Multi-agent collaborative education decision-making system and method

Abstract

The invention relates to the technical field of artificial intelligence education, and discloses a multi-agent collaborative education decision-making system and method. The method comprises the following steps: constructing a dynamic knowledge graph, and tracking the learning state of a student in real time; analyzing learning behavior data in parallel through a diagnosis agent, a resource agent and a path agent, and respectively outputting a knowledge weak point diagnosis vector, a personalized resource recommendation queue and an optimal learning path decision tree; based on a multi-target constraint strategy fusion algorithm, generating a dynamic teaching scheme in a collaborative decision center; and the execution engine monitors feedback in real time and triggers strategy re-planning. Through a multi-agent parallel collaboration mechanism, the technical problems of one-sided decision making of a single model and insufficient module collaboration are solved, the technical effects that the cognitive diagnosis accuracy is 72.3% and the path recommendation matching degree is 75.8% are achieved, and the personalized teaching precision and the resource utilization efficiency are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence education technology, and in particular to an educational decision-making system and method based on multi-agent collaboration, for realizing personalized learning path planning and precise teaching decisions. Background Technology

[0002] With the rapid development of artificial intelligence technology, intelligent education platforms have become an important support for the digital transformation of education. Existing intelligent education systems mainly use single algorithm models for learning recommendations, which makes it difficult to simultaneously meet the multidimensional requirements of cognitive diagnosis accuracy, resource matching efficiency, and path planning rationality.

[0003] There are already relevant technical solutions, such as patent CN120125394A which discloses a large-model-based assisted programming teaching method, which completes code correction tasks through multiple prompt word templates and a large model working together; patent CN120409650A which discloses a knowledge point mastery update method based on knowledge graphs, which updates mastery through the propagation of pre- and post-knowledge points; and patent CN120804282A which discloses an AI-based intelligent question generation practice method, which generates personalized questions based on behavioral data and question complexity.

[0004] However, existing technologies have the following prominent problems: First, a single intelligent agent architecture is difficult to collaboratively process multi-dimensional teaching decisions, resulting in biased recommendation results; second, diagnostic modules, resource libraries, and path planners often operate independently, forming an "information silo" effect and lacking a dynamic information sharing mechanism; third, traditional systems use preset rules for resource allocation, which cannot be dynamically adjusted according to real-time changes in learning progress; and finally, the decision-making process of most AI education systems lacks interpretability, reducing the credibility of the system. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a multi-agent collaborative education decision-making system and method. By constructing a dynamic knowledge graph hub and a parallel decision-making architecture of three functional agents, the entire process of education decision-making is made intelligent.

[0006] The technical solution of the present invention is achieved through the following steps:

[0007] Step 1: Construct a dynamic knowledge graph to track and update students' knowledge mastery status in real time;

[0008] Step 2: Analyze the learning behavior data through a diagnostic agent and output a diagnostic vector for knowledge gaps; the diagnostic agent uses a deep temporal network to analyze the learning behavior sequence and outputs a multi-dimensional diagnostic vector containing knowledge mastery, error patterns, and cognitive states.

[0009] Step 3: The resource agent evaluates the utility of the learned resources and generates a personalized resource recommendation queue. The resource agent uses the multi-armed gambling machine algorithm to evaluate the resource utility and combines the resource associations in the knowledge graph to generate a resource recommendation queue sorted by matching degree.

[0010] Step 4: Construct an optimal learning path decision tree using a path agent based on an improved Monte Carlo tree search algorithm; the path agent searches the knowledge graph for the optimal learning path that satisfies multiple objectives, including maximizing knowledge mastery, minimizing learning time, and maximizing learning interest matching.

[0011] Step 5: In the collaborative decision-making center, a multi-objective constraint strategy fusion algorithm is used to integrate the output results of the diagnostic agent, resource agent, and path agent to generate a dynamic teaching scheme; the collaborative decision-making center establishes a communication protocol and message bus between agents to realize strategy fusion under multi-objective constraints.

[0012] Step Six: Monitor the implementation effect of the teaching plan in real time through the execution engine, and trigger strategy replanning based on feedback data; the execution engine evaluates the strategy effect through A / B testing, forming a closed-loop control mechanism of decision-making-execution-feedback-optimization.

[0013] The diagnostic agent uses deep temporal network analysis to learn behavior sequences and outputs a multi-dimensional diagnostic vector containing knowledge mastery, error patterns, and cognitive states.

[0014] The resource agent uses a multi-armed gambling machine algorithm to evaluate resource utility and combines the resource associations in the knowledge graph to generate a resource recommendation queue sorted by matching degree.

[0015] The path agent employs an improved Monte Carlo tree search algorithm to search for the optimal learning path that satisfies multi-objective constraints in the knowledge graph.

[0016] The collaborative decision-making center establishes an inter-agent communication protocol and message bus to achieve policy fusion under multi-objective constraints.

[0017] The execution engine monitors the effectiveness of the teaching plan in real time, evaluates the effectiveness of the strategy through A / B testing, and forms a closed-loop control mechanism of decision-making, execution, feedback and optimization.

[0018] Compared with existing technologies, this invention effectively solves the technical bottlenecks in traditional education systems, such as the one-sided decision-making of single models, the "information silos" formed by the isolation of various functional modules, and the lack of interpretability of the decision-making process, by constructing a multi-agent collaborative decision-making architecture. It achieves a significant improvement in the accuracy of cognitive diagnosis and the matching degree of learning path recommendation. The system adopts a parallel processing mechanism to greatly optimize response efficiency and resource utilization. By introducing a dynamic knowledge graph and a closed-loop feedback mechanism, it constructs an integrated intelligent decision-making process of "diagnosis-recommendation-planning-execution-feedback", thereby improving the accuracy and adaptability of teaching while enhancing the transparency and interpretability of the system's decision-making process. Attached Figure Description

[0019] Figure 1 This is a core architecture diagram of a multi-agent collaborative decision-making system illustrated in an exemplary embodiment of the present invention.

[0020] Figure 2 This is a flowchart illustrating a multi-agent collaborative education decision-making method according to an exemplary embodiment of the present invention.

[0021] Figure 3 This is a schematic diagram illustrating the principle of a dynamic knowledge graph construction and updating method according to an exemplary embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Those skilled in the art, after reading this specification, can make various equivalent modifications and adaptations to the embodiments based on the principles and spirit of this invention, and these modifications and adaptations should also be considered within the scope of protection of this invention.

[0023] like Figure 1 As shown, the multi-agent collaborative education decision-making system of this invention adopts a layered distributed architecture design, mainly including four core layers: data acquisition layer, intelligent analysis layer, decision-making layer, and execution layer. Data exchange between each layer is achieved through well-defined interface protocols, ensuring loose coupling and high scalability of the system.

[0024] The data acquisition layer is responsible for collecting student learning behavior data in real time from multiple sources, including code submission records, test results, and resource usage. This data, after cleaning and standardization, is then passed to the intelligent analysis layer for in-depth processing. The intelligent analysis layer comprises three intelligent agent modules, employing a parallel processing mechanism to analyze student data simultaneously, effectively improving system response speed.

[0025] Reference Figure 2The flowchart of the multi-agent collaborative education decision-making method shown below illustrates the following six core steps in the specific implementation process of this invention:

[0026] First, the system executes the knowledge graph construction step, extracting knowledge points and their relationships from the course syllabus, textbook content, and historical teaching data to construct an initial knowledge graph structure. This graph is represented using an attribute graph model, where nodes represent entities such as knowledge points and learning resources, and edges represent semantic relationships and dependencies between entities.

[0027] Next, the diagnostic analysis step begins. The diagnostic agent uses a deep temporal network model to extract features and recognize patterns from the input learning behavior sequence. This network employs a multi-head self-attention mechanism to capture long-term dependencies, processes variable-length sequence inputs through gated recurrent units, and finally outputs a multi-dimensional diagnostic vector containing knowledge point mastery, error pattern classification, and cognitive state assessment.

[0028] In the resource recommendation step, the resource agent employs a context-aware multi-armed gambler algorithm to evaluate resource utility. The algorithm maintains a resource feature library, where each resource is represented as a high-dimensional feature vector, including attributes such as difficulty level, knowledge coverage, and media type. By balancing the exploration of new resources with the utilization of known, efficient resources, the algorithm dynamically updates the expected utility value of each resource, generating a recommendation queue sorted in descending order of matching degree.

[0029] The path planning step employs an improved Monte Carlo tree search algorithm, conducting heuristic exploration within a search space comprised of a knowledge graph. The algorithm defines a multi-objective optimization function, comprehensively considering the requirements of knowledge mastery, learning efficiency, and interest matching. During the search process, the algorithm simulates the learning effects of different paths, evaluates the comprehensive utility value of each candidate path, and ultimately outputs a Pareto optimal solution set.

[0030] The decision fusion step is executed in the collaborative decision center, which employs a multi-objective decision framework based on the analytic hierarchy process (AHP). The system first constructs a judgment matrix to assess the relative importance of each optimization objective, and ensures the rationality of weight allocation through consistency checks. Then, a weighted summation method is used to fuse the outputs of each agent to generate the final teaching plan. The decision center also maintains a policy knowledge base, recording historical decisions and their effects to provide a reference for subsequent decisions.

[0031] Finally, the monitoring and adjustment steps are executed, with the engine tracking the effectiveness of the teaching plan in real time through embedded data. The system is equipped with a multi-dimensional evaluation indicator system, including core indicators such as learning progress, mastery level, and participation. When a deviation in effectiveness is detected, the engine automatically triggers a strategy replanning mechanism, forming a complete closed-loop control system.

[0032] like Figure 3As shown, the dynamic knowledge graph of this invention adopts an incremental update mechanism to ensure that the graph content remains synchronized with students' learning progress. The graph update process includes two main stages: node state update and relationship weight adjustment.

[0033] Node state updates are based on a time decay model, which considers the knowledge forgetting curve and the reinforcement effect of practice. For each knowledge point node, the system periodically calculates its mastery score, using the following formula: ,in Indicates performance in historical exercises. As the weight of the question, As the attenuation factor, The time interval is defined by this model. This model allows the system to accurately reflect students' real-time mastery of each knowledge point.

[0034] The relation weight adjustment adopts an association rule-based approach, systematically analyzing co-occurrence patterns in student behavior data and dynamically updating the association strength between knowledge points. When it is found that certain knowledge points are frequently studied simultaneously or are easily confused, the system automatically strengthens the semantic relationships between these knowledge points, providing a more accurate reference for path planning.

[0035] The map update process also incorporates a conflict detection mechanism. When an inconsistent update operation is detected, the system will automatically trigger a consistency verification process to ensure the integrity and accuracy of the map data.

[0036] In actual deployment, the system adopts a containerized microservice architecture, with each functional module deployed and scaled independently. The knowledge graph service is deployed on a graph database cluster, using sharding technology to ensure query performance. The agent service is deployed on GPU computing nodes, utilizing model parallelism to improve inference speed.

[0037] The system establishes a comprehensive service governance mechanism, including load balancing, circuit breaking and degradation, and traffic control, to ensure stable operation under high-concurrency scenarios. Simultaneously, the system provides a complete management and monitoring interface, allowing administrators to easily monitor system status in real time and adjust system parameters promptly.

[0038] Through the above specific embodiments, this invention constructs a complete and efficient multi-agent collaborative education decision-making system, which can effectively support the implementation of personalized teaching. Those skilled in the art, after reading this specification, can make various modifications and adaptations to the embodiments without departing from the principles and spirit of this invention, and these modifications and adaptations should also be considered within the scope of protection of this invention.

Claims

1. A multi-agent collaborative educational decision-making method, characterized in that, Includes the following steps: A dynamic knowledge graph is constructed to track and update students' knowledge mastery status in real time; a diagnostic agent analyzes learning behavior data and outputs diagnostic vectors for knowledge gaps; a resource agent evaluates the utility of learning resources and generates a personalized resource recommendation queue; a path agent constructs an optimal learning path decision tree based on an improved Monte Carlo tree search algorithm; a multi-objective constraint strategy fusion algorithm is used in the collaborative decision center to integrate the output results of the three agents to generate a dynamic teaching plan; and an execution engine monitors the execution effect of the teaching plan in real time and triggers strategy replanning based on feedback data.

2. The method according to claim 1, characterized in that, The diagnostic agent uses deep temporal network analysis to learn behavior sequences and outputs a multi-dimensional diagnostic vector containing knowledge mastery, error patterns, and cognitive states.

3. The method according to claim 1, characterized in that, The resource agent uses a multi-armed gambling machine algorithm to evaluate resource utility and combines the resource associations in the knowledge graph to generate a resource recommendation queue sorted by matching degree.

4. The method according to claim 1, characterized in that, The path agent employs an improved Monte Carlo tree search algorithm to search for the optimal learning path in the knowledge graph that satisfies multiple objectives, including maximizing knowledge mastery, minimizing learning time, and maximizing the matching degree of learning interest.

5. The method according to claim 1, characterized in that, The collaborative decision-making center establishes an inter-agent communication protocol and message bus to achieve policy fusion under multi-objective constraints, including diagnostic accuracy, resource utilization, and path rationality.

6. The method according to claim 1, characterized in that, The execution engine monitors the effectiveness of the teaching plan in real time, evaluates the effectiveness of the strategy through A / B testing, and triggers strategy replanning based on real-time feedback data, forming a closed-loop control mechanism of decision-making-execution-feedback-optimization.

7. A multi-agent collaborative educational decision-making system, characterized in that, include: The knowledge graph construction module is used to build and maintain dynamic knowledge graphs; The diagnostic agent module analyzes learning behavior data and outputs diagnostic results; the resource agent module evaluates resource utility and generates a recommendation queue; and the path agent module plans the optimal learning path. The collaborative decision-making center module is used to integrate the outputs of multiple agents and generate teaching plans; the execution engine module is used to monitor the execution effect and trigger replanning.

8. A computing device, characterized in that, The method includes a memory and a processor, the memory storing computer-executable instructions, and the processor executing the computer-executable instructions to implement the method of any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, It stores computer-executable instructions that, when executed by a processor, implement the method of any one of claims 1-6.

Images