Education knowledge graph reasoning and complementing method based on dynamic updating

Abstract

The invention relates to an educational knowledge graph reasoning and complementing method based on dynamic updating, and belongs to the crossing field of artificial intelligence and educational technology. The method comprises the following steps: constructing a multi-dimensional initial education knowledge graph; capturing and analyzing multivariate interaction data of the learner into a structured event stream in real time; dynamically updating the cognitive state of the learner through a sequential cognitive diagnosis model, and optimizing a knowledge association relationship based on swarm intelligence analysis; a dynamic graph neural network is adopted to generate embedded representation, and cognitive state reasoning and structural knowledge completion are achieved. The system is composed of map initialization, event collection, a dynamic evolution engine and a deep inference engine, and a'data-updating-inference-intervention 'closed loop is formed. According to the method, the problems of cognitive distortion and structure stiffness of the static atlas are solved, adaptive evolution and accurate personalized service of the atlas are realized, and the intelligent level of an education system is remarkably improved.

Description

Technical Field

[0001] This invention relates to the intersection of artificial intelligence and educational technology, specifically to a method for reasoning and completing educational knowledge graphs based on dynamic updates. Background Technology

[0002] Educational knowledge graphs, as a core infrastructure for smart education, aim to represent knowledge, resources, and learners in a structured way, providing support for personalized learning. However, most existing educational knowledge graphs are static or quasi-static. Their construction relies on the prior knowledge of domain experts and offline batch processing analysis of educational resources; once built, their topology and parameters tend to be fixed.

[0003] This static nature creates a profound contradiction with the inherent dynamism of the educational process. Learners' cognitive states are fluid, constantly changing with learning, practice, and even forgetting. Simultaneously, the strength of connections between pieces of knowledge is not static but is influenced by teaching methods and the cognitive patterns of the learner group. Static mapping cannot capture these dynamic changes, leading to the following fundamental flaws:

[0004] 1. Distortion of cognitive state: It is impossible to accurately depict each learner's knowledge mastery at any given moment in real time, which leads to blind recommendations based on outdated or inaccurate profiles.

[0005] 2. Rigid knowledge structure: Unable to discover or correct the connections between knowledge points based on the actual learning behavior of the learner group, such as discovering better learning prerequisite relationships or identifying commonly confused knowledge point pairs.

[0006] 3. Limited reasoning ability: Reasoning based on static and incomplete maps makes it difficult to discover deep-seated cognitive obstacles and predict the learner's development trajectory.

[0007] 4. Lack of evolutionary capacity: The updating and improvement of the map heavily rely on human intervention, which is costly and slow to respond, and cannot form an intelligent system that can grow on its own through interaction with learners.

[0008] Therefore, there is an urgent need for a method that can transform a knowledge graph from a static data warehouse into a dynamic, evolving "living" knowledge agent, enabling it to resonate with the learning process. Summary of the Invention

[0009] Purpose of the invention: To provide a reasoning and completion method for educational knowledge graphs based on dynamic updates, so as to solve the above-mentioned problems existing in the prior art.

[0010] Technical solution: A reasoning and completion method based on dynamically updated educational knowledge graphs, comprising the following steps:

[0011] S1: Constructing a multi-dimensional initial educational knowledge graph: Integrating multi-source heterogeneous educational domain data, and through the joint extraction of entities, relationships, and attributes, constructing an initial educational knowledge graph that includes at least knowledge concept nodes, teaching resource nodes, and learner nodes;

[0012] S2: Capture and parse dynamic learning event streams: Collect learners’ multi-dimensional interaction data from one or more learning interaction terminals in real time, and parse the data into structured learning events containing temporal sequence, subject, object and interaction context information;

[0013] S3: Perform dynamic evolution update of the knowledge graph: Based on the structured learning events, drive a preset graph evolution model to update at least one item of the educational knowledge graph:

[0014] Update the node attributes associated with the learner node that represent its cognitive state;

[0015] Update the relationship parameters that characterize the strength of semantic association between knowledge concept nodes or between knowledge concept nodes and resource nodes;

[0016] S4: Implement deep reasoning and structural completion of the knowledge graph: On the dynamically updated educational knowledge graph, apply a graph representation learning model to generate low-dimensional vector embeddings of entities and relations, and perform at least one of the following tasks based on the embeddings:

[0017] Cognitive state reasoning: Based on the current embedding representation of learner nodes and their neighborhood structure in the graph, reason about their knowledge weaknesses, cognitive state transfer paths, or potential learning abilities.

[0018] Structural knowledge completion: Based on the global topological structure and entity embedding representation of the graph, predict the missing potential relationships with cognitive or pedagogical significance between knowledge concept nodes to achieve self-improvement of the graph structure.

[0019] In a further embodiment, in step S1, the initial educational knowledge graph is a hierarchical structure, including: a domain knowledge layer that defines knowledge concepts and their levels, predecessor and successor relationships; a teaching resource layer that associates teaching resources with knowledge concepts; and a learner cognition layer that represents the history and state of interaction between individual learners and knowledge concepts.

[0020] In a further embodiment, in step S2, the multi-dimensional interactive data includes not only explicit feedback data such as answer records and course learning duration, but also implicit feedback data such as page browsing trajectory, mouse hover duration, and video dragging behavior.

[0021] In a further embodiment, the method for updating the node attributes representing the cognitive state in step S3 is to apply a temporal cognitive diagnostic model, which comprehensively considers the learner's historical interaction sequence, the context of the current learning event (such as question difficulty and resource type), and a cognitive psychology model to update the state.

[0022] In a further embodiment, the cognitive psychology model includes the Ebbinghaus forgetting curve model, which is used to apply a time decay factor over time to cognitive state attributes.

[0023] In a further embodiment, the method for updating the relation parameters in step S3 is to perform group intelligence analysis on the learning trajectory data of a large group of learners to mine the learning transfer efficiency, co-occurrence patterns or confusion relationships between knowledge concepts, thereby dynamically adjusting their corresponding relation weights or types.

[0024] In step S4, the graph representation learning model used is a temporal graph neural network or a dynamic graph neural network model, in order to capture the dual dynamic characteristics of the graph in terms of time and structure.

[0025] In a further embodiment, the cognitive state reasoning task further includes: in the vector embedding space, by performing clustering or manifold analysis on the vector distribution of the learner's mastered and unmastered knowledge concepts, identifying the core obstacle points or key bridge knowledge points in their cognitive structure.

[0026] In a further embodiment, the structured knowledge completion task further includes: defining a relation plausibility scoring function, which not only considers the semantic similarity between entities, but also incorporates educational prior knowledge constraints, to evaluate the impact of adding new relations on the consistency and cognitive rationality of the entire knowledge graph.

[0027] In a further embodiment, steps S2, S3, and S4 form a continuously iterative closed-loop feedback system, wherein the results of reasoning and completion are used to generate new personalized learning interventions, which trigger new learning events and drive the next round of dynamic evolution of the knowledge graph.

[0028] In a further embodiment, the method for reasoning and completing the dynamically updated educational knowledge graph is implemented based on the dynamically updated educational knowledge graph reasoning and completion system.

[0029] The dynamically updated educational knowledge graph reasoning and completion system includes:

[0030] The graph initialization and management module is used to construct and store a multi-dimensional initial educational knowledge graph based on multi-source heterogeneous data.

[0031] The multi-channel learning event acquisition module is used to capture learners' diverse interactive data from various learning terminals in real time and parse it into structured learning events;

[0032] The graph dynamic evolution engine interacts with the graph initialization and management module to drive a graph evolution model based on the structured learning events, updating the node attributes and relationship parameters in the knowledge graph.

[0033] A deep reasoning and completion engine is used to run graph representation learning models on the updated knowledge graph and perform cognitive state reasoning and structural knowledge completion based on the generated embedding representations.

[0034] Beneficial Effects: This invention relates to a method for reasoning and completing educational knowledge graphs based on dynamic updates, which has the following beneficial effects:

[0035] 1. High-fidelity dynamic profile: It realizes real-time synchronization between the learner's cognitive profile and their actual learning status, providing a solid foundation for precise intervention.

[0036] 2. Adaptive evolution of knowledge graphs: This transforms knowledge graphs from a passive data structure into an intelligent agent that can learn and grow from interactions with the environment, possessing the ability to continuously self-optimize.

[0037] 3. Deep cognitive insight: Beyond simple knowledge point mastery judgment, it can perform more in-depth cognitive structure diagnosis, learning potential prediction and other complex reasoning.

[0038] 4. Full-process automated closed loop: An automated intelligent closed loop of "data-update-reasoning-service-re-data" has been constructed, which significantly improves the overall efficiency and intelligence level of the adaptive education system. Attached Figure Description

[0039] Figure 1 This is an overall flowchart of the dynamically updated educational knowledge graph reasoning and completion method described in this invention.

[0040] Figure 2 This is a schematic diagram of the composition framework of the dynamically updated educational knowledge graph reasoning and completion system described in this invention. Detailed Implementation

[0041] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.

[0042] The present invention relates to a method for reasoning and completing educational knowledge graphs based on dynamic updates, comprising the following steps:

[0043] S1: Constructing a multi-dimensional initial educational knowledge graph: Integrating multi-source heterogeneous educational domain data, and through the joint extraction of entities, relationships, and attributes, constructing an initial educational knowledge graph that includes at least knowledge concept nodes, teaching resource nodes, and learner nodes;

[0044] S2: Capture and parse dynamic learning event streams: Collect learners’ multi-dimensional interaction data from one or more learning interaction terminals in real time, and parse the data into structured learning events containing temporal sequence, subject, object and interaction context information;

[0045] S3: Perform dynamic evolution update of the knowledge graph: Based on the structured learning events, drive a preset graph evolution model to update at least one item of the educational knowledge graph:

[0046] Update the node attributes associated with the learner node that represent its cognitive state;

[0047] Update the relationship parameters that characterize the strength of semantic association between knowledge concept nodes or between knowledge concept nodes and resource nodes;

[0048] S4: Implement deep reasoning and structural completion of the knowledge graph: On the dynamically updated educational knowledge graph, apply a graph representation learning model to generate low-dimensional vector embeddings of entities and relations, and perform at least one of the following tasks based on the embeddings:

[0049] Cognitive state reasoning: Based on the current embedding representation of learner nodes and their neighborhood structure in the graph, reason about their knowledge weaknesses, cognitive state transfer paths, or potential learning abilities.

[0050] Structural knowledge completion: Based on the global topological structure and entity embedding representation of the graph, predict the missing potential relationships with cognitive or pedagogical significance between knowledge concept nodes to achieve self-improvement of the graph structure.

[0051] In step S1, the initial educational knowledge graph has a hierarchical structure, including: a domain knowledge layer that defines knowledge concepts and their levels, and their predecessor and successor relationships; a teaching resource layer that associates teaching resources with knowledge concepts; and a learner cognition layer that represents the history and state of interaction between individual learners and knowledge concepts.

[0052] In step S2, the multi-dimensional interactive data includes not only explicit feedback data such as answer records and course learning duration, but also implicit feedback data such as page browsing trajectory, mouse hover duration, and video dragging behavior.

[0053] In step S3, the method for updating the node attributes representing the cognitive state is to apply a temporal cognitive diagnostic model. This model comprehensively considers the learner's historical interaction sequence, the context of the current learning event, such as the difficulty of the question and the type of resources, as well as the cognitive psychology model to update the state.

[0054] The cognitive psychology model includes the Ebbinghaus forgetting curve model, which is used to apply a time decay factor over time to cognitive state attributes.

[0055] In step S3, the method for updating the relation parameters involves performing swarm intelligence analysis on the learning trajectory data of a large group of learners to uncover the learning transfer efficiency, co-occurrence patterns, or confusion relationships between knowledge concepts, thereby dynamically adjusting their corresponding relation weights or types. In step S4, the graph representation learning model used is a temporal graph neural network or a dynamic graph neural network model to capture the dual dynamic characteristics of the graph in both time and structure.

[0056] The cognitive state reasoning task further includes: in the vector embedding space, by performing clustering or manifold analysis on the vector distribution of the learner's mastered and unmastered knowledge concepts, identifying the core obstacle points or key bridge knowledge points in their cognitive structure.

[0057] The structured knowledge completion task further includes: defining a relation plausibility scoring function, which not only considers the semantic similarity between entities, but also incorporates educational prior knowledge constraints, to evaluate the impact of adding new relations on the consistency and cognitive rationality of the entire knowledge graph.

[0058] Steps S2, S3, and S4 form a continuously iterative closed-loop feedback system, in which the results of reasoning and completion are used to generate new personalized learning interventions, which trigger new learning events and drive the next round of dynamic evolution of the knowledge graph.

[0059] The proposed reasoning and completion method based on dynamically updated educational knowledge graphs is implemented using a dynamically updated educational knowledge graph reasoning and completion system.

[0060] The dynamically updated educational knowledge graph reasoning and completion system includes:

[0061] The graph initialization and management module is used to construct and store a multi-dimensional initial educational knowledge graph based on multi-source heterogeneous data.

[0062] The multi-channel learning event acquisition module is used to capture learners' diverse interactive data from various learning terminals in real time and parse it into structured learning events;

[0063] The graph dynamic evolution engine interacts with the graph initialization and management module to drive a graph evolution model based on the structured learning events, updating the node attributes and relationship parameters in the knowledge graph.

[0064] A deep reasoning and completion engine is used to run graph representation learning models on the updated knowledge graph and perform cognitive state reasoning and structural knowledge completion based on the generated embedding representations.

[0065] An adaptive learning service interface is used to receive the output of the deep reasoning and completion engine and transform it into specific personalized service instructions. These service instructions include, but are not limited to: adaptive learning path generation, intelligent exercise recommendation, learning warning and intervention suggestions, or learning community construction.

[0066] Furthermore, the specific operation method of the dynamically updated educational knowledge graph reasoning and completion method is as follows:

[0067] S1: Construct a multidimensional initial education knowledge graph.

[0068] Operating steps:

[0069] Data collection: Collecting data from multiple data sources in the education field, such as textbooks, online course platforms, and learning management systems. This data can include various formats such as text, images, and videos.

[0070] Entity, Relationship, and Attribute Extraction: Using natural language processing techniques (such as named entity recognition and relation extraction) and machine learning algorithms, entities (knowledge concepts, teaching resources, learners, etc.), relations (such as "belongs to", "associated with", "learning", etc.) and attributes (such as the difficulty of knowledge concepts, the type of teaching resources, etc.) are extracted from the collected data.

[0071] Graph Construction: Organize the extracted entities, relationships, and attributes into a hierarchical initial educational knowledge graph, including a domain knowledge layer, a teaching resource layer, and a learner cognition layer.

[0072] S2: Capture and parse the dynamic learning event stream.

[0073] Operating steps:

[0074] Data collection: By deploying data collection scripts on learning interaction terminals (such as web pages and mobile applications), learners' diverse interactive data are collected in real time, including answer records, course learning time, page browsing trajectory, mouse hover time, video dragging behavior, etc.

[0075] Data parsing: Parsing the collected raw data into structured learning events that include information on time sequence, subject, object, and interaction context. For example, parsing a question record into the form of "the learner answered question q at time t, the answer is a, and the question difficulty is d".

[0076] Let E be the set of structured learning events. For the i-th learning event, ,in Indicates the time when the event occurred. Indicates the subject (learner). It refers to the object (knowledge concepts, teaching resources, etc.). This indicates the context of the interaction.

[0077] S3: Perform dynamic evolution and updates of the knowledge graph.

[0078] Update the node attributes associated with the learner node that represent its cognitive state.

[0079] Operating steps:

[0080] Data preparation: Collect learners' historical interaction sequences and contextual information of current learning events.

[0081] Cognitive state update: Apply a temporal cognitive diagnostic model, comprehensively consider the learner's historical interaction sequence, the context of the current learning event (such as question difficulty and resource type), and cognitive psychology models (such as the Ebbinghaus forgetting curve model) to update the cognitive state attributes of the learner node.

[0082] set up This represents the cognitive state of learner s at time t. This represents the historical interaction sequence of learner s. This represents the learning event of learner s at time t. Represents learning events The context information. Therefore, the cognitive state update formula is:

[0083]

[0084] Where f is a temporal cognitive diagnostic model, It is a time decay factor calculated based on the Ebbinghaus forgetting curve model, and its calculation formula is as follows:

[0085]

[0086] in, It is the time when the last learning event occurred, and T is the forgetting time constant.

[0087] The specific steps for updating the relationship parameters that represent the strength of semantic association between knowledge concept nodes or between knowledge concept nodes and resource nodes are as follows:

[0088] 1. Data collection: Collect learning trajectory data of a large group of learners.

[0089] 2. Swarm intelligence analysis: Analyze learning trajectory data to uncover learning transfer efficiency, co-occurrence patterns, or confusion relationships between knowledge concepts.

[0090] 3. Relationship parameter update: Based on the analysis results, dynamically adjust the weight or type of the relationship between knowledge concept nodes or between knowledge concept nodes and resource nodes.

[0091] set up Let represent the relationship parameters between knowledge concept nodes i and j, T be the learning trajectory data set of a large group of learners, and g be the swarm intelligence analysis function. Then the formula for updating the relationship parameters is:

[0092]

[0093] in, These are the initial values ​​of the relational parameters.

[0094] S4: Implement deep reasoning and structural completion of the graph.

[0095] Cognitive state reasoning.

[0096] Operating steps:

[0097] Vector embedding generation: On the dynamically updated educational knowledge graph, a temporal graph neural network or a dynamic graph neural network model is applied to generate low-dimensional vector embedding representations of entities and relations.

[0098] Clustering or manifold analysis: In the vector embedding space, clustering or manifold analysis is performed on the vector distribution of learners' mastered and unmastered knowledge concepts to identify core obstacles or key bridge knowledge points in their cognitive structure.

[0099] Let V be the vector embedding space of entities and relations. Let represent the set of vectors representing the knowledge concepts that learner s has mastered. This represents the set of vectors representing the knowledge concepts that learner s has not yet mastered. Clustering algorithms (such as K-means clustering) are used to separate these vectors. and The data is divided into different clusters. Core obstacles and key bridge knowledge points can be identified by calculating the distance and density between clusters.

[0100] Complete the structural knowledge.

[0101] Operating steps:

[0102] Vector embedding generation: same as cognitive state reasoning step 1.

[0103] Relationship evaluation: Define a relation plausibility scoring function to evaluate the impact of adding new relations on the consistency and cognitive rationality of the entire knowledge graph.

[0104] Relationship addition: Based on the scoring results, select appropriate new relationships to add to the knowledge graph, thereby achieving self-improvement of the graph structure.

[0105] set up This represents the potential relationship between knowledge concept nodes i and j. Representing relations The plausibility score. The scoring function is defined as:

[0106]

[0107] in, Represents the vector embedding of nodes i and j and Semantic similarity between them (such as cosine similarity). This represents the relationship calculated based on prior knowledge in pedagogy. The prior probability. It is the weighting coefficient.

[0108] Furthermore, the specific details of the dynamically updated educational knowledge graph reasoning and completion system are as follows:

[0109] Map initialization and management module.

[0110] Operating steps:

[0111] Data processing: Cleaning, preprocessing and feature extraction of the collected multi-source heterogeneous data.

[0112] Graph Construction: Based on the extracted entities, relationships, and attributes, construct a multidimensional initial educational knowledge graph and store it in a graph database.

[0113] Map Management: Provides functions for querying, updating, and maintaining maps.

[0114] Multi-channel learning event acquisition module.

[0115] Operating steps:

[0116] Data collection: Deploy data collection scripts on various learning terminals to collect diverse interactive data from learners in real time.

[0117] Data transmission: The collected data is transmitted to the server for parsing and processing.

[0118] Data parsing: Parsing raw data into structured learning events.

[0119] Formula: Same as the formula in step S2.

[0120] Graph Dynamic Evolution Engine.

[0121] Operating steps:

[0122] Event reception: Receives structured learning events parsed by the multi-channel learning event acquisition module.

[0123] Graph Update: Based on structured learning events, drive the graph evolution model to update node attributes and relationship parameters in the knowledge graph.

[0124] Graph storage: Store the updated knowledge graph in a graph database.

[0125] Formula: Same as the formula in step S3.

[0126] Deep reasoning and completion engine.

[0127] Operating steps:

[0128] Vector embedding generation: On the updated knowledge graph, run a sequential graph neural network or dynamic graph neural network model to generate low-dimensional vector embedding representations of entities and relations.

[0129] Reasoning and Completion: Based on the generated embedded representations, perform cognitive state reasoning and structural knowledge completion tasks.

[0130] Output results: Output the results of inference and completion to the adaptive learning service interface.

[0131] Formula: Same as the formula in step S4.

[0132] Adaptive learning service interface.

[0133] Operating steps:

[0134] Result reception: Receives the output of the deep reasoning and completion engine.

[0135] Instruction generation: Transform the results of reasoning and completion into specific personalized service instructions, such as adaptive learning path generation, intelligent exercise recommendation, learning early warning and intervention suggestions, and learning community construction.

[0136] Command sending: Send personalized service commands to the learning interaction terminal to provide learners with personalized learning services.

[0137] Furthermore, steps S2, S3, and S4 form a continuously iterative closed-loop feedback system. The results of reasoning and completion are used to generate new personalized learning interventions, which trigger new learning events, thereby driving the next round of dynamic evolution of the knowledge graph. The specific process is as follows:

[0138] 1. Learning event collection: Collect learners' learning events through the multi-channel learning event collection module.

[0139] 2. Knowledge Graph Update: The knowledge graph dynamic evolution engine updates the knowledge graph based on learning events.

[0140] 3. Reasoning and Completion: The deep reasoning and completion engine performs reasoning and completion on the updated knowledge graph.

[0141] 4. Service Generation: The adaptive learning service interface generates personalized learning service instructions based on the results of reasoning and completion.

[0142] 5. Learning Intervention: Send personalized learning service instructions to the learning interaction terminal to trigger a new learning event and return to step 1.

[0143] As described above, although the invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the invention itself. Various changes in form and detail may be made without departing from the spirit and scope of the invention as defined in the appended claims.

Claims

1. A reasoning and completion method based on dynamically updated educational knowledge graphs, characterized by: Includes the following steps: S1: Constructing a multi-dimensional initial educational knowledge graph: Integrating multi-source heterogeneous educational domain data, and through the joint extraction of entities, relationships, and attributes, constructing an initial educational knowledge graph that includes at least knowledge concept nodes, teaching resource nodes, and learner nodes; S2: Capture and parse dynamic learning event streams: Collect learners’ multi-dimensional interaction data from one or more learning interaction terminals in real time, and parse the data into structured learning events containing temporal sequence, subject, object and interaction context information; S3: Perform dynamic evolution update of the knowledge graph: Based on the structured learning events, drive a preset graph evolution model to update at least one item of the educational knowledge graph: Update the node attributes associated with the learner node that represent its cognitive state; Update the relationship parameters that characterize the strength of semantic association between knowledge concept nodes or between knowledge concept nodes and resource nodes; S4: Implement deep reasoning and structural completion of the knowledge graph: On the dynamically updated educational knowledge graph, apply a graph representation learning model to generate low-dimensional vector embeddings of entities and relations, and perform at least one of the following tasks based on the embeddings: Cognitive state reasoning: Based on the current embedding representation of learner nodes and their neighborhood structure in the graph, reason about their knowledge weaknesses, cognitive state transfer paths, or potential learning abilities. Structural knowledge completion: Based on the global topological structure and entity embedding representation of the graph, predict the missing potential relationships with cognitive or pedagogical significance between knowledge concept nodes to achieve self-improvement of the graph structure.

2. The method for reasoning and completing an educational knowledge graph based on dynamic updates as described in claim 1, characterized in that: In step S1, the initial educational knowledge graph has a hierarchical structure, including: a domain knowledge layer that defines knowledge concepts and their levels, and their predecessor and successor relationships; a teaching resource layer that associates teaching resources with knowledge concepts; and a learner cognition layer that represents the history and state of interaction between individual learners and knowledge concepts.

3. The method for reasoning and completing an educational knowledge graph based on dynamic updates as described in claim 1, characterized in that: In step S2, the multi-dimensional interactive data includes not only explicit feedback data such as answer records and course learning duration, but also implicit feedback data such as page browsing trajectory, mouse hover duration, and video dragging behavior.

4. The method for reasoning and completing an educational knowledge graph based on dynamic updates according to claim 1, characterized in that: In step S3, the method for updating the node attributes representing the cognitive state is to apply a temporal cognitive diagnostic model. This model comprehensively considers the learner's historical interaction sequence, the context of the current learning event, and a cognitive psychology model to update the state.

5. The method for reasoning and completing an educational knowledge graph based on dynamic updates according to claim 1, characterized in that: The cognitive psychology model includes the Ebbinghaus forgetting curve model, which is used to apply a time decay factor over time to cognitive state attributes.

6. The method for reasoning and completing an educational knowledge graph based on dynamic updates according to claim 1, characterized in that: In step S3, the method for updating the relation parameters is to perform group intelligence analysis on the learning trajectory data of a large group of learners, mine the learning transfer efficiency, co-occurrence patterns or confusion relationships between knowledge concepts, and then dynamically adjust their corresponding relation weights or types. In step S4, the graph representation learning model used is a temporal graph neural network or a dynamic graph neural network model, in order to capture the dual dynamic characteristics of the graph in terms of time and structure.

7. The method for reasoning and completing an educational knowledge graph based on dynamic updates according to claim 1, characterized in that: The cognitive state reasoning task further includes: in the vector embedding space, by performing clustering or manifold analysis on the vector distribution of the learner's mastered and unmastered knowledge concepts, identifying the core obstacle points or key bridge knowledge points in their cognitive structure.

8. The method for reasoning and completing an educational knowledge graph based on dynamic updates according to claim 1, characterized in that: The structured knowledge completion task further includes: defining a relation scoring function that not only considers the semantic similarity between entities but also incorporates educational prior knowledge constraints to evaluate the impact of adding new relations on the consistency and cognitive rationality of the entire knowledge graph.

9. A method for reasoning and completing an educational knowledge graph based on dynamic updates, as described in any one of claims 1 to 8, characterized in that: Steps S2, S3, and S4 form a continuously iterative closed-loop feedback system, in which the results of reasoning and completion are used to generate new personalized learning interventions, which trigger new learning events and drive the next round of dynamic evolution of the knowledge graph.

10. The method for reasoning and completing an educational knowledge graph based on dynamic updates according to claim 1, characterized in that: The proposed reasoning and completion method based on dynamically updated educational knowledge graphs is implemented using a dynamically updated educational knowledge graph reasoning and completion system. The dynamically updated educational knowledge graph reasoning and completion system includes: The graph initialization and management module is used to construct and store a multi-dimensional initial educational knowledge graph based on multi-source heterogeneous data. The multi-channel learning event acquisition module is used to capture learners' diverse interactive data from various learning terminals in real time and parse it into structured learning events; The graph dynamic evolution engine interacts with the graph initialization and management module to drive a graph evolution model based on the structured learning events, updating the node attributes and relationship parameters in the knowledge graph. A deep reasoning and completion engine is used to run graph representation learning models on the updated knowledge graph and perform cognitive state reasoning and structural knowledge completion based on the generated embedding representations.

Images