Multi-agent cooperative reasoning system for intelligent teaching intervention

Abstract

The invention discloses a multi-agent collaborative reasoning system for intelligent teaching intervention, and relates to the technical field of artificial intelligence and education, and the system comprises a data collection and convergence module which collects data of learning behaviors, emotional states, academic scores and knowledge point mastering conditions of students by means of a classroom behavior analysis system, a camera, a microphone and a learning management system; according to the intelligent teaching intervention-oriented multi-agent collaborative reasoning system, the precision and individuation of teaching intervention are realized by constructing the intelligent teaching intervention-oriented multi-agent collaborative reasoning system, the system collects multi-dimensional data of students through the data acquisition and convergence module, and through the processing of the data conversion and cleaning module, the accuracy and availability of the data are ensured; the teaching strategy planning module generates a dynamic decision model by means of a large language model, a customized intervention strategy can be generated according to the learning condition and problem root of students, and the multi-agent collaborative reasoning module further improves the intelligent level of the system.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and educational technology, and specifically to a multi-agent collaborative reasoning system for intelligent teaching intervention. Background Art

[0002] With the rapid development of information technology, the field of education is gradually moving towards intelligence and personalization. In recent years, intelligent teaching systems have become a research hotspot in the field of education. By using artificial intelligence and big data analysis technologies, it aims to improve teaching efficiency and quality and achieve optimal allocation of educational resources. Intelligent teaching systems can collect students' learning data, analyze learning behaviors, and explore learning needs, thereby providing teachers with targeted teaching strategies and intervention measures. This technical background provides a solid foundation for the research and development of multi-agent collaborative reasoning systems for intelligent teaching intervention.

[0003] However, traditional intelligent teaching systems often have limitations when processing student data. On the one hand, they mostly rely on a single data source and analysis model, which makes it difficult to fully and accurately reflect students' learning status and needs. On the other hand, traditional systems lack sufficient flexibility and personalization when formulating teaching strategies, and often adopt a "one-size-fits-all" approach, ignoring the differences between students. In addition, traditional systems have limited capabilities in sentiment analysis and social interaction, making it difficult to capture students' emotional changes and social needs, thus affecting the improvement of teaching effectiveness. These shortcomings limit the effectiveness of traditional intelligent teaching systems in practical applications.

[0004] Therefore, developing a multi-agent collaborative reasoning system for intelligent teaching intervention will effectively promote the process of intelligent education and improve teaching quality and efficiency. Summary of the Invention

[0005] The purpose of this invention is to make up for the shortcomings of the existing technology and provide a multi-agent collaborative reasoning system for intelligent teaching intervention. The system integrates multiple modules of data collection, transformation and cleaning, strategy planning, collaborative reasoning, intervention execution and feedback, and realizes a comprehensive analysis of students' learning behavior, emotional state, academic performance and knowledge point mastery. By adopting a multi-agent collaborative reasoning mechanism, the system can accurately identify students' learning needs and root causes of problems, and generate personalized teaching intervention strategies. In addition, the system also has real-time feedback and adjustment functions, which can automatically optimize subsequent teaching strategies according to teaching results.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a multi-agent collaborative reasoning system for intelligent teaching intervention, the system comprising:

[0007] Data collection and aggregation module: Using classroom behavior analysis systems, cameras, microphones, and learning management systems, data on students' learning behaviors, emotional states, academic performance, and knowledge mastery is collected, cleaned, and stored in a database based on student identity.

[0008] Data conversion and cleaning module: For structured data, standardize the data field format; for unstructured text data, use natural language processing technology to perform structured conversion, and check and correct errors and missing values in the data;

[0009] Teaching Strategy Planning Module: This module uses a large language model to integrate theoretical models, quantitative models, and educational rubrics to generate a dynamic decision-making model with both explanatory and diagnostic components. Based on the collected data, it assesses students' learning status, uncovers the root causes of learning problems, and generates intervention strategies encompassing teaching content, methods, and tutoring plans. Furthermore, it transmits relevant information about the decision-making model, its analysis results, and the direction of teaching interventions to the multi-agent collaborative reasoning module.

[0010] Multi-agent collaborative reasoning module: It is composed of multiple agents, each responsible for data analysis in different dimensions. Through collaborative reasoning mechanism, they conduct joint reasoning and generate teaching intervention strategies.

[0011] Teaching intervention execution and feedback module: Converts teaching intervention suggestions into practical actions, collects feedback data from students and teachers through online questionnaires and learning platform feedback functions, and automatically adjusts subsequent teaching strategies and intervention measures based on the feedback data.

[0012] Furthermore, the specific content of the data on students' learning behavior, homework performance, and emotional state in the data collection and aggregation module is as follows:

[0013] The learning behavior data include the number of times students actively speak, the duration and contribution of group discussions, and the accuracy of their answers;

[0014] The academic performance data, including the completion time of each assignment, the answer status, and the differences in students' mastery of different question types;

[0015] The emotional state data: facial expression changes, voice and tone characteristics;

[0016] The knowledge point mastery data includes test scores and exercise completion status for different knowledge points.

[0017] Furthermore, the specific steps of using natural language processing technology to perform structured conversion in the data conversion and cleaning module are as follows:

[0018] (6) Collect unstructured text data related to student learning from data sources and store them centrally;

[0019] (7) Remove noise information from the text, split the continuous text into individual words or phrases, mark the part of speech for each word, and remove stop words in the text;

[0020] (8) Extract representative keywords from the text, divide the text into different topics through algorithms, and determine the emotional tendency expressed in the text;

[0021] (9) Convert text features into tables, vectors or graph structures and realize numerical processing through encoding;

[0022] Check data format and integrity, and optimize conversion quality based on evaluation indicators.

[0023] Furthermore, the explanation part of the decision-making model in the teaching strategy planning module explains the individual's current situation for each dimension of learning attitude, learning level, and thinking quality, and analyzes the individual's development process and overall situation.

[0024] Furthermore, the diagnostic part of the decision-making model in the teaching strategy planning module analyzes the relationship between variables through statistical inference based on the influencing factors in the theoretical model, finds out the variables that have the greatest impact and the closest relationship on the target attribute data, analyzes the reasons why students perform poorly on the target attribute data, and outputs specific action suggestions in combination with the educational scale.

[0025] Furthermore, the agents included in the multi-agent collaborative reasoning module are: cognitive agent, emotional agent, social agent, behavior prediction agent, learning resource recommendation agent, evaluation agent and collaborative agent, and each agent is implemented through prompt engineering and large model fine-tuning.

[0026] Furthermore, the specific functions and implementation methods of each agent in the multi-agent collaborative reasoning module are as follows:

[0027] The cognitive agent:

[0028] Function: Focuses on analyzing students' knowledge mastery. Based on data on students' knowledge point tests and homework completion, combined with subject knowledge graphs, it builds a personal knowledge graph for students, accurately infers learning progress, and predicts future learning paths.

[0029] Prompt project implementation: Design specific prompt information for the large model to guide the model to analyze students' knowledge mastery data;

[0030] Fine-tuning the large model: Utilizing student learning data and subject knowledge graph data, the large model is fine-tuned to enable the model to identify students' mastery levels of different knowledge points, as well as the relationships between knowledge points, and to predict future learning difficulties based on students' current knowledge status.

[0031] The emotional agent:

[0032] Function: With the help of sentiment analysis technology, it mines students' emotional state data, promptly detects students' emotional changes, and provides suggestions for adjusting teaching rhythm and selecting incentive methods;

[0033] Prompt project implementation: Design prompt information to allow the big model to analyze the emotional state of students from multiple sources of data such as facial expressions, voice intonation, and learning platform behavior;

[0034] Fine-tuning the large model: Using student behavior data with emotion annotations, fine-tuning the large model enables the model to identify behavioral characteristics corresponding to different emotional states and provide appropriate teaching adjustment suggestions based on students' emotional changes;

[0035] The social agent

[0036] Function: Analyze students' interactive data in group cooperation and classroom communication, evaluate students' communication and teamwork skills, and help teachers optimize teaching organization;

[0037] Prompt project implementation: Prompt information guides the large model to evaluate students' social interaction data;

[0038] Fine-tuning the large model: Using student social interaction data and corresponding evaluation results, fine-tuning the large model enables the model to assess students' social skills and provide teachers with effective optimization suggestions for teaching organization;

[0039] The behavior prediction agent:

[0040] Function: Uses algorithms to analyze students' historical learning behavior and current learning status data, predicts future learning difficulties or performance fluctuations, and issues early warnings;

[0041] Prompt project implementation: Design prompt information to allow the big model to make predictions based on students' historical grades, study time allocation, and learning resource utilization data;

[0042] Fine-tuning the large model: Using historical student learning data and actual learning difficulties or performance fluctuations to fine-tune the large model, the model can accurately predict and identify potential learning problems in advance.

[0043] The learning resource recommendation agent:

[0044] Function: Filter and recommend appropriate learning materials from the learning resource library based on students' learning progress, interests, preferences, and knowledge weaknesses;

[0045] Prompt project implementation: Design prompt information to guide the large model to recommend learning resources based on students' learning situation;

[0046] Fine-tuning the large model: Using students' feedback on their use of learning resources to fine-tune the large model, the model can recommend learning materials that meet students' needs, thereby improving the accuracy and pertinence of learning resource recommendations;

[0047] The evaluation agent:

[0048] Function: Using a multimodal assessment method, students' learning works are comprehensively evaluated from multiple dimensions such as content quality, innovation level, and skill application;

[0049] Prompt project implementation: Design prompt information to allow the big model to conduct multi-dimensional evaluation of students' learning works;

[0050] Fine-tuning the large model: Use student learning works and corresponding professional evaluation results to fine-tune the large model so that the model can accurately perform multimodal evaluation and provide objective and comprehensive evaluation opinions.

[0051] Furthermore, the steps for the collaborative agents in the multi-agent collaborative reasoning module to jointly reason with each other through a collaborative reasoning mechanism are as follows: after receiving the task assigned by the teaching strategy dynamic adjustment module, each agent independently analyzes the student data based on its own functions and responsibilities using prompt engineering and the fine-tuned capabilities of the large model. Each agent interacts and shares the analysis results through an information sharing platform. The collaborative agent comprehensively considers the suggestions of all agents and uses a joint reasoning algorithm to generate comprehensive teaching intervention suggestions. A supervision and feedback mechanism is set up to evaluate and verify the reasoning results of each agent.

[0052] Furthermore, the multi-agent collaborative reasoning module uses a joint reasoning algorithm to generate a teaching intervention suggestion I, and the calculation formula is: Among them, f is the total number of agents, and the output of each agent is O=[o1,o2,…,o f ] is obtained by the corresponding agents according to their respective analysis models, and the collaborative weight vector is φ=[φ1,φ2,…,φ f ] Dynamically adjust according to teaching objectives and the importance of each agent in different scenarios.

[0053] Compared with existing technologies, this multi-agent collaborative reasoning system for intelligent teaching intervention has the following beneficial effects:

[0054] 1. The present invention realizes the precision and personalization of teaching intervention by constructing a multi-agent collaborative reasoning system for intelligent teaching intervention. The system collects students' multi-dimensional data through the data acquisition and aggregation module, and ensures the accuracy and availability of the data through processing by the data conversion and cleaning module. The teaching strategy planning module generates a dynamic decision-making model with the help of a large language model, which can generate customized intervention strategies based on students' learning conditions and problem roots. The multi-agent collaborative reasoning module further improves the intelligence level of the system. Each agent is responsible for data analysis of different dimensions, and conducts joint reasoning through a collaborative reasoning mechanism, ultimately generating comprehensive and accurate teaching intervention suggestions. This multi-agent collaborative approach not only improves the pertinence and effectiveness of teaching intervention, but also promotes the optimal allocation of teaching resources, realizing the personalized and intelligent development of education.

[0055] 2. The present invention significantly improves the intelligent analysis capability and reasoning accuracy of the system by introducing prompt engineering and large model fine-tuning technology into the multi-agent collaborative reasoning module. By designing specific prompt information for the large model and using student learning data for fine-tuning, each agent can more accurately identify and analyze students' learning status, emotional changes, and behavioral characteristics of social interactions, enabling the system to have a deeper understanding of students' learning needs and problems, thereby providing teachers with more accurate and comprehensive teaching intervention suggestions. At the same time, the system's supervision and feedback mechanism can also evaluate and verify the reasoning results of each agent, ensuring the scientific nature and reliability of the teaching intervention suggestions, which not only promotes the development of intelligent teaching systems, but also provides new ideas and methods for the digital transformation of the education field.

[0056] Other advantages, objectives and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or can be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] To clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention, and those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0058] Figure 1 A flowchart of a multi-agent collaborative reasoning system for intelligent teaching intervention;

[0059] Figure 2 Schematic diagram of data collection and cleaning;

[0060] Figure 3 Generate and dynamically execute schematic diagrams for instructional intervention strategies. DETAILED DESCRIPTION

[0061] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0062] Example 1:

[0063] An implementation example of a multi-agent collaborative reasoning system for intelligent teaching intervention.

[0064] In a mathematics course teaching scenario in a middle school, the multi-agent collaborative reasoning system for intelligent teaching intervention plays a full role, providing strong support for improving teaching quality and student learning outcomes.

[0065] The data collection and aggregation module collects data through multiple devices and systems. The classroom behavior analysis system records the number of times students actively speak, the length of time they participate in group discussions and their contribution, as well as the accuracy of their answers to questions. The camera captures changes in students' facial expressions, and the microphone collects voice and intonation features for emotional state analysis. The learning management system obtains data on the completion time of each student's homework, their answers, the differences in their mastery of different question types, as well as their test scores and exercise completion status for different knowledge points. The collected data is cleaned and stored in the database according to student identity. For example, in this week's math class, Student A actively spoke three times, participated in group discussions for 20 minutes, and answered questions with an accuracy rate of 70%; it took 60 minutes to complete the homework, and there were many errors in the geometry questions in the homework; he scored 80 points in a function knowledge point test, and the completion rate of related exercises was 90%.

[0066] The data conversion and cleaning module processes the data. For structured data, such as grades and homework completion time, the data field format is standardized. For unstructured text data, such as students' discussions and descriptions of problem-solving ideas on the learning platform, natural language processing technology is used. First, relevant text data is collected and stored centrally, and then noise information is removed. The text is split into words or phrases and marked with parts of speech, stop words are removed, keywords are extracted from the text, topics are divided and emotional tendencies are judged, and text features are converted into tables, vectors or graph structures and processed numerically. Finally, the data format and integrity are checked to optimize the conversion quality. For example, when discussing the application of the Pythagorean theorem, a student mentioned, "I think this question is very difficult. It took me a long time to solve it." After processing, it can be analyzed that the student has a fear of this knowledge point and encounters certain difficulties in learning.

[0067] The teaching strategy planning module uses a large language model to integrate theoretical models, quantitative models and educational rubrics to generate a dynamic decision-making model. For example, student B's recent math scores have fluctuated greatly. Through analysis, it is found that his knowledge of functions is not solid. The explanatory part of the decision model analyzes student B's current difficulties in learning functions from the dimensions of learning attitude, learning level, and thinking quality. He had a problem of insufficient understanding when learning functions before, and his overall function learning level is in the lower middle level in the class. The diagnostic part finds out the variables that have a great impact on grades based on the influencing factors of the theoretical model, such as unclear understanding of function concepts and poor mastery of problem-solving methods. Combined with educational rubrics, action suggestions are given, such as strengthening the explanation of function concepts and adding targeted exercises. At the same time, the relevant information of the decision model is transmitted to the multi-agent collaborative reasoning module.

[0068] The multi-agent collaborative reasoning module is composed of multiple agents. The cognitive agent builds a personal knowledge graph for students based on the students' function knowledge point test and homework completion data, combined with the mathematics subject knowledge graph. For example, student C has a good grasp of the function monotonicity knowledge point, but is weak in the comprehensive application of function parity and monotonicity. The cognitive agent infers his learning progress and predicts that he may encounter difficulties in subsequent composite function learning. By designing prompt information for the large model, it guides him to analyze knowledge points and master data, and uses student learning data and subject knowledge graph data to fine-tune the large model.

[0069] The emotional agent uses sentiment analysis technology to mine data. If it finds that students have frowning facial expressions, low voice tone, and fewer questions on the learning platform during function learning, it determines that they may have negative emotions. By designing prompt information, the big model analyzes multi-source data and fine-tunes the big model with behavioral data containing emotion annotations, providing suggestions for adjusting the teaching rhythm and choosing incentive methods, such as appropriately slowing down the teaching progress and giving more encouragement.

[0070] The social intelligence agent analyzes the interactive data of students when they work in groups to learn functions, and evaluates their communication and teamwork skills. For example, when the group was discussing ideas for solving function application problems, Student D was not very involved and spoke less. The social intelligence agent assessed that his communication skills needed to be improved, and provided suggestions for teachers to optimize the teaching organization form, such as adjusting the group grouping.

[0071] The behavior prediction agent uses algorithms to analyze students' historical learning behaviors and current status data. If student E's recent homework completion time is extended and the number of wrong questions increases, it is predicted that his performance in the upcoming comprehensive function test may decline. The agent issues an early warning and designs prompt information to allow the large model to make predictions based on historical performance data. The large model is then fine-tuned using historical learning data and actual performance fluctuations.

[0072] The learning resource recommendation agent recommends materials based on students' learning progress, interest preferences, and knowledge weaknesses. For example, if student F is weak in function graph transformation and is interested in animation demonstration learning methods, the agent will select relevant animation tutorials and exercises from the learning resource library and recommend them to the student. It will guide the recommendation of the large model by designing prompt information and fine-tune the large model with feedback data from students' use of learning resources.

[0073] The evaluation agent adopts a multimodal evaluation method to comprehensively evaluate students' function learning works, such as problem-solving process and learning summary, from the dimensions of content quality, degree of innovation, and skill application. For example, the student's function problem-solving steps are evaluated to see whether their logic is rigorous, whether there are innovative solutions, and whether the function knowledge is applied accurately. By designing prompt information to allow the big model to evaluate, the big model is fine-tuned using students' learning works and professional evaluation results.

[0074] After receiving the task, each collaborative agent independently analyzes the data and interacts and shares the results through the information sharing platform. The collaborative agent comprehensively considers the suggestions of all agents and uses the joint reasoning algorithm to generate the teaching intervention suggestion I. The calculation formula is: Among them, the output of each agent is O = [o1, o2, ..., o f ] is obtained by the corresponding agents according to their respective analysis models, and the collaborative weight vector is Dynamic adjustments are made based on teaching objectives and the importance of each intelligent agent in different scenarios. For example, based on the overall function learning situation of the class, the collaborative weights of cognitive agents and emotional agents in different aspects are determined according to the analysis results of each intelligent agent, and teaching intervention suggestions are calculated, including adjusting the order of teaching content, adding interactive links, and recommending specific learning resources. A supervision and feedback mechanism is set up to evaluate and verify the reasoning results of each intelligent agent to ensure the effectiveness of the intervention suggestions.

[0075] The teaching intervention execution and feedback module converts teaching intervention suggestions into practical actions. Teachers adjust teaching content and methods based on the suggestions, such as adding more examples to function teaching and organizing group competitions; recommend learning resources to students, urge students to learn, and collect student and teacher feedback data through online questionnaires and learning platform feedback functions. If students feedback that the learning resources are too difficult or teachers feedback that the teaching activity schedule is unreasonable, the system will automatically adjust subsequent teaching strategies and intervention measures based on the feedback data, such as replacing learning resources and optimizing the teaching activity schedule, to continuously improve teaching effectiveness.

[0076] To summarize, in a certain middle school mathematics course teaching scenario, the various modules of the multi-agent collaborative reasoning system for intelligent teaching intervention work closely together. The data acquisition and aggregation module collects multi-source data, the data conversion and cleaning module processes the data, and the teaching strategy planning module generates decision models and intervention strategies. The various agents in the multi-agent collaborative reasoning module analyze from different dimensions and collaboratively generate comprehensive intervention suggestions. Finally, the teaching intervention execution and feedback module implements intervention measures and collects feedback, and adjusts the teaching strategy accordingly. This system achieves accurate analysis and effective intervention of students' learning status, which helps to improve teaching quality and student learning outcomes.

[0077] Example 2:

[0078] An implementation example of a multi-agent collaborative reasoning system for intelligent teaching intervention in primary school Chinese writing teaching.

[0079] In the Chinese writing teaching scenario of a primary school, this intelligent teaching intervention system plays an important role.

[0080] The data collection and aggregation module collects data through various channels. The classroom behavior analysis system records the number of times students actively share their writing ideas in writing classes, the length of time they participate in writing group discussions, and their contribution; the camera captures students' facial expressions when they are conceiving and writing, and the microphone collects their voice and intonation during discussions to analyze their emotional state; the learning management system obtains the completion time and score of each student's composition, the performance differences in different composition types (such as narrative and expository essays), and the completion status and test scores of exercises on various writing knowledge points (such as the use of rhetorical devices and paragraph structure arrangement). For example, Student A actively shared ideas twice in the writing class this week and participated in group discussions for 15 minutes; it took him 50 minutes to complete a narrative composition and scored 75 points, of which the score for the use of rhetorical devices was relatively low; the score in the rhetorical device knowledge point test was 60 points, and the completion rate of related exercises was 80%.

[0081] The data conversion and cleaning module processes the collected data. For structured data, such as grades and completion time, the data field format is standardized. For unstructured text data, such as students' comments in their essays and the content of drafts during the writing process, natural language processing technology is used to first collect and store them centrally, then remove redundant and erroneous information in the text, split the text into words or phrases and mark the parts of speech, remove stop words, then extract keywords, divide the topics and judge the emotional tendencies, convert text features into appropriate structures and perform numerical processing, finally check the format and integrity of the data, and optimize the conversion quality. For example, a student mentioned in the essay comments that "it was so difficult to write, I didn't know how to describe this scene." After processing, it can be analyzed that the student had difficulties in writing descriptions and had a fear of difficulties.

[0082] The teaching strategy planning module uses a large language model to integrate theoretical models, quantitative models and educational rubrics to generate a dynamic decision-making model. Taking Student B as an example, his recent composition scores have fluctuated greatly and he has performed poorly in describing characters. The explanatory part of the decision-making model analyzes Student B's current insufficient ability to describe characters from the dimensions of learning attitude, learning level, and thinking quality. He previously had a problem of single expression when describing the appearance and personality of characters, and his overall character description level was average in the class. The diagnostic part identified the variables that had a great impact on composition scores based on the influencing factors of the theoretical model, such as insufficient mastery of descriptive techniques and insufficiently detailed observation of characters. Combined with the educational rubric, action suggestions were given, such as adding special exercises on character description and guiding students to observe the details of characters in life. At the same time, the relevant information of the decision-making model was transmitted to the multi-agent collaborative reasoning module.

[0083] The various agents in the multi-agent collaborative reasoning module began to play a role. The cognitive agent built a personal knowledge graph for the students based on the students' writing knowledge point test and composition completion data, combined with the Chinese language knowledge graph. For example, Student C has a good grasp of the knowledge points on the use of rhetorical techniques, but is weak in character description and plot setting. The cognitive agent infers his learning progress and predicts that he may encounter difficulties in writing complex narrative essays in the future. By designing prompt information for the large model, it guides him to analyze the writing knowledge points and master the data, and uses the students' learning data and subject knowledge graph data to fine-tune the large model.

[0084] The emotional agent uses sentiment analysis technology to mine data. If it finds that students have anxious facial expressions, urgent voice tones, and frequently ask writing-related questions on the learning platform during writing classes, it will judge that they may be under great pressure and anxiety. By designing prompt information, the large model will analyze multi-source data and fine-tune the large model with behavioral data containing emotion annotations to provide suggestions for adjusting the teaching rhythm and choosing incentive methods, such as arranging a relaxing writing warm-up activity before the writing class begins and giving timely praise to students for their progress.

[0085] The social intelligence agent analyzes students' interactive data in writing group collaborations and assesses their communication and teamwork skills. For example, when the group was discussing essay ideas, student Ding rarely expressed his opinions and simply agreed with others when he spoke. The social intelligence agent assessed that his communication skills needed to be improved and provided teachers with suggestions for optimizing teaching organization forms, such as conducting group writing competitions to stimulate students' enthusiasm for participating in discussions.

[0086] The behavior prediction agent uses algorithms to analyze students' historical learning behaviors and current status data. If student E's recent essays are taking longer and longer to complete and there are more typos and grammatical errors in his essays, it is predicted that he may not perform well in the next essay test. The agent will issue a warning in advance and design prompt information to allow the large model to make predictions based on historical performance data. The large model can then be fine-tuned using historical learning data and actual performance fluctuations.

[0087] The learning resource recommendation agent recommends materials based on students' learning progress, interests, preferences, and knowledge weaknesses. For example, if a student is weak in describing scenery and likes to read storybooks with illustrations, the agent will select relevant excellent essays describing scenery and writing guidance picture books with illustrations from the learning resource library and recommend them to the student. It will guide the large model's recommendations by designing prompt information and fine-tune the large model with feedback data from students' use of learning resources.

[0088] The evaluation agent adopts a multimodal evaluation method to comprehensively evaluate students' compositions from the dimensions of content richness, language expression, and structural rationality. For example, it evaluates whether the story content of students' compositions is vivid and interesting, whether the sentences are fluent, and whether the paragraph structure is clear and reasonable. By designing prompt information to allow the large model to evaluate, the large model is fine-tuned using student compositions and professional evaluation results.

[0089] After receiving the task, each collaborative agent independently analyzes the data and interacts and shares the results through the information sharing platform. The collaborative agent comprehensively considers the suggestions of all agents and uses the joint reasoning algorithm to generate the teaching intervention suggestion I. The calculation formula is: Among them, the output of each agent is O = [o1, o2, ..., o f ] is obtained by the corresponding agents according to their respective analysis models, and the collaborative weight vector is Dynamic adjustments are made based on teaching objectives and the importance of each intelligent agent in different scenarios. For example, based on the analysis results of each intelligent agent in the overall writing situation of the class, the collaborative weights of each intelligent agent in different aspects are determined, and teaching intervention suggestions are calculated, including adjusting the order of writing teaching content, increasing writing practice activities, and recommending personalized learning resources. A supervision and feedback mechanism is set up to evaluate and verify the reasoning results of each intelligent agent to ensure the effectiveness of the intervention suggestions.

[0090] The teaching intervention execution and feedback module converts teaching intervention suggestions into practical actions. Teachers adjust teaching content and methods based on the suggestions, such as adding more explanations of descriptive skills in writing teaching and organizing writing practice activities; recommend learning resources to students, encourage students to read and learn, and collect feedback data from students and teachers through online questionnaires and learning platform feedback functions. If students feedback that the learning resources do not match their own level, and teachers feedback that the teaching activities are too difficult, the system will automatically adjust subsequent teaching strategies and intervention measures based on the feedback data, such as replacing learning resources, reducing the difficulty of teaching activities, and continuously improving the effectiveness of writing teaching.

[0091] To sum up, in the primary school Chinese writing teaching scenario, the multi-agent collaborative reasoning system for intelligent teaching intervention operates efficiently. The data collection and aggregation module comprehensively collects writing-related data, the data conversion and cleaning module organizes and optimizes the data, and the teaching strategy planning module generates targeted decision-making models and intervention strategies. The multi-agent collaborative reasoning module analyzes from multiple dimensions of cognition and emotion, and collaboratively gives comprehensive intervention suggestions. The teaching intervention execution and feedback module is responsible for implementation and collection of feedback, and then adjusts the teaching strategy. The system accurately locates students' writing problems, provides effective teaching intervention, and effectively improves the quality of writing teaching and students' writing level, bringing new vitality and effectiveness to Chinese teaching.

[0092] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A multi-agent collaborative reasoning system for intelligent teaching intervention, characterized by: The system includes: Data collection and aggregation module: Using classroom behavior analysis systems, cameras, microphones, and learning management systems, data on students' learning behaviors, emotional states, academic performance, and knowledge mastery is collected, cleaned, and stored in a database based on student identity. Data conversion and cleaning module: For structured data, standardize the data field format; for unstructured text data, use natural language processing technology to perform structured conversion, and check and correct errors and missing values in the data; Teaching Strategy Planning Module: This module uses a large language model to integrate theoretical models, quantitative models, and educational rubrics to generate a dynamic decision-making model with both explanatory and diagnostic components. Based on the collected data, it assesses students' learning status, uncovers the root causes of learning problems, and generates intervention strategies encompassing teaching content, methods, and tutoring plans. Furthermore, it transmits relevant information about the decision-making model, its analysis results, and the direction of teaching interventions to the multi-agent collaborative reasoning module. Multi-agent collaborative reasoning module: It is composed of multiple agents, each responsible for data analysis in different dimensions. Through collaborative reasoning mechanism, they conduct joint reasoning and generate teaching intervention strategies. Teaching intervention execution and feedback module: Converts teaching intervention suggestions into practical actions, collects feedback data from students and teachers through online questionnaires and learning platform feedback functions, and automatically adjusts subsequent teaching strategies and intervention measures based on the feedback data.

2. The multi-agent collaborative reasoning system for intelligent teaching intervention according to claim 1 is characterized in that: The specific content of the data on students' learning behavior, homework performance, and emotional state in the data collection and aggregation module is as follows: The learning behavior data include the number of times students actively speak, the duration and contribution of group discussions, and the accuracy of their answers; The academic performance data, including the completion time of each assignment, the answer status, and the differences in students' mastery of different question types; The emotional state data: facial expression changes, voice and tone characteristics; The knowledge point mastery data includes test scores and exercise completion status for different knowledge points.

3. The multi-agent collaborative reasoning system for intelligent teaching intervention according to claim 1 is characterized in that: The specific steps of using natural language processing technology to perform structured conversion in the data conversion and cleaning module are as follows: (1) Collect unstructured text data related to student learning from data sources and store them centrally; (2) Remove noise information from the text, split the continuous text into individual words or phrases, mark the part of speech for each word, and remove stop words in the text; (3) Extract representative keywords from the text, divide the text into different topics through algorithms, and determine the emotional tendency expressed in the text; (4) Convert text features into tables, vectors, or graph structures and implement numerical processing through encoding; (5) Check the data format and integrity, and optimize the conversion quality based on the evaluation indicators.

4. The multi-agent collaborative reasoning system for intelligent teaching intervention according to claim 1 is characterized in that: The explanation part of the decision-making model in the teaching strategy planning module explains the individual's current situation for each dimension of learning attitude, learning level, and thinking quality, and analyzes the individual's development process and overall situation.

5. The multi-agent collaborative reasoning system for intelligent teaching intervention according to claim 1 is characterized in that: The diagnostic part of the decision-making model in the teaching strategy planning module analyzes the relationship between variables through statistical inference based on the influencing factors in the theoretical model, finds the variables that have the greatest impact and the closest relationship with the target attribute data, analyzes the reasons why students perform poorly on the target attribute data, and outputs specific action suggestions in combination with the educational scale.

6. The multi-agent collaborative reasoning system for intelligent teaching intervention according to claim 1 is characterized in that: The agents included in the multi-agent collaborative reasoning module are: cognitive agent, emotional agent, social agent, behavior prediction agent, learning resource recommendation agent, evaluation agent and collaborative agent. Each agent is implemented through prompt engineering and large model fine-tuning.

7. The multi-agent collaborative reasoning system for intelligent teaching intervention according to claim 6 is characterized in that: The specific functions and implementation methods of each agent in the multi-agent collaborative reasoning module are as follows: The cognitive agent: Function: Focuses on analyzing students' knowledge mastery. Based on data on students' knowledge point tests and homework completion, combined with subject knowledge graphs, it builds a personal knowledge graph for students, accurately infers learning progress, and predicts future learning paths. Prompt project implementation: Design specific prompt information for the large model to guide the model to analyze students' knowledge mastery data; Fine-tuning the large model: Utilizing student learning data and subject knowledge graph data, the large model is fine-tuned to enable the model to identify students' mastery levels of different knowledge points, as well as the relationships between knowledge points, and to predict future learning difficulties based on students' current knowledge status. The emotional agent: Function: With the help of sentiment analysis technology, it mines students' emotional state data, promptly identifies students' emotional changes, and provides suggestions for adjusting teaching rhythm and selecting motivational methods; Prompt project implementation: Design prompt information to allow the big model to analyze the emotional state of students from multiple sources of data such as facial expressions, voice intonation, and learning platform behavior; Fine-tuning the large model: Using student behavior data with emotion annotations, fine-tuning the large model enables the model to identify behavioral characteristics corresponding to different emotional states and provide appropriate teaching adjustment suggestions based on students' emotional changes; The social agent Function: Analyze students' interactive data in group cooperation and classroom communication, evaluate students' communication and teamwork skills, and help teachers optimize teaching organization; Prompt project implementation: Prompt information guides the large model to evaluate students' social interaction data; Fine-tuning the large model: Using student social interaction data and corresponding evaluation results, fine-tuning the large model enables the model to assess students' social skills and provide teachers with effective optimization suggestions for teaching organization; The behavior prediction agent: Function: Uses algorithms to analyze students' historical learning behavior and current learning status data, predicts future learning difficulties or performance fluctuations, and issues early warnings; Prompt project implementation: Design prompt information to allow the big model to make predictions based on students' historical grades, study time allocation, and learning resource utilization data; Fine-tuning the large model: Using historical student learning data and actual learning difficulties or performance fluctuations to fine-tune the large model, the model can accurately predict and identify potential learning problems in advance. The learning resource recommendation agent: Function: Filter and recommend appropriate learning materials from the learning resource library based on students' learning progress, interests, preferences, and knowledge weaknesses; Prompt project implementation: Design prompt information to guide the large model to recommend learning resources based on students' learning situation; Fine-tuning the large model: Using students' feedback on their use of learning resources to fine-tune the large model, the model can recommend learning materials that meet students' needs, thereby improving the accuracy and pertinence of learning resource recommendations; The evaluation agent: Function: Using a multimodal assessment method, students' learning works are comprehensively evaluated from multiple dimensions such as content quality, innovation level, and skill application; Prompt project implementation: Design prompt information to allow the big model to conduct multi-dimensional evaluation of students' learning works; Fine-tuning the large model: Use student learning works and corresponding professional evaluation results to fine-tune the large model so that the model can accurately perform multimodal evaluation and provide objective and comprehensive evaluation opinions.

8. The multi-agent collaborative reasoning system for intelligent teaching intervention according to claim 6 is characterized in that: The steps of the collaborative agent in the multi-agent collaborative reasoning module to jointly reason with each agent through the collaborative reasoning mechanism are as follows: after receiving the task assigned by the teaching strategy dynamic adjustment module, each agent independently analyzes the student data based on its own functions and responsibilities using the prompt engineering and the fine-tuned capabilities of the large model. Each agent interacts and shares the analysis results through the information sharing platform. The collaborative agent comprehensively considers the suggestions of all agents and uses the joint reasoning algorithm to generate comprehensive teaching intervention suggestions. A supervision and feedback mechanism is set up to evaluate and verify the reasoning results of each agent.

9. The multi-agent collaborative reasoning system for intelligent teaching intervention according to claim 8 is characterized in that: The multi-agent collaborative reasoning module uses a joint reasoning algorithm to generate teaching intervention suggestions I, and the calculation formula is: Among them, f is the total number of agents, and the output of each agent is O=[o1,o2,…,o f ] is obtained by the corresponding agents according to their respective analysis models, and the collaborative weight vector is φ=[φ1,φ2,…,φ f ] Dynamically adjust according to teaching objectives and the importance of each agent in different scenarios.

Images