Multi-scene teaching AI intelligent agent application method, system, device, medium and product

Abstract

The invention relates to the technical field of multi-scene teaching and training AI agent application, and discloses a multi-scene teaching and training AI agent application method, system and device, a medium and a product. Through multi-source teaching and training demand data standardization processing, teaching and training scene adaptation model reasoning, business rule bidirectional verification, deployment optimization and whole-process iteration closed loop logic, target interaction request data output by a cue word engineering model serves as core input of a teaching and training adaptation model, and deep collaboration of the target interaction request data and the cue word engineering model is achieved; and formulating an optimization deployment scheme based on the target interaction result and the feedback data, and iterating the intelligent agent. According to the method, the problems of independence and disjunction of cue word engineering and model fine tuning, insufficient interaction continuity, insufficient data security, poor teaching business adaptability and the like in an existing intelligent agent are effectively solved, the accuracy and the real-time performance of teaching and training services are remarkably improved, the teaching management cost is reduced, and meanwhile, the service effect is continuously optimized through closed-loop iteration.

Description

Technical Field

[0001] This invention relates to the field of multi-scenario education and training AI intelligent agent application technology, and in particular to a multi-scenario education and training AI intelligent agent application method, system, device, medium and product. Background Technology

[0002] Currently, AI-powered educational assistants are intelligent educational support systems built on artificial intelligence technology. Their core objective is to break away from the traditional "one-size-fits-all" teaching model and achieve precision and personalization in the teaching process. By collecting students' learning data in real time and combining it with pre-set subject knowledge graphs and teaching logic, this intelligent assistant can dynamically analyze students' knowledge mastery, learning weaknesses, and learning habits.

[0003] However, existing intelligent agents are mostly limited to recommending single subjects or simple exercises, lacking multi-scenario adaptability, dynamic learning path adjustment mechanisms, and accurate quantitative evaluation models. At the same time, the prompt word engineering and model fine-tuning in existing intelligent agents are mostly independent links, without forming effective collaboration, and have obvious defects in terms of interaction coherence, data security, and adaptability to teaching business. Summary of the Invention

[0004] This invention provides a method, system, device, medium, and product for multi-scenario AI intelligent agent applications in education and training. It solves the technical problems that in existing intelligent agents, prompt word engineering and model fine-tuning are mostly independent links, without effective collaboration, and there are obvious defects in terms of interaction coherence, data security, and adaptability to teaching business.

[0005] The first aspect of this invention provides a method for applying multi-scenario AI intelligent agents in education and training, which is applied to AI intelligent agents. The method includes: Multi-source education and training demand data is input into a preset prompt word engineering model. The prompt word engineering model is used to standardize the multi-source education and training demand data to obtain target interaction request data. Input the target interaction request data into the preset target education and training adaptation model, and output the initial interaction result; The target interaction request data and the initial interaction result are reviewed and verified according to the preset business rule verification method to obtain the target interaction result and feedback data. Based on the target interaction results and the feedback data, an optimized deployment plan for the AI ​​agent is formulated, and the AI ​​agent is iteratively optimized according to the optimized deployment plan.

[0006] Optionally, the step of inputting multi-source education and training demand data into a preset prompt word engineering model, and standardizing the multi-source education and training demand data through the prompt word engineering model to obtain target interaction request data includes: Acquire the target subject, teaching scenario, and teaching objectives to obtain subject and scenario requirements data; Obtain preliminary data on students' current learning stage, mastery of knowledge points, and learning preferences to obtain basic data on student learning. Obtain learning requests initiated by students or teachers to get initial interaction request data; Using the subject and scenario demand data, the student learning foundation data, and the initial interaction request data, multi-source education and training demand data is obtained; Input the multi-source education and training demand data into a preset prompt word engineering model; The multi-source education and training demand data is standardized using the prompt word engineering model to output standardized interaction instructions, dynamic context information packages, safety prompt words, and dialogue consistency data. The target interaction request data is generated by combining the standardized interaction instructions, the dynamic context information package, the security prompt words, and the dialogue consistency data.

[0007] Optionally, the step of inputting the target interaction request data into a preset target education and training adaptation model and outputting the initial interaction result includes: A dataset is generated by using standardized interaction instructions, dynamic context information packets, security prompts, and dialogue consistency data based on local educational data and the target interaction request data. The dataset is divided into a training set and a test set according to a preset ratio; The training set is input into a preset initial education and training adaptation model for training, and the hyperparameters of the initial education and training adaptation model are optimized to obtain an updated education and training adaptation model. The test set is input into the updated education and training adaptation model for testing to obtain the target education and training adaptation model; The target interaction request data is input into the target education and training adaptation model, and the model outputs knowledge point explanations, exercise solutions, practical training guidance and learning path recommendations that are adapted to the target interaction request. The initial interaction results are obtained by using the knowledge point explanations, exercise solutions, practical training guidance, and learning path recommendations.

[0008] Optionally, the step of verifying the target interaction request data and the initial interaction result according to a preset business rule verification method to obtain the target interaction result and feedback data includes: Perform data preprocessing on the target interactive request data; Based on the preprocessed target interaction request data, the learning progress monitoring rules are used to evaluate the student's learning progress based on the initial interaction results, and subsequent learning content is recommended based on the evaluation results. Based on the preprocessed target interaction request data, the output content of the initial interaction result is quantitatively scored using teaching content compliance rules. Based on the preprocessed target interaction request data, the difficulty level of the initial interaction result and the student learning ability assessment result are verified using personalized adaptation rules. Based on the verification results, learning content corresponding to the personalized learning difficulty level that is adapted to the student is generated. Based on the preprocessed target interaction request data, the evaluation results, the quantitative scoring results, and the verification results, the multiple output results of the initial interaction results of the target education and training adaptation model are weighted and sorted to obtain the target interaction results; Feedback data is generated by using the quantitative scoring results, the verification results, and the data from the students' feedback on the target interaction results.

[0009] Optionally, the step of formulating an optimized deployment plan for the AI ​​agent based on the target interaction result and the feedback data, and iteratively optimizing the AI ​​agent according to the optimized deployment plan, includes: Based on the target interaction results and the feedback data, determine the model quantization parameters, graphics processor resource allocation parameters, high concurrency stability scheme, and security protection optimization method of the target education and training adaptation model; Using the model quantization parameters, the graphics processor resource allocation parameters, the high-concurrency stability scheme, and the security protection optimization method, an optimized deployment scheme for the AI ​​agent is formulated. Based on the optimized deployment scheme, the AI ​​agent is iteratively optimized.

[0010] The second aspect of this invention provides a multi-scenario education and training AI intelligent agent application system, applied to an AI intelligent agent, the system comprising: The processing module is used to input multi-source education and training demand data into a preset prompt word engineering model, and to standardize the multi-source education and training demand data through the prompt word engineering model to obtain target interaction request data. The input module is used to input the target interaction request data into a preset target education and training adaptation model and output the initial interaction result; The verification module is used to review and verify the target interaction request data and the initial interaction result according to the preset business rule verification method, so as to obtain the target interaction result and feedback data. The optimization module is used to formulate an optimized deployment plan for the AI ​​agent based on the target interaction results and the feedback data, and to iteratively optimize the AI ​​agent according to the optimized deployment plan.

[0011] Optionally, the processing module includes: The first acquisition submodule is used to acquire the target subject, teaching scenario, and teaching objective, and obtain subject and scenario requirement data; The second acquisition submodule is used to acquire students' current learning stage, preliminary data on their mastery of knowledge points, and learning preferences, thereby obtaining basic learning data for students. The third acquisition submodule is used to acquire learning requests initiated by students or teachers and obtain initial interaction request data; The multi-source submodule is used to obtain multi-source education and training demand data by using the subject and scenario demand data, the student learning foundation data, and the initial interaction request data. The input submodule is used to input the multi-source education and training demand data into a preset prompt word engineering model; The output submodule is used to standardize the multi-source education and training demand data through the prompt word engineering model, and output standardized interaction instructions, dynamic context information packages, safety prompt words and dialogue consistency data. The combined submodule is used to combine the standardized interaction instructions, the dynamic context information package, the security prompt words, and the dialogue consistency data to generate target interaction request data.

[0012] The third aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the multi-scenario education and training AI intelligent agent application method as described in any of the preceding claims.

[0013] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the multi-scenario education and training AI intelligent agent application method as described in any of the preceding claims.

[0014] The fifth aspect of the present invention provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein, when the program instructions are executed by a computer, the computer performs the multi-scenario education and training AI intelligent agent application method as described in any of the preceding claims.

[0015] As can be seen from the above technical solutions, the present invention has the following advantages: This invention employs a closed-loop logic encompassing standardized processing of multi-source training demand data, training scenario adaptation model reasoning, bidirectional verification of business rules, deployment optimization, and end-to-end iteration. It uses the target interaction request data output from the prompt word engineering model as the core input to the training adaptation model, achieving deep collaboration between the two. Relying on a pre-defined business rule verification method, it conducts bidirectional review combining the target interaction request data and initial interaction results to ensure service compliance and personalization. Based on the target interaction results and feedback data, it formulates optimized deployment plans and iterates the intelligent agent, forming end-to-end data interoperability. This invention effectively solves problems in existing intelligent agents such as the independent disconnect between prompt word engineering and model fine-tuning, insufficient interaction coherence, inadequate data security, and poor adaptability to teaching business. It significantly improves the accuracy and real-time performance of training services, reduces teaching management costs, and continuously optimizes service effectiveness through closed-loop iteration, providing an efficient, secure, and practically tailored intelligent solution for multi-disciplinary and multi-scenario training. Attached Figure Description

[0016] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the steps of a multi-scenario AI intelligent agent application method for education and training provided in Embodiment 1 of the present invention. Figure 2 This is a flowchart illustrating the steps of a multi-scenario AI intelligent agent application method for education and training provided in Embodiment 1 of the present invention. Figure 3 This is a structural block diagram of a multi-scenario education and training AI intelligent agent application system provided in Embodiment 1 of the present invention; Figure 4 This is a structural block diagram of an electronic device provided in Embodiment 4 of the present invention. Detailed Implementation

[0018] This invention provides a method, system, device, medium, and product for multi-scenario AI intelligent agent applications in education and training, which addresses the technical problem that existing intelligent agents often have independent steps for prompt word engineering and model fine-tuning, lacking effective collaboration, and exhibiting significant deficiencies in interactive coherence, data security, and adaptability to teaching business.

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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. It should be noted that in the optional embodiments of the present invention, the object information and other related data involved require the permission or consent of the object when the embodiments of the present invention are applied to specific products or technologies, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. That is to say, if the embodiments of the present invention involve data related to the object, it needs to be obtained with the authorization and consent of the object, the authorization and consent of the relevant departments, and in compliance with the relevant laws, regulations, and standards of the country and region. If personal information is involved in the embodiments, the acquisition of all personal information requires the consent of the individual. If sensitive information is involved, the separate consent of the information subject is required, and the embodiments also need to be implemented with the authorization and consent of the object.

[0020] Please see Figure 1 , Figure 1 This is a flowchart illustrating the steps of a multi-scenario AI intelligent agent application method for education and training provided in Embodiment 1 of the present invention.

[0021] This invention provides a method for applying AI intelligent agents in multi-scenario education and training, which is applied to AI intelligent agents and includes: Step 101: Input the multi-source education and training demand data into the preset prompt word engineering model, and standardize the multi-source education and training demand data through the prompt word engineering model to obtain the target interaction request data.

[0022] In this embodiment of the invention, multi-source education and training demand data refers to a collection of various input data that support the development of education and training services, including subject and scenario demand data, student learning basic data, student interaction request data, etc.

[0023] The prompt word engineering model refers to a model specifically designed for education and training scenarios, which has functions such as instruction generation, context construction, security defense, and dialogue consistency maintenance, and can transform non-standardized requirement data into standardized interactive data.

[0024] Standardization processing refers to the process of making raw data meet the needs of model processing and business requirements through operations such as format normalization, content security filtering, and scene attribute annotation.

[0025] Target interaction request data refers to interactive data that, after being standardized by the prompt word engineering model, possesses a unified structure, security and compliance attributes, and clear scenario labels, and can be directly input into the education and training adaptation model for reasoning.

[0026] Multi-source training demand data is input into a pre-defined prompt word engineering model. This multi-source training demand data includes subject and scenario demand data, student learning foundation data, and student interaction request data. Through the domain-specific 3D instruction generation logic, dynamic context construction algorithm, adversarial prompt defense mechanism, and multi-turn dialogue consistency maintenance rules built into the prompt word engineering model, the multi-source training demand data undergoes standardized processing, including format normalization, content security filtering, and scenario attribute annotation. Simultaneously, teaching progress data and domain-related knowledge are integrated to generate dynamic context information packages. Semantic similarity calculation is combined to ensure interaction coherence, ultimately resulting in target interaction request data that is structurally unified, safe and compliant, and includes scenario tags.

[0027] Step 102: Input the target interaction request data into the preset target education and training adaptation model and output the initial interaction results.

[0028] In this embodiment of the invention, the preset target education and training adaptation model refers to an AI model that is based on a basic large model, fine-tuned with education and training-specific data, configured with scenario-based hyperparameters, and optimized with an education adaptation layer, and has the ability to output professional knowledge and respond to needs in the education and training field.

[0029] The initial interaction result refers to the preliminary education and training service content generated by the target education and training adaptation model based on the target interaction request data, which has not undergone business rule verification and personalized optimization.

[0030] The target interaction request data and the dynamic context information package carried by the target interaction request data are synchronously input into the preset target education and training adaptation model. The target education and training adaptation model can accurately identify the scene tags and core needs in the target interaction request data. Combined with the student learning status, teaching progress and domain-related knowledge in the dynamic context information package, the model infers and outputs the initial interaction results that fit the education and training scenario, including knowledge point explanation, exercise answers, learning path recommendations and preliminary scoring of open questions.

[0031] Step 103: Review and verify the target interaction request data and initial interaction results according to the preset business rule verification method to obtain the target interaction results and feedback data.

[0032] In this embodiment of the invention, the preset business rule verification method refers to a method specifically designed for education and training scenarios, which includes rules for compliance of teaching content, personalized adaptation rules, and rules for monitoring learning progress.

[0033] The target interaction result refers to the final education and training service content that meets the needs of education and training business and is adapted to the individual circumstances of students after being verified by business rules and optimized and integrated.

[0034] Feedback data refers to the collection of information recorded during the verification process, such as adaptation deviations, compliance issues, and requirement matching degrees.

[0035] Based on the preset business rule verification method, the initial interaction results are reviewed and verified in multiple dimensions using information such as scene tags, learning ability annotations, and demand types in the target interaction request data as a benchmark. First, the initial interaction results are verified to meet the requirements of education policies, subject standards, and value orientations through the teaching content compliance rules. Then, the difficulty level and output format of the initial interaction results are compared with the target interaction request data according to the personalized adaptation rules. At the same time, the learning progress monitoring rules are combined to determine whether the student's current progress meets the standard and generate supplementary learning suggestions. The format of the verified initial interaction results is optimized and structured to form compliant and personalized target interaction results. The adaptation deviation, compliance issues, and demand matching degree information during the verification process are recorded simultaneously to generate feedback data for subsequent model optimization.

[0036] Step 104: Based on the target interaction results and feedback data, formulate an optimized deployment plan for the AI ​​agent, and iteratively optimize the AI ​​agent according to the optimized deployment plan.

[0037] In this embodiment of the invention, the optimized deployment scheme refers to a feasible deployment scheme for AI agents that combines model performance optimization, resource allocation strategies, cache configuration, and security protection measures, based on the actual application needs of the education and training scenario.

[0038] AI intelligent agents refer to intelligent education and training service carriers that integrate prompt word engineering models, target education and training adaptation models, and business rule verification mechanisms, and have the ability to respond to education and training needs, output professional content, and optimize and iterate services.

[0039] Iterative optimization refers to adjusting parameters, updating rules, and optimizing functions of core modules of AI agents, such as prompt word engineering models, target education and training adaptation models, and business rule verification mechanisms, based on feedback data.

[0040] Based on the target interaction results and feedback data, and combined with the concurrent requirements, operational performance indicators, and security protection requirements of the education and training scenario, an AI agent optimization and deployment plan was formulated, covering model quantization optimization, dynamic allocation of GPU resources, caching strategy configuration, and enhanced security protection. At the same time, based on the adaptation deviation, compliance issues, and demand matching information recorded in the feedback data, the instruction generation rules and dynamic context construction logic of the prompt word engineering model were adjusted in reverse. The hyperparameter configuration of the target education and training adaptation model and the association relationship of the educational knowledge graph were optimized. The verification threshold and adaptation standards in the preset business rule verification method were updated simultaneously to achieve iterative optimization of the core modules of the AI ​​agent throughout the entire process, and continuously improve the accuracy, stability, and scenario adaptability of its education and training services.

[0041] Please see Figure 2 , Figure 2 This is a flowchart illustrating the steps of a multi-scenario AI intelligent agent application method for education and training provided in Embodiment 2 of the present invention.

[0042] This invention provides a method for applying AI intelligent agents in multi-scenario education and training, which is applied to AI intelligent agents and includes: Step 201: Input the multi-source education and training demand data into the preset prompt word engineering model, and standardize the multi-source education and training demand data through the prompt word engineering model to obtain the target interaction request data.

[0043] In this embodiment of the invention, multi-source education and training demand data is input into a preset prompt word engineering model. The prompt word engineering model performs standardized processing on the multi-source education and training demand data, including format normalization, content security filtering, and scene attribute annotation. Simultaneously, teaching progress data and domain-related knowledge are integrated to generate dynamic context information packages, ultimately resulting in target interactive request data that is structurally unified, secure and compliant, and includes scene tags.

[0044] Further, step 201 includes the following sub-steps: S11. Obtain the target subject, teaching scenario, and teaching objectives to obtain subject and scenario requirement data.

[0045] In this embodiment of the invention, the target subject refers to the specific teaching subject that the AI ​​intelligent agent service focuses on, such as finance and commerce, mathematics, international trade, computer science, etc.

[0046] The teaching scenario refers to the specific form and environment in which teaching activities are carried out, including theoretical teaching, practical training, Q&A and tutoring, etc.

[0047] Teaching objectives refer to the expected outcomes of teaching activities in terms of ability development, such as knowledge explanation, skills training, and case analysis.

[0048] Subject and scenario requirements data refers to structured data that integrates the target subject, teaching scenario, teaching objectives and corresponding knowledge point tags, and scenario adaptation requirements.

[0049] By connecting with the teaching plan database, curriculum standard documents, and teacher preset configuration information of educational institutions, the knowledge system framework of the target subject and the implementation mode corresponding to the teaching scenario are clarified, such as theoretical teaching, practical training, Q&A tutoring, etc., as well as the ability cultivation dimensions of the teaching objectives, such as knowledge mastery, skill improvement, case application, etc., and further linking the core knowledge point tags of the subject with the scenario adaptation requirements to form structured and standardized subject and scenario demand data.

[0050] S12. Obtain preliminary data on students' current learning stage, mastery of knowledge points, and learning preferences to obtain basic data on student learning.

[0051] In this embodiment of the invention, the learning stage refers to the specific learning process node in which the student is in the corresponding target subject, such as the course chapter stage, the basic introductory stage, the advanced improvement stage, etc.

[0052] Preliminary data on knowledge mastery refers to data on students' mastery of learned knowledge points, obtained through historical learning records, basic assessments, and other means, including the chapters already studied and the accuracy rate of exercises.

[0053] Learning preferences refer to the tendencies students exhibit during the learning process, including preferences for the form of content presentation, such as a preference for text and image explanations or video teaching.

[0054] Student learning basic data refers to structured data that integrates students' learning stages, preliminary knowledge mastery data, and learning preferences.

[0055] By collecting students' historical learning records, pre-class feedback, and basic assessment results, we can clarify the students' current course progress and their mastery of the knowledge points they have learned, including the accuracy of the chapters they have learned and the pre-class preparation for the knowledge points they have not yet learned. Combined with students' past interaction choices, such as their preference for text and image explanations, video teaching, or voice Q&A, and their learning behavior characteristics, such as the pace of answering questions and the frequency of asking questions, we can integrate these to form structured basic student learning data.

[0056] S13. Obtain the learning request initiated by the student or teacher and get the initial interaction request data.

[0057] In this embodiment of the invention, a learning request refers to various requests initiated by students or teachers to meet teaching and learning needs, including knowledge point consultation, exercise solutions, practical training guidance, teaching assessment assistance, etc.

[0058] Initial interaction request data refers to raw data that has not undergone standardization and contains the original content of the learning request and related information.

[0059] The system acquires learning requests initiated by students or teachers. These requests support multiple formats, including text, voice, and images, and cover core needs such as knowledge point consultation, exercise solutions, practical training guidance, and teaching assessment assistance. The system collects the original content of the learning request, as well as the initiation time and associated courses, in real time through the receiving terminal, directly forming initial interactive request data that has not undergone standardized processing.

[0060] S14. Using subject and scenario demand data, student learning foundation data, and initial interaction request data, multi-source education and training demand data is obtained.

[0061] In this embodiment of the invention, a data association mapping mechanism is used to integrate three types of data: subject and scenario requirement data, student learning basic data, and initial interaction request data, according to the logical dimensions of scenario benchmark, student characteristics, and core demands. Among them, subject and scenario requirement data clarifies the service boundaries, student learning basic data supplements the basis for personalized adaptation, and initial interaction request data focuses on specific teaching and learning demands. At the same time, duplicate and redundant information is eliminated and data association fields are completed to form multi-source education and training demand data covering scenarios, students, and demands in all dimensions.

[0062] S15. Input multi-source education and training demand data into the preset prompt word engineering model.

[0063] In this embodiment of the invention, multi-source education and training demand data is input into a preset prompt word engineering model, which is pre-configured with three-dimensional instruction generation rules, dynamic context construction algorithms and security defense mechanisms specific to the education and training scenario.

[0064] S16. Standardize the multi-source education and training demand data through the prompt word engineering model, and output standardized interaction instructions, dynamic context information packages, safety prompt words and dialogue consistency data.

[0065] In this embodiment of the invention, standardized interaction instructions refer to interaction instructions specific to the education and training scenario generated based on role-task-output rules.

[0066] Dynamic context information packages refer to the integration of information such as students' learning status, teaching progress, and domain-related knowledge.

[0067] Security prompts refer to compliant request data that has been filtered by security defense mechanisms, contains no harmful or irrelevant content, and can be directly used for model interaction.

[0068] Dialogue consistency data refers to related data generated based on semantic similarity calculations, used to ensure the consistency of AI agent responses in multi-turn interactions.

[0069] Based on the 3D instruction generation rules, the role-task-output interaction logic adapted to the scenario is extracted, and standardized interaction instructions are output. Then, based on the dynamic context construction algorithm, student learning status, teaching progress and domain-related knowledge from multiple sources are integrated to generate dynamic context information packages. Through the security defense mechanism, harmful and irrelevant content in the data is filtered and screened, and security prompts are output. At the same time, semantic similarity calculation is combined to maintain the coherence of multi-round interactions and output dialogue consistency data. The four types of output data form a standardized processing result that is structurally unified, safe and compliant, and fits the education and training scenario.

[0070] Specifically, based on the input student learning data and subject requirements, teaching progress and domain-related knowledge are integrated to generate a dynamic context information package. The formula for the dynamic context information package is as follows:

[0071] In the formula, C Represents a set of contextual information; S It represents real-time learning status data for students, such as answer accuracy, knowledge mastery, and learning time. T This indicates teaching progress data, such as the current course chapter, teaching objectives, and planned completion milestones. K It represents domain-related knowledge, such as international trade regulations, financial systems, mathematical formulas and theorems.

[0072] The safety prompt words are based on a rule base for identifying harmful instructions. They are filtered through a dual detection process of keyword matching and semantic analysis to filter student interaction request data. The specific formula is as follows:

[0073] In the formula, Indicates safety warning words; This represents the student's original interaction request data; This represents the set of harmful / irrelevant prompts matched in the rule base.

[0074] Dialogue consistency data is calculated based on the semantic similarity between the dialogue history and the current request to maintain dialogue coherence. The specific formula is as follows:

[0075] In the formula, Indicates semantic similarity; This represents the dialogue history vector, which is generated from the previous dialogue content using a word embedding model. Indicates the current request vector; when When, output consistent interaction data related to historical dialogues, When a new question is generated, the initial interaction data is output.

[0076] S17. Combine standardized interaction instructions, dynamic context information packets, security prompts, and dialogue consistency data to generate target interaction request data.

[0077] In this embodiment of the invention, a data fusion algorithm is used to associate and integrate standardized interaction instructions, dynamic context information packages, security prompts, and dialogue consistency data according to the logic of instruction guidance, scenario support, core requirements, and coherence constraints. This process completes the data association fields, eliminates redundant information, and integrates the interaction rules of standardized interaction instructions, the scenario background of dynamic context information packages, the core requirements of security prompts, and the coherence requirements of dialogue consistency data into a unified whole, generating target interaction request data that is structurally consistent, logically complete, secure and compliant, and includes scenario tags and interaction constraints.

[0078] Step 202: Input the target interaction request data into the preset target education and training adaptation model and output the initial interaction results.

[0079] In this embodiment of the invention, the four core types of target interaction request data are: standardized interaction instructions, dynamic context information packets, etc. C Safety Tips Dialogue consistency data serves as the input basis and constraint conditions for the initial education and training adaptation model. The initial education and training adaptation model is then fine-tuned to obtain the target education and training adaptation model and the initial interaction results output for these four types of core data.

[0080] Further, step 202 includes the following sub-steps: S21. Generate a dataset by using standardized interaction instructions, dynamic context information packets, safety prompts, and dialogue consistency data based on local education data and target interaction request data.

[0081] In this embodiment of the invention, local education data refers to localized education-related data adapted to the training needs of the target service area, including past exam questions, student homework answers, classroom interaction records, teacher teaching cases, etc.

[0082] A dataset refers to a collection of educational and training-specific data that has been integrated, cleaned, and labeled, and has a unified data format, clear classification labels, and complete relationships.

[0083] Collect local educational data, such as past exam papers, student homework answers, classroom interaction records, and teacher teaching cases, and integrate the safety prompt words output in step S201. Standardized interaction commands and context information packets C This forms a dataset.

[0084] Specifically, the dataset needs to be cleaned and labeled. Data cleaning involves using an outlier detection algorithm (Z-score method) to remove duplicate data, erroneous data, and invalid interaction records, as shown in the following formula:

[0085] In the formula, X This represents data values, such as answering time, score, and interaction frequency. This represents the data mean; Indicates standard deviation; when When an outlier is detected, it is identified and removed to ensure the quality of the dataset.

[0086] The data annotation process involves: following three-dimensional annotation rules based on quality level, knowledge point association, and difficulty coefficient, and combining this with the domain subject classification and teaching scenario definition from step S201, the data is divided into high-quality samples (annotated as...). Medium-quality samples (labeled as) ), low-quality samples (labeled as) The sample is labeled with the corresponding knowledge point tags (such as financial management - budget) and the difficulty level (1-5).

[0087] S22. Divide the dataset into training set and test set according to the preset ratio.

[0088] In this embodiment of the invention, the training set refers to a dataset that is partitioned from the dataset and used by the target education and training adaptation model to learn education and training scenario knowledge, adapt interaction rules, and adjust model parameters.

[0089] The test set refers to a dataset that is divided from the main dataset to verify the output accuracy, scenario adaptability, and interaction coherence of the fine-tuned target education and training adaptation model, and to provide quantitative evaluation results.

[0090] The dataset is divided into training and test sets in a ratio of 7:3 or 8:2. During the division process, stratified sampling is used to ensure that the training and test sets are consistent in terms of subject distribution, knowledge point coverage, difficulty level, and scenario type, so as to avoid the impact of data distribution deviation on model training effect and test accuracy. The training set is used for parameter fine-tuning and ability learning of the target education and training adaptation model, while the test set is used to verify the demand response accuracy, content output accuracy, and interaction coherence of the fine-tuned model in education and training scenarios.

[0091] S23. Input the training set into the preset initial education and training adaptation model for training, and optimize the hyperparameters of the initial education and training adaptation model to obtain the updated education and training adaptation model.

[0092] In this embodiment of the invention, the initial education and training adaptation model refers to an initial model that is based on a general large model, with the addition of educational knowledge graphs and interactive logic, without undergoing fine-tuning and hyperparameter optimization of education and training scenario data, and possessing basic AI reasoning capabilities.

[0093] Hyperparameters refer to parameters that are used to control the initial training process of the training adaptation model and are automatically learned without model training.

[0094] Updating the education and training adaptation model refers to a model that, after being trained on a training set and optimized with hyperparameters, adapts to the data distribution of education and training scenarios and has preliminary education and training service capabilities.

[0095] The training set is input into a preset initial education and training adaptation model. The initial range of hyperparameters is preset in combination with the characteristics of education and training scenarios. The hyperparameters of the model are iteratively optimized through adaptive learning rate adjustment algorithm and cross-validation method. During the process, the accuracy of education and training demand response and content output matching degree of the training set are used as the core evaluation indicators. When the indicators reach the preset threshold and the model converges, the optimization stops, resulting in an updated education and training adaptation model that adapts to the data distribution of education and training scenarios and has the ability to accurately respond to demand and output professional content.

[0096] Hyperparameter optimization configuration of the education and training adaptation model: Based on the subject characteristics and teaching scenario requirements output in step S201, dynamically adjust the model's hyperparameters to improve the model's learning and adaptation capabilities to education and training interaction data. Adaptive learning rate adjustment, based on the difficulty coefficient distribution of the labeled data, employs an adaptive learning rate algorithm, as shown in the following formula:

[0097] In the formula, Indicates the first t Learning rate during round training; Indicates the initial learning rate; , For batches of high-difficulty samples (difficulty coefficient > 4), the learning rate is automatically increased by 10% to ensure that the model fully learns the interaction logic of complex knowledge points.

[0098] Dynamic batch size setting: The training batch size is adjusted based on the complexity (theoretical teaching scenario / practical training scenario) and difficulty level of the teaching scenario corresponding to the sample, using the following formula:

[0099] The construction of the educational knowledge graph is based on the domain subject classification and knowledge point tags in step S201. It builds a network of interconnected subject knowledge, where nodes represent knowledge points and edges represent the logical relationships (dependency, derivation, and application) between knowledge points. The formula is as follows:

[0100] In the formula, V Represents a set of knowledge point nodes; E This represents the set of edges representing relationships between knowledge points. The knowledge graph is linked in real time with the context information package C in step 201, providing structured support for the education and training adaptation model to understand the association of knowledge points.

[0101] Interaction logic adaptation is the embedding of multi-turn dialogue consistency rules and output format control standards in step 201, which enables the model to follow the interaction specifications of the teaching and training scenario during the learning process, ensures that the model output is consistent with the instructions in step one, and improves the continuity of interaction.

[0102] S24. Input the test set into the updated education and training adaptation model for testing, and obtain the target education and training adaptation model.

[0103] In this embodiment of the invention, a test set is input into the updated education and training adaptation model for testing. A test evaluation index system is set based on the core needs of the education and training scenario, focusing on verifying the model's demand response accuracy, content output accuracy, interaction coherence, and compliance under different subjects and teaching scenarios. By calculating the index scores and error ranges corresponding to the test set, it is determined whether the model performance meets the preset standards. If the indicators meet the standards, the updated education and training adaptation model is directly determined as the target education and training adaptation model. If the standards are not met, the process returns to adjusting the hyperparameter range and retraining and optimizing until the model test performance meets the education and training service requirements.

[0104] S24. Input the target interaction request data into the target education and training adaptation model, and output the knowledge point explanations, exercise solutions, practical training guidance and learning path recommendations that are adapted to the target interaction request.

[0105] In this embodiment of the invention, the explanation of knowledge points refers to the professional analysis content that is logically coherent and appropriately challenging, based on the curriculum standards and knowledge system of the target subject and for specific subject knowledge points raised by students. This includes the definition of the knowledge point, core principles, derivation process, application scope, and explanation of common mistakes.

[0106] Exercise solutions refer to complete solutions provided to students' needs for subject-specific exercises. These solutions include problem analysis, a breakdown of the problem-solving approach, step-by-step solution process, answer conclusions, and a summary of problem-solving techniques. The logic and steps of the solutions must conform to the teaching standards and answer requirements of the corresponding subject.

[0107] Practical training guidance refers to specialized guidance content that is adapted to the practical teaching needs in the education and training scenario. It includes the specific procedures for practical training operations, key operation steps, precautions for practical operation, equipment usage specifications, troubleshooting methods for common problems, and verification standards for practical training results.

[0108] Recommended learning path content refers to personalized learning plans generated based on students' learning stages, mastery of knowledge points, learning preferences, and teaching objectives, combined with the progressive logic of subject knowledge. It includes the recommended order of knowledge points, key learning modules, supporting learning resources, phased learning goals, and assessment node planning.

[0109] The target interaction request data is input into the target education and training adaptation model. The model first accurately analyzes the scenario tags, core demands and interaction constraints in the data. Combined with the student's basic learning data and teaching progress in the dynamic context information package, it calls on the professional knowledge system adapted to the target subject and teaching scenario to generate targeted knowledge point explanations that match the student's ability level, logically clear exercise solutions, and practical training guidance that fits the practical needs. At the same time, it plans personalized learning path recommendations based on the student's mastery of knowledge points. All output content matches the requirements of the curriculum standards and takes into account the continuity of interaction.

[0110] S25. Initial interactive results are obtained by using content such as knowledge point explanation, exercise solutions, practical training guidance, and learning path recommendations.

[0111] In this embodiment of the invention, knowledge point explanations, exercise solutions, practical training guidance, and learning path recommendations are used. Through a content integration mechanism, these four types of content are logically linked and formatted according to the core requirements of the target interaction request. The corresponding scenario annotations and adaptation instructions for each type of content are supplemented, and repetitive and redundant descriptions are eliminated to form an initial interaction result that covers all dimensions of demand response, is logically coherent, has a unified format, and is suitable for students' learning foundation.

[0112] Step 203: Review and verify the target interaction request data and initial interaction results according to the preset business rule verification method to obtain the target interaction results and feedback data.

[0113] In this embodiment of the invention, based on the preset business rule verification method, the initial interaction result is reviewed and verified in multiple dimensions using information such as scene tags, learning ability annotations, and requirement types in the target interaction request data as a benchmark, so as to form a compliant and personalized target interaction result. The adaptation deviation, compliance issues, and requirement matching degree information during the verification process are recorded simultaneously to generate feedback data for subsequent model optimization.

[0114] Furthermore, step 203 includes the following sub-steps: S31. Perform data preprocessing on the target interactive request data.

[0115] In this embodiment of the invention, data preprocessing refers to converting the voice and image content in the student or teacher interactive request data into a unified text format using ASR (speech-to-text) and OCR (optical character recognition) technologies.

[0116] Specifically, the voice and image content in the student interaction request data is converted into a unified text format using ASR (speech-to-text) and OCR (optical character recognition) technologies; according to the output format control standard preset in step 201, the same question expressed in different ways is unified into standardized text, for example, "How to solve this math problem?" and "Please provide the solution steps for this problem" are unified into "Request to analyze [knowledge point tag: mathematics - function] problem and output the solution steps").

[0117] Using the knowledge point tagging system and difficulty coefficient grading standard from step 202, the normalized student input data is labeled with scenario attributes, such as "Subject: Finance and Trade", "Knowledge Point: Budget", "Difficulty: Level 3", and "Need Type: Q&A / Learning Path Recommendation", to provide a basis for business rule verification.

[0118] S32. Based on the preprocessed target interaction request data, the learning progress monitoring rules are used to evaluate the student's learning progress based on the initial interaction results, and subsequent learning content is recommended based on the evaluation results.

[0119] In this embodiment of the invention, the learning progress monitoring rule refers to the standardized rules preset based on the curriculum standards and teaching plans of the education and training scenario, used to evaluate whether students' learning progress meets the standards and whether they have mastered the knowledge points completely.

[0120] The assessment results refer to the conclusions drawn from progress comparisons regarding whether students' learning progress meets the standards and whether there are any gaps in their mastery of knowledge points.

[0121] Set learning progress threshold (Dynamically adjusted according to the teaching plan), students' learning progress is assessed in real time through the model in step 202. P ,when When this occurs, the teacher reminder mechanism is triggered and supplementary learning suggestions are generated. The supplementary learning content is generated by the model based on the educational knowledge graph (knowledge point dependencies) from step 202; when In each case, the subsequent learning content recommended by the model is directly entered into the post-processing stage.

[0122] S33. Based on the preprocessed target interaction request data, the output content of the initial interaction result is quantitatively scored using the teaching content compliance rules.

[0123] In this embodiment of the invention, the teaching content compliance rules refer to a standardized rule system based on educational policies, target subject curriculum standards, and industry norms for determining whether the output content of education and training is compliant. It covers multiple dimensions such as content accuracy and curriculum standard fit.

[0124] Quantitative scoring refers to an evaluation method that transforms compliance rules into quantifiable scoring indicators and score standards, and scores each compliance dimension of the initial interaction results.

[0125] Establish an education compliance rule base (RuleBase), which includes education policy requirements, subject standards, and value orientation requirements. Use a compliance scoring function to quantitatively score the model's output, as shown in the following formula:

[0126] In the formula, R Indicates the review result (pass / fail); This represents a compliance scoring function; Indicates the model's output content; when If the time interval is determined to be passed, then step 202 is returned and the model is re-optimized for output.

[0127] S34. Based on the preprocessed target interaction request data, the difficulty level of the initial interaction result and the student's learning ability assessment result are verified using personalized adaptation rules. Based on the verification results, learning content corresponding to the personalized learning difficulty level that is suitable for the student is generated.

[0128] In this embodiment of the invention, the personalized adaptation rule refers to the standardized rule preset based on the principle of teaching according to aptitude in the teaching and training scenario, which is used to verify the matching degree between the difficulty of the output content and the student's ability and to ensure that the content is adapted to the student's personalized needs.

[0129] Difficulty level classification refers to the hierarchical division of the difficulty level of educational and training content.

[0130] Student learning ability assessment results refer to the judgment conclusions drawn from students' basic learning data regarding their core learning abilities, such as knowledge acquisition ability, problem-solving ability, and practical skills.

[0131] Personalized learning difficulty level refers to a specific difficulty level that is tailored to the student's current learning ability.

[0132] Based on the difficulty level grading criteria from step 202 and the student learning ability assessment result A, verify whether the learning path and exercise difficulty recommended by the model are reasonable; if the model is based on learning ability... If the recommended difficulty level for students is ≥4 (average level of students in the same batch), it is deemed unreasonable and the difficulty level is automatically lowered to 2-3.

[0133] S35. Based on the preprocessed target interaction request data, evaluation results, quantitative scoring results, and verification results, the multiple output results of the initial interaction results of the target education and training adaptation model are weighted and sorted to obtain the target interaction results.

[0134] In this embodiment of the invention, weighted sorting refers to a processing method that sets corresponding weight coefficients for different evaluation dimensions, calculates a comprehensive score through weighted calculation, and then sorts multiple output results according to the score.

[0135] Based on the preprocessed target interaction request data, key information such as the priority of core demands and scenario adaptation requirements is extracted. Combining the evaluation results (learning progress adaptability), quantitative scoring results (teaching content compliance), and verification results (personalization adaptability), weight coefficients matching the training service goals are set for each dimension of the results. The comprehensive adaptability score of multiple output results in the initial interaction results is obtained through weighted calculation. Then, the comprehensive scores are sorted from high to low, and the output results with the highest comprehensive scores and meeting the preset qualification threshold are selected as the core content. Simultaneously, optimization suggestions from the verification of each dimension are integrated to standardize the format and refine the details. Finally, the target interaction results that take into account progress adaptability, compliance, and personalized needs are obtained. At the same time, the weighted sorting process, scoring status, and other information are recorded in the feedback data.

[0136] Specifically, based on the multiple Q&A answers / learning resource recommendations generated by the model, a quality score and ranking are performed by assigning weight coefficients based on authority in the education field, as shown in the following formula:

[0137] In the formula, It is the weight of the matching degree of the teaching materials. The weight of the teacher's experience matching degree; This is an extended weight; This indicates the score based on the degree of matching between the output content and the knowledge points in the textbook. A score is awarded based on the degree of matching with the teaching experience of front-line teachers; The content is scored for its extensibility; the top 3 results with the highest weighted scores are selected as the final recommendations.

[0138] S36. Generate feedback data by using quantitative scoring results, verification results, and student feedback on the target interaction results.

[0139] In this embodiment of the invention, the quantitative scoring results include sub-item compliance scores, comprehensive compliance scores and compliance judgments, and verification results. These include data on difficulty adaptation deviation, personalized matching issues and adjustment basis, as well as student feedback on the target interaction results, such as content satisfaction, comprehension difficulty evaluation, and error correction suggestions. The three types of information are associated and aligned through a data integration mechanism, data association fields are supplemented, invalid and redundant information is eliminated, and the interaction session ID and scene tag corresponding to each piece of information are simultaneously labeled to generate structured feedback data covering multiple dimensions of information such as compliance issues, personalized adaptation deviation, and student subjective evaluation.

[0140] It's worth mentioning that, in addition to obtaining the target interaction results and feedback data, a structured learning report is also generated. This report integrates the student's current learning data (answer accuracy, learning progress, etc.). P Knowledge Point Mastery M The model evaluation results (weakness labels, competency levels), business rule verification records, and the "report output format" preset in step 201 are used to generate a structured learning report containing a learning overview, weakness analysis, improvement suggestions, and teacher reference data.

[0141] Based on the learning behavior analysis in step 202, which incorporates student learning style tags, the overall output presentation is adjusted. For example, visual learners are prioritized for Q&A answers combining text and images, while auditory learners are offered audio explanations. Therefore, the output primarily includes personalized learning paths, Q&A answers, recommended exercises, and suggestions for improving weaknesses for students; and class-wide learning reports, detailed learning profiles for individual students, and progress alerts for teachers.

[0142] Step 204: Based on the target interaction results and feedback data, determine the model quantization parameters, graphics processor resource allocation parameters, high concurrency stability scheme, and security protection optimization methods for the target education and training adaptation model.

[0143] In this embodiment of the invention, the model quantization parameter refers to the parameter used to quantify the target education and training adaptation model.

[0144] Graphics processor resource allocation parameters refer to the parameters used to plan the computing power resources of a graphics processor (GPU).

[0145] High concurrency stability solutions refer to solutions designed to ensure the stable operation of a system during peak periods in education and training scenarios (such as after-class Q&A and exam assessment stages) when a large number of users access the system simultaneously.

[0146] Security protection optimization methods refer to optimized security protection measures based on security risks (such as malicious requests and content leakage risks) discovered in feedback data.

[0147] Based on the output content, such as Q&A answers, learning reports, personalized recommendations, and feedback data, such as student satisfaction ratings, compliance review feedback, and system load data, the performance tuning of AI agents includes aspects such as model quantization parameters of the target education and training adaptation model, graphics processor resource allocation parameters, high concurrency stability solutions, and security protection optimization methods.

[0148] Step 205: Develop an optimized deployment plan for the AI ​​agent by using model quantization parameters, graphics processor resource allocation parameters, high concurrency stability schemes, and security protection optimization methods.

[0149] In this embodiment of the invention, the optimized deployment scheme for the AI ​​agent is as follows: 1) Model Quantization: The model parameters of the training adaptation model output in step two are compressed using INT8 quantization technology to reduce storage and computation costs. The quantization formula is as follows:

[0150] In the formula, Indicates the quantized parameters; Represents the original floating-point parameters; , These represent the minimum and maximum values ​​of the parameter, respectively. Indicates the number of bits used for quantization.

[0151] Graphics processor resource allocation (i.e., dynamic GPU resource allocation) is based on the scene complexity Y (levels 1-3) fed back from step 202 and the obtained number of concurrent users N, allocating GPU resources according to the following formula:

[0152] In the formula, GPU represents the number of GPU cores allocated; This represents the floor function, used to report model response speed in high-concurrency scenarios.

[0153] 2) High concurrency stability guarantee: Caching mechanism: Redis is used to cache the local educational data (textbook knowledge points, answers to high-frequency exercises) and the output results of the education and training adaptation model output in step 202. The cache validity period is set according to the resource update frequency (7200s for textbook knowledge points, 3600s for answers to high-frequency exercises, and 600s for real-time interactive answers) to reduce redundant calculations.

[0154] Degradation strategy: Based on the system load data fed back from step three, when the system load... When this happens, the service quality of non-core functions (such as extended knowledge recommendations and multimedia displays) will be automatically reduced, while core functions such as Q&A tutoring and learning progress management will be prioritized.

[0155] 3) Security protection optimization: Prompt injection attack defense: Perform syntax analysis and code detection on the interactive request data input by students to filter malicious injection instructions. Combined with the security prompt word filtering mechanism in step 201, a double protection is formed.

[0156] Access control: The RBAC (Role-Based Access Control) model is adopted, dividing the system into three roles: student, teacher, and administrator. Access to core modules (model parameters, private knowledge base) is restricted to protect the core data of the system.

[0157] Optimized deployment solution output: The optimized deployment solution for the AI ​​agent includes quantized model files, GPU resource allocation strategies, cache configuration parameters, security protection rules, etc., to ensure that the system runs efficiently and stably in actual education and training scenarios.

[0158] Step 206: Iteratively optimize the AI ​​agent according to the optimized deployment plan.

[0159] In this embodiment of the invention, the system receives operational data of the optimized deployment scheme (such as system response speed, concurrent processing capability, and security protection logs), as well as student learning feedback data and full-process user interaction logs (question content, answer records, and learning duration).

[0160] Domain Data Barrier Construction: Private Knowledge Base Construction: Integrate real business data of enterprises (such as financial data of financial and trade enterprises), teaching experience of front-line teachers, and industry standards and norms, and transform them into teaching cases and training data, which correspond to and are adapted to the domain three-dimensional instruction system in step 201; establish a regular update mechanism, with monthly updates for policy-sensitive domains (finance, law) and quarterly updates for knowledge-stable domains (mathematics, physics).

[0161] User behavior data mining: Analyze the received full-process user interaction logs to extract student learning preferences (such as question frequency, question type, and learning duration distribution) and model output performance data (such as answer satisfaction and learning path fit) to form a user behavior feature dataset. .

[0162] Dynamic optimization iteration: Optimization of the education and training adaptation model: based on user behavior feature dataset Through formula Calculate the parameter update amount (where, Indicates the amount of model parameter updates. This is the iteration step size (set to 0.01). L For loss function, (As the current parameters of the model), fine-tune the parameters of the education and training adaptation model in step 202; optimize the knowledge graph association logic of the education adaptation layer to improve the accuracy of model output.

[0163] Optimization of the prompt word engineering model: Based on student feedback data, adjust the "role-task" matching relationship in step 201 (e.g., add the "customs broker" role instruction in the international trade practice training); optimize the context construction rules and adjust... The weights of each data point (e.g., for students lagging behind in their learning progress, increase the weight of S (learning status)).

[0164] Business rules and deployment scheme optimization: Based on learning effect data, adjust the "learning progress threshold" in step 203. Based on student feedback, optimize the "weighted sorting weight" in step 203; and adjust the GPU resource allocation strategy and cache validity period in step 205 based on deployment and operation data.

[0165] The output of the iterative agent is a dynamically optimized AI agent that has more accurate personalized service capabilities, higher operating efficiency, and stronger scene adaptability, forming a closed-loop iterative system of "building-fine-tuning-application-optimization".

[0166] It's worth noting that the dynamically optimized AI agent can guide students to answer questions and provide feedback based on the teaching steps set by the teacher, monitoring learning progress and knowledge acquisition in real time, providing teachers with precise teaching data; dynamically adjusting learning paths and difficulty based on student performance, providing personalized learning plans; and offering timely feedback and suggestions to help students correct mistakes, consolidate knowledge, and improve their self-learning abilities. Each student can also be assigned a dedicated AI teacher to ask and answer questions about knowledge points, assess their level of mastery, and provide one-on-one tutoring; providing real-time answers to student questions, detailed problem-solving approaches and methods, tracking learning outcomes and generating reports to help teachers understand student progress and optimize teaching strategies. When students request interactive data, the AI ​​agent responds in real time, using multiple answer methods to meet different learning needs, recommending learning resources and extension content based on questions and learning progress, broadening students' knowledge and enhancing the depth and breadth of their learning. The dynamically optimized AI agent can also engage in role-playing, interactive discussions, or human-computer interaction, integrating knowledge points into vivid scenarios to stimulate learning interest; and allowing students to apply knowledge through scenario simulations, introducing cooperation and competition mechanisms to cultivate teamwork and a competitive spirit. Furthermore, it provides personalized role settings based on students' characteristics and interests, intelligently adjusts the scenarios and difficulty, and provides real-time feedback and encouragement on performance, helping students build confidence and stimulate motivation.

[0167] Please see Figure 3 , Figure 3 This is a structural block diagram of a multi-scenario education and training AI intelligent agent application system provided in Embodiment 3 of the present invention.

[0168] This invention provides a multi-scenario education and training AI intelligent agent application system, applied to AI intelligent agents, including: Processing module 301 is used to input multi-source education and training demand data into a preset prompt word engineering model, and to standardize the multi-source education and training demand data through the prompt word engineering model to obtain target interaction request data; Input module 302 is used to input target interaction request data into a preset target education and training adaptation model and output initial interaction results; The verification module 303 is used to audit and verify the target interaction request data and the initial interaction result according to the preset business rule verification method, and to obtain the target interaction result and feedback data. The optimization module 304 is used to formulate an optimized deployment plan for the AI ​​agent based on the target interaction results and feedback data, and to iteratively optimize the AI ​​agent according to the optimized deployment plan.

[0169] Furthermore, the processing module includes: The first acquisition submodule is used to acquire the target subject, teaching scenario, and teaching objective, and obtain subject and scenario requirement data; The second acquisition submodule is used to acquire students' current learning stage, preliminary data on their mastery of knowledge points, and learning preferences, thereby obtaining basic learning data for students. The third acquisition submodule is used to acquire learning requests initiated by students or teachers and obtain initial interaction request data; The multi-source submodule is used to obtain multi-source education and training demand data by using subject and scenario demand data, student learning foundation data, and initial interaction request data. The input submodule is used to input multi-source education and training demand data into a preset prompt word engineering model; The output submodule is used to standardize multi-source education and training demand data through the prompt word engineering model, and output standardized interactive instructions, dynamic context information packages, safety prompt words and dialogue consistency data. Combined with sub-modules, it is used to generate target interaction request data by combining standardized interaction instructions, dynamic context information packets, security prompts, and dialogue consistency data.

[0170] Furthermore, the input module 302 includes: The dataset submodule is used to generate a dataset by using standardized interaction instructions, dynamic context information packets, security prompts, and dialogue consistency data from local educational data and target interaction request data. The partitioning module is used to divide the dataset into training and testing sets according to a preset ratio; The update submodule is used to train the training set into the preset initial education and training adaptation model and optimize the hyperparameters of the initial education and training adaptation model to obtain the updated education and training adaptation model. The testing submodule is used to input the test set into the updated education and training adaptation model for testing, and obtain the target education and training adaptation model. The recommendation submodule is used to input the target interaction request data into the target education and training adaptation model and output knowledge point explanations, exercise solutions, practical training guidance and learning path recommendations that are adapted to the target interaction request. The interactive submodule is used to obtain initial interactive results by using content such as knowledge point explanations, exercise solutions, practical training guidance, and learning path recommendations.

[0171] Furthermore, the verification module 303 includes: The preprocessing submodule is used to preprocess the target interactive request data; The evaluation submodule is used to evaluate the student's learning progress based on the preprocessed target interaction request data and the learning progress monitoring rules, and recommend subsequent learning content based on the evaluation results. The scoring submodule is used to quantitatively score the output of the initial interaction results based on the preprocessed target interaction request data and the teaching content compliance rules. The verification submodule is used to verify the difficulty level of the initial interaction result and the student's learning ability assessment result based on the preprocessed target interaction request data and the personalized adaptation reasonable rules. Based on the verification result, it generates learning content corresponding to the personalized learning difficulty level that is adapted to the student. The sorting submodule is used to perform weighted sorting of multiple output results of the initial interaction results of the target education and training adaptation model based on the preprocessed target interaction request data, evaluation results, quantitative scoring results and verification results, to obtain the target interaction results; The feedback submodule is used to generate feedback data by taking quantitative scoring results, verification results, and student feedback on the target interaction results.

[0172] Furthermore, optimization module 304 includes: The quantization submodule is used to determine the model quantization parameters, graphics processor resource allocation parameters, high-concurrency stability schemes, and security protection optimization methods of the target education and training adaptation model based on the target interaction results and feedback data. A submodule is defined to develop an optimized deployment plan for the AI ​​agent by using model quantization parameters, graphics processor resource allocation parameters, high concurrency stability schemes, and security protection optimization methods. The optimization submodule is used to iteratively optimize the AI ​​agent based on the optimization deployment plan.

[0173] Please see Figure 4 , Figure 4 This is a structural block diagram of an electronic device provided in Embodiment 4 of the present invention.

[0174] An electronic device according to an embodiment of the present invention includes: a memory 401 and a processor 402. The memory 401 stores a computer program. When the computer program is executed by the processor 402, the processor 402 executes the multi-scenario education and training AI intelligent agent application method as described in any of the above embodiments.

[0175] Memory 401 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Memory 401 has storage space 403 for program code 413 for performing any of the method steps described above. For example, storage space 403 for program code may include individual program codes 413 for implementing the various steps in the methods described above. This program code may be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When run by a computing processing device, this code causes the computing processing device to perform the various steps in the methods described above. This program code may be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When this code is run by a computing device, it causes the computing device to perform the various steps in the multi-scenario education and training AI agent application method described above.

[0176] Embodiment 5 of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multi-scenario education and training AI intelligent agent application method as described in any of the above embodiments.

[0177] Embodiment 6 of the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer performs the multi-scenario education and training AI intelligent agent application method as described in any of the above embodiments.

[0178] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0179] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0180] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0181] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0182] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0183] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for applying AI intelligent agents in multi-scenario education and training, characterized in that, Applied to AI intelligent agents, the method includes: Multi-source education and training demand data is input into a preset prompt word engineering model. The prompt word engineering model is used to standardize the multi-source education and training demand data to obtain target interaction request data. Input the target interaction request data into the preset target education and training adaptation model, and output the initial interaction result; The target interaction request data and the initial interaction result are reviewed and verified according to the preset business rule verification method to obtain the target interaction result and feedback data. Based on the target interaction results and the feedback data, an optimized deployment plan for the AI ​​agent is formulated, and the AI ​​agent is iteratively optimized according to the optimized deployment plan.

2. The multi-scenario education and training AI intelligent agent application method according to claim 1, characterized in that, The process involves inputting multi-source education and training demand data into a preset prompt word engineering model, and then standardizing the multi-source education and training demand data through the prompt word engineering model to obtain target interaction request data, including: Acquire the target subject, teaching scenario, and teaching objectives to obtain subject and scenario requirements data; Obtain preliminary data on students' current learning stage, mastery of knowledge points, and learning preferences to obtain basic data on student learning. Obtain learning requests initiated by students or teachers to get initial interaction request data; Using the subject and scenario demand data, the student learning foundation data, and the initial interaction request data, multi-source education and training demand data is obtained; Input the multi-source education and training demand data into a preset prompt word engineering model; The multi-source education and training demand data is standardized using the prompt word engineering model to output standardized interaction instructions, dynamic context information packages, safety prompt words, and dialogue consistency data. The target interaction request data is generated by combining the standardized interaction instructions, the dynamic context information package, the security prompt words, and the dialogue consistency data.

3. The multi-scenario education and training AI intelligent agent application method according to claim 1, characterized in that, The step of inputting the target interaction request data into a preset target education and training adaptation model and outputting the initial interaction result includes: A dataset is generated by using standardized interaction instructions, dynamic context information packets, security prompts, and dialogue consistency data based on local educational data and the target interaction request data. The dataset is divided into a training set and a test set according to a preset ratio; The training set is input into a preset initial education and training adaptation model for training, and the hyperparameters of the initial education and training adaptation model are optimized to obtain an updated education and training adaptation model. The test set is input into the updated education and training adaptation model for testing to obtain the target education and training adaptation model; The target interaction request data is input into the target education and training adaptation model, and the model outputs knowledge point explanations, exercise solutions, practical training guidance and learning path recommendations that are adapted to the target interaction request. The initial interaction results are obtained by using the knowledge point explanations, exercise solutions, practical training guidance, and learning path recommendations.

4. The multi-scenario education and training AI intelligent agent application method according to claim 1, characterized in that, The step of reviewing and verifying the target interaction request data and the initial interaction result according to the preset business rule verification method to obtain the target interaction result and feedback data includes: Perform data preprocessing on the target interactive request data; Based on the preprocessed target interaction request data, the learning progress monitoring rules are used to evaluate the student's learning progress based on the initial interaction results, and subsequent learning content is recommended based on the evaluation results. Based on the preprocessed target interaction request data, the output content of the initial interaction result is quantitatively scored using teaching content compliance rules. Based on the preprocessed target interaction request data, the difficulty level of the initial interaction result and the student learning ability assessment result are verified using personalized adaptation rules. Based on the verification results, learning content corresponding to the personalized learning difficulty level that is adapted to the student is generated. Based on the preprocessed target interaction request data, the evaluation results, the quantitative scoring results, and the verification results, the multiple output results of the initial interaction results of the target education and training adaptation model are weighted and sorted to obtain the target interaction results; Feedback data is generated by using the quantitative scoring results, the verification results, and the data from the students' feedback on the target interaction results.

5. The multi-scenario education and training AI intelligent agent application method according to claim 1, characterized in that, The step of formulating an optimized deployment plan for the AI ​​agent based on the target interaction results and the feedback data, and iteratively optimizing the AI ​​agent according to the optimized deployment plan, includes: Based on the target interaction results and the feedback data, determine the model quantization parameters, graphics processor resource allocation parameters, high concurrency stability scheme, and security protection optimization method of the target education and training adaptation model; Using the model quantization parameters, the graphics processor resource allocation parameters, the high-concurrency stability scheme, and the security protection optimization method, an optimized deployment scheme for the AI ​​agent is formulated. Based on the optimized deployment scheme, the AI ​​agent is iteratively optimized.

6. A multi-scenario education and training AI intelligent agent application system, characterized in that, The system, applied to AI intelligent agents, includes: The processing module is used to input multi-source education and training demand data into a preset prompt word engineering model, and to standardize the multi-source education and training demand data through the prompt word engineering model to obtain target interaction request data. The input module is used to input the target interaction request data into a preset target education and training adaptation model and output the initial interaction result; The verification module is used to review and verify the target interaction request data and the initial interaction result according to the preset business rule verification method, so as to obtain the target interaction result and feedback data. The optimization module is used to formulate an optimized deployment plan for the AI ​​agent based on the target interaction results and the feedback data, and to iteratively optimize the AI ​​agent according to the optimized deployment plan.

7. The multi-scenario education and training AI intelligent agent application system according to claim 6, characterized in that, The processing module includes: The first acquisition submodule is used to acquire the target subject, teaching scenario, and teaching objective, and obtain subject and scenario requirement data; The second acquisition submodule is used to acquire students' current learning stage, preliminary data on their mastery of knowledge points, and learning preferences, thereby obtaining basic learning data for students. The third acquisition submodule is used to acquire learning requests initiated by students or teachers and obtain initial interaction request data; The multi-source submodule is used to obtain multi-source education and training demand data by using the subject and scenario demand data, the student learning foundation data, and the initial interaction request data. The input submodule is used to input the multi-source education and training demand data into a preset prompt word engineering model; The output submodule is used to standardize the multi-source education and training demand data through the prompt word engineering model, and output standardized interaction instructions, dynamic context information packages, safety prompt words and dialogue consistency data. The combined submodule is used to combine the standardized interaction instructions, the dynamic context information package, the security prompt words, and the dialogue consistency data to generate target interaction request data.

8. An electronic device, characterized in that, The system includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the multi-scenario education and training AI intelligent agent application method as described in any one of claims 1-5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the multi-scenario education and training AI intelligent agent application method as described in any one of claims 1-5.

10. A computer program product, characterized in that, The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, wherein when the program instructions are executed by a computer, the computer performs the multi-scenario education and training AI intelligent agent application method as described in any one of claims 1-5.

Images