Education cognitive engine adaptive optimization method based on large model

Abstract

The invention discloses an educational cognitive engine adaptive optimization method based on a large model, which comprises the steps of S1, formulating an acquisition specification, acquiring teaching behavior data from an educational database, forming a uniform format through an ETL tool, and mapping the behavior data into a cognitive state vector through a cognitive signal encoder in combination with a preset cognitive dimension; s2, cognitive association is extracted from the cognitive state vector through BERT, and an evolutionary knowledge graph with evolutionary markers is obtained through an evolutionary marker mechanism optimized by cognitive association weight; s3, based on the evolutionary knowledge graph, introducing an improved graph neural network to carry out cognitive evolutionary link modeling to obtain a cognitive link evolutionary model, and generating an optimal candidate strategy list through the cognitive link evolutionary model; and S4, accessing a basic education cognitive engine based on the optimal candidate strategy list to automatically match teaching resources, thereby effectively satisfying learning requirements of different students, and improving pertinence and effectiveness of teaching.

Description

Technical Field

[0001] The present invention relates to the technical field of educational cognitive optimization, and in particular to a large-model-based adaptive optimization method for an educational cognitive engine. Background Art

[0002] With the penetration of technologies like artificial intelligence and knowledge graphs into education, the digital transformation of education is accelerating. The collection and analysis of learning data has become possible, enabling data on students' cognitive states and knowledge connections to support precise instruction. Furthermore, teaching strategies must be dynamically adjusted as students' cognitive development evolves, establishing a complete closed loop of "teaching-learning-evaluation" and achieving iterative upgrades in the educational ecosystem. All of this is driving the development of education towards intelligent and personalized learning.

[0003] At present, in terms of cognitive modeling, traditional methods such as simple neural networks have obvious shortcomings. They often ignore the complex associations between "knowledge point-knowledge point" and "student-knowledge point" in the knowledge graph, and only focus on the individual cognitive state of students. They cannot reflect the impact of knowledge associations on cognitive development. For example, it is difficult to consider how the mastery of knowledge point B affects the complex application learning of knowledge point C. Moreover, the model has poor interpretability. After outputting the probability of the cognitive path, it is impossible to explain the underlying cause. It is difficult for teachers to understand the reason for the high probability, and it cannot be effectively applied to teaching decisions. It cannot be combined with knowledge graph updates and students' new learning data to optimize the model in real time, and cannot keep up with the dynamic rhythm of students' cognitive evolution. Therefore, an adaptive optimization method for an educational cognitive engine based on a large model is proposed here. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned objectives, the present invention proposes the following technical solutions: A large-model-based adaptive optimization method for an educational cognitive engine, comprising: S1: Develop collection specifications to collect teaching behavior data from the education database, use ETL tools to form a unified format, and use cognitive signal encoders to combine preset cognitive dimensions to map the behavior data into cognitive state vectors; S2: Extract cognitive associations from cognitive state vectors using BERT, and obtain an evolutionary knowledge graph with evolutionary labels through an evolutionary labeling mechanism optimized with cognitive association weights; S3: Based on the evolutionary knowledge graph, an improved graph neural network is introduced to model the cognitive evolutionary links to obtain a cognitive link evolution model, and the optimal candidate strategy list is generated through the cognitive link evolution model; The improved graph neural network is implemented by introducing node-level attention and relationship-level attention into the basic graph neural network to perform feature optimization fusion coding; S4: Access the basic education cognitive engine to automatically match teaching resources based on the optimal candidate strategy list.

[0005] The cognitive state vector acquisition process is: Use ETL tools to clean and transform the collected initial teaching behavior data to form a standardized data set; Through the cognitive signal encoder, the standardized teaching behavior data is disassembled into preset cognitive dimensions, including memory dimension, understanding dimension, and application dimension. Rules are designed for each dimension, and the rule engine of the cognitive signal encoder is used to calculate the dimension value and splice it into a cognitive state vector to achieve the quantitative conversion of behavioral data to cognitive state. The implementation formula is expressed as follows: ,in, is the cognitive state vector of the i-th cognitive dimension, is the behavioral data corresponding to the i-th cognitive dimension, is the weight of the i-th dimension, is a rule function; Output cognitive state vector , where n is the number of cognitive dimensions.

[0006] The implementation process of extracting cognitive associations is as follows: Use the pre-trained BERT model to identify entity types in the data. Entity types include students, knowledge points, and error types. The BERT model outputs the entity probability distribution P. Pre-training BERT model to determine the relationship between entities by calculating the relationship probability accomplish: ; Output the relationship type and label the attributes to generate a set of triples of cognitive associations ,in For the main entity, is the object entity, r is the relationship, and a is the attribute.

[0007] The implementation process of the evolutionary marking mechanism based on cognitive association weight optimization is as follows: Add evolution metadata to each triple and record the time when the association was generated Update times c teaching effect correlation value , get the triple form after marking ; Define the relationship weight w to represent the influence of the association, the initial weight According to the statistical frequency setting, each time the teaching effect feedback data is obtained, the weight is updated according to the following formula: ; in, , is the score weight, is the learning rate, To improve the test scores, For the change in the accuracy of the operation, It represents comprehensive teaching indicators; By updating the formula Adjust the weight of cognitive association, if If is positive, the association weight increases. When it is negative, the association weight decreases, and after each weight update, an evolutionary mark is added to the cognitive association; Finally, the relationship weights are calculated and updated according to the weight update formula, evolutionary markers are added, and the extracted and processed entities, relationships, and marker information are stored in the graph database. By developing query statements and establishing indexes, the dynamic evolution of the knowledge graph is realized, and an evolved knowledge graph is obtained.

[0008] The improved graph neural network implementation process is as follows: Introducing node-level attention and relationship-level attention into the basic graph neural network and performing feature optimization fusion encoding The node features and relationship features processed by the attention mechanism are fused, and the current path node is processed through multi-layer GNN convolution operation. And the graph G encoding, get the evolutionary potential feature GNN of each path node p p,G); GNN of each path node based on GNN encoding output p,G), use the softmax function to calculate the probability distribution of future paths: ; in, represents the conditional probability, Represents the GNN path node and the encoded output of the knowledge graph G, represents the exponential function, The P in represents the set of all possible cognitive path nodes. Represents a specific cognitive path node in the set P; Based on the output probability distribution results, the improved graph neural network is obtained by combining the weight information of the attention mechanism.

[0009] The process of introducing node-level attention is as follows: Calculate the current path node The attention weight with other nodes in the knowledge graph is expressed as: ; in 、 is the feature vector of nodes i and j, for The learnable weight matrix, for The learnable bias term, is the attention weight of node i to node j; The process of introducing relation-level attention is as follows: For different relationship types in the knowledge graph, learn the relationship attention weights, the formula is: ; in, is the relation feature vector, for The learnable weight matrix, for The learnable bias term, is the attention weight of relation r.

[0010] The process of acquiring the cognitive link evolution model is as follows: The process of outputting probability distribution results based on the improved graph neural network is used as the model output process. Historical teaching data is collected, including students' cognitive state vectors at different time periods, corresponding knowledge graph associations, and real cognitive evolution paths, to construct training and validation sets. The cross entropy loss function is used to measure the difference between the predicted probability distribution and the true path The Adam optimizer is used to iteratively optimize the model parameters. The learning rate is initially set to 0.001 and dynamically adjusted according to the training situation. The validation set is used to evaluate the model performance. The model structure is adjusted according to the evaluation results to complete the cognitive evolution link modeling and obtain the cognitive link evolution model.

[0011] The process of outputting the optimal candidate strategy list is as follows: According to the cognitive link evolution model, the path probability distribution output by the cognitive evolution link modeling and the entity association data of the evolutionary knowledge graph are input into the inference engine, and the matching strategy is carried out according to the following steps: Extract cognitive paths with a probability greater than 50%, match the path rules, traverse the path-strategy mapping rules, and check whether the current knowledge graph data meets the preset rule conditions. If the conditions are met, trigger the teaching strategy corresponding to the rule to form a candidate strategy list The present invention has the following beneficial effects: In the present invention, firstly, the improved GNN is used to fuse the knowledge graph association with the student's cognitive state to achieve accurate prediction of the cognitive path. On the one hand, it can dynamically track the evolutionary path of students from basic concepts to complex applications and other cognitive stages, and the prediction accuracy is improved compared with traditional methods. On the other hand, combined with specific node-level attention and relationship-level attention, it can explain the results of high probability of complex application paths, such as explaining that "the mastery relationship weight of knowledge point B is high", so that teachers can improve their understanding rate and help teachers accurately grasp students' cognitive development; Secondly, the knowledge graph is transformed from a static knowledge base to a dynamic cognitive engine. It not only stores knowledge associations, but also uses GNN to explore the dynamic impact of students' cognition-knowledge associations, deepening the value of knowledge associations. Finally, based on the path probability distribution output by the cognitive link evolution model and combined with the entity association data of the evolutionary knowledge graph, the inference engine can screen out high-probability paths and match them with corresponding teaching strategies. For example, for students who are predicted to have a high probability of entering the complex application path but have a low weight in mastering knowledge point B, the complex application reinforcement package of knowledge point B can be accurately pushed to effectively meet the learning needs of different students and improve the pertinence and effectiveness of teaching. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a method step diagram of a large-model-based educational cognitive engine adaptive optimization method proposed by the present invention. DETAILED DESCRIPTION

[0013] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0014] Example 1 like Figure 1 As shown, the present invention proposes an adaptive optimization method for an educational cognitive engine based on a large model, comprising: S1: Develop collection specifications to collect teaching behavior data from the education database, use ETL tools to form a unified format, and use cognitive signal encoders to combine preset cognitive dimensions to map the behavior data into cognitive state vectors; Deploy multiple types of collection terminals (such as smart terminals and teaching system interfaces) to cover all teaching scenarios. By formulating collection specifications, collect teacher and student operation behaviors, interaction content, and time series raw data as teaching behavior data. The specific process is as follows: Collection specifications include: defining data collection fields (student ID, behavior type, timestamp), formats (such as standardized storage in JSON format), and setting data collection frequency. Data collection is triggered by submission events, and system logs are collected incrementally every hour. Based on teaching scenario requirements and data analysis goals, the fields that need to be collected for each data collection terminal and interface are determined. Data format conversion tools (JSONSchema validation tools) are used to ensure that the collected data format complies with the specifications. After the collection specifications are formulated, each collection terminal performs collection from the education database according to the collection specifications and outputs the initial teaching behavior data; Use ETL tools (such as Apache NiFi) to clean (remove noise and duplicate data) and transform (unify format and encoding) the collected initial teaching behavior data to form a standardized data set; Through the cognitive signal encoder, the standardized teaching behavior data is disassembled into preset cognitive dimensions, including memory dimension, understanding dimension, and application dimension. Rules are designed for each dimension, and the rule engine of the cognitive signal encoder is used to calculate the dimension value and splice it into a cognitive state vector to achieve the quantitative conversion of behavioral data to cognitive state. The implementation formula is expressed as follows: ,in, is the cognitive state vector of the i-th cognitive dimension, is the behavioral data corresponding to the i-th cognitive dimension, is the weight of the i-th dimension (set according to teaching objectives and expert experience, and remember the dimension weight , understand dimension weights , apply dimension weights ), It is a rule function that outputs a quantitative value between 0 and 1 based on logic such as thresholds and intervals; Specifically, the definition of cognitive dimensions is determined by collaborating with subject teachers and education experts to cover all aspects of teaching, such as: The memory dimension is related to behaviors such as recitation question scores and knowledge point repetition accuracy; Understand the correct rate of explanation questions related to the concept of dimension, the logic score of classroom questions and answers, etc. The application dimension is related to the completion rate of homework application questions, the degree of achievement of practical tasks, and other behaviors; The final output cognitive state vector , where n is the number of cognitive dimensions.

[0015] S2: Extract cognitive associations from cognitive state vectors using BERT, and obtain an evolutionary knowledge graph with evolutionary labels through an evolutionary labeling mechanism optimized with cognitive association weights; Use the pre-trained BERT model to identify entity types in the data. Entity types include students, knowledge points, and error types. The BERT model outputs the entity probability distribution P, which is calculated using the softmax function: ; in, is the probability of entity e, is the weight matrix, b is the bias, when hour,( is the confidence threshold, preset to 0.8), extract the entity; For example, from the situation where student A answers knowledge point B incorrectly, we can identify entities e1 = student A (type: student) and e2 = knowledge point B (type: knowledge point). Pre-training BERT model to determine the relationship between entities by calculating the relationship probability accomplish: ; Output relation types (e.g., error grasp association) and label attributes (e.g., error type concept misunderstanding) to generate a set of triples of cognitive associations ,in For the main entity, is the object entity, r is the relationship, and a is the attribute; Example: For entity student A and knowledge point B, extract the relation r = error - knowledge point, attribute a = concept misunderstanding, triple = (student A, error-knowledge point, knowledge point B, concept misunderstanding); The implementation process of the evolutionary marking mechanism based on cognitive association weight optimization is as follows: Add evolution metadata to each triple and record the time when the association was generated (e.g., homework submission time) Update times c (initial value is 1, each weight update +1) Teaching effect correlation value (such as changes in student grades, changes in knowledge point mastery), and obtain the triple form after marking: ; Define the relationship weight w to represent the influence of the association, the initial weight Set by statistical frequency (such as error-knowledge point initial weight ), each time to obtain teaching effect feedback data (such as test score improvement value , changes in operation accuracy ), update the weights according to the following formula: ; in, , is the score weight, such as 0.6, integrating multi-dimensional feedback, is the learning rate (the empirical value is 0.1-0.3, which controls the amplitude of weight changes), It represents comprehensive teaching indicators; By updating the formula Adjust the weight of cognitive association. If student A's test score on knowledge point B improves significantly ( is positive), then the association weight of student A-mastery-knowledge point B increases, reflecting the positive contribution of this association to learning progress; If the accuracy of the work decreases ( is negative), then the association weight of student A-error-knowledge point B is reduced (or reversely marked), and the influence judgment of the association is corrected; After each weight update, an evolutionary mark is added to the cognitive association, including: Time stamp: records the time point of weight update (such as 2025-07-15 10:00) and traces the associated evolution nodes; Feedback source: Mark the type of feedback that triggered this update (e.g., improved test scores, decreased homework accuracy), and clarify the driving force of evolution; Version iteration: records the number of weight updates (e.g., the third update) to reflect the maturity of the association; Effect Notes: Briefly describe the impact of the update on the teaching analysis (e.g., the priority of this association for warning of weakness in knowledge point B is from high to medium); Synchronize new weights and evolutionary markers of cognitive associations to the knowledge graph in real time, including: New associations: If new associations are discovered in teaching feedback (e.g., student A - application - knowledge point B), they are extracted according to the initial process, weighted, and marked as first created; Existing association iteration: If the association weight or label changes (e.g., student A - error - knowledge point B weight decreases), update the attributes of the corresponding edge in the graph (weight value, evolution label field); Specifically, we selected a graph database (such as Neo4j) to store the knowledge graph. Leveraging its graph structure query advantages, we developed Cypher query statements to support rapid retrieval of complex relationships, such as the optimal teaching strategy for a student's weak points and the evolutionary path of a particular knowledge point within the class. We also established indexes (such as student ID indexes and knowledge point code indexes) to optimize query performance and ensure a single query response time of ≤500ms. Finally, the relationship weights are calculated and updated according to the weight update formula, evolutionary markers are added, and the extracted and processed entities, relationships, and marker information are stored in the graph database. By developing query statements and establishing indexes, the dynamic evolution of the knowledge graph is realized, and an evolved knowledge graph is obtained.

[0016] S3: Based on the evolutionary knowledge graph, an improved graph neural network is introduced to model the cognitive evolutionary links to obtain a cognitive link evolution model, and the optimal candidate strategy list is generated through the cognitive link evolution model; Extract student cognition-related entity and relationship data from the evolutionary knowledge graph to form a cognitive association set ),in For the main entity, is the object entity, r is the relationship, including student cognitive nodes (such as student A-cognitive state: basic concept learning), knowledge point mastery relationship (such as student A-mastery-knowledge point B), and teaching behavior association (such as teacher C-explanation-knowledge point B); The student's current cognitive state vector V and cognitive association set G are modeled into cognitive evolution links through an improved graph neural network learning path evolution law; Integrate the node attribute features of the knowledge graph (such as the difficulty coefficient of the knowledge point, the student's learning time, etc.) into the input. For student nodes, concatenate the cognitive state vector with the learning time and the number of wrong questions in the history test. For knowledge point nodes, concatenate the knowledge point ID with the difficulty, the number of related knowledge points, and other features to construct a spatiotemporal feature sequence. Specifically, if there are students’ cognitive state vectors for multiple consecutive periods, they are spliced ​​in chronological order to form a sequence input to assist the model in learning the temporal trend of cognitive evolution; The improved graph neural network acquisition process is: Based on a graph neural network that integrates node-level attention and relationship-level attention and performs feature optimization fusion encoding, the implementation process is as follows: Incorporating node-level attention: Calculate the current path node The attention weights of other nodes in the knowledge graph highlight the nodes that have a great impact on cognitive evolution, expressed as: ; in 、 is the feature vector of nodes i and j, for The learnable weight matrix, for The learnable bias term, is the attention weight of node i to node j, through which neighbor node features are aggregated to enhance the influence of key associations; Specifically, by calculating the current path node Attention weights with other nodes Nodes with high weights have greater influence during feature aggregation, which allows us to focus on nodes that play a key role in cognitive evolution, such as knowledge points that students have just learned and have high error rates in their homework. This strengthens the impact of these nodes on subsequent path predictions, making predictions more in line with actual learning situations, thereby avoiding interference from irrelevant nodes and improving the ability to capture key cognitive nodes. Then incorporate relation-level attention: For different relationship types in the knowledge graph (such as mastery, association, and misunderstanding), the relationship attention weights are learned, and the formula is: ; in, is the relation feature vector, for The learnable weight matrix, for The learnable bias term, is the attention weight of relation r, which is used to focus on key relations when aggregating relation information; Specifically, by learning the relational attention weights When aggregating relationship information, key relationships (such as students' misunderstanding of a certain knowledge point) are given a larger proportion. When analyzing students' cognitive paths, if there is a misunderstanding relationship, the strategy package generation process will focus on this relationship to determine whether students will "correct misunderstandings and deepen their cognition" or "fall into a learning bottleneck due to misunderstandings", thereby improving the accuracy of cognitive path branch prediction; Then perform feature optimization fusion coding, and the implementation process is as follows: The node features and relationship features processed by the attention mechanism are fused, and the current path node is convolved through multi-layer GNN convolution operation (GraphSAGE convolution). And the graph G encoding, get the evolutionary potential feature GNN of each path node p p,G) is expressed as: , is the feature of node i in the kth layer, is the new feature vector of node i after the k+1th layer of convolution, which is the result of splicing, N i) is the neighbor set of node i, and the deep features of nodes and associations are learned through iterative convolution; Specifically, by integrating the features processed by node-level and relationship-level attention, features of different dimensions are spliced ​​together to form richer input features. These features contain key information about nodes and relationships in the knowledge graph, providing comprehensive materials for subsequent encoding. At the same time, through multi-layer GNN convolution operations, node features are iteratively updated Each layer of convolution combines the features of neighboring nodes, allowing node features to continuously integrate into surrounding related information, and progressively learn more abstract features that can better reflect the trend of cognitive evolution. p,G), improve the effectiveness of path prediction; GNN of each path node based on GNN encoding output p,G), use the softmax function to calculate the probability distribution of future paths: ; in, Represents the conditional probability, that is, at the current path node Under the conditions of knowledge graph G, the next step is to evolve to the path node The probability of Represents GNN (graph neural network) pair path node Based on the encoded output of the knowledge graph G, GNN will analyze the node relationships in the knowledge graph (such as the relationship between students' mastery of knowledge points and the relationship between knowledge points) and score each path. The higher the score, the greater the evolutionary potential. represents the exponential function, The P in represents the set of all possible cognitive path nodes. Represents a specific cognitive path node in the set P; Based on the output probability distribution results and combined with the weight information of the attention mechanism, the paths with higher probabilities are explained, including which nodes and relationships contribute most to the prediction of the path (for example, "the 'mastery' relationship of knowledge point B has a high attention weight and contributes significantly to the probability of the 'complex application' path"). This helps teachers understand the prediction logic and obtain improved graph neural networks. Using the above process of outputting probability distribution results based on the improved graph neural network as the model output process, we collected historical teaching data, including students' cognitive state vectors at different time periods, corresponding knowledge graph associations, and real cognitive evolution paths (manually annotated or derived from teaching records), to construct training and validation sets. The cross entropy loss function is used to measure the difference between the predicted probability distribution and the true path The Adam optimizer was used to iteratively optimize the model parameters. The learning rate was initially set to 0.001 and dynamically adjusted according to the training situation. The validation set was used to evaluate the model performance. Based on the evaluation results, the model structure was adjusted (increasing the number of convolutional layers and adjusting the attention mechanism parameters) to complete the cognitive evolution link modeling and obtain the cognitive link evolution model. The process of outputting the optimal candidate strategy list is: According to the cognitive link evolution model, the path probability distribution output by the cognitive evolution link modeling and the entity association data of the evolutionary knowledge graph are input into the inference engine, and the matching strategy is carried out according to the following steps: Screening high-probability paths: Extract cognitive paths with a probability greater than a threshold (50%) as the core target of strategic reasoning (e.g., the "complex application" path with a probability of 65% is selected); Matching path rules: traverse the "path-strategy mapping rules" to check whether the current knowledge graph data meets the preset rule conditions (for example, whether the student's mastery weight of knowledge point B is <0.6); Generate candidate strategies: If the conditions are met, trigger the teaching strategy corresponding to the rule (such as pushing the "Knowledge Point B Complex Application Enhancement Package") to form a candidate strategy list.

[0017] S4: Access the basic education cognitive engine to automatically match teaching resources based on the optimal candidate strategy list; Connect the optimal candidate strategy list to the basic learning engine and execute it according to strategy type: Resource recommendation strategy: call the resource management interface to push designated learning packages to students; Supervision strategy: Trigger the teaching supervision system to generate to-do tasks for teachers (such as "focus on checking student A's mastery of the complex application of knowledge point B").

[0018] The teaching effect data after the strategy is implemented (such as changes in student resource access rate and knowledge point mastery rate) is collected and sent back to: Evolving the knowledge graph: updating association weights (e.g., the weight of “student A – mastered – knowledge point B” increased from 0.5 to 0.7); Retrain the model with new data to optimize the accuracy of cognitive path predictions.

[0019] In the application, several formulas involved are calculated by taking their numerical values ​​after removing the dimensions, and the formulas are established by collecting a large amount of data and performing software simulation to obtain a formula for the most recent real situation. Some coefficients or weights in the formulas are set by technical personnel in this field according to actual conditions, so they will not be elaborated here.

[0020] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0021] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A self-adaptive optimization method for an educational cognitive engine based on a large model, characterized in that: include: S1: Develop collection specifications to collect teaching behavior data from the education database, use ETL tools to form a unified format, and use cognitive signal encoders to combine preset cognitive dimensions to map the behavior data into cognitive state vectors; S2: Extract cognitive associations from cognitive state vectors using BERT, and obtain an evolutionary knowledge graph with evolutionary labels through an evolutionary labeling mechanism optimized with cognitive association weights; S3: Based on the evolutionary knowledge graph, an improved graph neural network is introduced to model the cognitive evolutionary links to obtain a cognitive link evolution model, and the optimal candidate strategy list is generated through the cognitive link evolution model; The improved graph neural network is implemented by introducing node-level attention and relationship-level attention into the basic graph neural network to perform feature optimization fusion coding; S4: Access the basic education cognitive engine to automatically match teaching resources based on the optimal candidate strategy list.

2. The method for adaptive optimization of an educational cognitive engine based on a large model according to claim 1, characterized in that: The cognitive state vector acquisition process is: Use ETL tools to clean and transform the collected initial teaching behavior data to form a standardized data set; Through the cognitive signal encoder, the standardized teaching behavior data is disassembled into preset cognitive dimensions, including memory dimension, understanding dimension, and application dimension. Rules are designed for each dimension, and the rule engine of the cognitive signal encoder is used to calculate the dimension value and splice it into a cognitive state vector to achieve the quantitative conversion of behavioral data to cognitive state. The implementation formula is expressed as follows: ,in, is the cognitive state vector of the i-th cognitive dimension, is the behavioral data corresponding to the i-th cognitive dimension, is the weight of the i-th dimension, is a rule function; Output cognitive state vector , where n is the number of cognitive dimensions.

3. The method for adaptive optimization of an educational cognitive engine based on a large model according to claim 1, characterized in that: The implementation process of extracting cognitive associations is as follows: Use the pre-trained BERT model to identify entity types in the data. Entity types include students, knowledge points, and error types. The BERT model outputs the entity probability distribution P. Based on the pre-trained BERT model, the relationship between entities is judged by calculating the relationship probability accomplish: ; Output the relationship type and label the attributes to generate a set of triples of cognitive associations ,in For the main entity, is the object entity, r is the relationship, and a is the attribute.

4. The method for adaptive optimization of an educational cognitive engine based on a large model according to claim 3, characterized in that: The implementation process of the evolutionary marking mechanism based on cognitive association weight optimization is as follows: Add evolution metadata to each triple and record the time when the association was generated Update times c teaching effect correlation value , get the triple form after marking ; Define the relationship weight w to represent the influence of the association, the initial weight According to the statistical frequency setting, each time the teaching effect feedback data is obtained, the weight is updated according to the following formula: ； in, , is the score weight, is the learning rate, To improve the test scores, For the change in the accuracy of the operation, It represents comprehensive teaching indicators; By updating the formula Adjust the weight of cognitive association, if If is positive, the association weight increases. When it is negative, the association weight decreases, and after each weight update, an evolutionary mark is added to the cognitive association; Finally, the relationship weights are calculated and updated according to the weight update formula, evolutionary markers are added, and the extracted and processed entities, relationships, and marker information are stored in the graph database. By developing query statements and establishing indexes, the dynamic evolution of the knowledge graph is realized, and an evolved knowledge graph is obtained.

5. The method for adaptive optimization of an educational cognitive engine based on a large model according to claim 1, characterized in that: The improved graph neural network implementation process is as follows: Introducing node-level attention and relationship-level attention into the basic graph neural network and performing feature optimization fusion encoding The node features and relationship features processed by the attention mechanism are fused, and the current path node is processed through multi-layer GNN convolution operation. And the graph G encoding, get the evolutionary potential feature GNN of each path node p p,G); GNN of each path node based on GNN encoding output p,G), use the softmax function to calculate the probability distribution of future paths: ； in, represents the conditional probability, Represents the GNN path node and the encoded output of the knowledge graph G, represents the exponential function, The P in represents the set of all possible cognitive path nodes. Represents a specific cognitive path node in the set P; Based on the output probability distribution results, the improved graph neural network is obtained by combining the weight information of the attention mechanism.

6. The method for adaptive optimization of an educational cognitive engine based on a large model according to claim 5, characterized in that: The process of introducing node-level attention is as follows: Calculate the current path node The attention weight with other nodes in the knowledge graph is expressed as: ； in 、 is the feature vector of nodes i and j, for The learnable weight matrix, for The learnable bias term, is the attention weight of node i to node j; The process of introducing relation-level attention is as follows: For different relationship types in the knowledge graph, learn the relationship attention weights, the formula is: ； in, is the relation feature vector, for The learnable weight matrix, for The learnable bias term, is the attention weight of relation r.

7. The method for adaptive optimization of an educational cognitive engine based on a large model according to claim 6, characterized in that: The process of acquiring the cognitive link evolution model is as follows: The process of outputting probability distribution results based on the improved graph neural network is used as the model output process. Historical teaching data is collected, including students' cognitive state vectors at different time periods, corresponding knowledge graph associations, and real cognitive evolution paths, to construct training and validation sets. The cross entropy loss function is used to measure the difference between the predicted probability distribution and the true path The Adam optimizer is used to iteratively optimize the model parameters. The learning rate is initially set to 0.001 and dynamically adjusted according to the training situation. The validation set is used to evaluate the model performance. The model structure is adjusted according to the evaluation results to complete the cognitive evolution link modeling and obtain the cognitive link evolution model.

8. The method for adaptive optimization of an educational cognitive engine based on a large model according to claim 7, characterized in that: The process of outputting the optimal candidate strategy list is as follows: According to the cognitive link evolution model, the path probability distribution output by the cognitive evolution link modeling and the entity association data of the evolutionary knowledge graph are input into the inference engine, and the matching strategy is carried out according to the following steps: Extract cognitive paths with a probability greater than 50%, match the path rules, traverse the path-strategy mapping rules, and check whether the current knowledge graph data meets the preset rule conditions. If the conditions are met, trigger the teaching strategy corresponding to the rule to form a candidate strategy list.

Images