An Adaptive Intelligent Training Agent Method and System Based on Capability Atom Units

Abstract

This invention provides an adaptive intelligent training agent method and system based on capability atomic units, relating to the field of data processing technology. It determines the target capability atomic units based on the user's learning request, selects the training order according to the difficulty level of the target capability atomic units, and obtains the target training order. It then executes the corresponding training steps sequentially according to the target training order, and performs evaluation and adaptive decision-making in each training step to obtain the step training state. After all training steps are completed, the training state is persistently stored based on the step training state, and cross-session recovery is achieved based on the interval between new sessions. This allows for customized training order execution according to the actual unit modules trained by the user, improving the training efficiency of the user's transferability practice.

Description

Technical Field

[0001] This invention relates to data processing technology, and more particularly to an adaptive intelligent training agent method and system based on capability atomic units. Background Technology

[0002] With the rapid development of artificial intelligence technology, intelligent learning guidance systems have become an important tool in the field of education and training.

[0003] Currently, traditional intelligent training generally adopts a fixed-process teaching model to guide users' training, which has many shortcomings. Among them, because the teaching behavior of existing systems is usually driven by the general generation ability of large language models, the answer content is highly random and cannot be customized according to the difficulty of the actual teaching modules. This makes it impossible for users to obtain a complete training experience from problem stimulation to ability solidification, affecting the training efficiency of users' practice of transfer ability.

[0004] Therefore, how to execute a customized training sequence based on the actual unit modules trained by the user, and improve the training efficiency of the user's ability to transfer skills, has become an urgent problem to be solved. Summary of the Invention

[0005] This invention provides an adaptive intelligent training agent method and system based on capability atomic units, which can execute a customized training order according to the actual training unit modules of the user, thereby improving the training efficiency of the user's practice of transfer capabilities.

[0006] A first aspect of this invention provides an adaptive intelligent training agent method based on capability atomic units, comprising: The target capability atomic units are determined based on the user's learning request, and the training order is selected according to the difficulty level of the target capability atomic units to obtain the target training order. According to the target training order, the corresponding training steps are executed sequentially, and evaluation and adaptive decision-making are performed in each training step to obtain the step training state; Once all training steps have been completed, the training state is persistently stored based on the training state of each step, and cross-session recovery is achieved based on the number of days between new sessions.

[0007] Optionally, in one possible implementation of the first aspect, determining the target capability atomic unit based on the user's learning request and selecting the training order according to the difficulty level of the target capability atomic unit to obtain the target training order includes: Based on the user's learning request, the corresponding capability atomic unit in the capability atomic unit library is selected as the target capability atomic unit; Extract the metadata of the target capability atomic unit, and obtain the difficulty level of the target capability atomic unit based on the metadata. The difficulty level includes a basic level and an advanced level. When determining the difficulty level as the base level, the first training order is taken as the target training order; When the difficulty level is determined to be high level, the second training order is used as the target training order, and the training order includes the first training order and the second training order.

[0008] Optionally, in one possible implementation of the first aspect, the step of sequentially executing corresponding training steps according to the target training order, and performing evaluation and adaptive decision-making in each training step to obtain the step training state includes: The user's response training data is obtained based on the current training steps; The corresponding cognitive level is determined according to the training steps, and the cognitive diagnostic strategy is retrieved based on the cognitive level to perform cognitive scoring on the answer training data to obtain an evaluation score. The updated mastery level corresponding to the cognitive level is obtained based on the assessment score; Based on the updated mastery, adaptive decisions are made for the training steps to obtain the training state of each step.

[0009] Optionally, in one possible implementation of the first aspect, obtaining the user's answer training data based on the current training step includes: When it is determined that the training step is not a discussion or practice step, the corresponding training content is retrieved and displayed. Based on the training content and the intensity of the prompt, a prompt question is generated, and the user's answer training data is received based on the prompt question. When the training step is determined to be a discussion practice step, training practice data is allocated based on the user's level by retrieving independent practice units, and independent practice data of the user is received based on the training practice data. Users are grouped into discussion groups, discussion content is generated based on the discussion groups and training content, and discussion responses from each user in the discussion group are received based on the discussion content. The answer training data is obtained based on the independent practical data and the discussion responses.

[0010] Optionally, in one possible implementation of the first aspect, the step of retrieving a cognitive diagnostic strategy based on the cognitive level to perform cognitive scoring on the answer training data to obtain an evaluation score includes: The cognitive diagnostic algorithm is used to perform structured scoring on the answer training data to obtain a structured cognitive score; When the cognitive level is determined to be the first cognitive level, the structural cognitive score is used as the evaluation score corresponding to the first cognitive level. When the cognitive level is determined to be the second cognitive level, the answer training data is reviewed to obtain the review result. The review result and the structural cognitive score are weighted and fused to obtain the evaluation score of the second cognitive level. The cognitive level includes the first cognitive level and the second cognitive level.

[0011] Optionally, in one possible implementation of the first aspect, obtaining the updated mastery level corresponding to the cognitive level based on the assessment score includes: When the cognitive level is determined to be the first cognitive level, the updated mastery of the first cognitive level is obtained based on the assessment score and the historical mastery of the corresponding cognitive level. When the cognitive level is determined to be the second cognitive level, the discussion answers of each user in the discussion group are displayed based on the discussion content, and the multi-dimensional feedback results corresponding to each user are obtained according to the discussion answers and the evaluation dimension list. Based on the assessment scores and multidimensional feedback results, the updated mastery level of the second cognitive level is obtained.

[0012] Optionally, in one possible implementation of the first aspect, the step of making adaptive decisions on the training steps based on the updated mastery to obtain the step training state includes: When the updated mastery is determined to be greater than or equal to the first threshold, the next training step is executed based on the target training order; If the update mastery is determined to be less than the first threshold and greater than or equal to the second threshold, and the number of attempts is less than the maximum number of attempts, the number of attempts is incremented to obtain the current number of attempts; Adjust the cue strength based on the current number of attempts, and repeat the above steps of obtaining updated mastery and making adaptive decisions within the current training step according to the cue strength; If the mastery level is determined to be less than the first threshold but greater than or equal to the second threshold, and the number of attempts is greater than or equal to the maximum number of attempts, then revert to the basic content of the current training step. If the updated mastery level is less than the second threshold, revert to the basic content of the current training step. Record the practice data corresponding to the training steps, and obtain the training status of the steps based on the current number of attempts and the practice data.

[0013] Optionally, in one possible implementation of the first aspect, the persistent storage of the training state based on the training state in the step includes: The training states of the target capability atomic units are obtained by statistical analysis of the training states in the above steps. The user's training state is updated based on the unit training states, and the current training state is persistently stored.

[0014] Optionally, in one possible implementation of the first aspect, the method of implementing cross-session recovery based on the interval of days between new sessions includes: The knowledge retention rate is obtained based on the interval in days and the stability coefficient. When the knowledge retention rate is determined to be less than a preset threshold, the corresponding capability atom unit is marked as needing review. The restoration and display are performed based on the atomic units that retrieve the capability of the state to be reviewed.

[0015] A second aspect of this invention provides an adaptive intelligent training agent system based on capability atomic units, comprising: The sequence selection module is used to determine the target capability atomic unit based on the user's learning request, and select the training order according to the difficulty level of the target capability atomic unit to obtain the target training order; The training execution module is used to execute the corresponding training steps sequentially according to the target training order, and to perform evaluation and adaptive decision-making in each training step to obtain the step training state. The storage recovery module is used to determine that all training steps have been completed, persistently store the training state based on the training state of the steps, and realize cross-session recovery based on the number of days between new sessions.

[0016] A third aspect of the present invention provides an electronic device, comprising: a memory, a processor, and a computer program, wherein the computer program is stored in the memory, and the processor executes the computer program to perform the methods described in the first aspect of the present invention and various possible methods related to the first aspect.

[0017] A fourth aspect of the present invention provides a storage medium storing a computer program, which, when executed by a processor, is used to implement the first aspect of the present invention and various methods possibly involved in the first aspect.

[0018] The beneficial effects of this invention are as follows: 1. This invention can execute a customized training sequence based on the user's actual training modules, improving the training efficiency of the user's transfer learning ability. Specifically, this invention can use a sequence selection module to retrieve the corresponding ability atomic units according to the user's learning request, and select a customized training sequence as the target training sequence based on the difficulty level of the ability atomic units. Then, it executes each training step in the training execution module according to the target training sequence, so that different execution sequences are adopted for different teaching and training units, conforming to the user's cognitive learning patterns and facilitating the improvement of the training efficiency of the user's transfer learning ability.

[0019] 2. This invention can implement multiple stages, including independent practice, group discussion, and peer evaluation, in the core practical discussion and practice steps. Within the training steps, it makes adaptive decisions based on the user's mastery level at the corresponding cognitive level, improving the internalization efficiency of the user's transfer ability. Specifically, this invention can make adaptive decisions regarding the user's training process based on the updated mastery level of the current training step.

[0020] 3. This invention can realize the storage and cross-session recovery of training data throughout the entire training process through the storage and recovery module. Specifically, this invention can first summarize the training status of each step to form a unit training status, complete the persistent storage of the overall training progress of the learner, avoid the loss of training data, and prioritize the push of review content according to the knowledge retention rate from low to high, so as to realize the connection of training progress and targeted review. Attached Figure Description

[0021] Figure 1 A flowchart illustrating an adaptive intelligent training agent method based on capability atomic units provided by this invention; Figure 2 A flowchart of an execution order selection module provided by the present invention; Figure 3 A flowchart for performing adaptive decision-making in each training step based on updated mastery, provided by the present invention; Figure 4 A schematic diagram of the structure of an adaptive intelligent training agent system based on capability atomic units provided by the present invention; Figure 5 This is a schematic diagram of the hardware structure of an electronic device provided by the present invention. Detailed Implementation

[0022] 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, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0024] It should be noted that this invention is not limited to programming education scenarios. Its six-step training framework is designed based on general cognitive principles and can be widely applied to various fields requiring skills training, including but not limited to programming education, vocational skills training, corporate training, K-12 education, medical training, and legal practice. The implementation methods in each field are similar to those in programming education; simply replacing the training content with the professional knowledge and cases of the corresponding field will achieve the technical effects of this invention.

[0025] The six steps constitute the skills training process. Its core is to guide users through a complete closed loop of problem introduction, practical demonstration, thought process breakdown, discussion and practice, knowledge point organization, and terminology explanation to internalize skills. Step four, discussion and practice, is the core practical component, where users complete multiple sub-steps including independent practice, group discussion, and mutual evaluation. The other five steps lay the groundwork and support for this core practical component. Furthermore, all skills units do not execute these six steps in the same order; instead, they are pre-matched according to the difficulty level of each skill unit. That is, each skill unit can be pre-classified by the course designer in terms of difficulty level and bound to a corresponding training order. This differentiated process design aligns with cognitive principles: basic skills require imitation before understanding, while advanced skills can be challenged first and then verified. Regardless of the order, the complete set of six steps is always preserved, ensuring that the training of each skill unit covers the entire process from motivation to terminology solidification. Knowledge transfer serves only as an auxiliary means of skills training, interspersed throughout the various steps.

[0026] like Figure 1 As shown, the present invention provides a flowchart of an adaptive intelligent training agent method based on capability atomic units. This adaptive intelligent training agent method based on capability atomic units includes: S1, determine the target capability atomic unit based on the user's learning request, select the training order according to the difficulty level of the target capability atomic unit, and obtain the target training order, which includes a first training order and a second training order.

[0027] It should be noted that traditional intelligent training agents generally use a fixed sequence of steps for teaching, which cannot adapt to the cognitive patterns of ability atomic units with different difficulty levels. In other words, they cannot adapt different training paths based on the difficulty of the metadata corresponding to the ability atomic unit. Figure 2As shown, the sequence selection module in the Agent system of the present invention can determine the target capability atomic unit according to the user's learning request and obtain the difficulty level of the target capability atomic unit, thereby determining the target training order according to the difficulty level. When the difficulty level is basic, the first training order is selected as the target training order. When the difficulty level is advanced, the second sequence order is selected as the target training order, so that the training path conforms to the cognitive law and maximizes the improvement of the user's learning effect.

[0028] Understandably, the sequence selection module is a module for determining the execution order of training steps. It is a functional module used to analyze the difficulty level of the capability atomic unit and select the corresponding training order. The learning request is an instruction request initiated by the user to select a capability atomic unit for training. The target capability atomic unit is the capability atomic unit that needs to be trained. The difficulty level is a level that classifies the difficulty of the capability atomic unit. It is preset and includes at least a basic level and an advanced level. The training order is the order in which the six steps in the capability atomic unit are executed, including the first training order and the second training order. The target training order is the training order that corresponds to and matches the target capability atomic unit.

[0029] The first training sequence corresponds to the basic level, namely, problem introduction, practical demonstration, problem breakdown, discussion and practice, knowledge point organization, and terminology explanation. The second training sequence corresponds to the advanced level, namely, problem introduction, problem breakdown, practical demonstration, discussion and practice, knowledge point organization, and terminology explanation.

[0030] In some embodiments, step S1 includes: S11, select the corresponding capability atom unit in the capability atom unit library as the target capability atom unit based on the user's learning request.

[0031] It is understandable that the capability atom unit library is a database that stores multiple capability atom units. It can be pre-built. The capability atom unit is a pre-built minimum capability training unit with teaching functions. It is the minimum capability training unit with standardized description and systematic teaching support.

[0032] For example, when user A uses the system, if they click to select the "Variables and Data Types" capability atom in the capability atom library, then the capability atom corresponding to "Variables and Data Types" can be used as the target capability atom.

[0033] S12, extract the metadata of the target capability atomic unit, and obtain the difficulty level of the target capability atomic unit based on the metadata. The difficulty level includes at least a basic level and an advanced level.

[0034] Understandably, the metadata of the target capability atom unit contains all the data about the corresponding capability atom unit, such as teaching content and difficulty level. Therefore, the corresponding difficulty level can be determined based on the metadata of the target capability atom unit, so that the appropriate training order can be selected based on the difficulty level.

[0035] Metadata consists of structured data representing the atomic unit attributes of capabilities, including but not limited to difficulty level, teaching content, application scenarios, and core capability types. The basic level is a pre-set level based on the difficulty of the practice, while the advanced level is a pre-set level based on the difficulty of the practice.

[0036] For example, when the variable and data type correspond to the pre-set difficulty level of the base level, the system will obtain the base difficulty level when reading the metadata of the target capability atomic unit to determine the variable and data type selected by user A.

[0037] S13, when the difficulty level is determined to be the basic level, the first training order is taken as the target training order.

[0038] It is understandable that different difficulty levels have pre-configured training sequences. Therefore, when the difficulty level is determined as the base level, the first training sequence is used as the target training sequence.

[0039] S14, when the difficulty level is determined to be high level, the second training order is used as the target training order.

[0040] Understandably, when the difficulty level is determined to be high, the second training order will be used as the target training order.

[0041] S2, execute the corresponding training steps sequentially according to the target training order, and perform evaluation and adaptive decision-making in each training step to obtain the step training state.

[0042] Understandably, once the target training sequence is determined, the entire training process can be executed through the training execution module, and an evaluation and adaptive decision-making process can be performed in each training step. This facilitates the realization of a complete process of training, evaluation, and adjustment, so as to adjust the user's state in a timely manner during the training process and improve the effectiveness of the user's training transfer ability.

[0043] The training execution module is the core operating unit of the system. It is responsible for executing training steps in the order of target training, receiving user feedback, making evaluation decisions, and outputting training status. The training steps are the capability training process steps, including six steps: problem introduction, thought process breakdown, practical demonstration, discussion and practice, knowledge point organization, and terminology explanation. The step training status is the status information output after a single training step is completed, including information such as user mastery and number of attempts.

[0044] In some embodiments, step S2 includes: S21, Obtain user response training data based on the current training step.

[0045] It is understandable that, during the execution of the current training step, the user's training answer data is received.

[0046] Among them, the training data consists of the user's responses in each training step.

[0047] In some embodiments, step S21 includes: S211, when it is determined that the training step is not a discussion practice step, the corresponding training content is retrieved and displayed, a prompt question is generated based on the training content and the prompt intensity, and the user's answer training data is received based on the prompt question.

[0048] It is understandable that the practice content and training methods corresponding to different training steps are different. Therefore, when it is determined that the training step is not a discussion practice step, the pre-stored training content can be retrieved and displayed according to the current training step. Corresponding guiding prompt questions can be generated according to the training content and prompt intensity so that users can answer the guiding prompt questions and obtain answer training data. The system receives the user's answer training data for subsequent evaluation and adjustment.

[0049] The training content consists of pre-set content data corresponding to each training step, such as professional knowledge and cases that match the current training step. The prompt intensity is the degree of guidance that leads the user to answer the prompt questions, which is pre-set. The prompt questions are questions generated by the system in response to the displayed training content, which can be pre-set to match the training content. The answer training data is the user's feedback answers to the prompt questions.

[0050] For example, when performing the first training step of importing questions: the Agent presents the training content, such as a real case: an auditor needs to process the names and ages of 100 customers, how to store this data using variables, and generates a prompt question: what problems will you encounter if you don't use variables, and receives the user's answer. The training data is: the data has to be rewritten every time, which is very troublesome.

[0051] It should be noted that each training step within the capability atom unit is divided into multiple difficulty levels, such as L1 basic, L2 advanced, and L3 challenge. The user's capability level can be obtained through pre-tests or historical learning records, so that the corresponding level of content in each training step can be automatically selected for display and training based on the user's capability level.

[0052] S212, when the training step is determined to be a discussion practice step, the user's independent practice unit is retrieved and training practice data is allocated according to the user's level, and the user's independent practice data is received based on the training practice data.

[0053] Understandably, discussion and practice are the core practical components. Due to differences in user skill levels, using uniform training content would be too difficult for basic users and too easy for advanced users. Therefore, in the discussion and practice step, users can be assigned corresponding independent practice units based on their skill level. These independent practice units can be equipped with appropriate training content and receive users' independent practice data, ensuring that each user can practice within their own ability range and improving the effectiveness of skill training.

[0054] Among them, user level refers to the user's practice level, which can be obtained through pre-tests or historical learning records. Assuming a learner is a beginner, the level is L1. Independent practice unit is a system unit for users to practice alone. That is, it assigns practice tasks of different difficulties according to user level, receives code or solutions submitted by users for real-time code review, and gradually increases the intensity of prompts according to the number of attempts. Training practice data consists of the training content and prompt questions of the independent practice part in the discussion practice steps. Independent practice data consists of the user's answers in the independent practice training session.

[0055] For example, when conducting independent practice in the core practical session, if the user's level is L1, the Agent can assign basic training content to define three variables to store your name, age, and height and print them. If the user's level is L2, the advanced training content swaps the values ​​of two variables (without using the third variable).

[0056] It is worth mentioning that, since discussion and practice are the core practical components, the core practical support module in the training execution module can provide independent practical support, group discussion guidance, and mutual evaluation mechanisms in the discussion and practice steps.

[0057] It is easy to understand that during the discussion and practice steps, trainees can try and modify their independent practice multiple times. The system records the status of each attempt and gradually increases the intensity of the prompts. However, this process data is only used for training status tracking and is not included in the evaluation score of this step. When the trainee confirms that the practice has been completed, the system will present a special evaluation question at the end of the step. This evaluation question is different from the practice question and is used to measure the trainee's true mastery of the current cognitive level. This prevents trainees from giving up exploration for fear of losing points, while ensuring the authenticity of the ability assessment.

[0058] S213, group users into discussion groups, generate discussion content based on the discussion groups and training content, and receive discussion responses from each user in the discussion group based on the discussion content.

[0059] Understandably, in order to improve training effectiveness, a group discussion guidance unit can be added to the core practical session to automatically group users, generate discussion topics, guide the discussion, and provide a discussion framework to avoid going off-topic.

[0060] Among them, the group discussion guidance unit is an intelligent unit responsible for user group management and discussion content generation. The discussion group is a group that combines multiple users for collaborative discussion. For example, user A and user B are divided into a discussion group. The discussion content is the content matched to the discussion group. It can be generated based on the training content in independent practice and the user's level. The discussion topics or tasks guide the discussion group to conduct in-depth thinking and communication. The discussion answers are the answers of the users in the discussion group to the discussion content.

[0061] It's easy to understand that through group communication, users can learn from each other and improve the efficiency of internalizing skills.

[0062] S214, Obtain answer training data based on the independent practical data and the discussion answers.

[0063] Understandably, in the discussion and practice steps, the training data includes both independent practice data and discussion responses.

[0064] S22, determine the corresponding cognitive level according to the training steps, and retrieve the cognitive diagnostic strategy based on the cognitive level to perform cognitive scoring on the answer training data to obtain an evaluation score.

[0065] Understandably, different training steps correspond to different levels of cognitive improvement for users. For example, after completing the first question-introduction training step, the system will evaluate the user's score at the comprehension level. After completing the second practical demonstration training step, the system will evaluate the user's score at the application level. The system will determine the corresponding cognitive level evaluation score for each training step in turn to assess the user's practice effect. Since the sub-steps in each training step are somewhat different, the methods for obtaining the corresponding cognitive level evaluation scores for different training steps are also different. Therefore, the cognitive level of the answer training data in each training step can be evaluated according to the cognitive diagnosis strategy, so as to determine the mastery level corresponding to each training step based on the evaluation score, and thus make adaptive adjustments within each training step.

[0066] The cognitive level refers to the level of cognition in the internalization of abilities, including evaluation, analysis, application, understanding, memory, and creation. Different training steps have corresponding matching cognitive levels, which are pre-mapped and matched. The evaluation score is the score of the cognitive level corresponding to each training step.

[0067] In some embodiments, step S22 includes: S221, invoke the cognitive diagnostic algorithm to perform structured scoring on the answer training data to obtain a structured cognitive score.

[0068] Understandably, after receiving the user's answer training data, the system can call a cognitive diagnostic algorithm to score the answer training data in order to obtain a structural cognitive score.

[0069] It's easy to understand that the cognitive diagnostic algorithm is an algorithm that uses a large language model (LLM) to perform structured evaluation of user answers based on a pre-set Bloom hierarchical evaluation matrix. The Bloom hierarchical evaluation matrix is ​​pre-set and contains feature knowledge points corresponding to each cognitive level. The LLM uses a Transformer architecture, and the core modules that play a role in the evaluation logic include a semantic encoder, a self-attention mechanism, and an inference chain module. When evaluating the answer training data, the current cognitive level and its corresponding Bloom hierarchical evaluation matrix parameters, along with the training background, question content, and user answer training data, are input into the LLM as multi-dimensional input parameters. Internally, the LLM uses a Transformer architecture to evaluate the evaluation results. The semantic encoder in the ansformer architecture transforms text into high-dimensional vectors, utilizes a self-attention mechanism to extract hidden logical operators in the answers, such as causal relationships or value judgment features, and combines them with the thought chain reasoning module to perform layer-by-layer comparisons based on the feature knowledge points defined in the Bloom hierarchical evaluation matrix. Then, the system performs multi-dimensional weighted calculations, weighting and fusing the accuracy of knowledge identified by the model, logical rigor, and achievement of the target level, and introducing hierarchical consistency penalties or reward factors for non-linear correction. Finally, it generates an evaluation report that includes both structural cognition scores and structured cognitive ability diagnoses. The structural cognition score is obtained from the evaluation report. The data processing process of the large language model is existing technology and will not be elaborated here.

[0070] The hierarchical evaluation matrix is ​​a Bloom hierarchical evaluation matrix, which can be a pre-set matrix containing scoring dimensions, scoring criteria, and weight allocation for each cognitive level. The structural cognitive score is the score corresponding to each cognitive level.

[0071] S222, when the cognitive level is determined to be the first cognitive level, the structural cognitive score is used as the evaluation score corresponding to the first cognitive level.

[0072] It should be noted that since the training data for the answers corresponding to the first cognitive level are only simple basic text answers, there is no need for system review and verification. Therefore, the obtained structural cognitive score can be directly used as the evaluation score for the corresponding first cognitive level.

[0073] The first cognitive level refers to the cognitive level corresponding to the training steps that do not require the system to review the training data. This can include the cognitive levels of understanding, application, analysis, evaluation, and memorization corresponding to problem introduction, practical demonstration, thought process breakdown, knowledge point organization, and terminology explanation.

[0074] S223, when the cognitive level is determined to be the second cognitive level, the answer training data is reviewed to obtain the review result, and the review result and the structural cognitive score are weighted and fused to obtain the evaluation score of the second cognitive level. The cognitive level includes the first cognitive level and the second cognitive level.

[0075] Understandably, the independent practice unit can receive users' answer training data, review the answer training data in real time, obtain the review results, and calculate the weighted sum of the review results and the structural cognition score according to two preset weights. That is, the result of multiplying the review result by the corresponding review weight is added to the result of multiplying the structural cognition score by the corresponding cognition weight to obtain the evaluation score of the second cognitive level.

[0076] Among them, the review prompts are the specific prompts obtained after system review, the review results are the scores for training and evaluating the answer training data after review, and can be pre-matched and set based on the deviation between the answer training data and the standard correct answer. The second cognitive level is the cognitive level that needs to be combined with the review results to calculate the evaluation score, including the innovation cognitive level.

[0077] For example, after user A submits code, the Agent can review the submitted code in real time. When a single-letter variable appears in the submitted code, the Agent can provide a review prompt: "Variable naming is not standardized," and retrieve the review result corresponding to the degree of deviation, which is 0.6.

[0078] It's easy to understand that when the user submits code as training data, the system can automatically run the code, compare the output with the expected output, and calculate a score based on the code's success rate and the degree of matching of the output format.

[0079] S23, based on the assessment score, obtain the updated mastery level corresponding to the cognitive level.

[0080] Understandably, in order to determine in real time the user's level of mastery at each cognitive level in each training step, and to make adaptive adjustments based on the user's current actual practice situation, for example, when the user's mastery of the current training content is low, the difficulty of the corresponding training step can be reduced, and when the mastery is determined to be high, the next training step can be advanced so that the user can practice.

[0081] Among them, the updated mastery level is the current mastery level after updating the historical mastery level.

[0082] In some embodiments, step S23 includes: S231, when the cognitive level is determined to be the first cognitive level, the updated mastery of the first cognitive level is obtained based on the assessment score and the historical mastery of the corresponding cognitive level.

[0083] Understandably, since there are no sub-steps such as group discussions in the training steps corresponding to the first cognitive level, the updated mastery can be calculated directly based on the assessment score of the corresponding cognitive level and the historical stored mastery.

[0084] Among them, historical mastery refers to the mastery of historical moments that have not yet been updated.

[0085] For example, the updated mastery of the corresponding cognitive level can be obtained by using the exponential moving average algorithm based on the cognitive state tracking module in the system. That is, updated mastery = learning rate × assessment score + (1 - learning rate) × historical mastery, where the learning rate is preset and can be 0.3.

[0086] S232, when the cognitive level is determined to be the second cognitive level, display the discussion answers of each user in the discussion group based on the discussion content, and obtain the multi-dimensional feedback results corresponding to each user according to the discussion answers and the evaluation dimension list.

[0087] Understandably, in order to accurately assess the learning and training status of each user, the discussion responses from each user within the discussion group can be shared and displayed to facilitate mutual learning among users and to provide multi-dimensional evaluations of group members based on the evaluation dimensional list.

[0088] The evaluation dimension list is a pre-set form with multi-dimensional evaluation information, which can include multiple dimensions such as practicality and simplicity. The multi-dimensional feedback result is the result of evaluating the discussion and answers of users in the group based on the dimension indicators in the evaluation dimension list.

[0089] S233, based on the assessment score and multidimensional feedback results, the updated mastery level of the second cognitive level is obtained.

[0090] Understandably, the second cognitive level corresponds to the creative cognitive level of the discussion and practice steps. This cognitive level needs to be combined with the multi-dimensional feedback results of the group members' evaluations in order to comprehensively assess the user's mastery of the creative cognitive level.

[0091] For example, the obtained multidimensional feedback results can be compared with a preset scoring form to determine the multidimensional score of the multidimensional feedback results, and then weighted with the evaluation score to obtain the updated mastery level of the second cognitive level.

[0092] S24, make adaptive decisions on the training steps based on the updated mastery, and obtain the training state of the steps.

[0093] It's understandable that different users may have varying levels of mastery within the same training step. For example, some users can quickly move on to the next step after one practice session, while others find their mastery low on their first attempt. This indicates that the user lacks the skills needed for training in that particular step. Therefore, they can revert to the baseline content for repeated practice to train the corresponding cognitive level. However, if the training steps are not adaptively adjusted based on updated mastery levels, users with weak foundations will fall behind, while more capable users will repeat inefficient practice. Therefore, if... Figure 3 The diagram shows a flowchart of adaptive decision-making in each training step based on updated mastery. This means that the user's current state can be determined by updating mastery, thus deciding whether the user should proceed to the next step, repeat the current step, or revert to basic content. During the execution process, practice data and the number of attempts are recorded to obtain the step training state, so as to determine the training process based on the user's actual learning state, thereby improving training efficiency.

[0094] Among them, the step training status is the training status data of each training step, including data information such as the number of repeated attempts for the same training step, mastery level, and completion status of training content.

[0095] In some embodiments, step S24 includes: S241, when it is determined that the updated mastery is greater than or equal to the first threshold, the next training step is executed based on the target training order.

[0096] Understandably, when the updated mastery level reaches or exceeds the preset threshold range, it means that the user's cognitive level ability required to be trained in the current step has met the requirements and is ready to enter the next training step. At this time, the system can unlock the next training step according to the target training order, so that the user can train the cognitive ability corresponding to the next training step, avoid ineffective repetitive practice, and improve training efficiency.

[0097] The first threshold is a baseline value for determining whether a training step can proceed to the next step. It can be preset, such as 0.8.

[0098] For example, when the Agent evaluates and calculates the training data of the answers imported in the first training step, and finds that the updated mastery of the corresponding understanding level is 0.85, which is greater than the first threshold of 0.8, it can proceed to the next training step, allowing the user to enter the second training step of practical demonstration to train the application level of ability.

[0099] S242, if it is determined that the update mastery is less than the first threshold and greater than or equal to the second threshold, and the number of attempts is less than the maximum number of attempts, the number of attempts is incremented to obtain the current number of attempts.

[0100] Understandably, when the update mastery is less than the first threshold but greater than or equal to the second threshold and the number of attempts is less than the maximum number of attempts, it indicates that the user's cognitive ability at the current training level is at a medium level but has not reached the standard requirements. With repeated practice, there is room for improvement in ability training. Therefore, the number of attempts can be incremented to obtain the current number of attempts, so that the guidance level of the training content displayed in the training steps can be adjusted according to the current number of attempts, which can help improve the user's understanding and practice efficiency.

[0101] The second threshold is a pre-set minimum judgment value, which can be 0.5. The number of attempts is the number of times the corresponding training step is repeated. The maximum number of attempts is the maximum number of times the same step is repeated, which can be pre-set, such as 8. The current number of attempts is the number of attempts after incrementing. For example, if the number of attempts stored in the system is 2, and the mastery is determined to be 0.7 after training, since 0.8>0.7>0.5 and the number of attempts is 2 less than 8, the current number of attempts is 2+1=3.

[0102] S243, adjust the cue strength based on the current number of attempts, and repeat the above steps of obtaining updated mastery and making adaptive decisions within the current training step according to the cue strength.

[0103] Understandably, in the current training step, the practice is repeated, and since the number of attempts has changed, the intensity of the prompts is increased. That is, the agent provides more detailed explanations and generates prompt questions again to ask questions. The user answers, and the answers are evaluated to obtain the updated mastery. Adaptive decisions are then made based on the updated mastery. For example, when the updated mastery after repeated training exceeds the first threshold, the next training step can be started.

[0104] For example, in the third training step's breakdown, the Agent displays a flowchart of variable assignment, breaking it down into three steps: variable declaration, assignment, and usage. It then generates a prompt question: "Please analyze the differences in how variables 'name' and 'age' are stored in memory." The user answers: "'Name' is stored as a string, and 'age' as an integer." The Agent evaluates the analysis level's mastery score at 0.70 (between 0.5 and 0.8), performs two attempts, triggering an adaptive decision-making repetition within the step. Simultaneously, the prompt strength is increased, prompting the Agent to provide a more detailed explanation: strings and integers occupy different amounts of space in memory, strings are variable-length, and the question is asked again. After the user provides further answers, the updated mastery score is 0.85 > 0.8, thus advancing to the next training step for further practice.

[0105] S244, if the update mastery is less than the first threshold and greater than or equal to the second threshold, and the number of attempts is greater than or equal to the maximum number of attempts, then revert to the basic content of the current training step.

[0106] Understandably, if a user's mastery level still hasn't reached the level required for stable mastery after reaching the maximum number of attempts, it means that the current training difficulty exceeds their temporary capacity to accept the material. Continuing to repeat the practice would be of limited value. Therefore, the system can revert to the basic content of the current step for further learning and consolidation.

[0107] It should be noted that the training content used by users during training differs from the basic content. The basic content consists of fundamental knowledge points that correspond to the selected ability units, while the training content focuses on knowledge that trains and cultivates the user's cognitive abilities, and its difficulty level differs from that of the basic content.

[0108] S245, if the updated mastery is less than the second threshold, revert to the basic content of the current training step.

[0109] Understandably, when the update mastery level is significantly lower than the second threshold, it indicates that the user's understanding of the current content is insufficient and that they still have the corresponding cognitive transfer ability. Therefore, the system can directly revert to the basic content for display and teach the user the basic content again.

[0110] S246, record the practice data corresponding to the training step, and obtain the step training status based on the current number of attempts and the practice data.

[0111] Understandably, the user's practice status at each training step is recorded to obtain practice data, which is then combined with the current number of attempts to form the step training status, so as to facilitate subsequent cross-session recovery.

[0112] The practice data includes all data from each training step, such as modification records and completion status. The step training status is the practice status data corresponding to each step.

[0113] S3, determine that all training steps have been completed, persist the training state based on the training state of the steps, and realize cross-session recovery based on the number of days between new sessions.

[0114] It is understandable that users' skill training is usually not completed in a single session. There may be situations where users exit midway, log in across devices, or resume training after a period of time. However, as time goes by, users will forget the training content they have already mastered. Therefore, the overall training status of users can be persistently stored by summarizing the training status of each step to ensure that training progress is not lost. At the same time, the knowledge retention rate can be calculated by combining the forgetting curve, so that the system can determine the skill atomic units that need to be reviewed based on the knowledge retention rate, and realize training recovery across sessions.

[0115] Among them, the storage recovery module is a functional module responsible for the storage and recovery of training data, the new session is a session started when the user logs back into the system, and the interval number of days is the time interval between the new session and the previous session.

[0116] In some embodiments, step S3 includes: S31, the training state of the target capability atomic unit is obtained by statistical analysis of the training state in the previous step, the user's training state is updated based on the unit training state, and the current training state is persistently stored.

[0117] Understandably, once all training processes in the training execution module have been completed, the training states corresponding to each training step can be comprehensively statistically analyzed to obtain the unit training state of the target capability atomic unit. This allows the data of the originally stored training states to be replaced with the current unit training states and persistently saved.

[0118] Among them, the unit training state is the training data containing all training steps in the capability atomic unit, the training state is the historical data stored in the system, and the current training state is the training state stored at the current moment.

[0119] S32, the knowledge retention rate is obtained based on the interval number of days and the stability coefficient.

[0120] It is understandable that the knowledge retention rate can be obtained from the forgetting formula:

[0121] in, s is the knowledge retention rate, and s is the stability coefficient. This represents the number of days between intervals.

[0122] It is easy to understand that the stability coefficient can be obtained by multiplying the user's historical mastery of the training by the preset initial coefficient. For example, the stability coefficient = initial coefficient × (1 + historical average mastery). Since the forgetting curves of different people are highly consistent in basic trends, the preset initial coefficient can be obtained by collecting a large number of users' review performance at different time intervals for big data analysis in the early stage of system testing. The big data analysis methods can be various methods such as survival analysis and cluster analysis, which will not be elaborated here.

[0123] S33, when the knowledge retention rate is determined to be less than a preset threshold, the corresponding capability atom unit is marked as pending review.

[0124] It is understandable that when the knowledge retention rate is less than the preset threshold, it means that when a new session starts, the training ability of the corresponding capability atom unit in the system may be forgotten or lost. Therefore, it is necessary to restore the training review content of the corresponding capability atom unit in a timely manner in order to strengthen the corresponding training ability and practice knowledge. Therefore, the corresponding capability atom unit with a knowledge retention rate less than the preset threshold can be marked as pending review.

[0125] Among them, the "to be reviewed" status indicates that the ability atomic unit needs to be reviewed.

[0126] It is easy to understand that, since there are a large number of capability atomic units in the system, when a user starts a new session, only the knowledge retention rate of the capability atomic units that the user has trained is calculated in order to reduce the amount of data processing.

[0127] S34, restore and display based on the atomic unit of the ability to retrieve the state to be reviewed.

[0128] Understandably, when it is determined that there are capability atomic units in the system that need to be reviewed, the corresponding capability atomic units can be displayed in a new session so that users can enter each training step in sequence according to the corresponding training order to review and train.

[0129] It is easy to understand that when there are multiple ability atomic units in a state to be reviewed, they can be sorted according to the knowledge retention rate, and the ability atomic unit in the state to be reviewed with the lowest knowledge retention rate can be retrieved first so that users can review and train in a timely manner.

[0130] See Figure 4 This is a schematic diagram of the structure of an adaptive intelligent training agent system based on capability atomic units provided in an embodiment of the present invention. The adaptive intelligent training agent system based on capability atomic units includes: The sequence selection module is used to determine the target capability atomic unit based on the user's learning request, and select the training order according to the difficulty level of the target capability atomic unit to obtain the target training order, which includes a first training order and a second training order.

[0131] The training execution module is used to execute the corresponding training steps sequentially according to the target training order, and to perform evaluation and adaptive decision-making in each training step.

[0132] The storage recovery module is used to determine that all training steps have been completed, update and persistently store the user's training state, and perform content recovery based on the forgetting curve when a new session begins.

[0133] See Figure 5 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. The electronic device 50 includes: a processor 51, a memory 52, and a computer program; wherein... The memory 52 is used to store the computer program, and the memory may also be flash memory. The computer program is, for example, an application program or functional module that implements the above method.

[0134] The processor 51 is configured to execute the computer program stored in the memory to implement the various steps performed by the device in the above method. For details, please refer to the relevant descriptions in the preceding method embodiments.

[0135] Alternatively, the memory 52 can be either standalone or integrated with the processor 51.

[0136] When the memory 52 is a device independent of the processor 51, the device may further include: Bus 53 is used to connect the memory 52 and the processor 51.

[0137] The present invention also provides a readable storage medium storing a computer program, which, when executed by a processor, is used to implement the methods provided in the various embodiments described above.

[0138] The readable storage medium can be a computer storage medium or a communication medium. A communication medium includes any medium that facilitates the transfer of computer programs from one location to another. A computer storage medium can be any available medium accessible to a general-purpose or special-purpose computer. For example, a readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application-Specific Integrated Circuit (ASIC). Alternatively, the ASIC can be located in a user equipment. Of course, the processor and the readable storage medium can also exist as discrete components in a communication device. The readable storage medium can be a read-only memory (ROM), random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0139] The present invention also provides a program product including executable instructions stored in a readable storage medium. At least one processor of the device can read the executable instructions from the readable storage medium, and the at least one processor executes the executable instructions to cause the device to implement the methods provided in the various embodiments described above.

[0140] In the embodiments of the above-described device, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; 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 or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An adaptive intelligent training agent method based on capability atomic units, characterized in that, include: The target capability atomic units are determined based on the user's learning request, and the training order is selected according to the difficulty level of the target capability atomic units to obtain the target training order. According to the target training order, the corresponding training steps are executed sequentially, and evaluation and adaptive decision-making are performed in each training step to obtain the step training state; Once all training steps have been completed, the training state is persistently stored based on the training state of each step, and cross-session recovery is achieved based on the number of days between new sessions.

2. The method according to claim 1, characterized in that, The process of determining the target capability atomic unit based on the user's learning request and selecting the training order according to the difficulty level of the target capability atomic unit to obtain the target training order includes: Based on the user's learning request, the corresponding capability atomic unit in the capability atomic unit library is selected as the target capability atomic unit; Extract the metadata of the target capability atomic unit, and obtain the difficulty level of the target capability atomic unit based on the metadata. The difficulty level includes a basic level and an advanced level. When determining the difficulty level as the base level, the first training order is taken as the target training order; When the difficulty level is determined to be high level, the second training order is used as the target training order, and the training order includes the first training order and the second training order.

3. The method according to claim 1, characterized in that, The step of sequentially executing corresponding training steps according to the target training order, and performing evaluation and adaptive decision-making in each training step to obtain the step training state includes: The user's response training data is obtained based on the current training steps; The corresponding cognitive level is determined according to the training steps, and the cognitive diagnostic strategy is retrieved based on the cognitive level to perform cognitive scoring on the answer training data to obtain an evaluation score. The updated mastery level corresponding to the cognitive level is obtained based on the assessment score; Based on the updated mastery, adaptive decisions are made for the training steps to obtain the training state of each step.

4. The method according to claim 3, characterized in that, The training data obtained based on the current training step, including the user's answer, includes: When it is determined that the training step is not a discussion or practice step, the corresponding training content is retrieved and displayed. Based on the training content and the intensity of the prompt, a prompt question is generated, and the user's answer training data is received based on the prompt question. When the training step is determined to be a discussion practice step, training practice data is allocated based on the user's level by retrieving independent practice units, and independent practice data of the user is received based on the training practice data. Users are grouped into discussion groups, discussion content is generated based on the discussion groups and training content, and discussion responses from each user in the discussion group are received based on the discussion content. The answer training data is obtained based on the independent practical data and the discussion responses.

5. The method according to claim 4, characterized in that, The cognitive scoring of the response training data based on the cognitive level retrieval cognitive diagnostic strategy yields an evaluation score, including: The cognitive diagnostic algorithm is used to perform structured scoring on the answer training data to obtain a structured cognitive score; When the cognitive level is determined to be the first cognitive level, the structural cognitive score is used as the evaluation score corresponding to the first cognitive level. When the cognitive level is determined to be the second cognitive level, the answer training data is reviewed to obtain the review result. The review result and the structural cognitive score are weighted and fused to obtain the evaluation score of the second cognitive level. The cognitive level includes the first cognitive level and the second cognitive level.

6. The method according to claim 5, characterized in that, The process of obtaining the updated mastery level corresponding to the cognitive level based on the assessment score includes: When the cognitive level is determined to be the first cognitive level, the updated mastery of the first cognitive level is obtained based on the assessment score and the historical mastery of the corresponding cognitive level. When the cognitive level is determined to be the second cognitive level, the discussion answers of each user in the discussion group are displayed based on the discussion content, and the multi-dimensional feedback results corresponding to each user are obtained according to the discussion answers and the evaluation dimension list. Based on the assessment scores and multidimensional feedback results, the updated mastery level of the second cognitive level is obtained.

7. The method according to claim 6, characterized in that, The step of making adaptive decisions on training steps based on the updated mastery to obtain the step training state includes: When the updated mastery is determined to be greater than or equal to the first threshold, the next training step is executed based on the target training order; If the update mastery is determined to be less than the first threshold and greater than or equal to the second threshold, and the number of attempts is less than the maximum number of attempts, the number of attempts is incremented to obtain the current number of attempts; Adjust the cue strength based on the current number of attempts, and repeat the above steps of obtaining updated mastery and making adaptive decisions within the current training step according to the cue strength; If the mastery level is determined to be less than the first threshold but greater than or equal to the second threshold, and the number of attempts is greater than or equal to the maximum number of attempts, then revert to the basic content of the current training step. If the updated mastery level is less than the second threshold, revert to the basic content of the current training step. Record the practice data corresponding to the training steps, and obtain the training status of the steps based on the current number of attempts and the practice data.

8. The method according to claim 7, characterized in that, The step of persistently storing the training state based on the training state in the above steps includes: The training states of the target capability atomic units are obtained by statistical analysis of the training states in the above steps. The user's training state is updated based on the unit training states, and the current training state is persistently stored.

9. The method according to claim 8, characterized in that, The method for cross-session recovery based on the interval of days between new sessions includes: The knowledge retention rate is obtained based on the interval in days and the stability coefficient. When the knowledge retention rate is determined to be less than a preset threshold, the corresponding capability atom unit is marked as needing review. The restoration and display are performed based on the atomic units that retrieve the capability of the state to be reviewed.

10. An adaptive intelligent training agent system based on capability atomic units, characterized in that, include: The sequence selection module is used to determine the target capability atomic unit based on the user's learning request, and select the training order according to the difficulty level of the target capability atomic unit to obtain the target training order, which includes a first training order and a second training order. The training execution module is used to execute the corresponding training steps sequentially according to the target training order, and to perform evaluation and adaptive decision-making in each training step; The storage recovery module is used to determine that all training steps have been completed, update and persistently store the user's training state, and perform content recovery based on the forgetting curve when a new session begins.

Images