Knowledge tracking method based on multi-dimensional cognitive space layering decoupling and related equipment

Abstract

The application discloses a knowledge tracking method based on multi-dimensional cognitive space layering decoupling and related equipment, which can be applied to the technical field of intelligent education. The application inputs a historical interaction sequence into a knowledge tracking model based on multi-dimensional cognitive space layering decoupling, so as to respectively extract and fuse feature vectors from semantic, structural and statistical aspects through a multi-view feature extraction and fusion unit in the model to obtain target fusion features, thereby the multi-dimensional attributes of knowledge points can be described. Then, a Gaussian mixture autoencoder is used to decouple the knowledge points in the historical interaction sequence into cognitive attribute category probabilities and latent semantic vectors according to the target fusion feature vectors, thereby the latent cognitive dimensions are deeply mined. Then, a sequence interaction encoding and closed-loop state tracking unit is used to perform knowledge dynamic updating and closed-loop tracking according to the cognitive attribute category probabilities and the latent semantic vectors, so that the explainability and pertinence of knowledge tracking are effectively improved, and the accuracy of the knowledge tracking prediction effect is improved.

Description

Technical Field

[0001] This application relates to the field of intelligent education technology, and in particular to a knowledge tracking method and related equipment based on multidimensional cognitive space hierarchical decoupling. Background Technology

[0002] Among related technologies, knowledge tracing (KT) is a key technology in the field of intelligent education. It aims to track students' knowledge mastery in real time based on their historical learning interaction sequences (such as problem-solving records) and predict their performance in future exercises. With the popularization of online education platforms, massive amounts of learning data are recorded and applied to personalized recommendations and precise tutoring. Although deep learning technology has achieved significant results in KT tasks, most existing methods treat knowledge points merely as simple IDs (one-hot encoding) or single-dimensional embedding vectors. Furthermore, existing knowledge tracing methods lack in-depth mining and utilization of the potential cognitive attributes of knowledge points, resulting in a lack of interpretability and specificity in current cognitive state tracking. Simultaneously, because existing methods still rely on historical interaction data to learn knowledge point representations, the predictive performance of existing knowledge tracing methods deteriorates when new cognitive points or systems emerge during the knowledge tracing process.

[0003] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0004] The main objective of this application is to propose a knowledge tracing method and related equipment based on multidimensional cognitive space tomographic decoupling, which can effectively improve the interpretability and relevance of knowledge tracing and improve the accuracy of knowledge tracing prediction.

[0005] To achieve the above objectives, one aspect of this application proposes a knowledge tracing method based on multidimensional cognitive space hierarchical decoupling, the method comprising the following steps: Obtain the historical interaction sequence of the target student during the history problem-solving process and the current unfinished exercises; The historical interaction sequence is input into a knowledge tracking model based on multidimensional cognitive space hierarchical decoupling, so that after the knowledge tracking model is trained on the historical interaction sequence, it can predict the probability of answering the current exercise. The knowledge tracking model includes a multi-view feature extraction and fusion unit, a Gaussian mixture autoencoder, and a sequence interactive coding and closed-loop state tracking unit. The multi-view feature extraction and fusion unit includes a semantic feature extraction module, a graph structure feature extraction module, a question-answering result statistical feature extraction module, and a gated feature mixer. The semantic feature extraction module is used to extract high-dimensional semantic vectors corresponding to knowledge points in the historical interaction sequence. The graph structure feature extraction module is used to extract structured embedding vectors of knowledge points in the historical interaction sequence. The question-answering result statistical feature extraction module is used to extract question-answering statistical feature vectors for each knowledge point in the historical interaction sequence. The gated feature mixer is used to perform a fusion operation on the high-dimensional semantic vectors, the structured embedding vectors, and the question-answering statistical feature vectors to obtain the target fusion feature. The Gaussian mixture autoencoder is used to decouple the knowledge points in the historical interaction sequence into cognitive attribute category probabilities and latent semantic vectors based on the target fusion feature vector; The sequence interaction encoding and closed-loop state tracking unit includes an attribute-specialized attention mechanism module, a cognitive gating update module, and a prediction module. The attribute-specialized attention mechanism module is used to construct a context vector based on the cognitive attribute category probability, the latent semantic vector, the historical interaction sequence, and the state-driven weights. The context vector is used to represent the target cognitive surface information of the target student. The prediction module is used to predict the probability of answering the current exercise based on the context vector and the attribute mastery state. The cognitive gating update module is used to update the attribute mastery state of the target student based on the prediction error between the cognitive attribute category probability and the probability of answering correctly. The attribute mastery state is used to update the state-driven weights.

[0006] In some embodiments, extracting the high-dimensional semantic vectors corresponding to knowledge points in the historical interaction sequence includes: Obtain a semantic view of the historical interaction sequence; The textual descriptions of knowledge points in the semantic view are encoded using a large language model to obtain the high-dimensional semantic vectors; the textual descriptions include the knowledge point names, parsing information, the cognitive dimensions and competencies assessed by the exercises.

[0007] In some embodiments, extracting the structured embedding vectors of knowledge points from the historical interaction sequence includes: A knowledge point association graph is constructed as a graph view based on the historical interaction sequence; The graph attention network extracts structured embedding vectors of knowledge points from the graph view, which are used to capture the logical dependencies between knowledge points.

[0008] In some embodiments, extracting the statistical feature vector of each knowledge point in the historical interaction sequence includes: Statistically analyze the global difficulty, discrimination, or average accuracy of each knowledge point in the historical interaction sequence. The question-answering statistical feature vector is constructed based on the global difficulty, the discrimination, or the average accuracy. The question-answering statistical feature vector is used to capture the empirical difficulty attribute of each knowledge point.

[0009] In some embodiments, constructing a context vector based on the cognitive attribute category probability, the latent semantic vector, the historical interaction sequence, and state-driven weights includes: Calculate multidimensional distance and basic similarity based on the cognitive attribute category probability, the latent semantic vector, and the historical interaction sequence; Calculate the weighted distance penalty term based on the multidimensional distance and the state-driven weights; The attention score is calculated based on the basic similarity and the weighted distance penalty term. Construct a context vector based on the state-driven weights and the attention score.

[0010] In some embodiments, calculating the multidimensional distance based on the cognitive attribute category probability, the latent semantic vector, and the historical interaction sequence includes: Calculate the semantic distance based on the latent semantic vector; Calculate attribute distance, difficulty distance, and time distance based on the cognitive attribute category probability and the historical interaction sequence; The multidimensional distance includes the semantic distance, the attribute distance, the difficulty distance, and the time distance.

[0011] In some embodiments, updating the attribute mastery status of the target student based on the prediction error of the cognitive attribute category probability and the correct answer probability includes: The gating coefficient is calculated based on the prediction error of the cognitive attribute category probability, the correct answer probability, and the attribute mastery status at the previous time point. Update the attribute mastery status at the current time point based on the gating coefficient.

[0012] To achieve the above objectives, another aspect of this application proposes a knowledge tracking device based on multidimensional cognitive space tomographic decoupling, the device comprising: The first module is used to obtain the historical interaction sequence of the target student during the historical problem-solving process and the current exercises to be done. The second module is used to input the historical interaction sequence into a knowledge tracking model based on multidimensional cognitive space hierarchical decoupling, so that after the knowledge tracking model is trained on the historical interaction sequence, it can predict the probability of answering the current exercise. The knowledge tracking model includes a multi-view feature extraction and fusion unit, a Gaussian mixture autoencoder, and a sequence interactive coding and closed-loop state tracking unit. The multi-view feature extraction and fusion unit includes a semantic feature extraction module, a graph structure feature extraction module, a question-answering result statistical feature extraction module, and a gated feature mixer. The semantic feature extraction module is used to extract high-dimensional semantic vectors corresponding to knowledge points in the historical interaction sequence. The graph structure feature extraction module is used to extract structured embedding vectors of knowledge points in the historical interaction sequence. The question-answering result statistical feature extraction module is used to extract question-answering statistical feature vectors for each knowledge point in the historical interaction sequence. The gated feature mixer is used to perform a fusion operation on the high-dimensional semantic vectors, the structured embedding vectors, and the question-answering statistical feature vectors to obtain the target fusion feature. The Gaussian mixture autoencoder is used to decouple the knowledge points in the historical interaction sequence into cognitive attribute category probabilities and latent semantic vectors based on the target fusion feature vector; The sequence interaction encoding and closed-loop state tracking unit includes an attribute-specialized attention mechanism module, a cognitive gating update module, and a prediction module. The attribute-specialized attention mechanism module is used to construct a context vector based on the cognitive attribute category probability, the latent semantic vector, the historical interaction sequence, and the state-driven weights. The context vector is used to represent the target cognitive surface information of the target student. The prediction module is used to predict the probability of answering the current exercise based on the context vector and the attribute mastery state. The cognitive gating update module is used to update the attribute mastery state of the target student based on the prediction error between the cognitive attribute category probability and the probability of answering correctly. The attribute mastery state is used to update the state-driven weights.

[0013] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0014] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods described above.

[0015] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method.

[0016] The embodiments of this application include at least the following beneficial effects: This application provides a knowledge tracking method and related equipment based on multidimensional cognitive space layered decoupling. This scheme inputs the acquired historical interaction sequence into a knowledge tracking model based on multidimensional cognitive space layered decoupling. The multi-view feature extraction and fusion unit in the model extracts and fuses feature vectors from semantic, structural, and statistical aspects to obtain target fusion features. The target fusion features can then characterize the multidimensional attributes of knowledge points. Then, the Gaussian mixture autoencoder in the model decouples the knowledge points in the historical interaction sequence into cognitive attribute category probabilities and latent semantic vectors based on the target fusion feature vectors, thereby deeply mining the latent cognitive dimensions. Finally, the sequence interaction encoding and closed-loop state tracking unit in the model performs dynamic knowledge updates and closed-loop tracking based on cognitive attribute category probabilities and latent semantic vectors, thereby effectively improving the interpretability and targeting of knowledge tracking, and thus improving the accuracy of knowledge tracking prediction. Attached Figure Description

[0017] Figure 1 This is a flowchart of the knowledge tracking method based on multidimensional cognitive space hierarchical decoupling provided in the embodiments of this application; Figure 2 This is a schematic diagram of the modules of the knowledge tracking model provided in the embodiments of this application; Figure 3 This is a data processing diagram of the attribute specialization attention mechanism module provided in an embodiment of this application; Figure 4 This is a data processing diagram of the cognitive gating update module provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of the knowledge tracking device based on multidimensional cognitive space tomography decoupling provided in the embodiments of this application; Figure 6 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0019] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0020] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0022] Among related technologies, Knowledge Tracing (KT) is a key technology in the field of intelligent education. It aims to track students' knowledge mastery in real time based on their historical learning interaction sequences (such as problem-solving records) and predict their performance in future exercises. With the popularization of online education platforms, massive amounts of learning data have been recorded, and how to utilize this data to achieve personalized recommendations and precise tutoring has become a research hotspot. Although deep learning technology has achieved significant results in KT tasks, most existing methods treat knowledge points merely as simple IDs (one-hot encoding) or single-dimensional embedding vectors, ignoring the rich semantic information and structural relationships of the knowledge points themselves. More importantly, existing methods often lack in-depth mining and utilization of the potential cognitive attributes of knowledge points. For example, whether the knowledge point mainly tests logical reasoning ability, spatial imagination ability, or memory ability, students often show consistency in their learning performance for knowledge points or exercises with the same or similar attributes. This lack of deep attributes limits the performance and interpretability of the model. Furthermore, the cold start problem is also a major challenge in this field. When faced with newly appearing knowledge points or exercises in the system, traditional models struggle to quickly construct accurate representations due to the lack of historical interaction data, thus affecting prediction performance. To address these issues, existing methods such as PEBG (Pre-training Embeddings based on Bipartite Graphs) utilize bipartite graph embeddings to capture the association between questions and concepts; GKT (Graph Knowledge Tracing) uses graph neural networks to model the structural relationships between knowledge points; EKT (Exercise-enhanced Knowledge Tracing) incorporates question text information to enhance representation; QIKT (Question-centric Interpretable Knowledge Tracing) focuses on question-level features to improve interpretability; AKT (Context-Aware Attentive Knowledge Tracing) uses attention mechanisms to capture contextual dependencies; and RKT (Relation-Aware Knowledge Tracing) introduces relational contextual information. However, while these methods introduce external information (such as text or graph structures), most are simply feature fusions, failing to explicitly decouple the discrete cognitive attribute categories behind knowledge points, nor establishing an explicit correspondence between "cognitive attributes and student states." Meanwhile, existing methods still rely on historical interaction data or pre-trained static embeddings when dealing with the cold start problem, but they cannot quickly generate dynamic feature representations that adapt to the current task for new knowledge points, which leads to a decline in the predictive performance of knowledge tracing.

[0023] In view of this, this application provides a knowledge tracing method and related equipment based on multidimensional cognitive space layer-by-layer decoupling, which can effectively improve the interpretability and pertinence of knowledge tracing and improve the accuracy of knowledge tracing prediction.

[0024] The knowledge tracing method based on multidimensional cognitive space hierarchical decoupling provided in this application relates to the field of intelligent education technology. This method can be applied to terminals, servers, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or in-vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the knowledge tracing method based on multidimensional cognitive space hierarchical decoupling, but is not limited to the above forms.

[0025] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0026] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0027] The embodiments of this application will be described in detail below with reference to the accompanying drawings: Figure 1 This is an optional flowchart of a knowledge tracing method based on multidimensional cognitive space hierarchical decoupling provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S110 to S120: Step S110: Obtain the historical interaction sequence of the target student during the history problem-solving process and the current unfinished exercises; Step S120: Input the historical interaction sequence into the knowledge tracking model based on multidimensional cognitive space hierarchical decoupling, so that after the knowledge tracking model is trained on the historical interaction sequence, it can predict the probability of answering the current exercise.

[0028] It is understandable that, such as Figure 2As shown, the knowledge tracking model based on multidimensional cognitive space layered decoupling in this embodiment includes a multi-view feature extraction and fusion unit, a Gaussian mixture autoencoder, and a sequence interaction coding and closed-loop state tracking unit. Specifically, the multi-view feature extraction and fusion unit includes a semantic feature extraction module, a graph structure feature extraction module, a question-answering result statistical feature extraction module, and a gated feature mixer. The semantic feature extraction module is used to extract high-dimensional semantic vectors corresponding to knowledge points in the historical interaction sequence (Key / Value History). The graph structure feature extraction module is used to extract structured embedding vectors of knowledge points in the historical interaction sequence. The question-answering result statistical feature extraction module is used to extract question-answering statistical feature vectors of each knowledge point in the historical interaction sequence. The gated feature mixer is used to perform a fusion operation on the high-dimensional semantic vector, the structured embedding vector, and the question-answering statistical feature vector to obtain the target fused feature. The Gaussian mixture autoencoder is used to decouple the knowledge points in the historical interaction sequence into cognitive attribute category probabilities p and latent semantic vectors z according to the target fused feature vector. The sequence interaction encoding and closed-loop state tracking unit includes an attribute-specialized attention mechanism module, a cognitive gating update module, and a prediction module. The attribute-specialized attention mechanism module is used to drive weights based on the cognitive attribute category probability p, the latent semantic vector z, the historical interaction sequence, and the state. A context vector is constructed to represent the target student's cognitive information. The prediction module predicts the probability of answering the current exercise based on the context vector and the attribute mastery state. The cognitive gating update module updates the target student's attribute mastery state s_u based on the prediction error between the cognitive attribute category probability and the probability of answering correctly (PredictionProbability). The attribute mastery state s_u is used to update the state-driven weights. .

[0029] Understandably, the multi-view feature extraction and fusion unit is responsible for extracting features of knowledge points from three different dimensions to comprehensively characterize the attributes of knowledge points. Specifically, in this embodiment, during the process of obtaining the high-dimensional semantic vectors corresponding to knowledge points in the historical interaction sequence, after obtaining the semantic view of the historical interaction sequence, a pre-trained large language model (such as BERT) is used to encode the textual descriptions of knowledge points in the semantic view to obtain high-dimensional semantic vectors, thereby capturing the conceptual connotations of knowledge points. The textual descriptions include the knowledge point name, parsing information, the cognitive dimension and competency being assessed corresponding to the exercises. During the extraction of structured embedding vectors of knowledge points in the historical interaction sequence, a knowledge point association graph can be constructed as a graph view based on the historical interaction sequence. Then, a graph attention network (GAT) is used to perform message passing in the graph view to extract the structured embedding vectors of knowledge points, thereby capturing the logical dependencies between knowledge points. In the process of extracting the statistical feature vector of each knowledge point in the historical interaction sequence, we can construct the statistical feature vector of each knowledge point based on the global difficulty (error rate), discrimination or average accuracy rate of each knowledge point by statistically analyzing the large-scale question-answering data in the historical interaction sequence, so as to capture the empirical difficulty attribute of each knowledge point.

[0030] Understandably, in order to adaptively select effective information from different views, this embodiment designs a gated feature mixer. This gated feature mixer adopts a residual gating structure, which allows the knowledge tracing model to dynamically decide whether to retain the original spliced ​​features or to superimpose the enhanced features after nonlinear transformation. This allows for the fusion operation of high-dimensional semantic vectors, structured embedding vectors, and question-solving statistical feature vectors to obtain high-quality target fused features.

[0031] Understandably, to uncover the underlying cognitive dimensions (such as "logical reasoning" or "spatial computation") behind knowledge points, this application utilizes a Gaussian mixture autoencoder to process the target fusion feature vector. The Gaussian mixture autoencoder assumes that the data distribution is a mixture of K Gaussian distributions, each representing a potential cognitive attribute category. Specifically, the encoder includes an inference network and a generator network. The inference network, on the one hand, predicts the probability of a knowledge point belonging to each cognitive attribute category using a multilayer perceptron and employs the Gumbel-Softmax technique for differentiable discrete sampling to obtain a category distribution vector (one-hot approximation), thus explicitly classifying knowledge points to specific cognitive dimensions. On the other hand, given the input and the inferred category, it infers the specific performance characteristics (mean and variance) of the knowledge point under a specific cognitive dimension and uses a reparameterization technique to sample continuous latent variables. The generator network is responsible for generating the prior distribution of latent variables based on the categories and reconstructing the original input features based on the latent variables. Unlike traditional unsupervised clustering, the Gaussian mixture autoencoder in this application is jointly trained with the subsequent knowledge tracking task. The prediction error gradient of the knowledge tracing task is fed back to the Gaussian mixture autoencoder, forcing the "cognitive categories" learned by the Gaussian mixture autoencoder to not only cluster in the feature space, but also to have discriminative power in "predicting student performance". The final output is the latent semantic vector z and the cognitive attribute category probability p.

[0032] It is understood that the sequence interaction encoding and closed-loop state tracking unit in this embodiment includes an attribute-specialized attention mechanism module, a cognitive gating update module, and a prediction module. Specifically, in order to accurately capture students' historical performance on specific cognitive dimensions, this embodiment designs an attention mechanism that combines explicit distance penalty and student state gating. Figure 3 As shown, in the attribute-specialized attention module, after calculating the multidimensional distance and basic similarity based on the cognitive attribute category probability, latent semantic vector, and historical interaction sequence, a weighted distance penalty term (Penalty=Sum( *Distance), and then calculate the attention score (Final=Base-Penalty) based on the base similarity (Base) and the weighted distance penalty term (Distance), and then construct the context vector based on the state-driven weights and attention score. Specifically, in this embodiment, when predicting the accuracy of the current exercise (Query), the relevant information of the current exercise (Query) and the historical interaction sequence (key / value) can be input together into the attribute specialization attention mechanism module. This attribute specialization attention mechanism module calculates the prediction accuracy of the current exercise by introducing dynamic penalty weight generation and multi-dimensional distance penalty. The knowledge tracking model of this embodiment maintains an explicit student attribute mastery state vector. The dimension of the attribute mastery state vector is consistent with the number of cognitive attribute categories K. The multilayer perceptron is used to map the student attribute mastery vector to the penalty weight of each head in the multi-head attention, so that the penalty weight represents the student's current cognitive state and dynamically determines which feature dimensions the model focuses on. When calculating the attention score, this embodiment uses dot product similarity to subtract four explicit multi-dimensional distances to filter out historical records with weak relevance. The multidimensional distance calculation process includes calculating the semantic distance based on the latent semantic vector z-axis extracted by a Gaussian mixture autoencoder, and calculating the attribute distance, difficulty distance, and temporal distance of the current exercise based on the cognitive attribute category probability p and historical interaction sequences. In this embodiment, the output of the multi-head attention is weighted and fused through a gating network controlled by the student's attribute mastery state, ensuring that the final context vector reflects the cognitive aspect that the student currently needs to focus on most.

[0033] Understandably, in the Cognitive Gated Update Module (CGU), such as Figure 4As shown, in this embodiment, after calculating the gating coefficient based on the prediction error of the cognitive attribute category probability, the probability of answering correctly, and the attribute mastery state at the previous time point, the attribute mastery state at the current time point is updated based on the gating coefficient. The cognitive gating update module of this embodiment can achieve a closed loop of "prediction-feedback-update," not only predicting the student's answer but also updating the student's cognitive state in real time. Specifically, this embodiment first concatenates the attention context, the question features of the current exercise, and the student's current attribute mastery state, and inputs them into the prediction module. To strengthen the role of the student's attribute mastery state, residual mastery injection is introduced, and finally, the probability of answering correctly for the current exercise is obtained through the Sigmoid activation function. Secondly, after obtaining the actual answer result, the prediction error is calculated, and the student's attribute mastery state is updated using a cognitive gating update unit similar to GRU. In this embodiment, the new quantity of the cognitive gating update module is jointly determined by the error and the question attribute probability. For example, if the question mainly belongs to the k-th cognitive attribute, then the prediction error will mainly be used to correct the k-th component in the student's attribute mastery vector. This mechanism ensures that the updating of student status is attribute-decoupled, enabling accurate tracking of students' abilities in various dimensions.

[0034] As described above, the method in this application integrates multi-view features of knowledge points (semantic text embedding, graph structure embedding, and statistical features), and decouples the representations through a Gaussian mixture variational autoencoder (GMVAE) jointly trained with the knowledge tracking main task. This allows the output discrete cognitive attribute distribution and continuous semantic variables to form "task-related attributes" that are more discriminative of student performance under the supervision of prediction error. Furthermore, in cold-start scenarios, for new knowledge points lacking or without accumulated interaction records, this application can at least generate semantic view features based on their text descriptions (the text can be expanded by a large language model to complete the description before embedding and encoding), thereby obtaining initial knowledge point representations that can be used for tracking. The GMVAE can then further infer their potential attribute attribution. In addition, this application proposes an attribute-specialized attention mechanism, explicitly introducing multiple distance penalties (including semantic distance, attribute distance, difficulty difference, and time interval) in attention scoring, and dynamically generating penalty weights based on the student's cognitive state, thereby achieving "state-guided attention." Regarding state updates, this application uses a cognitive gating update unit (CGU) to allocate prediction errors according to the probability of question attributes and update the corresponding dimensions, forming a closed-loop tracking mechanism of "prediction-feedback-update", thereby improving the pertinence and interpretability of predictions and providing a clearer basis for cognitive diagnosis.

[0035] Please see Figure 5 This application also provides a knowledge tracking device based on multidimensional cognitive space tomographic decoupling, the device comprising: The first module is used to obtain the historical interaction sequence of the target student during the historical problem-solving process and the current exercises to be done. The second module is used to input the historical interaction sequence into the knowledge tracking model based on the hierarchical decoupling of the multidimensional cognitive space. After the knowledge tracking model is trained based on the historical interaction sequence, it predicts the probability of answering the current exercise. The knowledge tracking model includes a multi-view feature extraction and fusion unit, a Gaussian mixture autoencoder, and a sequence interaction coding and closed-loop state tracking unit. The multi-view feature extraction and fusion unit includes a semantic feature extraction module, a graph structure feature extraction module, a question-answering result statistical feature extraction module, and a gated feature mixer. The semantic feature extraction module is used to extract high-dimensional semantic vectors corresponding to knowledge points in the historical interaction sequence. The graph structure feature extraction module is used to extract structured embedding vectors of knowledge points in the historical interaction sequence. The question-answering result statistical feature extraction module is used to extract question-answering statistical feature vectors for each knowledge point in the historical interaction sequence. The gated feature mixer is used to fuse the high-dimensional semantic vectors, structured embedding vectors, and question-answering statistical feature vectors to obtain the target fused feature. Gaussian mixture autoencoders are used to decouple knowledge points in historical interaction sequences into cognitive attribute category probabilities and latent semantic vectors based on target fusion feature vectors; The sequence interaction encoding and closed-loop state tracking unit includes an attribute-specialized attention mechanism module, a cognitive gating update module, and a prediction module. The attribute-specialized attention mechanism module is used to construct a context vector based on the cognitive attribute category probability, latent semantic vector, historical interaction sequence, and state-driven weights, wherein the context vector is used to represent the target cognitive surface information of the target student. The prediction module is used to predict the probability of answering the current exercise based on the context vector and the attribute mastery state. The cognitive gating update module is used to update the attribute mastery state of the target student based on the prediction error between the cognitive attribute category probability and the probability of answering correctly, wherein the attribute mastery state is used to update the state-driven weights.

[0036] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0037] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0038] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0039] Please see Figure 6 , Figure 6 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 610 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 620 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 620 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 620 and is called and executed by the processor 610 using the methods described in the embodiments of this application. The input / output interface 630 is used to realize information input and output; The communication interface 640 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 650 transmits information between various components of the device (e.g., processor 610, memory 620, input / output interface 630, and communication interface 640); The processor 610, memory 620, input / output interface 630 and communication interface 640 are connected to each other within the device via bus 650.

[0040] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0041] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0042] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0043] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0044] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0045] This application provides a knowledge tracking method and related equipment based on multidimensional cognitive space layered decoupling. The method involves inputting the acquired historical interaction sequence into a knowledge tracking model based on multidimensional cognitive space layered decoupling. The model's multi-view feature extraction and fusion unit extracts and fuses feature vectors from semantic, structural, and statistical aspects to obtain target fusion features. These target fusion features characterize the multidimensional attributes of knowledge points. Then, the model's Gaussian mixture autoencoder decouples the knowledge points in the historical interaction sequence into cognitive attribute category probabilities and latent semantic vectors based on the target fusion feature vectors, thereby deeply mining latent cognitive dimensions. Finally, the model's sequence interaction encoding and closed-loop state tracking unit dynamically updates knowledge and performs closed-loop tracking based on cognitive attribute category probabilities and latent semantic vectors. This effectively improves the interpretability and targeting of knowledge tracking, thereby enhancing the accuracy of knowledge tracking predictions.

[0046] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0047] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0048] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0049] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0050] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0051] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0052] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above 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 apparatuses or units may be electrical, mechanical, or other forms.

[0053] The units described above 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.

[0054] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0055] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0056] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A knowledge tracing method based on multidimensional cognitive space hierarchical decoupling, characterized in that, The method includes the following steps: Obtain the historical interaction sequence of the target student during the history problem-solving process and the current unfinished exercises; The historical interaction sequence is input into a knowledge tracking model based on multidimensional cognitive space hierarchical decoupling, so that after the knowledge tracking model is trained on the historical interaction sequence, it can predict the probability of answering the current exercise. The knowledge tracking model includes a multi-view feature extraction and fusion unit, a Gaussian mixture autoencoder, and a sequence interactive coding and closed-loop state tracking unit. The multi-view feature extraction and fusion unit includes a semantic feature extraction module, a graph structure feature extraction module, a question-answering result statistical feature extraction module, and a gated feature mixer. The semantic feature extraction module is used to extract high-dimensional semantic vectors corresponding to knowledge points in the historical interaction sequence. The graph structure feature extraction module is used to extract structured embedding vectors of knowledge points in the historical interaction sequence. The question-answering result statistical feature extraction module is used to extract question-answering statistical feature vectors for each knowledge point in the historical interaction sequence. The gated feature mixer is used to perform a fusion operation on the high-dimensional semantic vectors, the structured embedding vectors, and the question-answering statistical feature vectors to obtain the target fusion feature. The Gaussian mixture autoencoder is used to decouple the knowledge points in the historical interaction sequence into cognitive attribute category probabilities and latent semantic vectors based on the target fusion feature vector; The sequence interaction encoding and closed-loop state tracking unit includes an attribute-specialized attention mechanism module, a cognitive gating update module, and a prediction module. The attribute-specialized attention mechanism module is used to construct a context vector based on the cognitive attribute category probability, the latent semantic vector, the historical interaction sequence, and the state-driven weights. The context vector is used to represent the target cognitive surface information of the target student. The prediction module is used to predict the probability of answering the current exercise based on the context vector and the attribute mastery state. The cognitive gating update module is used to update the attribute mastery state of the target student based on the prediction error between the cognitive attribute category probability and the probability of answering correctly. The attribute mastery state is used to update the state-driven weights.

2. The method according to claim 1, characterized in that, The step of extracting the high-dimensional semantic vectors corresponding to the knowledge points in the historical interaction sequence includes: Obtain a semantic view of the historical interaction sequence; The textual descriptions of knowledge points in the semantic view are encoded using a large language model to obtain the high-dimensional semantic vectors; the textual descriptions include the knowledge point names, parsing information, the cognitive dimensions and competencies assessed by the exercises.

3. The method according to claim 1, characterized in that, The extraction of structured embedding vectors for knowledge points from the historical interaction sequence includes: A knowledge point association graph is constructed as a graph view based on the historical interaction sequence; The graph attention network extracts structured embedding vectors of knowledge points from the graph view, which are used to capture the logical dependencies between knowledge points.

4. The method according to claim 1, characterized in that, The step of extracting the question-answering statistical feature vector for each knowledge point in the historical interaction sequence includes: Statistically analyze the global difficulty, discrimination, or average accuracy of each knowledge point in the historical interaction sequence. The question-answering statistical feature vector is constructed based on the global difficulty, the discrimination, or the average accuracy. The question-answering statistical feature vector is used to capture the empirical difficulty attribute of each knowledge point.

5. The method according to claim 1, characterized in that, The step of constructing a context vector based on the cognitive attribute category probability, the latent semantic vector, the historical interaction sequence, and the state-driven weights includes: Calculate multidimensional distance and basic similarity based on the cognitive attribute category probability, the latent semantic vector, and the historical interaction sequence; Calculate the weighted distance penalty term based on the multidimensional distance and the state-driven weights; The attention score is calculated based on the basic similarity and the weighted distance penalty term. Construct a context vector based on the state-driven weights and the attention score.

6. The method according to claim 5, characterized in that, The step of calculating the multidimensional distance based on the cognitive attribute category probability, the latent semantic vector, and the historical interaction sequence includes: Calculate the semantic distance based on the latent semantic vector; Calculate attribute distance, difficulty distance, and time distance based on the cognitive attribute category probability and the historical interaction sequence; The multidimensional distance includes the semantic distance, the attribute distance, the difficulty distance, and the time distance.

7. The method according to claim 1, characterized in that, The step of updating the target student's attribute mastery status based on the prediction error of the cognitive attribute category probability and the correct answer probability includes: The gating coefficient is calculated based on the prediction error of the cognitive attribute category probability, the correct answer probability, and the attribute mastery status at the previous time point. Update the attribute mastery status at the current time point based on the gating coefficient.

8. An electronic device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method as described in any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images