitive state is fine-tuned through a sparse attention mechanism to obtain the final cognitive state.

[0013] Furthermore, step five includes: The final cognitive state is projected onto the latent space through a fully connected layer, utilizing... The activation function filters out noise signals and enhances higher-order feature interactions, then outputs a scalar value through a compression layer, and then... Sigmoid The activation function generates probabilistic predictions.

[0014] Secondly, the present invention provides a computer-readable storage medium comprising a stored program, wherein, when the program is running, it controls the device on which the computer-readable storage medium is located to execute the knowledge tracing method based on hierarchical data and cognitive behavior modeling described above.

[0015] The beneficial effects of this invention are as follows: This invention first performs data analysis and cleaning on student learning sequence data. Then, it enhances the representation of question features by hierarchically processing the characteristics of question difficulty and response time. Next, by combining forgetting factors, it systematically simulates the laws of students' knowledge acquisition and forgetting, models the student cognitive behavior module, and optimizes the knowledge state through a knowledge state fine-tuning mechanism. This helps to avoid problems such as knowledge representation degradation caused by continuous state updates, unreasonable fluctuations in knowledge state, and significant deviations between non-physiological forgetting phenomena and human cognitive laws. Attached Figure Description

[0016] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0017] Figure 1 A flowchart of a knowledge tracing method based on hierarchical data and cognitive behavior modeling according to the present invention; Figure 2This is a flowchart illustrating a knowledge tracing method based on hierarchical data and cognitive behavior modeling according to the present invention. Figure 3 This is a schematic diagram of the structure of a knowledge tracking method based on hierarchical data and cognitive behavior modeling according to the present invention. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, each technical and scientific term used in these embodiments has the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0020] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0021] In this invention, terms such as "upper," "lower," "left," "right," "front," "back," "vertical," "horizontal," "side," and "bottom" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used only to facilitate the description of the structural relationships of the various components or elements of this invention and do not specifically refer to any component or element in this invention. They should not be construed as limiting the invention.

[0022] In this invention, terms such as "fixed connection," "connected," and "linked" should be interpreted broadly, indicating a fixed connection, an integral connection, or a detachable connection; a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can determine the specific meaning of these terms in this invention based on the specific circumstances, and they should not be construed as limitations on the invention.

[0023] Example 1: like Figures 1 to 3 As shown, this embodiment provides a knowledge tracing method based on hierarchical data and cognitive behavior modeling, including the following steps: S1: Obtain the learning sequence and perform data analysis and cleaning on the obtained learning sequence to extract learning features, including: question difficulty features, response time features, question features, concept features, and response features; The difficulty of a problem is determined by its error rate on a given set of data.

[0024] S2: Construct interactive embedding vectors based on the extracted learning features.

[0025] Specifically, the following steps are included: S21: Perform hierarchical processing on the characteristics of question difficulty and response time to obtain hierarchical representations of question difficulty and response time.

[0026] Specifically, it includes the following steps: S21-1: Based on the error rate, the difficulty features of the questions are hierarchically divided, and an appropriate difficulty label is assigned to each question to obtain a hierarchical representation of the question difficulty. The specific process is as follows: ; in, It is the level of difficulty of the problem. Is it the correct answer to the question? Number of times, It is to answer questions The total number of times.

[0027] S21-2: The response time characteristics are hierarchically divided according to the response time, and a label is assigned to each time stage. The specific process is as follows: ; in, It is the current moment. t Time allotted for answering questions It is a scaling factor that hierarchically represents the response time. It refers to the current moment t A hierarchical representation of the response time.

[0028] From the perspective of learners' answering time, there is a positive correlation between the difficulty of the question and the time learners spend answering it. More difficult questions usually lead to learners spending more time answering them, while less difficult questions have the opposite effect, with an average answering time that is significantly reduced.

[0029] S22: Embed the problem features, hierarchical representation of problem difficulty, hierarchical representation of response time, concept features, and response features to obtain the problem feature embedding vector, the hierarchical embedding vector of problem difficulty, the hierarchical embedding vector of response time, the concept feature embedding vector, and the response feature embedding vector.

[0030] Specifically, the problem features are embedded to obtain the problem feature embedding matrix. ,in, Through questions IDQuery To obtain problem characteristics t Problem feature embedding vector at time step ,in, Represents the total number of questions. d Represents a dimension.

[0031] The hierarchical representation of problem difficulty is embedded to obtain the hierarchical representation embedding matrix of problem difficulty. ,in, Then, the problem difficulty is hierarchically represented by the embedding matrix. Perform a query to obtain a hierarchical representation of the problem's difficulty level. t Hierarchical embedding vector of problem difficulty at each time step ,in, It is the level of difficulty of the problem. d Represents dimensions; The hierarchical representation of response time is embedded to obtain the hierarchical representation of response time embedding matrix. ,in, Then, the hierarchical representation of the response time is embedded in the matrix. Perform a query to obtain a hierarchical representation of the response time. t Hierarchical embedding vector of response time corresponding to each moment ;in, Z The level representing the time taken to respond. d Represents dimensions; Conceptual features are embedded to obtain a conceptual feature embedding matrix. ,in, Through conceptual features ID Embedding matrix of concept features Perform a query to obtain concept features. t Multiple concept feature embedding vectors corresponding to time step ,in, The total number of representative concepts, d Represents dimensions, 1、2 and max This indicates the sequence number of multiple conceptual features corresponding to a single issue.

[0032] The response features are embedded to obtain the response feature embedding matrix. ,in, , for response feature embedding matrix Perform a query to obtain the characteristics of the problem response. t Problem response feature embedding vector at time step , d Represents a dimension.

[0033] S23: The enhanced question embedding vector is obtained based on the question feature embedding vector, the hierarchical question difficulty embedding vector, and the hierarchical response time embedding vector. The formula is as follows: ； in, They represent the current time. t The enhanced question embedding vector, question feature embedding vector, question difficulty hierarchical embedding vector, and answer time hierarchical embedding vector are used.

[0034] By hierarchically defining the difficulty and response time characteristics of questions, and integrating a dynamic labeling system for question difficulty with differentiated representations of response time, this approach overcomes the limitations of traditional methods that rely solely on knowledge point labels and binary response results. By utilizing hierarchical difficulty labels based on error rate statistics and a dynamic time segmentation strategy, it achieves refined modeling of learners' multidimensional behavioral characteristics (such as cognitive load and time efficiency), effectively improving the discriminativeness and interpretability of feature representations.

[0035] S24: Aggregate the concept feature embedding vectors to obtain multi-concept aggregated embedding vectors.

[0036] Each problem's corresponding one or more concepts are converted into a vector representation, facilitating model processing and analysis. Simultaneously, the relationships between multiple concepts need to be considered, such as correlation and hierarchy; therefore, the embedding vectors of each concept need to be aggregated.

[0037] Specifically, the following steps are included: By using a multilayer perceptron (MLP) network, the conceptual features of multiple concepts under the same problem are embedded into vectors and mapped to dimensions. d The multi-concept aggregated embedding vector is obtained, and the specific formula is as follows: ; ； in, This indicates a splicing operation. This represents the maximum number of concepts a problem can contain. This represents the concatenation and embedding of feature vectors from multiple concepts. , It is a weight matrix. 、 It is a bias term. Represents the feature dimension; This represents a multi-concept aggregation embedding vector.

[0038] It should be noted that, This represents the concatenation and embedding of feature vectors from multiple concepts, for example, max If it is 7, then The dimension is ,from M The embedding vectors of the seven corresponding concepts are found. A multilayer perceptron (MLP) network is used to map the feature vector dimensions of each concept under the same question to the feature dimension of a single concept. This approach achieves feature alignment between the question, concepts, and response, avoids information distortion that may result from directly accumulating or averaging multi-concept features, and automatically learns the complex relationships and weight allocations between multiple concepts, retaining more useful feature information.

[0039] S25: Integrate the enhanced question embedding vector, multi-concept aggregation embedding vector, and response feature embedding vector to construct an interactive embedding vector, thereby improving the accuracy and discriminative ability of knowledge tracking. The specific formula is as follows: .

[0040] in, Current time t Interactive embedding vectors.

[0041] S3: Systematically simulate the learner's knowledge acquisition and forgetting patterns, construct a multidimensional memory-enhancing cognitive behavior model, and update the learner's cognitive state from multiple dimensions through the constructed multidimensional memory-enhancing cognitive behavior model and the learner's interactive embedding.

[0042] These include multiple dimensions such as problem cognition, concept cognition, and global cognition.

[0043] The core role of forgetting factors is to model the learner's forgetting process. In traditional knowledge tracing models (such as the classic Markov chain-based model), learners' mastery of concepts is usually static; that is, once a concept is mastered, it is considered to be mastered forever. However, in real learning, learners may gradually forget the knowledge they have acquired due to lack of review or the passage of time.

[0044] Specifically, based on the forgetting mechanism, we construct forgetting effects at the problem cognition level, concept state level, and global cognition level to understand learners' knowledge mastery and to analyze the forgetting benefits. It dynamically reflects the learner's forgetting process of knowledge, with time intervals. The longer the duration, the greater the forgetting effect. The more significant.

[0045] S31: Based on the forgetting mechanism, construct forgetting effects targeting the problem cognition level, concept cognition level, and global cognition level respectively. The specific formulas are as follows:

[0046]

[0047]

[0048] in, This indicates the forgetting effect at the cognitive level of the problem; This indicates the forgetting effect at the conceptual cognitive level. The forgetting effect at the global cognitive level Indicates the historical state of problem cognition. Represents the historical state of conceptual cognition. Represents the overall historical state of cognition. This indicates the time interval between two interactions that represents the same question. This indicates the embedding of the same concept between two interactions. Embed the time interval for each time step. , and This represents the weight matrix that undergoes a linear transformation on the vector concatenated with historical states and time intervals. , and Indicates the bias term. represent Sigmoid Activation function.

[0049] S32: Based on the forgetting effect at the problem cognition level, concept cognition level, and global cognition level, update the cognitive state at each level respectively.

[0050] S32-1: Update the cognitive state at the problem-solving level. The specific update rules are as follows: ； ； in, This indicates the memory state when the problem last occurred. After the forgetting effect Filtered historical memory For the current moment t Interactive embedding vectors, It is a revised historical memory With current time t Interactive embedding vectors The fusion result, that is, the result after cognitive state update, Represents the learnable parameter matrix. It is a bias term. It is an activation function.

[0051] S32-2: Update the conceptual cognitive state at the conceptual cognition level, with the following specific rules: ; ; Since the problem involves multiple concepts, among which, This represents the result of the aggregation of the memory state of each concept corresponding to the current problem at the time of its last occurrence, after the forgetting effect at the conceptual cognitive level. For example, the first concept... The memory state of the concept when it last appeared. Through the forgetting effect at the conceptual cognitive level After filtering, one of the concept memory states is obtained. This refers to the updated cognitive state corresponding to the current problem. Represents the learnable parameter matrix. It is a bias term.

[0052] S32-3: Update the global cognitive state at the global cognitive level, with the following specific rules: ; ; in, The global cognitive state at the previous moment Through the forgetting effect at the global cognitive level The filtered result Represents the learnable parameter matrix. It is a bias term.

[0053] S4: Constrain and optimize the updated cognitive state through a cognitive state fine-tuning mechanism to obtain the final cognitive state.

[0054] Specifically, the following steps are included: S41: Aggregate the updated multi-dimensional cognitive states, as follows: ； ; in, express t+1 Time-based aggregation of multi-dimensional cognitive state representations and t+1 Knowledge embedding that combines characteristic information such as the time of the question, the concept, the difficulty of the question, and the response time. It represents the combination of initial cognitive state representations at all times.

[0055] S42: Fine-tune the aggregated cognitive state through a sparse attention mechanism to obtain the final cognitive state; The cognitive state dynamic fine-tuning method based on sparse attention mechanism can achieve refined modeling of learners' knowledge mastery, specifically including the following steps: S42-1: Calculate attention for cognitive states at each time step. The specific formula is as follows: ; in, For activation function, Represents the scaling factor. d Represents the dimension of the key vector. As a dynamic sparse masking function, attention is focused by setting a threshold strategy. This local attention focuses on the most relevant historical states. The specific implementation is as follows: ; in, Indicates time step Regarding the historical state The level of attention, Indicates taking the first The attention matrix is ​​changed from a dense matrix to one where each row retains only the largest element. One non-zero element, and all other elements are set to zero. By preserving the maximum response value, interference from irrelevant historical states is suppressed.

[0056] Therefore, the cognitive state after fine-tuning through the sparse attention mechanism is as follows: ; in, This represents the cognitive state after fine-tuning.

[0057] S42-2: By performing residual connections and layer normalization operations, the initial cognitive state representation and the fine-tuned cognitive state representation are concatenated to obtain the final cognitive state representation. The specific operations are as follows: ; Residual connections and layer normalization operations help ensure the stability of model training and prevent the loss of cognitive states due to increased network depth.

[0058] S5: Make predictions based on the final cognitive state to output the prediction results, and construct the loss function.

[0059] In obtaining the final cognitive state Then, use the final knowledge state. Predict learners' answers t+1 The learner's performance on the first question is mapped to the next question using a fully connected neural network. Correct answer rate ∈(0,1).

[0060] Specifically, the following steps are included: S51: The final cognitive state is transmitted through a fully connected layer. Projected into the hidden space, using The activation function filters out noisy signals and enhances higher-order feature interactions, while a random deactivation layer is introduced to improve robustness to sparse learning behavior. Finally, a scalar value is output through a compression layer and then... Sigmoid The activation function generates probabilistic predictions, and the specific formula is as follows: ; in, , For learnable parameters, for Sigmoid Activation function.

[0061] S52: Constructing the loss function: Specifically, the following steps are included: First, the model's assessment error of the learner's cognitive state is quantified using a binary cross-entropy loss function, whereby... ∈(0,1) express t+1 The accuracy of prediction at any given time. ∈(0,1) represents the true answer label. The Adam optimizer's dynamic learning rate adjustment mechanism is used to efficiently optimize the knowledge state encoding network parameters. The specific formula is as follows: .

[0062] Example 2: This example provides a knowledge tracing system based on hierarchical data and cognitive behavior modeling, including: The data acquisition and processing module is used to acquire learning sequences and perform data analysis and cleaning on the acquired learning sequences to extract learning features, including: question difficulty features, response time features, question features, concept features, and response features.

[0063] Embedding building blocks are used to construct interactive embedding vectors based on extracted learned features.

[0064] This embedding vector building module includes: Hierarchical data representation unit is used to hierarchically process the characteristics of question difficulty and response time to obtain hierarchical representations of question difficulty and response time. The feature embedding unit is used to embed question features, hierarchical representation of question difficulty, hierarchical representation of response time, concept features, and response features to obtain question feature embedding vectors, hierarchical embedding vectors of question difficulty, hierarchical embedding vectors of response time, concept feature embedding vectors, and response feature embedding vectors. The question enhancement unit is used to obtain an enhanced question embedding vector based on the obtained question feature embedding vector, question difficulty hierarchical embedding vector, and response time hierarchical embedding vector; The concept aggregation unit is used to aggregate concept feature embedding vectors to obtain multi-concept aggregated embedding vectors; Interactive Embedding Vector Building Unit: Used to integrate enhanced question embedding vectors, multi-concept aggregation embedding vectors, and response feature embedding vectors to build interactive embedding vectors.

[0065] The cognitive behavior modeling module, which integrates forgetting mechanisms, is used to systematically simulate learners' knowledge acquisition and forgetting patterns, construct a multidimensional memory-enhanced cognitive behavior model, and update learners' cognitive states from multiple dimensions through the constructed multidimensional memory-enhanced cognitive behavior model of interactive embedding.

[0066] The knowledge state fine-tuning module based on sparse attention is used to constrain and optimize the updated cognitive state through the cognitive state fine-tuning mechanism to obtain the final cognitive state. The result prediction module is used to make predictions based on the obtained final cognitive state to output the prediction results and construct the loss function.

[0067] Experimental Validation: Using the area under the curve and accuracy as evaluation metrics, the model was validated using the method described in this application, as follows: The model was validated using three benchmark datasets from real-world educational scenarios: ASSISTments2012, ASSISTments2017, and the large-scale online learning interaction dataset EdNet-KT1. During the data preprocessing stage, learning behavior records with missing concept labels (KC, Knowledge Component) were removed to ensure data quality and the effectiveness of model training.

[0068] In the experiment, the dataset was standardized: the length of the learning sequences was uniformly set to 200 time steps, sequences exceeding the length were truncated, and insufficient sequences were padded. The dataset was divided into training, validation, and test sets, with 80% of the data used for training and validation (8:2 ratio) and the remaining 20% ​​used as an independent test set. During training, the Adam optimizer was used with an initial learning rate of 0.002, and an early stopping mechanism (30 epochs of patience) was introduced to prevent overfitting. To further improve the model's generalization ability, a Dropout layer was added to the network with a dropout rate of 0.4. Regarding the model architecture, the embedding layer dimension d was set to 128, and the problem difficulty and complexity levels were set to 10. A sparse attention mechanism was used in the model, with the sparse attention method focusing only on the top 5 elements of the attention score for adjustment.

[0069] The results of comparing this case with other models on three benchmark datasets, including AUC (Area Under Curve) and ACC (Accuracy), are shown in Table 1.

[0070] Table 1

[0071] Notes: DKT (Deep Knowledge Tracing): Deep knowledge tracing model; DKVMN (Dynamic Key-Value Memory Networks for Knowledge Tracing): Knowledge tracing model based on dynamic key-value memory networks; Sakt (Self-Attentive Knowledge Tracing): Knowledge tracing model based on self-attention mechanism; DeepIRT (Deep-IRT: Make Deep Learning Based Knowledge Tracing Explainable Using Item Response Theory): Explainable deep knowledge tracing model using item response theory; Akt (Context-Aware Attentive Knowledge Tracing): Context-aware knowledge tracing model based on attention mechanism; qDkt (Question-centric Deep Knowledge Tracing): Question-centric deep knowledge tracing model; Hawkeskt (Temporal Cross-Effects in Knowledge Tracing): Knowledge tracing model considering temporal cross-effects; Lpkt (Learning Process-consistent Knowledge Tracing): Knowledge tracing model considering learning process consistency; Dtransformer (Tracing Knowledge Instead of Patterns: Stable Knowledge Tracing with Diagnostic) Transformer: A stable knowledge tracing model based on diagnostic Transformer; Folibikt (Forgetting-aware Linear Bias for Attentive Knowledge Tracing): An attention-based knowledge tracing model that considers the influence of linear bias in forgetting perception; Sparsekt (Towards Robust Knowledge Tracing Models via k-Sparse Attention): A robust knowledge tracing model based on k-sparse attention; Stablekt (Enhancing Length Generalization for Attention Based Knowledge Tracing Models with Linear Biases): An attention-based knowledge tracing model under linear bias.Extrakt (Extending Context Window of Attention Based Knowledge Tracing Models via Length Extrapolation): An attention-based knowledge tracing model that extends the context window via length extrapolation.

[0072] Example 3: This embodiment provides a computer-readable storage medium, which includes a stored program, wherein, when the program is running, it controls the device where the computer-readable storage medium is located to execute the knowledge tracking method based on hierarchical data and cognitive behavior modeling described in Embodiment 1.

[0073] The same or similar parts between the various embodiments in this specification can be referred to mutually. In particular, the terminal embodiments are basically similar to the method embodiments, so the description is relatively simple, and the relevant parts can be referred to the description in the method embodiments.

[0074] In the several embodiments provided by this invention, it should be understood that the disclosed systems and methods can... This can be achieved through other means. For example, the system embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0075] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0076] Additionally, it should be noted that the flowcharts in the accompanying drawings illustrate methods according to embodiments of this disclosure. In the descriptions corresponding to the flowcharts or block diagrams in the drawings, the operations or steps corresponding to different blocks may occur in a different order than disclosed in the description; sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, or sometimes in reverse order, depending on the function involved. Each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions.

[0077] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A knowledge tracing method based on hierarchical data and cognitive behavior modeling, characterized in that, Includes the following steps: Step 1: Obtain the learning sequence and perform data analysis and cleaning on the obtained learning sequence to extract learning features, including: question difficulty features, response time features, question features, concept features, and response features; Step 2: Construct interactive embedding vectors based on the extracted learning features; Step 3: Systematically simulate the learner's knowledge acquisition and forgetting patterns, construct a multidimensional memory enhancement cognitive behavior model, and update the learner's cognitive state from multiple dimensions through the constructed multidimensional memory enhancement cognitive behavior model and the learner's interactive embedding; Step 4: Constrain and optimize the updated cognitive state through a cognitive state fine-tuning mechanism to obtain the final cognitive state; Step 5: Make predictions based on the obtained final cognitive state to output the prediction results, and construct the loss function.

2. The knowledge tracing method based on hierarchical data and cognitive behavior modeling as described in claim 1, characterized in that, Step two includes: The characteristics of question difficulty and response time are processed hierarchically to obtain hierarchical representations of question difficulty and response time. The problem features, hierarchical representation of problem difficulty, hierarchical representation of response time, conceptual features, and response features are embedded to obtain problem feature embedding vectors, hierarchical embedding vectors of problem difficulty, hierarchical embedding vectors of response time, conceptual feature embedding vectors, and response feature embedding vectors. Enhance the question embedding representation based on the obtained question feature embedding vector, question difficulty hierarchical embedding vector, and response time hierarchical embedding vector; The concept feature embedding vectors are aggregated to obtain multi-concept aggregated embedding vectors; We integrate enhanced question embedding representation, multi-concept aggregation embedding representation, and response embedding representation to construct an interactive embedding vector.

3. The knowledge tracing method based on hierarchical data and cognitive behavior modeling as described in claim 2, characterized in that, The hierarchical processing of question difficulty features and response time features to obtain hierarchical representations of question difficulty and response time includes: The difficulty features of problems are hierarchically divided based on error rate, and each problem is assigned an appropriate difficulty label to obtain a hierarchical representation of problem difficulty. The specific process is as follows: ； in, It is the level of difficulty of the problem. Is it the correct answer to the question? Number of times, It is to answer questions Total number of times; The response time characteristics are hierarchically divided according to the response time, and a label is assigned to each time stage. The specific process is as follows: ； in, It is the current moment. t Time allotted for answering questions It is a scaling factor that hierarchically represents the response time. It refers to the current moment t A hierarchical representation of the response time.

4. The knowledge tracing method based on hierarchical data and cognitive behavior modeling as described in claim 3, characterized in that, The embedding of question features, hierarchical representation of question difficulty, hierarchical representation of response time, conceptual features, and response features to obtain question feature embedding vectors, hierarchical embedding vectors of question difficulty, hierarchical embedding vectors of response time, conceptual feature embedding vectors, and response feature embedding vectors includes: The problem features are embedded to obtain the problem feature embedding matrix. ,in, Through questions ID Query To obtain problem characteristics t Problem feature embedding vector at time step ,in, Represents the total number of questions. d Represents dimensions; The hierarchical representation of problem difficulty is embedded to obtain the hierarchical representation embedding matrix of problem difficulty. ,in, Then, the problem difficulty is hierarchically represented by the embedding matrix. Perform a query to obtain a hierarchical representation of the problem's difficulty level. t Hierarchical embedding vector of problem difficulty at each time step ,in, It is the level of difficulty of the problem. d Represents dimensions; The hierarchical representation of response time is embedded to obtain the hierarchical representation of response time embedding matrix. ,in, Then, the hierarchical representation of the response time is embedded in the matrix. Perform a query to obtain a hierarchical representation of the response time. t Hierarchical embedding vector of response time corresponding to each moment , Z The level representing the time taken to respond; Conceptual features are embedded to obtain a conceptual feature embedding matrix. ,in, Through conceptual features ID Embedding matrix of concept features Perform a query to obtain concept features. t Multiple concept feature embedding vectors corresponding to time step ,in, The total number of representative concepts, d Represents dimension, 1、2 and max Indicates the sequence number of multiple conceptual features corresponding to a problem; The response features are embedded to obtain the response feature embedding matrix. ,in, , for response feature embedding matrix Perform a query to obtain the characteristics of the problem response. t Problem response feature embedding vector at time step .

5. The knowledge tracing method based on hierarchical data and cognitive behavior modeling according to claim 2, characterized in that, The aggregation of concept feature embedding vectors to obtain multi-concept aggregated embedding vectors includes: By using a multilayer perceptron network, the conceptual features of multiple concepts under the same problem are embedded into vectors and mapped to dimensions. d The multi-concept aggregated embedding vector is obtained, and the specific formula is as follows: ； ； in, This indicates a splicing operation. This represents the maximum number of concepts a problem can contain. This represents the concatenation and embedding of feature vectors from multiple concepts. , It is a weight matrix. 、 It is a bias term. Represents the feature dimension; This represents a multi-concept aggregation embedding vector.

6. The knowledge tracing method based on hierarchical data and cognitive behavior modeling according to claim 1, characterized in that, Step three includes: Based on the forgetting mechanism, forgetting effects are constructed targeting the problem cognition level, the concept cognition level, and the global cognition level, respectively. The specific formulas are shown below: ； ； ； in, This indicates the forgetting effect at the cognitive level of the problem; This indicates the forgetting effect at the conceptual cognitive level. The forgetting effect at the global cognitive level Indicates the historical state of problem cognition. Represents the historical state of conceptual cognition. Represents the overall historical state of cognition. This indicates the time interval between two interactions that represents the same question. This indicates the embedding of the same concept between two interactions. Embed the time interval for each time step. , and This represents the weight matrix that undergoes a linear transformation on the vector concatenated with historical states and time intervals. , and Indicates the bias term. represent Sigmoid Activation function; Based on the forgetting effect at the problem cognition level, concept cognition level, and global cognition level, the cognitive state at each level is updated respectively.

7. The knowledge tracing method based on hierarchical data and cognitive behavior modeling according to claim 6, characterized in that, The forgetting effect based on the problem cognition level, concept cognition level, and global cognition level, and the updating of the cognitive state at each level respectively, includes: A: Update the cognitive state at the problem-solving level. The specific update rules are as follows: ； ； in, This indicates the memory state when the problem last occurred. After the forgetting effect The filtered result For the current moment t Interactive embedding vectors, It is the result after the cognitive state has been updated. Represents the learnable parameter matrix. It is a bias term. It is an activation function; B: Update the conceptual cognitive state at the conceptual cognition level, with the following specific rules: ； ； Since the problem involves multiple concepts, among which This represents the result of the aggregation of memory states of each concept corresponding to the current problem after the forgetting effect, i.e., the... The memory state of the concept when it last appeared. After the forgetting effect After filtering, one of the concept memory states is obtained. This refers to the updated cognitive state corresponding to the current problem. C: Update the global cognitive state at the global cognitive level, with the following specific rules: ； ； in, The global cognitive state at the previous moment After the forgetting effect The filtered result Represents the learnable parameter matrix. It is a bias term; This is the result of updating the global cognitive state of the current problem.

8. The knowledge tracing method based on hierarchical data and cognitive behavior modeling according to claim 1, characterized in that, Step four includes: Aggregate the updated multi-dimensional cognitive states; The aggregated cognitive state is fine-tuned through a sparse attention mechanism to obtain the final cognitive state.

9. The knowledge tracing method based on hierarchical data and cognitive behavior modeling according to claim 1, characterized in that, Step five includes: The final cognitive state is projected onto the latent space through a fully connected layer, utilizing... The activation function filters out noise signals and enhances higher-order feature interactions, then outputs a scalar value through a compression layer, and then... Sigmoid The activation function generates probabilistic predictions.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium resides to perform the knowledge tracing method based on hierarchical data and cognitive behavior modeling as described in any one of claims 1 to 9.

Images