Self-adaptive learning path generation method and device based on cognitive diagnosis, equipment and medium

Abstract

The invention relates to an adaptive learning path generation method and device based on cognitive diagnosis, equipment and a medium. The method comprises the following steps: acquiring a cognitive state vector of a user; the cognitive state vector is used for including a mastering degree value of the user to each knowledge point; based on the cognitive state vector and a preset knowledge point mapping knowledge domain, selecting knowledge points to form a learnable knowledge point set; inputting the cognitive state vector and the knowledge point mapping knowledge domain into a preset knowledge learning potential prediction model to obtain a learning potential vector; the learning potential vector comprises a future potential value of each knowledge point; the knowledge learning potential prediction model is constructed based on a graph neural network model; generating an optimal learning path based on the learning potential vector, the cognitive state vector, the knowledge point mapping knowledge domain and the learnable knowledge point set; the optimal learning path is used for indicating the learning sequence of the knowledge points. By adopting the method, the learning path balancing the short-term learning efficiency and the long-term development potential can be generated.

Description

Technical Field

[0001] This invention belongs to the field of intelligent education, and in particular relates to a method, apparatus, device and medium for generating adaptive learning paths based on cognitive diagnosis. Background Technology

[0002] With the development of intelligent education technology, adaptive learning path generation technology based on cognitive diagnosis has emerged. This technology can dynamically plan personalized learning sequences for learners based on their real-time knowledge mastery status, thus forming a traditional path generation method with real-time cognitive diagnosis and multi-objective optimization as its core.

[0003] Traditional techniques typically employ a phased modeling and flattening optimization approach. First, learners' historical response data is analyzed using item response theory or multidimensional cognitive diagnostic models to obtain their mastery vectors for each knowledge point. Then, this vector is combined with a predefined knowledge graph to filter out the set of all currently learnable knowledge points that satisfy prerequisite relationships. Finally, a multi-objective evolutionary algorithm is applied to this set, simultaneously optimizing multiple objectives such as learning efficiency, cognitive load, and immediate mastery. Through non-dominated ranking and selection mechanisms, a path is selected from the Pareto optimal solution set and recommended to the learner.

[0004] However, this traditional method optimizes goals at different time scales on the same level, and its search scope is strictly limited to the current "learnable" set. This results in the system only being able to identify and recommend the path that is most effective in the short term. This method is essentially a local optimization and cannot assess the potential strategic value of a seemingly inefficient learning choice in unlocking multiple key knowledge modules in the future and enhancing the long-term connectivity of the knowledge network. In other words, in a knowledge system with complex dependencies, it cannot balance the fundamental contradiction between short-term learning efficiency and long-term development potential, so the generated learning path may hinder the development of learners' knowledge transfer ability and deep understanding. Summary of the Invention

[0005] Therefore, it is necessary to provide a cognitive diagnostic-based adaptive learning path generation method, apparatus, device, and medium that can generate learning paths that balance short-term learning efficiency and long-term development potential, addressing the aforementioned technical problems.

[0006] Firstly, this application provides an adaptive learning path generation method based on cognitive diagnosis, including:

[0007] Obtain the user's cognitive state vector; the cognitive state vector is used to include the user's level of mastery of each knowledge point.

[0008] Based on the cognitive state vector and the pre-set knowledge point knowledge graph, knowledge points are selected to form a set of learnable knowledge points; the knowledge point knowledge graph includes graph nodes and directed edges; graph nodes are used to represent each knowledge point, and directed edges are used to represent the relationship between each knowledge point;

[0009] The cognitive state vector and the knowledge graph of knowledge points are input into a pre-defined knowledge learning potential prediction model to obtain a learning potential vector. The learning potential vector includes the future potential value of each knowledge point. The knowledge learning potential prediction model is constructed based on a graph neural network model.

[0010] Based on the learning potential vector, cognitive state vector, knowledge point knowledge graph, and set of learnable knowledge points, an optimal learning path is generated; the optimal learning path is used to indicate the learning order of each knowledge point.

[0011] Furthermore, the knowledge graph of knowledge points was obtained through the following methods:

[0012] Obtain a preset set of knowledge points; the set of knowledge points includes the correlation feature vectors between each knowledge point, as well as the learning value parameters and mastery difficulty values ​​of each knowledge point;

[0013] Based on the set of knowledge points, each knowledge point is used as a graph node, and the learning value parameter and mastery difficulty value of each knowledge point are used as the graph node features to obtain the first knowledge point knowledge graph.

[0014] Based on the correlation feature vectors between knowledge points in the knowledge point set and the first knowledge point knowledge graph, directed connection edges between graph nodes are constructed to obtain the second knowledge point knowledge graph; the direction of the directed connection edges is used to represent the direction of the correlation relationship between knowledge points.

[0015] Based on the graph node features of the graph nodes at both ends of the directed connection edges and the preset association weights, the weights of each directed connection edge are calculated to obtain the edge weights of each directed connection edge; and based on the edge weights of each directed connection edge, the knowledge graph of the second knowledge point is adjusted to obtain the knowledge graph of the knowledge point.

[0016] Furthermore, based on the cognitive state vector and the pre-defined knowledge graph, knowledge points are selected to form a set of learnable knowledge points, including:

[0017] Based on the cognitive state vector and the preset mastery threshold, the knowledge points in the cognitive state vector whose mastery value is less than the mastery threshold are identified as unfamiliar knowledge points.

[0018] For each unfamiliar knowledge point, extract the knowledge points from the knowledge point knowledge graph that are connected by directed edges and point to the unfamiliar knowledge point, and form the set of prerequisite knowledge points for the unfamiliar knowledge point.

[0019] For each unfamiliar knowledge point, based on the cognitive state vector and mastery threshold, the unfamiliar knowledge points whose mastery value is greater than the mastery threshold are identified as learnable knowledge points.

[0020] Based on each learnable knowledge point, a set of learnable knowledge points is obtained.

[0021] Furthermore, the knowledge learning potential prediction model was obtained through the following method:

[0022] Acquire preset historical learning trajectory data and knowledge graph of knowledge points; the historical learning trajectory data includes the execution timestamps, cognitive state vectors and sets of learned knowledge points for each historical learning event; the set of learned knowledge points is used to represent the knowledge points learned after executing the historical learning event;

[0023] For each historical learning event, select other historical learning events whose execution timestamps meet the preset forward return calculation time window from the historical learning trajectory data to form a potential reference event set for each historical learning event;

[0024] For each historical learning event, based on the knowledge graph of knowledge points, the influence of the cognitive state vector on the set of learning knowledge points of each historical learning event in the potential reference event set is quantified to obtain the future potential value of each knowledge point in each historical learning event.

[0025] For each historical learning event, a model is constructed using a knowledge graph of knowledge points and a cognitive state vector as input features and future potential as the prediction target. The model construction dataset is then divided according to a preset partitioning ratio to obtain a model training set and a model test set.

[0026] An initial knowledge learning potential prediction model is constructed based on a graph neural network model, and the initial knowledge learning potential prediction model is iteratively trained using a model training set to obtain a preliminary knowledge learning potential prediction model.

[0027] The preliminary knowledge learning potential prediction model is tested using a model test set until it meets the preset accuracy requirements, thus obtaining the knowledge learning potential prediction model.

[0028] Furthermore, based on the learning potential vector, cognitive state vector, knowledge point knowledge graph, and set of learnable knowledge points, an optimal learning path is generated, including:

[0029] For each knowledge point in the set of learnable knowledge points, select the future potential value of the knowledge point from the learning potential vector to form a set of learnable knowledge potential values.

[0030] The future potential values ​​in the set of learnable knowledge potential values ​​are sorted in descending order, and the knowledge points whose sorting position matches the preset sampling position are selected to form a set of candidate knowledge points.

[0031] Based on the candidate knowledge point set, the knowledge points corresponding to the future potential values ​​are sorted in descending order of future potential value to obtain the initial learning path;

[0032] Based on the learning potential vector, knowledge point knowledge graph, and cognitive state vector, the initial learning path is adjusted to obtain the optimal learning path.

[0033] Furthermore, based on the learning potential vector, knowledge point knowledge graph, and cognitive state vector, the initial learning path is adjusted to obtain the optimal learning path, including:

[0034] Based on the cognitive state vector, the simulated cognitive state vector before completing each knowledge point is calculated using the following formula:

[0035]

[0036] in, It is any position in the initial learning path. It is the position in completing the initial learning path. The simulated cognitive state vector before the knowledge point It is a state simulation function. It is the position in completing the initial learning path. The simulated cognitive state vector before the knowledge point It is the position in the initial learning path. Knowledge points;

[0037] Based on the simulated cognitive state vector, knowledge graph of knowledge points, and learning potential vector, the fitness of the initial learning path is calculated using the following formula:

[0038]

[0039] in, It is the fitness of the initial learning path. This is the total number of knowledge points in the initial learning path. It is any position in the initial learning path. It is the position in completing the initial learning path. The simulated cognitive state vector before the knowledge point It is a load estimation function. It is the position in the initial learning path. Knowledge points, It is the load weight. It is an efficiency estimation function. It is an efficiency weight. It is the gain estimation function. It is the gain weight. It is a potential weight. It is the position in the initial learning path. The future potential value of the knowledge points. It is the discount factor;

[0040] Based on the knowledge graph of knowledge points, the order of knowledge points in the initial learning path is adjusted to obtain a new initial learning path;

[0041] Calculate the fitness of each new initial learning path, and determine the initial learning path with the highest fitness value as the optimal learning path.

[0042] Secondly, this application also provides an adaptive learning path generation device based on cognitive diagnosis, comprising:

[0043] The data acquisition module is used to acquire the user's cognitive state vector; the cognitive state vector includes the user's level of mastery of various knowledge points.

[0044] The learnable knowledge point extraction module is used to select knowledge points to form a set of learnable knowledge points based on the cognitive state vector and the preset knowledge point knowledge graph. The knowledge point knowledge graph includes graph nodes and directed edges. Graph nodes are used to represent each knowledge point, and directed edges are used to represent the relationship between each knowledge point.

[0045] The learnable knowledge point potential analysis module is used to input the cognitive state vector and the knowledge point knowledge graph into the preset knowledge learning potential prediction model to obtain the learning potential vector; the learning potential vector includes the future potential value of each knowledge point; the knowledge learning potential prediction model is constructed based on the graph neural network model.

[0046] The learning path generation module is used to generate the optimal learning path based on the learning potential vector, cognitive state vector, knowledge point knowledge graph, and set of learnable knowledge points; the optimal learning path is used to indicate the learning order of each knowledge point.

[0047] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the adaptive learning path generation methods based on cognitive diagnosis described in the first aspect of this application.

[0048] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any of the adaptive learning path generation methods based on cognitive diagnosis described in the first aspect.

[0049] The aforementioned adaptive learning path generation method, apparatus, device, and medium based on cognitive diagnosis acquires the user's cognitive state vector. This cognitive state vector includes the user's mastery level of each knowledge point. Based on the cognitive state vector and a pre-set knowledge point knowledge graph, knowledge points are selected to form a set of learnable knowledge points. The knowledge point knowledge graph includes graph nodes and directed edges. Graph nodes represent each knowledge point, and directed edges represent the relationships between knowledge points. The cognitive state vector and the knowledge point knowledge graph are input into a pre-set knowledge learning potential prediction model to obtain a learning potential vector. This learning potential vector includes the future potential value of each knowledge point. The knowledge learning potential prediction model is constructed based on a graph neural network model. Based on the learning potential vector, cognitive state vector, knowledge point knowledge graph, and set of learnable knowledge points, an optimal learning path is generated. The optimal learning path indicates the learning order of each knowledge point. This overcomes the short-sightedness of traditional path generation by guiding micro-search through macro-potential, balancing immediate efficiency and long-term benefits, thereby improving the scientific nature and adaptability of the learning path and generating a learning path that balances short-term learning efficiency and long-term development potential. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 A flowchart illustrating an adaptive learning path generation method based on cognitive diagnosis, provided in one embodiment of this application;

[0052] Figure 2 A schematic diagram of the structure of an adaptive learning path generation device based on cognitive diagnosis provided in one embodiment of this application;

[0053] Figure 3 This is a schematic diagram of the structure of a computer device for an adaptive learning path generation method based on cognitive diagnosis, provided in one embodiment of this application. Detailed Implementation

[0054] 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 and not intended to limit the scope of this application.

[0055] In one embodiment, such as Figure 1As shown, an adaptive learning path generation method based on cognitive diagnosis is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following S101-S104, wherein:

[0056] S101, Obtain the user's cognitive state vector; the cognitive state vector is used to include the user's level of mastery of each knowledge point.

[0057] Specifically, the terminal acquires the user's cognitive state vector; this vector includes the user's level of mastery over various knowledge points. The user's cognitive state vector is a multi-dimensional real-valued vector, denoted as... , It is a cognitive state vector. It represents the total number of knowledge points, and each element... Indicates the user's opinion on the first The mastery level value for each knowledge point is denoted as 0, representing no mastery at all, and 1, representing complete mastery. The cognitive state vector is obtained based on the user's historical learning interaction data and is dynamically calculated through a preset cognitive diagnostic model. For example, the user's historical learning interaction data can be such as answer records or learning time series; the preset cognitive diagnostic model can adopt a multidimensional latent trait model that integrates item response theory (IRT) and fuzzy set theory, which estimates the mastery parameter by maximizing the likelihood function. This embodiment does not further limit the construction process of this preset cognitive diagnostic model.

[0058] S102, based on the cognitive state vector and the preset knowledge point knowledge graph, select knowledge points to form a set of learnable knowledge points; the knowledge point knowledge graph includes graph nodes and directed connecting edges; graph nodes are used to represent each knowledge point, and directed connecting edges are used to represent the relationship between each knowledge point.

[0059] Specifically, the pre-defined knowledge graph is a directed heterogeneous graph. .in, It is a collection of graph nodes, where each graph node corresponds to a knowledge point and each graph node is accompanied by graph node features; It is a set of directed connecting edges; It is a set of directed connection edge association types. The terminal inputs the mastery level values ​​of each knowledge point in the cognitive state vector into the preset knowledge point knowledge graph, and based on the directed connection edges in the preset knowledge point knowledge graph, it deduces the knowledge points that can be associated and learned according to the user's current mastery of each knowledge point, forming a set of learnable knowledge points.

[0060] S103, input the cognitive state vector and knowledge point knowledge graph into the preset knowledge learning potential prediction model to obtain the learning potential vector; the learning potential vector includes the future potential value of each knowledge point; the knowledge learning potential prediction model is constructed based on the graph neural network model.

[0061] Specifically, the pre-defined knowledge learning potential prediction model is a core component of hierarchical optimization, used to quantify the long-term strategic value of each knowledge point. This model is a pre-trained graph neural network, denoted as... Its input is the user's current cognitive state vector. and knowledge graph of knowledge points The output is a learning potential vector. ,in Representing knowledge points The future potential value (i.e., the predicted gain in the future knowledge exploration space after learning this point). For example, the preset knowledge learning potential prediction model adopts a Graph Attention Network (GAT) architecture, aggregating neighbor information through multi-layer message passing by setting a preset number of layers: each layer's node representation is updated to... ,in It is any node in the knowledge graph of knowledge points. It is the first Nodes in a convolutional layer of a layer graph The output vector. It is attention weight. It is an activation function. This indicates a graph node Iterate through all neighboring nodes and itself to sum them. It is the first Nodes in a convolutional layer of a layer graph or graph node The output vector of the neighboring nodes. It is the first The weight matrix of the layered convolutional layer. The prediction head concatenates the node representations with the mastery levels using a multilayer perceptron (MLP): To obtain knowledge points Its future potential.

[0062] S104 generates an optimal learning path based on the learning potential vector, cognitive state vector, knowledge point knowledge graph, and set of learnable knowledge points; the optimal learning path is used to indicate the learning order of each knowledge point.

[0063] Specifically, by integrating four types of information—the learning potential vector (reflecting long-term strategic value), the cognitive state vector (reflecting the current knowledge state), the knowledge point knowledge graph (reflecting the constraints between knowledge points), and the set of learnable knowledge points (reflecting the current feasible options)—an optimized decision-making process generates an ordered sequence of knowledge points, i.e., the optimal learning path. The path length is ,in Indicates recommendation in the first The knowledge points learned step by step. The core of the generation process is to weigh and optimize multiple objectives while satisfying the prior learning constraints of the learning graph. These objectives include at least: maximizing short-term learning efficiency, minimizing cognitive load, maximizing immediate knowledge gain, and maximizing long-term cumulative benefits quantified by the learning potential vector.

[0064] This embodiment provides an adaptive learning path generation method based on cognitive diagnosis. It quantifies the user's knowledge state by acquiring cognitive state vectors, generates a learnable set constraining the search space based on a knowledge graph, uses a GNN model to predict learning potential vectors to introduce long-term goals, and finally generates the optimal path through multi-objective optimization. This method overcomes the short-sightedness of traditional path generation by guiding micro-searches with macro-potential, balancing immediate efficiency and long-term benefits, thereby improving the scientific rigor and adaptability of the learning path and generating a learning path that balances short-term learning efficiency and long-term development potential.

[0065] In one embodiment, the knowledge graph of knowledge points is obtained through the following method:

[0066] S201, Obtain the preset set of knowledge points; the set of knowledge points includes the correlation feature vectors between each knowledge point, and the learning value parameters and mastery difficulty values ​​of each knowledge point.

[0067] Specifically, the preset set of knowledge points is a structured dataset, where each knowledge point corresponds to a unique identifier and contains the following fields: an association feature vector, used to represent the type of association between knowledge points; a learning value parameter, used to characterize the importance of the knowledge point, with a value ranging from [0,1]; and a mastery difficulty value, used to characterize the learner's difficulty with the knowledge point, with a value ranging from [0,1]. For example, the learning value parameter can be set based on exam scores or scalar values ​​annotated by experts; the mastery difficulty value can be obtained based on statistics of historical learner performance; and the association feature vector can be a binary vector, in the form of... Used for knowledge points Points of knowledge The association type is a prerequisite relationship, indicating that the learned knowledge points must be completed first. Only then can we proceed with the study of knowledge points. The learning. Used for knowledge points Points of knowledge The association type is similarity, representing knowledge points. and knowledge points This is a general concept similarity relationship.

[0068] S202, based on the set of knowledge points, with each knowledge point as a graph node, and with the learning value parameter and mastery difficulty value of the knowledge points as the graph node features, the first knowledge point knowledge graph is obtained.

[0069] Specifically, the terminal uses each knowledge point in the knowledge point set as a graph node. The graph node features are composed of learning value parameters and mastery difficulty values, resulting in a first knowledge point knowledge graph. This first knowledge point knowledge graph is an undirected graph that only contains the node set and node feature matrix, and does not yet include edge information.

[0070] S203. Based on the correlation feature vectors between knowledge points in the knowledge point set and the knowledge graph of the first knowledge point, a directed connection edge is constructed between graph nodes to obtain the knowledge graph of the second knowledge point; the direction of the directed connection edge is used to represent the direction of the correlation relationship between knowledge points.

[0071] Specifically, based on the association types between knowledge points represented by the associated feature vectors, the terminal constructs directed connections between the graph nodes corresponding to the knowledge points in the first knowledge point knowledge graph to obtain the second knowledge point knowledge graph. For example, there exists any associated feature vector... Representing knowledge points Points of knowledge The association type is a prerequisite relationship, and the terminal is based on... Find knowledge points in the knowledge graph of the first knowledge point Graph nodes and knowledge points The graph nodes, and construct a path from the knowledge points The knowledge points pointed to There exists a unidirectional directed connection edge. There exists any associated feature vector. Representing knowledge points Points of knowledge The association type is ordinary concept similarity relationship, and the terminal determines the relationship based on this. Find knowledge points in the knowledge graph of the first knowledge point Graph nodes and knowledge points The graph nodes, and construct a knowledge point and knowledge points A bidirectional directed connection edge.

[0072] S204. Based on the graph node features of the graph nodes at both ends of the directed connection edges and the preset association weights, calculate the weight of each directed connection edge to obtain the edge weight of each directed connection edge; and based on the edge weight of each directed connection edge, adjust the knowledge graph of the second knowledge point to obtain the knowledge graph of the knowledge point.

[0073] Specifically, the preset association weights represent the degree of mandatory requirements for each knowledge point with a prerequisite relationship. The terminal traverses the second knowledge point knowledge graph. For bidirectional directed edges, the terminal calculates the cosine similarity of the graph node features of the graph nodes at both ends of the directed edge and determines this similarity as the edge weight of the bidirectional directed edge. For unidirectional directed edges, the terminal queries the corresponding knowledge point pair's association weight from the preset association weights and determines this association weight as the edge weight of the unidirectional directed edge. The terminal assigns the edge weights of each directed edge to the corresponding directed edge to obtain the knowledge point knowledge graph. For example, the preset association weights can take the form of: .in, It is a preset association weight. , These are knowledge points that have prerequisite relationships. Representation of knowledge points Points of knowledge The association type is a prerequisite relationship. It's about building relationships first. The association relationship weights. The preset association relationship weights can be set according to actual work, and this embodiment does not further limit the setting of these preset association relationship weights.

[0074] This embodiment provides an adaptive learning path generation method based on cognitive diagnosis. By constructing a knowledge graph from a set of knowledge points, including node featureization and edge weighting, it ensures the structured and quantifiable representation of knowledge. This graph serves as the search space for path generation, and its rich semantic and weight information improves the accuracy of path optimization.

[0075] In one embodiment, based on the cognitive state vector and a preset knowledge graph, knowledge points are selected to form a set of learnable knowledge points, including:

[0076] S301, based on the cognitive state vector and the preset mastery threshold, the knowledge points in the cognitive state vector whose mastery value is less than the mastery threshold are identified as unfamiliar knowledge points.

[0077] Specifically, a preset mastery threshold is used to distinguish between knowledge points that have been mastered and those that have not. The terminal compares the mastery level value of each knowledge point in the cognitive state vector with the preset mastery threshold. If the mastery level value is greater than the preset mastery threshold, the knowledge point is marked as an unfamiliar knowledge point. For example, the preset mastery threshold can be set according to the actual work requirements for the level of knowledge mastery. The default setting is 0.7, and this embodiment does not further limit the setting of this preset mastery threshold.

[0078] S302. For each unfamiliar knowledge point, extract the knowledge points from the knowledge point knowledge graph that are connected by directed edges and point to the unfamiliar knowledge point, and form the set of prerequisite knowledge points for the unfamiliar knowledge point.

[0079] Specifically, for each unfamiliar knowledge point, the terminal traverses all knowledge points in the knowledge graph that are connected by unidirectional directed edges and point to the unfamiliar knowledge point, forming a set of prerequisite knowledge points for each unfamiliar knowledge point.

[0080] S303: For each unfamiliar knowledge point, based on the cognitive state vector and mastery threshold, the unfamiliar knowledge points whose mastery value is greater than the mastery threshold in the prerequisite knowledge point set are identified as learnable knowledge points.

[0081] Specifically, for each unfamiliar knowledge point, the terminal extracts the mastery level value of each knowledge point in its prerequisite knowledge point set based on the cognitive state vector. If the unfamiliar knowledge points of all knowledge points in the prerequisite knowledge point set of the unfamiliar knowledge point are greater than the mastery level threshold, then the unfamiliar knowledge point is considered a learnable knowledge point.

[0082] S304, based on each learnable knowledge point, obtain the set of learnable knowledge points.

[0083] Specifically, the terminal integrates all learnable knowledge points into a set, resulting in a set of learnable knowledge points.

[0084] This embodiment provides an adaptive learning path generation method based on cognitive diagnosis. It generates a learnable set through threshold filtering and prior relationship checks, strictly adhering to knowledge dependency logic, avoiding invalid recommendations, and providing a high-quality candidate space for subsequent optimization.

[0085] In one embodiment, the knowledge learning potential prediction model is obtained through the following method:

[0086] S401, Obtain preset historical learning trajectory data and knowledge point knowledge graph; the historical learning trajectory data includes the execution timestamp of each historical learning event, cognitive state vector, and set of learning knowledge points; the set of learning knowledge points is used to represent the knowledge points learned after executing the historical learning event.

[0087] Specifically, the pre-set historical learning trajectory data serves as the training sample source for the potential prediction model, recording the historical learning process. Each historical learning event is a triple: an execution timestamp indicating the time the historical learning event occurred; a cognitive state vector, including the mastery level of each knowledge point before executing the historical learning event; and a set of learned knowledge points, which is a list of knowledge points learned after the mastery level was reached. For example, the pre-set historical learning trajectory data can be obtained by acquiring historical logs from several users, and then anonymizing and cleaning them. The pre-set historical learning trajectory data can be collected according to actual work requirements.

[0088] S402, for each historical learning event, select other historical learning events whose execution timestamps meet the preset forward return calculation time window from the historical learning trajectory data, and form a potential reference event set for each historical learning event.

[0089] Specifically, the potential reference event set is used to quantify the long-term impact of each historical learning event. For each historical learning event, historical learning events with timestamps following it and falling within a preset long-term benefit calculation time window are selected to form the potential reference event set for that historical learning event. The potential reference event set represents the future learning activities that the potential reference event set may influence. For example, the preset long-term benefit calculation time window can be set according to the course length in actual work and the historical learning events in the historical learning trajectory data. The default setting is 30 days, and this embodiment does not further limit the setting of this preset long-term benefit calculation time window.

[0090] S403, for each historical learning event, based on the knowledge graph of knowledge points, quantifies the influence of the cognitive state vector on the set of learning knowledge points of each historical learning event in the potential reference event set, and obtains the future potential value of each knowledge point in each historical learning event.

[0091] Specifically, the calculation of future potential value serves as a supervisory signal for model training. For each historical learning event, its true potential value is defined as follows: after completing the learning of any knowledge point of that historical learning event, within a preset time window for calculating future returns, using the following formula: The total value of the newly unlocked set of knowledge points is calculated to obtain the future potential value of any knowledge point in this historical learning event. Any historical learning event Any knowledge point in the set of learning knowledge points. These are the key knowledge points in this historical learning event. Its future potential. It is any node in the knowledge graph that represents any knowledge point. It is a set of graph nodes in a knowledge graph of knowledge points. It is an indicator function. It is a historical learning event The execution timestamp. This is the preset time window for calculating forward returns; if knowledge points Knowledge points in historical learning events When something becomes learnable, the indicator function value is 1, meaning it is within the historical learning event. Before the execution timestamp, knowledge points It is unlearnable, but through studying historical events... Learned the knowledge points After that, knowledge points When it becomes a learnable state, The value is 1. It is a knowledge point The learning value parameters can be obtained based on the features of graph nodes in the knowledge graph of knowledge points.

[0092] S404: For each historical learning event, a model is constructed using knowledge point knowledge graphs and cognitive state vectors as input features and future potential values ​​as the prediction target. The model construction dataset is then divided according to a preset partitioning ratio to obtain a model training set and a model test set.

[0093] Specifically, the terminal constructs a corresponding model construction sample for each historical learning event. Each model construction sample includes input features and a prediction target: the input features include the cognitive state vector and knowledge graph of the historical learning event, and the prediction target is the future potential value of each knowledge point calculated by S403. Each sample fully describes the mapping relationship between the learning context and long-term benefits. The terminal combines the model construction samples into a model construction sample and divides the model construction dataset according to a preset partitioning ratio to obtain a model training set and a model test set. For example, the preset partitioning ratio can be set according to the requirements of the training dataset and test dataset in actual work. The default setting is model training set: model test set = 7:3. This embodiment does not further limit the setting of this preset partitioning ratio.

[0094] S405. An initial knowledge learning potential prediction model is constructed based on a graph neural network model, and the initial knowledge learning potential prediction model is iteratively trained using a model training set to obtain a preliminary knowledge learning potential prediction model.

[0095] Specifically, the initial knowledge learning potential prediction model is a graph neural network model with an encoder-predictor head architecture. The encoder uses a multi-layer graph attention network to aggregate structural information from the knowledge graph; the predictor head is a multi-layer perceptron that outputs a learning potential vector. The training process uses a model training set for supervised learning: for each model construction sample in the training set, the model inputs a cognitive state vector and a knowledge graph of knowledge points, and outputs a predicted learning potential vector. The model's loss function is constructed using mean squared error, calculated for the predicted value of the learned knowledge point to approximate the true label. The terminal iteratively updates the parameters of the initial knowledge learning potential prediction model using a gradient descent algorithm based on the model loss from the training set until the loss converges. For example, the form of the model's loss function can be... ,in It is the model loss. It is the number of model construction samples in the model training set. It is a sample constructed from any model. It is a model building sample Any knowledge point Its future potential. It is the initial knowledge learning potential prediction model for model construction samples Any knowledge point The projected value of future potential.

[0096] S406, Test the preliminary knowledge learning potential prediction model with the model test set until the preliminary knowledge learning potential prediction model meets the preset accuracy requirements, and obtain the knowledge learning potential prediction model.

[0097] Specifically, the terminal verifies the accuracy of the preliminary knowledge learning potential prediction model using a model test set. When the accuracy of the preliminary knowledge learning potential prediction model reaches a preset threshold, the final preliminary knowledge learning potential prediction model is obtained. The accuracy verification process involves inputting the cognitive state vector and knowledge point knowledge graph from the model test set into the preliminary knowledge learning potential prediction model, comparing the predicted future potential values ​​with the actual future potential values, and calculating the overall classification accuracy. For example, the preset accuracy requirement is set to 90% by default, but it can be set according to actual work. This embodiment does not further limit the setting of this preset threshold.

[0098] This embodiment provides an adaptive learning path generation method based on cognitive diagnosis. By training a GNN model with historical trajectory data, it achieves accurate prediction of learning potential and provides a scientific long-term indicator for hierarchical optimization.

[0099] In one embodiment, an optimal learning path is generated based on a learning potential vector, a cognitive state vector, a knowledge graph of knowledge points, and a set of learnable knowledge points, including:

[0100] S501: For each knowledge point in the set of learnable knowledge points, select the future potential value of the knowledge point from the learning potential vector to form a set of learnable knowledge potential values.

[0101] Specifically, for each knowledge point in the set of learnable knowledge points, the terminal extracts its corresponding future potential value from the learning potential vector to form a set of learnable knowledge potential values. For example, the set of learnable knowledge potential values ​​can take the form of... .in, It is a set of potential values ​​for learnable knowledge. It is a collection of learnable knowledge points. It is any knowledge point in the set of learnable knowledge points. It is a knowledge point Its future potential.

[0102] S502, sort the future potential values ​​in the set of learnable knowledge potential values ​​in descending order, and select the knowledge points whose sorting position of the future potential values ​​matches the preset sampling position to form a candidate knowledge point set.

[0103] Specifically, the preset sampling position indicates the number of positions for selecting high-potential knowledge points, and is set to the top 20% by default. The terminal sorts the future potential values ​​in the set of learnable knowledge potential values ​​in descending order, and selects the knowledge points corresponding to the future potential values ​​at the top preset sampling positions to form a candidate knowledge point set. For example, when the preset sampling position is the top 20%, the terminal sorts the future potential values ​​in the set of learnable knowledge potential values ​​in descending order, and selects the knowledge points corresponding to the top 20% of the future potential values ​​according to this sorting to form a candidate knowledge point set. The preset sampling position can be set according to actual work, and this embodiment does not further limit the setting of this preset sampling position.

[0104] S503: Based on the candidate knowledge point set, sort the knowledge points corresponding to the future potential values ​​in descending order of future potential value to obtain the initial learning path.

[0105] Specifically, for each knowledge point in the candidate knowledge point set, the terminal sorts the knowledge points in descending order of their corresponding future potential values ​​to obtain an initial learning path. For example, if the future potential value is... And the future potential values ​​are ranked as follows: The terminal sorts the knowledge points within the future potential value based on the future potential value to obtain the initial learning path:

[0106] S504 adjusts the initial learning path based on the learning potential vector, knowledge point knowledge graph, and cognitive state vector to obtain the optimal learning path.

[0107] Specifically, the terminal evaluates the initial learning path based on the learning potential vector, knowledge graph of knowledge points, and cognitive state vector, considering short-term learning efficiency, cognitive load, immediate knowledge gain, and long-term cumulative benefits. It also adjusts the order of knowledge points in the learning path to obtain the optimal learning path that balances maximizing short-term learning efficiency, minimizing cognitive load, maximizing immediate knowledge gain, and maximizing long-term cumulative benefits quantified by the learning potential vector.

[0108] This embodiment provides an adaptive learning path generation method based on cognitive diagnosis. It strategically focuses on the learnable set through potential prediction results, significantly reducing the search space; it generates a heuristic path based on potential as an optimization starting point; and it integrates short-term and long-term goals through an adjustment process while satisfying the hard constraints of the knowledge structure. This achieves a balance between short-term and long-term goals, generating a scientifically sound and personalized path.

[0109] In one embodiment, based on the learning potential vector, the knowledge graph of knowledge points, and the cognitive state vector, the initial learning path is adjusted to obtain the optimal learning path, including:

[0110] S601, based on the cognitive state vector, uses the following formula to simulate the cognitive state vector before completing each knowledge point when executing the initial learning path:

[0111]

[0112] in, It is any position in the initial learning path. It is the position in completing the initial learning path. The simulated cognitive state vector before the knowledge point It is a state simulation function. It is the position in completing the initial learning path. The simulated cognitive state vector before the knowledge point It is the position in the initial learning path. The knowledge points.

[0113] Specifically, the simulated cognitive state vector is used for state recursion in path evaluation. In the formula, any position in the initial learning path... Used to indicate the position of the path. ,and It was obtained from S101. State simulation function It is a simplified model that receives two inputs: one is the input for performing the first step. Simulated cognitive state vector before learning knowledge points at the unit level One is rank. The knowledge points of learning The state simulation function simulates the cognitive state vector of the previous order of input before learning the knowledge point. And the knowledge points I just learned Rule-based mastery updates are performed, outputting a new simulated cognitive state vector to quickly and efficiently predict learning progress during path optimization iterations. The state simulation function does not perform precise cognitive diagnosis, but rather employs a simplified mechanism to simulate the impact of learning behavior on knowledge state, thus providing a feasible data foundation for fitness assessment. For example, if the simulated learning path... Knowledge points for units digit Previously, the simulated cognitive state vector was: Next, we will simulate learning the knowledge points. (Assuming its corresponding vector is the first) The state change brought about by each component, the state simulation function implement: That is to The middle corresponds to the knowledge point The mastery level component is directly set to 1, while The mastery level components of all other knowledge points remain unchanged, thus obtaining .

[0114] S602, based on the simulated cognitive state vector, knowledge point knowledge graph, and learning potential vector, calculates the fitness of the initial learning path using the following formula:

[0115]

[0116] in, It is the fitness of the initial learning path. This is the total number of knowledge points in the initial learning path. It is any position in the initial learning path. It is the position in completing the initial learning path. The simulated cognitive state vector before the knowledge point It is a load estimation function. It is the position in the initial learning path. Knowledge points, It is the load weight. It is an efficiency estimation function. It is an efficiency weight. It is the gain estimation function. It is the gain weight. It is a potential weight. It is the position in the initial learning path. The future potential value of the knowledge points. It is the discount factor.

[0117] Specifically, fitness The calculation of this value is the core of the multi-objective evolutionary algorithm, integrating short-term learning experience and long-term development potential into an optimizable scalar indicator. The four summation terms in the formula correspond to short-term learning efficiency, cognitive load, immediate knowledge gain, and long-term cumulative benefit, respectively. Here, the total number of knowledge points in the initial learning path is... This can be obtained from the initial learning path. Load estimation function. The form is ,in It refers to any knowledge point. It is a knowledge point The difficulty level is determined based on the knowledge graph of the knowledge points. It is a knowledge point With simulated cognitive state vector Cognitive distance, calculated by knowledge points The cosine similarity between the graph node features and the graph node features of knowledge points whose mastery value is greater than the preset mastery threshold in the simulated cognitive state vector is obtained. and These are weighting coefficients, summing to 1; efficiency estimation function. The form is ,in It is a knowledge point The difficulty level of mastering it. It is based on a pre-defined cognitive diagnostic model to estimate the simulated cognitive state vector. The next step is to master the knowledge points The probability of this can be estimated using a pre-defined cognitive diagnostic model, such as the DINA model (Deterministic Inputs, Noisy "And" gate model), a discrete cognitive diagnostic model used to assess students' mastery of knowledge points. This model uses a Q-matrix and student responses to model the knowledge points required for correct answers and the probability of errors. and These are the weighting coefficients, summing to 1; the gain estimation function. The form is ,in It is a knowledge point The learning value parameters are obtained based on the knowledge graph of the knowledge points. This is the current knowledge gap.

[0118] S603, based on the knowledge graph of knowledge points, adjusts the order of knowledge points in the initial learning path to obtain a new initial learning path.

[0119] Specifically, the terminal adjusts the order of knowledge points in the initial learning path to obtain a new initial learning path. This order adjustment is achieved through evolutionary algorithm operators, aiming to explore better path sequences based on the constraints of the knowledge graph. The adjustment operations mainly include crossover and mutation: crossover (such as sequential crossover) randomly selects two paths, swaps one continuous segment, and generates a new path; mutation randomly selects a knowledge point in the path and replaces it with another knowledge point from the candidate set that satisfies the prerequisite relationship. All these operations must be based on the knowledge graph of the knowledge points to ensure that the newly generated path sequence does not violate the prerequisite dependencies of the knowledge points (i.e., for each knowledge point in the path, all its prerequisite knowledge points must appear before it).

[0120] S604, calculate the fitness of each new initial learning path, and determine the initial learning path with the highest fitness value as the optimal learning path.

[0121] Specifically, for all new initial learning paths generated through evolutionary operations, the terminal calculates their fitness and selects the path with the highest fitness from all initial learning paths as the optimal learning path. For example, if multiple paths have the same highest fitness, other rules can be used to assist in the decision-making process.

[0122] This embodiment provides an adaptive learning path generation method based on cognitive diagnosis. It achieves recursive prediction of the learning process through a state simulation function, quantifies short-term goals using clearly defined load, efficiency, and gain functions, and characterizes long-term goals by combining discounted potential values. An evolutionary algorithm is then used for multi-objective optimization decision-making. This effectively balances optimization goals across multiple time scales, thereby generating personalized learning paths with strategic value.

[0123] In the aforementioned adaptive learning path generation method based on cognitive diagnosis, the user's cognitive state vector is obtained. This vector includes the user's mastery level of each knowledge point. Based on the cognitive state vector and a pre-set knowledge point knowledge graph, knowledge points are selected to form a set of learnable knowledge points. The knowledge point knowledge graph includes graph nodes and directed edges. Graph nodes represent each knowledge point, and directed edges represent the relationships between knowledge points. The cognitive state vector and the knowledge point knowledge graph are input into a pre-set knowledge learning potential prediction model to obtain a learning potential vector. This vector includes the future potential value of each knowledge point. The knowledge learning potential prediction model is constructed based on a graph neural network model. Based on the learning potential vector, cognitive state vector, knowledge point knowledge graph, and set of learnable knowledge points, an optimal learning path is generated. This optimal learning path indicates the learning order of each knowledge point. This method overcomes the short-sightedness of traditional path generation by guiding micro-level searches with macro-level potential, balancing immediate efficiency and long-term benefits, thereby improving the scientific nature and adaptability of the learning path and generating a learning path that balances short-term learning efficiency and long-term development potential.

[0124] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0125] Based on the same inventive concept, this application also provides an adaptive learning path generation device based on cognitive diagnosis for implementing the aforementioned adaptive learning path generation method based on cognitive diagnosis. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the adaptive learning path generation device based on cognitive diagnosis provided below can be found in the limitations of the adaptive learning path generation method based on cognitive diagnosis described above, and will not be repeated here.

[0126] In one exemplary embodiment, such as Figure 2 As shown, an adaptive learning path generation device 200 based on cognitive diagnosis is provided, comprising:

[0127] The data acquisition module 201 is used to acquire the user's cognitive state vector; the cognitive state vector includes the user's level of mastery of various knowledge points.

[0128] The learnable knowledge point extraction module 202 is used to select knowledge points to form a set of learnable knowledge points based on the cognitive state vector and the preset knowledge point knowledge graph. The knowledge point knowledge graph includes graph nodes and directed connecting edges. Graph nodes are used to represent each knowledge point, and directed connecting edges are used to represent the relationship between each knowledge point.

[0129] The learnable knowledge point potential analysis module 203 is used to input the cognitive state vector and the knowledge point knowledge graph into the preset knowledge learning potential prediction model to obtain the learning potential vector; the learning potential vector includes the future potential value of each knowledge point; the knowledge learning potential prediction model is constructed based on the graph neural network model.

[0130] The learning path generation module 204 is used to generate an optimal learning path based on the learning potential vector, cognitive state vector, knowledge point knowledge graph, and set of learnable knowledge points; the optimal learning path is used to indicate the learning order of each knowledge point.

[0131] Furthermore, the device also includes a knowledge point knowledge graph construction module, used for:

[0132] Obtain a preset set of knowledge points; the set of knowledge points includes the correlation feature vectors between each knowledge point, as well as the learning value parameters and mastery difficulty values ​​of each knowledge point;

[0133] Based on the set of knowledge points, each knowledge point is used as a graph node, and the learning value parameter and mastery difficulty value of each knowledge point are used as the graph node features to obtain the first knowledge point knowledge graph.

[0134] Based on the correlation feature vectors between knowledge points in the knowledge point set and the first knowledge point knowledge graph, directed connection edges between graph nodes are constructed to obtain the second knowledge point knowledge graph; the direction of the directed connection edges is used to represent the direction of the correlation relationship between knowledge points.

[0135] Based on the graph node features of the graph nodes at both ends of the directed connection edges and the preset association weights, the weights of each directed connection edge are calculated to obtain the edge weights of each directed connection edge; and based on the edge weights of each directed connection edge, the knowledge graph of the second knowledge point is adjusted to obtain the knowledge graph of the knowledge point.

[0136] Furthermore, the learnable knowledge point extraction module is also used for:

[0137] Based on the cognitive state vector and the preset mastery threshold, the knowledge points in the cognitive state vector whose mastery value is less than the mastery threshold are identified as unfamiliar knowledge points.

[0138] For each unfamiliar knowledge point, extract the knowledge points from the knowledge point knowledge graph that are connected by directed edges and point to the unfamiliar knowledge point, and form the set of prerequisite knowledge points for the unfamiliar knowledge point.

[0139] For each unfamiliar knowledge point, based on the cognitive state vector and mastery threshold, the unfamiliar knowledge points whose mastery value is greater than the mastery threshold are identified as learnable knowledge points.

[0140] Based on each learnable knowledge point, a set of learnable knowledge points is obtained.

[0141] Furthermore, the device also includes a prediction model building module for:

[0142] Acquire preset historical learning trajectory data and knowledge graph of knowledge points; the historical learning trajectory data includes the execution timestamps, cognitive state vectors and sets of learned knowledge points for each historical learning event; the set of learned knowledge points is used to represent the knowledge points learned after executing the historical learning event;

[0143] For each historical learning event, select other historical learning events whose execution timestamps meet the preset forward return calculation time window from the historical learning trajectory data to form a potential reference event set for each historical learning event;

[0144] For each historical learning event, based on the knowledge graph of knowledge points, the influence of the cognitive state vector on the set of learning knowledge points of each historical learning event in the potential reference event set is quantified to obtain the future potential value of each knowledge point in each historical learning event.

[0145] For each historical learning event, a model is constructed using a knowledge graph of knowledge points and a cognitive state vector as input features and future potential as the prediction target. The model construction dataset is then divided according to a preset partitioning ratio to obtain a model training set and a model test set.

[0146] An initial knowledge learning potential prediction model is constructed based on a graph neural network model, and the initial knowledge learning potential prediction model is iteratively trained using a model training set to obtain a preliminary knowledge learning potential prediction model.

[0147] The preliminary knowledge learning potential prediction model is tested using a model test set until it meets the preset accuracy requirements, thus obtaining the knowledge learning potential prediction model.

[0148] Furthermore, the learning path generation module includes:

[0149] The potential determination unit is used to select the future potential value of each knowledge point from the learning potential vector for each knowledge point in the set of learnable knowledge points, and form a set of learnable knowledge potential values.

[0150] The knowledge point candidate unit is used to sort the future potential values ​​in the set of learnable knowledge potential values ​​in descending order, and select the knowledge points whose sorting position of the future potential values ​​matches the preset sampling position to form a candidate knowledge point set.

[0151] The initial learning path determination unit is used to sort the knowledge points corresponding to the future potential values ​​in descending order based on the candidate knowledge point set to obtain the initial learning path;

[0152] The optimal learning path determination unit is used to adjust the initial learning path based on the learning potential vector, knowledge point knowledge graph, and cognitive state vector to obtain the optimal learning path.

[0153] Furthermore, the optimal learning path determination unit is also used for:

[0154] Based on the cognitive state vector, the simulated cognitive state vector before completing each knowledge point is calculated using the following formula:

[0155]

[0156] in, It is any position in the initial learning path. It is the position in completing the initial learning path. The simulated cognitive state vector before the knowledge point It is a state simulation function. It is the position in completing the initial learning path. The simulated cognitive state vector before the knowledge point It is the position in the initial learning path. Knowledge points;

[0157] Based on the simulated cognitive state vector, knowledge graph of knowledge points, and learning potential vector, the fitness of the initial learning path is calculated using the following formula:

[0158]

[0159] in, It is the fitness of the initial learning path. This is the total number of knowledge points in the initial learning path. It is any position in the initial learning path. It is the position in completing the initial learning path. The simulated cognitive state vector before the knowledge point It is a load estimation function. It is the position in the initial learning path. Knowledge points, It is the load weight. It is an efficiency estimation function. It is an efficiency weight. It is the gain estimation function. It is the gain weight. It is a potential weight. It is the position in the initial learning path. The future potential value of the knowledge points. It is the discount factor;

[0160] Based on the knowledge graph of knowledge points, the order of knowledge points in the initial learning path is adjusted to obtain a new initial learning path;

[0161] Calculate the fitness of each new initial learning path, and determine the initial learning path with the highest fitness value as the optimal learning path.

[0162] In one embodiment, such as Figure 3 A computer device is provided, comprising:

[0163] At least one processor 301, and a memory 302 communicatively connected to at least one of the processors 301: the memory stores application code executable by at least one of the processors, the application code being executed by at least one of the processors to enable at least one of the processors to perform the cognitive diagnostic-based adaptive learning path generation method as described above.

[0164] Computer equipment may also include: sensor 303.

[0165] The processor 301, memory 302 and sensor 303 can be connected via a bus or other means, with the bus being an example in the figure.

[0166] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0167] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and 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 modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0168] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. An adaptive learning path generation method based on cognitive diagnosis, characterized in that, The method includes: Obtain the user's cognitive state vector; the cognitive state vector is used to include the user's level of mastery of each knowledge point; Based on the cognitive state vector and the preset knowledge point knowledge graph, the knowledge points are selected to form a set of learnable knowledge points; the knowledge point knowledge graph includes graph nodes and directed connecting edges; the graph nodes are used to represent each of the knowledge points, and the directed connecting edges are used to represent the relationship between the knowledge points. The cognitive state vector and the knowledge point knowledge graph are input into a preset knowledge learning potential prediction model to obtain a learning potential vector; the learning potential vector includes the future potential value of each knowledge point; the knowledge learning potential prediction model is constructed based on a graph neural network model. Based on the learning potential vector, the cognitive state vector, the knowledge point knowledge graph, and the set of learnable knowledge points, an optimal learning path is generated; the optimal learning path is used to indicate the learning order of each of the knowledge points.

2. The method according to claim 1, characterized in that, The knowledge graph of the knowledge points was obtained through the following methods: Obtain a preset set of knowledge points; the set of knowledge points includes the correlation feature vectors between each knowledge point, and the learning value parameters and mastery difficulty values ​​of each knowledge point; Based on the set of knowledge points, with each knowledge point as a graph node and the learning value parameter and the mastery difficulty value of the knowledge point as graph node features, a first knowledge point knowledge graph is obtained. Based on the association feature vectors between the knowledge points in the knowledge point set and the first knowledge point knowledge graph, directed connection edges between the graph nodes are constructed to obtain the second knowledge point knowledge graph; the direction of the directed connection edges is used to represent the direction of the association relationship between the knowledge points; Based on the graph node features of the graph nodes at both ends of the directed connection edge and the preset association weights, the weight of each directed connection edge is calculated to obtain the edge weight of each directed connection edge. Based on the edge weights of each directed connection edge, the second knowledge point knowledge graph is adjusted to obtain the knowledge point knowledge graph.

3. The method according to claim 2, characterized in that, The step of selecting knowledge points to form a set of learnable knowledge points based on the cognitive state vector and a preset knowledge graph includes: Based on the cognitive state vector and the preset mastery threshold, the knowledge points in the cognitive state vector whose mastery value is less than the mastery threshold are identified as unfamiliar knowledge points. For each unfamiliar knowledge point, extract the knowledge points that are connected by the directed edges and point to the unfamiliar knowledge point from the knowledge point knowledge graph to form the prerequisite knowledge point set for the unfamiliar knowledge point. For each unfamiliar knowledge point, based on the cognitive state vector and the mastery threshold, the unfamiliar knowledge points whose mastery value is greater than the mastery threshold in the prerequisite knowledge point set are identified as learnable knowledge points. Based on each of the learnable knowledge points, the set of learnable knowledge points is obtained.

4. The method according to claim 2, characterized in that, The knowledge learning potential prediction model was obtained through the following method: Obtain preset historical learning trajectory data and the knowledge graph of the knowledge points; the historical learning trajectory data includes the execution timestamps of each historical learning event, the cognitive state vector, and the set of learning knowledge points; The set of learning knowledge points is used to represent the knowledge points learned after executing the historical learning event; For each historical learning event, other historical learning events whose execution timestamps satisfy the preset long-term return calculation time window are selected from the historical learning trajectory data to form a potential reference event set for each historical learning event; For each historical learning event, based on the knowledge graph of the knowledge points, the influence of the cognitive state vector on the set of learning knowledge points of each historical learning event in the potential reference event set is quantified to obtain the future potential value of each knowledge point in each historical learning event. For each historical learning event, a model construction dataset is constructed using the knowledge graph of knowledge points and the cognitive state vector as input features and the future potential value as the prediction target. The model construction dataset is then divided according to a preset partitioning ratio to obtain a model training set and a model test set. An initial knowledge learning potential prediction model is constructed based on a graph neural network model, and the initial knowledge learning potential prediction model is iteratively trained using the model training set to obtain a preliminary knowledge learning potential prediction model. The preliminary knowledge learning potential prediction model is tested using the model test set until it meets the preset accuracy requirements, thus obtaining the knowledge learning potential prediction model.

5. The method according to claim 1, characterized in that, The process of generating an optimal learning path based on the learning potential vector, the cognitive state vector, the knowledge point knowledge graph, and the set of learnable knowledge points includes: For each knowledge point in the set of learnable knowledge points, the future potential value of the knowledge point is selected from the learning potential vector to form a set of learnable knowledge potential values. The future potential values ​​in the set of learnable knowledge potential values ​​are sorted in descending order, and the knowledge points whose sorting positions of the future potential values ​​match the preset sampling positions are selected to form a set of candidate knowledge points. Based on the candidate knowledge point set, the knowledge points corresponding to the future potential values ​​are sorted in descending order according to the future potential values ​​to obtain the initial learning path; Based on the learning potential vector, the knowledge graph of knowledge points, and the cognitive state vector, the initial learning path is adjusted to obtain the optimal learning path.

6. The method according to claim 5, characterized in that, The step of adjusting the initial learning path based on the learning potential vector, the knowledge graph of knowledge points, and the cognitive state vector to obtain the optimal learning path includes: Based on the cognitive state vector, the simulated cognitive state vector before completing each knowledge point is calculated using the following formula: in, It is any position in the initial learning path. It is the position in completing the initial learning path. The simulated cognitive state vector before the knowledge point It is a state simulation function. It is the position in completing the initial learning path. The simulated cognitive state vector before the knowledge point It is the position in the initial learning path. Knowledge points; Based on the simulated cognitive state vector, the knowledge graph of knowledge points, and the learning potential vector, the fitness of the initial learning path is calculated using the following formula: in, It is the fitness of the initial learning path. This is the total number of knowledge points in the initial learning path. It is any position in the initial learning path. It is the position in completing the initial learning path. The simulated cognitive state vector before the knowledge point It is a load estimation function. It is the position in the initial learning path. Knowledge points, It is the load weight. It is an efficiency estimation function. It is an efficiency weight. It is the gain estimation function. It is the gain weight. It is a potential weight. It is the position in the initial learning path. The future potential value of the knowledge points. It is the discount factor; Based on the knowledge graph of the knowledge points, the order of the knowledge points in the initial learning path is adjusted to obtain a new initial learning path; Calculate the fitness corresponding to each of the new initial learning paths, and determine the initial learning path corresponding to the highest fitness value as the optimal learning path.

7. An adaptive learning path generation device based on cognitive diagnosis, characterized in that, The device includes: The data acquisition module is used to acquire the user's cognitive state vector; the cognitive state vector includes the user's level of mastery of various knowledge points. The learnable knowledge point extraction module is used to select knowledge points to form a set of learnable knowledge points based on the cognitive state vector and a preset knowledge point knowledge graph; the knowledge point knowledge graph includes graph nodes and directed connecting edges; the graph nodes are used to represent each of the knowledge points, and the directed connecting edges are used to represent the relationship between the knowledge points. The learnable knowledge point potential analysis module is used to input the cognitive state vector and the knowledge point knowledge graph into a preset knowledge learning potential prediction model to obtain a learning potential vector; the learning potential vector includes the future potential value of each knowledge point; the knowledge learning potential prediction model is constructed based on a graph neural network model. The learning path generation module is used to generate an optimal learning path based on the learning potential vector, the cognitive state vector, the knowledge point knowledge graph, and the set of learnable knowledge points; the optimal learning path is used to indicate the learning order of each of the knowledge points.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images