A multi-agent based adaptive learning path planning method and system

Abstract

This invention discloses an adaptive learning path planning method and system based on multi-agent intelligence. The method includes: employing a cognitive diagnostic agent to calculate a student's mastery score for each knowledge point in a given knowledge graph, and employing a multimodal agent to analyze the student's answer records, question content, and question images to generate multimodal analysis results; employing a path planning agent to generate an implementation plan for the learning path planning; employing a path generation agent to generate a preliminary learning path following the implementation plan; and employing a path verification agent to verify the preliminary learning path and output the final learning path plan. This invention, in the learning path planning process, improves the illusion problem existing in previous large-scale model inference by constructing a multi-agent system and deeply integrating various types of agents, thereby enhancing the reliability and executability of large-scale model learning path planning.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and education technology, and more specifically to an adaptive learning path planning method and system based on multi-agent systems. Background Technology

[0002] In recent years, with the rapid development of modern information technology, emerging technologies such as artificial intelligence have been widely applied in the field of education, triggering profound changes in learning concepts and methods. Against this backdrop, online learning has transcended the limitations of time and space, providing learners with more possibilities for "anytime, anywhere" learning, thus experiencing rapid development. However, the separation of teachers and students in time and space during online learning makes it difficult for teachers to promptly grasp students' learning status, which to some extent restricts the improvement of teaching quality in online learning. Faced with diversified learning needs and massive learning resources, how to quickly achieve learning goals, reduce learning costs, and rationally allocate learning resources have become major issues limiting individual and societal development. The traditional "one-size-fits-all" education model can no longer meet people's needs for knowledge acquisition; learners need a more efficient and scientific personalized education model to help them achieve learning goals to the maximum extent with minimal learning costs. Based on this background, how to automatically and efficiently identify learner characteristics, efficiently organize and allocate learning resources, and plan personalized paths for each learner has become an urgent problem to be solved in the research of personalized, precise educational resource matching mechanisms.

[0003] Adaptive learning path planning refers to dynamically adjusting learning content and sequence by analyzing users' learning behavior and needs to provide a personalized and efficient learning experience. With the application of deep learning and reinforcement learning, learners' needs can be identified more accurately, allowing for the design of the most suitable learning paths. Advanced algorithms and models enable intelligent management of learning paths, real-time monitoring of learners' progress, and adjustments to subsequent learning tasks based on their performance, ensuring that each learner achieves optimal learning outcomes in the shortest possible time. However, developing adaptive learning path planning systems still faces challenges: traditional systems struggle to accurately diagnose learners' cognitive states and dynamically generate suitable paths; large language models are prone to "illusory output" during deep reasoning in STEM subjects; and single-agent architectures struggle to collaboratively handle multimodal teaching decisions.

[0004] Therefore, reducing the "illusion" interference during reasoning in large language models, ensuring that large language models can accept multimodal information, and adaptively generating personalized learning paths based on students' knowledge mastery and historical answer records are problems that urgently need to be solved by those skilled in the art. Summary of the Invention

[0005] In view of the above problems, this invention is proposed to provide a multi-agent-based adaptive learning path planning method and system that overcomes or at least partially solves the above problems. It adopts a cognitive diagnostic agent and a multimodal agent to learn the personalized characteristics of students, combines knowledge graphs, and uses a multi-agent collaboration of a path planning agent, a path generating agent, and a path verification agent to ensure that the generated plan is rigorous, executable, and adapted to the students' learning situation.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, embodiments of the present invention provide an adaptive learning path planning method based on multiple agents, comprising the following steps: S1. A cognitive diagnostic agent is used to calculate the student's mastery score for each knowledge point in a given knowledge graph, and a multimodal agent is used to analyze the student's answer records, question content, and question images to generate multimodal analysis results; the multimodal analysis results are the comprehensive cognitive diagnostic results of the student's mastery of each knowledge point. S2. The mastery score and the multimodal analysis results are used together as the student's knowledge point mastery status information. Combined with the student's answer records and the given knowledge graph, a path planning agent is used to generate an implementation plan for the learning path planning. The implementation plan includes the basic principles of path planning, student ability grading standards, the range of the number of learning knowledge points corresponding to different ability levels, and knowledge point priority strategies. S3. Using path generation, an intelligent agent is generated, following the implementation plan, and combining the student's knowledge point mastery status information and the knowledge point dependencies in the given knowledge graph to generate a preliminary learning path. S4. Employ a path verification agent to verify the preliminary learning path, checking whether it conforms to the pre- and post-dependencies of the given knowledge graph and whether it meets the requirements for the number of knowledge points to be learned in the implementation plan. Correct any errors found and output the final learning path plan.

[0007] Furthermore, in step S1, the mastery score ranges from 0 to 1; where 0 indicates no mastery and 1 indicates complete mastery.

[0008] Furthermore, in step S1, the cognitive diagnostic agent is a deep cognitive diagnostic model KaNCD, whose inputs include the student's answer records, question difficulty, and given knowledge graph, and outputs the student's mastery score for each knowledge point.

[0009] Furthermore, in step S1, the multimodal agent adopts a multimodal neural network architecture that integrates a text encoder, an image encoder, and a question-answering behavior sequence encoder. It extracts features from the question content text, the image attached to the question, and the student's answer time sequence, respectively, and integrates the three types of features through a cross-modal attention mechanism to generate fine-grained mastery status supplementary information for each knowledge point.

[0010] Furthermore, in step S2, the student ability grading standard specifically includes: classifying students with an average mastery score of less than 0.5 for all knowledge points as having poor ability, those with a score between 0.5 and 0.55 as having medium ability, and those with a score higher than 0.55 as having high ability.

[0011] Furthermore, in step S3, when generating the initial learning path, for multiple knowledge points with the same prerequisite knowledge points, the knowledge points with lower mastery scores are prioritized for learning.

[0012] Furthermore, in step S4, the verification process of the path verification agent includes structural verification, consistency verification, and logical verification. The structural verification is used to ensure that, in the initial learning path, for any knowledge point, all the prerequisite knowledge points defined in the given knowledge graph have been arranged to be learned before that knowledge point. The consistency verification is used to check whether the total number of knowledge points contained in the preliminary learning path conforms to the range of the number of learning knowledge points determined according to the student's ability level in the implementation plan. The logical verification is used to evaluate whether the order of knowledge points in the preliminary learning path follows the knowledge point priority strategy defined in the implementation plan.

[0013] Furthermore, the path planning agent, path generation agent, and path verification agent all provide explanations for the reasons behind the content they generate or verify when performing their respective tasks.

[0014] Secondly, embodiments of the present invention provide an adaptive learning path planning system based on multiple agents, comprising the following modules: Cognitive Diagnosis Module: This module uses a cognitive diagnostic agent to calculate the student's mastery score for each knowledge point in a given knowledge graph, and uses a multimodal agent to analyze the student's answer records, question content, and question images to generate multimodal analysis results. The multimodal analysis results are a comprehensive cognitive diagnostic result of the student's mastery of each knowledge point. The path planning module is used to combine the mastery score and the multimodal analysis results as information on the student's mastery of knowledge points. It combines the student's answer records and the given knowledge graph, and uses a path planning agent to generate an implementation plan for the learning path planning. The implementation plan includes the basic principles of path planning, student ability grading standards, the range of the number of learning knowledge points corresponding to different ability levels, and knowledge point priority strategies. Path generation module: Used to generate an agent using a path, following the implementation plan, and combining the student's knowledge point mastery status information and the knowledge point dependencies in the given knowledge graph to generate a preliminary learning path; Path verification module: Used by a path verification agent to verify the preliminary learning path, check whether it conforms to the pre- and post-dependencies of the given knowledge graph, whether it meets the requirements of the implementation plan regarding the number of learning knowledge points, correct errors when they are found, and output the final learning path plan.

[0015] As can be seen from the above technical solution, compared with the prior art, the present invention discloses an adaptive learning path planning method and system based on multiple agents, which has the following beneficial effects: This invention integrates a cognitive diagnostic agent and a multimodal agent to generate student knowledge mastery data. This allows the probabilistic output of the large language model to corroborate the results calculated by the deep model based on historical data, enhancing the reliability of the content generated by the large language model. Furthermore, the addition of a knowledge graph controls the scope of the generated content. The multimodal agent can acquire question information, including the corresponding text, data, and images. Based on this rich information, the generated results can capture students' personalized characteristics, meeting the needs of adaptive path planning.

[0016] Secondly, this invention employs three path planning agents, breaking down the path planning task into three sub-tasks. The large language model can focus on these sub-tasks, resulting in better performance. Furthermore, the verification agent can evaluate the generated path plans and adjust those that do not meet the constraints, enhancing the executability of the learned paths. Simultaneously, all agents provide explanations for the generated content during content generation, facilitating subsequent error correction. The invention can also be extended to adapt to different learning difficulties by modifying the knowledge graph, demonstrating strong scalability. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0018] Figure 1 This is a flowchart of the adaptive learning path planning method based on multiple agents provided in this embodiment of the invention; Figure 2 This is a schematic diagram of the adaptive learning path planning method based on multiple agents provided in this embodiment of the invention. Figure 3 This is a distribution chart of student knowledge point mastery scores in a simulation experiment provided in this embodiment of the invention; Figure 4 This is a scoring chart of some knowledge points for students in a simulated experiment provided in this embodiment of the invention; Figure 5 This is a diagram showing the student learning path planning results in a simulation experiment provided in this embodiment of the invention. Figure 6 This is a verification diagram of student learning path planning in a simulation experiment provided in this embodiment of the invention; Figure 7 This is a modified diagram of student learning path planning in a simulation experiment provided in this embodiment of the invention; Figure 8 This is a structural diagram of the multi-agent-based adaptive learning path planning system provided in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] This invention discloses an adaptive learning path planning method based on multiple agents, comprising the following steps: S1. A cognitive diagnostic agent is used to calculate the student's mastery score for each knowledge point in a given knowledge graph, and a multimodal agent is used to analyze the student's answer records, question content, and question images to generate multimodal analysis results; the multimodal analysis results are the comprehensive cognitive diagnostic results of the student's mastery of each knowledge point. S2. The mastery score and the multimodal analysis results are used together as the student's knowledge point mastery status information. Combined with the student's answer records and the given knowledge graph, a path planning agent is used to generate an implementation plan for the learning path planning. The implementation plan includes the basic principles of path planning, student ability grading standards, the range of the number of learning knowledge points corresponding to different ability levels, and knowledge point priority strategies. S3. Using path generation, an intelligent agent is generated, following the implementation plan, and combining the student's knowledge point mastery status information and the knowledge point dependencies in the given knowledge graph to generate a preliminary learning path. S4. Employ a path verification agent to verify the preliminary learning path, checking whether it conforms to the pre- and post-dependencies of the given knowledge graph and whether it meets the requirements for the number of knowledge points to be learned in the implementation plan. Correct any errors found and output the final learning path plan.

[0021] This embodiment is based on the Junyi dataset from a simulation experiment. The Junyi dataset was collected from the Junyi Education Platform and is geared towards K-12 education scenarios. The Junyi dataset contains detailed student learning records, including exercise knowledge points, exercise content (including images), answer time, and answer accuracy. Furthermore, the relationships between knowledge points are clearly defined, with corresponding knowledge graphs. The dataset is rich in data, containing over 80,000 student records and approximately three million records. This experiment collected information from fifty students with more than 10 answer records.

[0022] Step 1: Use a cognitive diagnostic agent to calculate the student's knowledge point mastery score. The student's knowledge point mastery score refers to the score given to the student's mastery of each knowledge point within a given knowledge graph, with a score range of 0-1.

[0023] Step 2: Use a multimodal agent to generate a comprehensive cognitive diagnostic result. This cognitive diagnostic result refers to the score calculated by the multimodal agent, student answer records, knowledge graph, and question content and images, providing a final evaluation of the student's mastery of the knowledge points.

[0024] Step 3: Use the path planning agent to generate an implementation plan for the learning path planning. The implementation plan refers to the path planning agent planning the specific implementation methods of the learning path planning, including how to process cognitive diagnostic results, how to differentiate student learning abilities, and what conditions the learning path should meet, ultimately generating the plan.

[0025] Step 4: Use the path generation agent to generate a learning path. The key to generating the learning path is to follow the path planning agent's plan, combining cognitive diagnostic results and student answer records, and generating the learning path based on the knowledge points contained in the input knowledge graph.

[0026] Step 5: Use a path verification agent to verify the generated learning path. The key to verification is ensuring that the generated learning path strictly follows the pre- and post-positional relationships of knowledge points in the knowledge graph. Simultaneously, verify whether the path-generating agent correctly estimated the student's learning level and planned the number of knowledge points to be learned based on that level.

[0027] Reference Figure 2 The implementation steps of this embodiment are described in detail below: Step 1: Use a cognitive diagnostic agent to calculate the students' mastery of knowledge points.

[0028] This embodiment uses the KaNCD deep cognitive diagnostic model to model student answer records, question difficulty, and a given knowledge graph, aggregating multi-dimensional information, and finally outputting the student's knowledge mastery score and the predicted answer results. The input to the cognitive diagnostic agent is: student answer records, question difficulty, and given knowledge graph.

[0029] The student's answer record includes the question ID, answer result, and answer timestamp; the question difficulty is calculated based on the historical accuracy rate of the answers; the given knowledge graph contains the dependencies and hierarchical structure between knowledge points. All knowledge points involved in the questions in the student's answer record exist in the knowledge graph.

[0030] The data processing procedure for the cognitive diagnostic agent in this embodiment is as follows: The answer records are serialized to construct a student-question interaction matrix; the question difficulty is associated with knowledge points in the knowledge graph to form a multi-dimensional feature vector; the attention mechanism and graph neural network layer in the KaNCD model are used to aggregate student behavior, question difficulty, and knowledge structure information. Finally, the student's mastery score for each knowledge point (a continuous value from 0 to 1) and the predicted accuracy rate for questions not answered by the student are output. The scores output in this step will serve as one of the basic data for subsequent agent evaluation.

[0031] Step 2: Use a multimodal agent to generate comprehensive cognitive diagnostic results.

[0032] This embodiment guides a multimodal agent to perform comprehensive diagnosis by designing structured prompts.

[0033] Input information includes: knowledge point mastery scores output by the KaNCD model, students' answer records (including question text, options, images, and other multimedia content), the structured representation of the given knowledge graph, and information on the relationship between question difficulty and knowledge points.

[0034] The internal processing flow of the multimodal intelligent agent includes: parsing multimodal input, extracting and aligning features from text and image information; verifying the correspondence between questions and knowledge points using a knowledge graph; and performing logical reasoning and consistency checks by combining KaNCD scores with the actual content of the questions. The final comprehensive cognitive diagnostic result is output, including the mastery score for each knowledge point and a brief diagnostic explanation.

[0035] The prompt words for the multimodal agent in this embodiment are designed as follows: "You are an expert in student cognitive diagnosis. Now, please conduct a cognitive diagnosis based on the student's answer records and question information, scoring each of the entered knowledge points. You need to follow these rules:" 1. First, consider and provide the points to note when diagnosing students' cognitive abilities; 2. The score for the student's mastery of the knowledge points is a number from 0 to 1, with the higher the number, the higher the student's mastery. 3. I will provide a sample rating for reference. Please rate according to the pattern of the sample. 4. In the answer record, I will provide the student's answer record, which includes the question information, and some questions will have corresponding pictures; 5. If there are historical messages, I will provide them as an assistant. Please indicate whether you have received the historical messages and use them as a reference. 6. Please summarize your answer at the end, compiling the scores from past messages under the subheading 'Scores for Student Cognitive Diagnosis'. Step 3: Use the path planning agent to generate an implementation plan for the learning path planning.

[0036] In this embodiment, the path planning agent is responsible for formulating the top-level strategies and constraints for path planning, rather than generating specific path sequences. The path planning agent makes decisions based on the following information: the comprehensive cognitive diagnostic results generated in step 2; the student's historical answer records; the structured representation of the knowledge graph; and the overall learning time framework or number of days target.

[0037] Its core task is to generate a "plan," which needs to clarify: the basic principles and considerations for path planning; the ability grading standard based on students' overall average mastery scores (e.g., high, medium, and poor levels) and the corresponding recommended learning volume range; the priority strategy for knowledge points with different levels of mastery in path arrangement (e.g., prioritizing weak points); and the decision-making logic for the learning order when dealing with knowledge points in the knowledge graph that have no clear dependencies. This plan serves as a guiding document for subsequent path generation, ensuring the standardization and consistency of the planning process.

[0038] The prompt words for the path planning agent in this embodiment are designed as follows: "You are an expert in student learning path planning. Now, please design a plan for student learning paths, without generating specific paths, only the plan itself. I will provide the number of days for the learning path plan and the knowledge points involved. You need to follow these rules:" 1. First, consider and provide the points to note when planning student learning paths; 2. Students are divided into three levels: high, medium, and low. Students with an average score below 0.5 are classified as low, those between 0.5 and 0.55 as medium, and those above 0.55 as high. High-achieving students will learn approximately 35 knowledge points, medium-achieving students approximately 28, and low-achieving students approximately 20. The exact number may vary. 3. Student mastery level is rated on a scale of 0 to 1, with higher numbers indicating greater mastery. Please evaluate whether students have mastered the knowledge points. 4. When considering the order of learning knowledge points that have no sequential relationship, it is necessary to take into account the student's mastery score for each knowledge point; 5. Please summarize your answer at the end and provide a plan, with the subheading 'A Plan for Student Learning Path Planning'. Step 4: Use the path generation agent to generate a learning path.

[0039] In this embodiment, the path generation agent generates a daily or phased sequence of learning knowledge points based on the "plan" formulated by the path planning agent. The inputs received by the path generation agent include: the "plan" generated by the path planning agent; the comprehensive cognitive diagnostic results generated by the multimodal agent; the structured representation of the given knowledge graph; and the student's historical answer records.

[0040] The generation process in this embodiment strictly follows the plan: First, based on the comprehensive cognitive diagnosis results, determine the students' ability levels, and then determine the total number of knowledge points that this round of planning should cover based on the students' ability levels.

[0041] Secondly, based on the topological structure of the knowledge graph, a preliminary learning sequence that conforms to the dependency relationship is generated.

[0042] Then, based on this sequence, dynamic adjustments are made according to the mastery scores of the knowledge points. For example, knowledge points with lower scores are appropriately moved forward to strengthen learning.

[0043] Finally, output a formatted learning path, with each path item containing a sequence number, the name of the knowledge point, and its mastery score.

[0044] The prompt words for the path generation agent in this embodiment are designed as follows: "You are an expert in student learning path planning. Now, please plan a student learning path based on the given plan. I will provide the number of days the student will study, the knowledge points involved in the path planning and their prerequisites, the student's mastery of the knowledge points (represented by a number in the interval (0, 1), and the student's answer records to assist in plan generation. You need to follow these rules:" 1. Strictly follow the plan when planning the route; 2. Remove the "Unknown" knowledge points; 3. Strictly follow the input knowledge points and their sequential relationships in the planning process; 4. Students are divided into three levels: high, medium, and low. Students with an average score below 0.5 are classified as low-achieving, those between 0.5 and 0.55 as medium-achieving, and those above 0.55 as high-achieving. High-achieving students will learn approximately 35 knowledge points, medium-achieving students approximately 28, and low-achieving students approximately 20. The exact number may vary. 5. The student's mastery level score is a number from 0 to 1. The higher the number, the higher the student's mastery level. I will provide a large language model to score the student's mastery of knowledge points and a cognitive diagnostic model to score the student's mastery of knowledge points. Please combine the two scores to evaluate whether the student has mastered the knowledge points well. 6. The generated student learning path plan must be logically sound; 7. All knowledge points are represented using the original text of this knowledge point, in the format of 21sequence_and_series&Topic; 8. Simply plan the learning order of the knowledge points and attach a knowledge point mastery score after each knowledge point, in the format of 1.21 sequence_and_series&Topic (0.5), 2.40 circle-properties&Topic (0.4), without needing to design the details; 9. Finally, summarize the findings, with the subheading 'Learning Path Planning Generation'. Step 5: Use the path verification agent to verify the generated learning path. The path verification agent in this embodiment is a key component in ensuring the quality and executability of the generated path. The path verification agent automatically reviews the learning path output by the path generation agent. The inputs received by the path verification agent include: the learning path generated by the path generation agent; the comprehensive cognitive diagnostic results generated by the multimodal agent; the structured representation of the knowledge graph; and the student's historical answer records.

[0045] First, the path verification agent generates the necessary verification considerations. Then, logical verification is performed, including but not limited to structural verification: checking whether all prerequisite knowledge points for each knowledge point in the path have been learned beforehand to ensure compliance with the dependency constraints of the knowledge graph. Consistency verification: verifying whether the total length of the path (number of knowledge points) matches the plan requirements corresponding to the student's ability level. Logical verification: evaluating whether the order of knowledge points in the path is consistent with established strategies such as "weakness priority". The verification process is carried out step by step. If any violation of rules or logical inconsistency is found, the agent automatically locates the error, provides reasons for modification, and makes partial adjustments to the path to ensure compliance. Finally, a verified and revised final version of the learning path plan is output and ready for direct use.

[0046] The prompt words for the path verification agent in this embodiment are designed as follows: "You are an expert in the field of learning path planning validation. Now, please validate the generated learning path plan. You need to follow these rules:" 1. First, consider and provide the points to note when verifying the learning path plan; 2. Strictly follow the precautions to check the path step by step, and provide the verification process and results; 3. If there are errors in the route planning, please indicate the cause of the error and make corrections; 4. Summarize the verified learning paths and title it 'Verified Student Learning Path Planning'. Based on the above steps, this embodiment conducts a specific simulation experiment.

[0047] The cognitive diagnostic agent KaNCD captures features such as student answer records, question content, and question difficulty to output a score indicating the student's mastery of knowledge points and a prediction of the accuracy rate of the student's answer to the next question. Since the accuracy rate of the student's mastery score is difficult to quantify, the evaluation criterion for cognitive diagnosis accuracy is the accuracy rate of predicting the correctness of the student's answer to the next question. Experimental results show that the cognitive diagnosis accuracy rate reaches 68.8%. Subsequently, a multimodal scoring agent was introduced for comprehensive scoring to further improve performance.

[0048] Reference Figure 3 The figure shows the student knowledge point mastery scores generated by the multimodal intelligent agent based on the cognitive diagnostic model. The horizontal axis represents the student knowledge point mastery scores, and the vertical axis represents the number of students. As can be seen, the student knowledge point mastery scores in this embodiment are concentrated between 0.46 and 0.56. The mean and median in the figure are close to 0.5, indicating a reasonable assessment of student abilities. This approach effectively differentiates student learning abilities.

[0049] This embodiment uses a simulation experiment based on the Junyi dataset for verification. To illustrate the actual operating mechanism and intermediate results of each step of the system, the complete processing flow of a student with ID 17 is used as an example for step-by-step explanation. This student has 97 valid answer records, involving multiple knowledge points in the knowledge graph.

[0050] Step 1: The cognitive diagnostic agent calculates the knowledge point mastery score. Input: 97 historical answer records of student 17, question difficulty coefficients, and a knowledge graph (including knowledge point dependencies). Processing: The KaNCD model models the interaction sequence, aggregating student behavior, question difficulty, and knowledge structure information. Output: Obtain the student's mastery score (0~1) for each knowledge point.

[0051] Step 2: The multimodal agent generates a comprehensive cognitive diagnostic result. Input: Student knowledge point mastery scores from Step 1, student historical answer records (including question text and images), and a knowledge graph. Processing: The agent parses the question images and text, verifies the knowledge point connections using the knowledge graph, and performs logical review and consistency calibration on the KaNCD scores. Output: Generates the final comprehensive score and a brief diagnosis. For example, confirming good mastery of "understanding time" (score 0.780), and multiple correct answers to "decimals" questions demonstrating good comprehension, the score increases by 0.750. (See reference...) Figure 4 As shown in the figure, the comprehensive cognitive diagnostic results for some knowledge points are presented. The knowledge point "Understanding Time" scored 0.780, and "Properties of Triangles" scored 0.770, etc.

[0052] Step 3: Generate a path planning agent to implement the plan. Input: Comprehensive diagnostic results from Step 2, student's historical answer records, structured representation of the knowledge graph, and overall learning time frame or number of days target. Processing: Calculate the student's average mastery level as 0.609, and classify them as a high-ability student according to the rule (above 0.55 is "advanced"). Output: Generate a planning plan, clarifying: (1) The path must strictly follow the knowledge graph dependencies; (2) The student belongs to the "high-ability" level, and it is recommended to learn a total of about 35 knowledge points; (3) Among the same prerequisite knowledge points, prioritize learning those with lower scores; (4) When there are no explicit dependencies, arrange them in ascending order of scores.

[0053] Step 4: Path Generation - The agent generates an initial learning path. Input: The plan from Step 3, the comprehensive cognitive diagnosis results from Step 2, the structured representation of the knowledge graph, and the student's historical answer records. Processing: Select approximately 35 knowledge points based on ability levels; initially sort them based on graph dependencies; prioritize low-scoring knowledge points according to the "weak points first" principle. Output: Generate a daily learning sequence containing 35 knowledge points. (Refer to...) Figure 5As shown, this initial path is presented in list form, for example: Day 1: "Addition and Subtraction (0.502)"; Day 2: "Decimal (0.504)" etc.

[0054] Step 5: Path Verification. The agent verifies and corrects the path. Inputs: The learning path generated in Step 4, the comprehensive cognitive diagnostic results generated in Step 2, the structured representation of the knowledge graph, and the student's historical answer records. Processing: The agent checks step by step. A violation is found: Quadrilaterals and polygons on day 28 must be learned after the triangle properties on day 29, violating the prerequisite relationship. Output: The verification process indicates the error and its cause.

[0055] Reference Figure 6 As shown, for example: Modify path: Error reason: "23 Triangle property" must precede "45 Quadrilateral and Polygon" (strong dependency strength 0.2), but in the path 45 is at position 28 and 23 is at position 29.

[0056] Modification: Swap positions 28 and 29 to ensure the correct dependency order. Original location 28:45 Quadrilaterals and Polygons & Key Points → Changed to location 28:23 Properties of Triangles & Key Points Original position 29:23 Properties and knowledge points of triangles → changed to position 29:45 Quadrilaterals and polygons & knowledge points After the revision, the core stage knowledge points 12-29 will be updated in order (dependency compliance), while other parts remain unchanged.

[0057] Subsequently, referring to Figure 7 As shown, the agent automatically adjusts the order, placing the "triangle" first, and adds annotations to the final path to generate a verified final path.

[0058] Based on the same inventive concept, this invention also provides a multi-agent adaptive learning path planning system. Since the principle of solving the problem by these systems is similar to the aforementioned multi-agent adaptive learning path planning method, the implementation of this system can refer to the implementation of the aforementioned method, and the repeated parts will not be described again.

[0059] Reference Figure 8 As shown, it includes the following modules: Cognitive Diagnosis Module: This module uses a cognitive diagnostic agent to calculate the student's mastery score for each knowledge point in a given knowledge graph, and uses a multimodal agent to analyze the student's answer records, question content, and question images to generate multimodal analysis results. The multimodal analysis results are a comprehensive cognitive diagnostic result of the student's mastery of each knowledge point. The path planning module is used to combine the mastery score and the multimodal analysis results as information on the student's mastery of knowledge points. It combines the student's answer records and the given knowledge graph, and uses a path planning agent to generate an implementation plan for the learning path planning. The implementation plan includes the basic principles of path planning, student ability grading standards, the range of the number of learning knowledge points corresponding to different ability levels, and knowledge point priority strategies. Path generation module: Used to generate an agent using a path, following the implementation plan, and combining the student's knowledge point mastery status information and the knowledge point dependencies in the given knowledge graph to generate a preliminary learning path; Path verification module: Used by a path verification agent to verify the preliminary learning path, check whether it conforms to the pre- and post-dependencies of the given knowledge graph, whether it meets the requirements of the implementation plan regarding the number of learning knowledge points, correct errors when they are found, and output the final learning path plan.

[0060] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0061] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-agent-based adaptive learning path planning method, characterized in that, Includes the following steps: S1. A cognitive diagnostic agent is used to calculate the student's mastery score for each knowledge point in a given knowledge graph, and a multimodal agent is used to analyze the student's answer records, question content, and question images to generate multimodal analysis results; the multimodal analysis results are the comprehensive cognitive diagnostic results of the student's mastery of each knowledge point. S2. The mastery score and the multimodal analysis results are used together as the student's knowledge point mastery status information. Combined with the student's answer records and the given knowledge graph, a path planning agent is used to generate an implementation plan for the learning path planning. The implementation plan includes the basic principles of path planning, student ability grading standards, the range of the number of learning knowledge points corresponding to different ability levels, and knowledge point priority strategies. S3. Using path generation, an intelligent agent is generated, following the implementation plan, and combining the student's knowledge point mastery status information and the knowledge point dependencies in the given knowledge graph to generate a preliminary learning path. S4. Employ a path verification agent to verify the preliminary learning path, checking whether it conforms to the pre- and post-dependencies of the given knowledge graph and whether it meets the requirements for the number of knowledge points to be learned in the implementation plan. Correct any errors found and output the final learning path plan.

2. The method as described in claim 1, characterized in that, In step S1, the mastery score ranges from 0 to 1; where 0 indicates no mastery and 1 indicates complete mastery.

3. The method as described in claim 1, characterized in that, In step S1, the cognitive diagnostic agent is the deep cognitive diagnostic model KaNCD. Its inputs include the student's answer records, the difficulty of the questions, and the given knowledge graph, and it outputs the student's mastery score for each knowledge point.

4. The method as described in claim 1, characterized in that, In step S1, the multimodal agent adopts a multimodal neural network architecture that integrates a text encoder, an image encoder, and a question-answering behavior sequence encoder. It extracts features from the question text, the attached image, and the student's answer time series, and then fuses these three types of features through a cross-modal attention mechanism. Generate detailed supplementary information on the mastery status of each knowledge point.

5. The method as described in claim 1, characterized in that, In step S2, the student ability grading standard specifically includes: classifying students with an average mastery score of less than 0.5 for all knowledge points as poor ability, those with a score between 0.5 and 0.55 as medium ability, and those with a score higher than 0.55 as high ability.

6. The method as described in claim 1, characterized in that, In step S3, when generating the initial learning path, for multiple knowledge points with the same prerequisite knowledge points, the knowledge points with lower mastery scores are prioritized for learning.

7. The method as described in claim 1, characterized in that, In step S4, the verification process of the path verification agent includes structural verification, consistency verification, and logical verification. The structural verification is used to ensure that, in the initial learning path, for any knowledge point, all the prerequisite knowledge points defined in the given knowledge graph have been arranged to be learned before that knowledge point. The consistency verification is used to check whether the total number of knowledge points contained in the preliminary learning path conforms to the range of the number of learning knowledge points determined according to the student's ability level in the implementation plan. The logical verification is used to evaluate whether the order of knowledge points in the preliminary learning path follows the knowledge point priority strategy defined in the implementation plan.

8. The method as described in claim 1, characterized in that, The path planning agent, path generation agent, and path verification agent all provide explanations for the reasons behind the content they generate or verify when performing their respective tasks.

9. A multi-agent adaptive learning path planning system, characterized in that, Includes the following modules: Cognitive Diagnosis Module: This module uses a cognitive diagnostic agent to calculate the student's mastery score for each knowledge point in a given knowledge graph, and uses a multimodal agent to analyze the student's answer records, question content, and question images to generate multimodal analysis results. The multimodal analysis results are a comprehensive cognitive diagnostic result of the student's mastery of each knowledge point. The path planning module is used to combine the mastery score and the multimodal analysis results as information on the student's mastery of knowledge points. It combines the student's answer records and the given knowledge graph, and uses a path planning agent to generate an implementation plan for the learning path planning. The implementation plan includes the basic principles of path planning, student ability grading standards, the range of the number of learning knowledge points corresponding to different ability levels, and knowledge point priority strategies. Path generation module: Used to generate an agent using a path, following the implementation plan, and combining the student's knowledge point mastery status information and the knowledge point dependencies in the given knowledge graph to generate a preliminary learning path; Path verification module: Used by a path verification agent to verify the preliminary learning path, check whether it conforms to the pre- and post-dependencies of the given knowledge graph, whether it meets the requirements of the implementation plan regarding the number of learning knowledge points, correct errors when they are found, and output the final learning path plan.

Images