Course structure self-adaptive optimization method, system, electronic device and storage medium

Abstract

The application provides a course structure adaptive optimization method and system, electronic equipment and a storage medium, and relates to the technical field of education and artificial intelligence. The application realizes real-time adaptive control of the course structure of a single learner on the basis of an existing course structure, in combination with a learner portrait, ability estimation and cognitive load estimation, realizes fine-grained personalized control of the course structure, and systematically calculates the support strength parameter of a dynamic support / gradual disappearance strategy, thereby realizing dynamic control of the cognitive load.

Description

Technical Field

[0001] This invention relates to the fields of educational technology and artificial intelligence technology, and in particular to a method, system, electronic device and storage medium for adaptive optimization of curriculum structure. Background Technology

[0002] With the development of the internet and mobile devices, online courses, corporate training, and lifelong learning platforms have rapidly become widespread. Many teachers and training institutions are using various course-building tools to convert documents, web pages, videos, and other content into structured courses. Some platforms already possess capabilities similar to "Files-to-Course," allowing them to extract knowledge points from multiple sources and generate course outlines during the course design phase, or to align course modules according to competency standards.

[0003] However, research has found that current technology still has the following technical shortcomings in the online delivery phase of the learning process, that is, when learners actually use the course to learn: 1. Course structures are generally static and lack fine-grained adaptation to individual learner states.

[0004] Most existing online courses use a uniform syllabus and a fixed chapter order. Even if some systems offer "customizable paths" and "chapter skipping," the choice is still primarily made by the learner, and the system itself lacks the ability to adapt to the learner's progress. Specifically, this manifests as: all learners receiving the same knowledge unit divisions and durations; a fixed ratio of explanation to practice; a fixed intensity of examples / hints; and fixed review points. This can lead to overload for learners with weak foundations, while those with stronger foundations may find it redundant and boring.

[0005] 2. Cognitive load is not systematically modeled, making it impossible to dynamically adjust the curriculum structure based on cognitive load.

[0006] Cognitive load theory suggests that both excessively high and low workloads on working memory during learning tasks can negatively impact learning outcomes. Traditional online courses often only control the "duration of lessons" at a macro level, lacking mechanisms for estimating and regulating the actual cognitive load during the learning process. Even when there are adaptive difficulty levels for individual questions, they are mostly simple rules, such as "a certain number of consecutive wrong answers will result in a lower level," without dynamically adjusting the granularity, practice density, and support intensity at the course structure level.

[0007] 3. The capability estimation method is simple and does not fully utilize IRT / 3PL and knowledge tracing models.

[0008] Many systems only roughly assess learners' levels based on accuracy rates, failing to incorporate the 3-Parameter Logistic Model (3PL) from Item Response Theory (IRT) to model question discrimination, difficulty, and guessing parameters. They also lack the ability to decompose learners' capabilities into knowledge point dimensions and continuously track the mastery probability of each knowledge point through a knowledge tracking model. Consequently, these systems only know "correct / incorrect" answers, lacking a more refined, structured understanding of the "learner-knowledge point-question" three-way relationship, making it difficult to support sophisticated adaptive learning.

[0009] 4. There is a lack of unified modeling and optimization mechanisms between learners' explicit profiles, implicit competence states, and cognitive load.

[0010] Existing learner profiles typically only include static attributes such as age, education level, occupation, and region, as well as simple learning behavior statistics, such as study time and number of logins. This information lacks linkage with learners' actual abilities in each knowledge point and their current cognitive load, and there is no unified representation of learner status to optimize it by linking it with course structure parameters (unit granularity, practice ratio, number of examples, support strength, and review pace).

[0011] 5. The dynamic support / fading strategy has not been systematized into computable structural-level parameters.

[0012] In teaching practice, teachers often employ a "worked examples + fading" strategy: initially providing numerous complete examples and detailed step-by-step explanations, then gradually reducing prompts and increasing independent practice. This helps control cognitive load at different stages. However, existing online platforms typically do not abstract this strategy into system-level support strength parameters, nor do they automatically "fade" based on learners' mastery and cognitive load levels. Teachers still need to manually design the ratio of examples to practice. Summary of the Invention

[0013] The purpose of this invention is to provide a course structure adaptive optimization method, system, electronic device and storage medium to at least solve one of the above-mentioned problems existing in the existing system.

[0014] In a first aspect, the present invention provides a method for adaptive optimization of course structure, comprising: The learner state vector is constructed by integrating at least the learner profile vector constructed based on learner static profile data and historical behavior data, the learner's ability estimation result and knowledge mastery state vector obtained based on IRT / 3PL, and the cognitive load estimate obtained based on learning behavior log prediction. Based on the learner state vector and the course graph of the target course, a multi-objective optimization decision model is used to determine the next learning unit and its corresponding course structure parameter configuration from the set of candidate knowledge units; the course structure parameters include at least the support strength parameter. Based on the learner's knowledge mastery state vector and cognitive load estimate in the learner's state vector, the support strength parameter for the next learning unit is dynamically adjusted. Based on the adjusted course structure parameter configuration for the next learning unit, generate and present the learning content for the next learning unit. Based on the learner's newly generated behavioral data in the next learning unit, the learner's state vector is updated, and the course graph based on the learner's state vector and the target course is re-executed. Through a multi-objective optimization decision model, the steps of determining the next learning unit and its corresponding course structure parameter configuration from the candidate knowledge unit set are carried out until the learning objective is achieved, so as to realize closed-loop adaptive optimization.

[0015] In an optional implementation, the learner state vector is constructed, including: Acquire learners' static profile data and historical behavior data to construct learner profile vectors; the static profile data includes basic attribute information, learning objectives, and time constraint information. Obtain the course diagram and course structure parameter definitions for the target course; wherein, the course structure parameters for each knowledge unit in the course diagram shall include at least the granularity parameter, practice ratio parameter, example ratio parameter, support strength parameter, review parameter, and difficulty level parameter; The IRT / 3PL is used to calibrate the parameters of questions related to the target course in the pre-set question bank, and the learner's overall ability parameters and knowledge point-level local ability parameters are estimated by combining historical answer records in historical behavior data to obtain ability estimation results. The knowledge mastery state vector of each knowledge point at each time step is determined by the knowledge tracing model. Based on the learning behavior logs in historical behavioral data, the cognitive load sequence is determined by the cognitive load prediction model. The cognitive load sequence includes the cognitive load estimates of the most recent preset number of knowledge units. Learner state vectors are constructed based on learner profile vectors, ability estimation results, knowledge mastery state vectors, and cognitive load sequences.

[0016] In an optional implementation, the cognitive load prediction model is a supervised learning model or a two-stream neural network model; The knowledge mastery state vector for each knowledge point at each time step is determined using a knowledge tracing model, including: The answer sequence, ordered by time, corresponding to the historical answer records is input into the knowledge tracing model to obtain the knowledge mastery state vector corresponding to each knowledge point; wherein, the knowledge tracing model is a deep knowledge tracing model based on a recurrent neural network or Transformer architecture.

[0017] In an optional implementation, based on the learner's state vector and the course graph of the target course, a multi-objective optimization decision model is used to determine the next learning unit and its corresponding course structure parameter configuration from the set of candidate knowledge units, including: Based on the knowledge mastery state vector in the learner's state vector and the curriculum graph, determine the set of candidate knowledge units that can be learned at present; For each candidate knowledge unit in the candidate knowledge unit set, generate different course structure parameter configurations as a candidate action set; Based on the learner's state vector, the impact of each candidate action in the candidate action set on the expected knowledge gain, expected cognitive load, and time cost is evaluated through a multi-objective optimization function. The optimal candidate action is then selected from the candidate action set as the next learning unit and its course structure parameter configuration.

[0018] In an optional implementation, the multi-objective optimization function is: ; in, Learner state vector s t Candidate knowledge units u Course structure parameter configuration p u The objective function value is below. For the expected knowledge gain, For the expected cognitive load, CLtarget The target cognitive load level is preset. C ( p u (Time cost) α , β , γ These are the preset weighting coefficients.

[0019] In an optional implementation, the support strength parameter for the next learning unit is dynamically adjusted based on the learner's knowledge mastery state vector and cognitive load estimate in the learner's state vector, including: The support strength parameter is defined as multiple discrete support strength levels; If the mastery probability of the first knowledge point is higher than the preset first threshold for N consecutive times and the cognitive load estimate is within the preset appropriate range, then the support strength level will be reduced in the next learning unit related to the first knowledge point; where N is a preset positive integer. If the mastery probability of the second knowledge point is lower than the preset second threshold and the cognitive load estimate is higher than the preset third threshold, then the support strength level will be maintained or increased in the next learning unit related to the second knowledge point. If the mastery probability of the third knowledge point is higher than the first threshold and the estimated cognitive load is lower than the preset fourth threshold, then the support strength level will be reduced and the proportion of self-practice will be increased in the next learning unit related to the third knowledge point.

[0020] In an optional implementation, the learning content for the next learning unit is generated and presented based on the adjusted course structure parameter configuration, including: Based on the next learning unit and its adjusted course structure parameters, a content generation template is constructed. Call the content generation module to generate or rewrite the corresponding learning content according to the content generation template to obtain the target learning content; The target learning content is presented to learners after quality verification.

[0021] Secondly, this invention provides a curriculum structure adaptive optimization system, comprising: The learner state construction module is used to construct learner state vectors. The learner state vectors are at least integrated with learner profile vectors constructed based on learner static profile data and historical behavior data, learner ability estimation results and knowledge mastery state vectors obtained based on IRT / 3PL, and cognitive load estimates obtained based on learning behavior log prediction. The adaptive curriculum structure engine module is used to determine the next learning unit and its corresponding curriculum structure parameter configuration from the set of candidate knowledge units based on the learner's state vector and the curriculum graph of the target curriculum through a multi-objective optimization decision model. The curriculum structure parameters include at least the support strength parameter. The support strength parameter of the next learning unit is dynamically adjusted according to the knowledge mastery state vector and the cognitive load estimate in the learner's state vector. The content generation and presentation module is used to generate and present the learning content for the next learning unit based on the adjusted course structure parameters configured for the next learning unit. The adaptive optimization module is used to update the learner's state vector based on the new behavioral data generated by the learner in the next learning unit, and to call the course structure adaptive engine module again until the learning objective is achieved, so as to realize closed-loop adaptive optimization.

[0022] Thirdly, the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the course structure adaptive optimization method of any of the foregoing embodiments.

[0023] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, performs the course structure adaptive optimization method of any of the foregoing embodiments.

[0024] The adaptive optimization method, system, electronic device, and storage medium for curriculum structure provided by this invention can construct a learner state vector. This learner state vector integrates at least a learner profile vector constructed from learner static profile data and historical behavioral data, learner ability estimation results and knowledge mastery state vectors obtained based on IRT / 3PL, and cognitive load estimates predicted based on learning behavior logs. Based on the learner state vector and the curriculum graph of the target curriculum, a multi-objective optimization decision model is used to determine the next learning unit and its corresponding curriculum structure parameter configuration from a set of candidate knowledge units. The curriculum structure parameters include at least a support strength parameter. The support strength parameter of the next learning unit is dynamically adjusted based on the knowledge mastery state vector and cognitive load estimate in the learner state vector. The learning content of the next learning unit is generated and presented based on the adjusted curriculum structure parameter configuration. The learner state vector is updated based on newly generated behavioral data in the next learning unit, and the steps of determining the next learning unit and its corresponding curriculum structure parameter configuration from the set of candidate knowledge units using the multi-objective optimization decision model, based on the learner state vector and the curriculum graph of the target curriculum, are re-executed until the learning objective is achieved, thus realizing closed-loop adaptive optimization.

[0025] Based on the existing curriculum structure, and combined with learner profiles, ability estimation, and cognitive load estimation, the curriculum structure for individual learners can be adaptively controlled in real time. This enables fine-grained personalized control of the curriculum structure and systematizes the dynamic support / fading strategy into calculable support strength parameters, thereby achieving dynamic control of cognitive load. Attached Figure Description

[0026] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0027] Figure 1 A flowchart illustrating an adaptive optimization method for course structure provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a curriculum structure adaptive optimization system provided in an embodiment of the present invention; Figure 3A flowchart illustrating another adaptive optimization method for course structure provided in an embodiment of the present invention; Figure 4 A schematic diagram of a two-stream neural network structure for joint modeling of knowledge tracking and cognitive load is provided in an embodiment of the present invention; Figure 5 A schematic diagram illustrating a dynamic support / fading-out strategy provided in an embodiment of the present invention; Figure 6 A schematic diagram of the system architecture of a curriculum structure adaptive optimization system provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0028] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0029] This invention provides a method, system, electronic device, and storage medium for adaptive optimization of curriculum structure based on learner profiles and cognitive load. Building upon existing curriculum structures, it combines learner profiles, ability estimation, and cognitive load estimation to perform real-time adaptive control of the curriculum structure for individual learners. This method is applicable to scenarios such as online education platforms, enterprise training platforms, and intelligent learning assistants, enabling real-time adaptive control of curriculum structure and learning paths for individual learners based on existing curriculum structures.

[0030] The purpose of this invention is to solve the following technical problems during the actual learning process of learners, without changing the initial course outline generation mode: 1. How to unify the learner explicit profile, implicit ability state based on IRT / 3PL and knowledge tracing, and cognitive load state into a computable learner state representation; 2. How to map the learner's state to course structure parameters (including unit granularity, explanation duration, practice ratio, number of examples, prompt intensity, review frequency, etc.) based on the existing course graph structure, so as to achieve personalized course structure adaptation for individual learners; 3. How to automatically implement the "dynamic support / fading" strategy based on the learner's knowledge mastery and cognitive load changes during the learning process, so as to ensure learning effectiveness while avoiding cognitive overload or underload.

[0031] To facilitate understanding of this embodiment, a course structure adaptive optimization method disclosed in this embodiment of the invention will first be described in detail.

[0032] This invention provides a method for adaptive optimization of course structure. This method can be executed by an electronic device with data processing capabilities and can be applied to online learning platforms or intelligent learning systems. See also... Figure 1 The diagram shows a flowchart of an adaptive optimization method for curriculum structure. This method, assuming the curriculum map has been generated and is available, mainly includes the following steps S110 to S150: Step S110: Construct learner state vector. The learner state vector integrates at least the learner profile vector constructed based on learner static profile data and historical behavior data, the learner's ability estimation result and knowledge mastery state vector obtained based on IRT / 3PL, and the cognitive load estimate obtained based on learning behavior log prediction.

[0033] In this embodiment, learner static profile data (such as age, education level, and learning goals), dynamic behavioral data (such as historical answer records), real-time ability levels and knowledge mastery assessed based on the IRT / 3PL model and knowledge tracing model, and cognitive load values ​​predicted from learning behavior logs are integrated into a unified, quantifiable state representation, namely, the learner state vector. This learner state vector provides precise data for subsequently automatically adjusting structural parameters such as course difficulty, content granularity, practice ratio, and prompt intensity.

[0034] In some possible embodiments, step S110 above may include the following sub-steps S111 to S115: Sub-step S111: Obtain learner static profile data and historical behavior data to construct learner profile vectors; wherein, static profile data includes basic attribute information, learning objectives and time constraint information.

[0035] It can collect static profile data such as learners' basic attribute information, learning objectives, and time constraints to form learners' static profiles; collect learners' historical behavior data on the platform, including learning duration, number of course visits, chapter completion status, and answer records, to form learners' behavioral profiles; and construct learner profile vectors based on static and behavioral profiles for subsequent adaptive decision-making.

[0036] Sub-step S112: Obtain the course diagram and course structure parameter definition of the target course; wherein, the course structure parameters of each knowledge unit in the course diagram shall include at least the granularity parameter, practice ratio parameter, example ratio parameter, support strength parameter, review parameter and difficulty level parameter.

[0037] A course diagram for the target course can be obtained from a course development system or through manual configuration. The course diagram can consist of multiple knowledge units and their prerequisite relationships. A set of adjustable course structure parameters is predefined for each knowledge unit, including but not limited to: granularity parameter g (number of knowledge points, estimated duration) and practice ratio parameters. r pr (Ratio of practice questions to explanation content), example ratio parameters r ex (Number of worked examples), support strength parameters h sup (Detail level suggestion), Review parameters f rev (Installation strategy or frequency of review units) and difficulty level parameters lvl (Including the overall difficulty of knowledge points or questions), etc. The above course structure parameters are defined as the course structure parameter space, providing adjustable dimensions for subsequent adaptive optimization.

[0038] Sub-step S113 uses IRT / 3PL to calibrate the parameters of questions related to the target course in the preset question bank, and estimates the learner's overall ability parameters and knowledge point-level local ability parameters by combining historical answer records in historical behavior data, to obtain the ability estimation results, and determines the knowledge mastery state vector of each knowledge point at each time step through the knowledge tracing model.

[0039] For IRT / 3PL-based competency estimation and knowledge tracing, 3PL parameters can be calibrated on questions related to the target course in a pre-set question bank to obtain the discrimination parameter for each question. a q Difficulty parameters b q and guessing parameters c q Based on learners' existing response records (a subset extracted from historical behavioral data), the overall ability of learners is assessed using the 3PL model. θ The model estimates the local capabilities of each knowledge point; it inputs the learner's time-ordered response sequence (a specific form of sorting existing response records by time) into the knowledge tracing model to obtain the mastery probability vector of each knowledge point at each time step. K t (i.e., the knowledge mastery state vector).

[0040] In one possible implementation, determining the knowledge mastery state vector of each knowledge point at each time step through the knowledge tracing model may include: inputting the answer sequence corresponding to the historical answer records in time order into the knowledge tracing model to obtain the knowledge mastery state vector corresponding to each knowledge point; wherein, the knowledge tracing model can be a deep knowledge tracing model based on RNN (Recurrent Neural Network) / LSTM (Long Short-Term Memory) or Transformer architecture, where LSTM is a special form of RNN.

[0041] Sub-step S114: Based on the learning behavior log in the historical behavior data, determine the cognitive load sequence through the cognitive load prediction model. The cognitive load sequence includes the cognitive load estimates of the most recent preset number of knowledge units.

[0042] When estimating cognitive load, cognitive load-related features can be extracted from learning behavior logs (real-time, fine-grained data sources that, after accumulation and aggregation, become part of historical behavior data), including task difficulty, unit dwell time, number of pauses / reviews, question answering time, error frequency and type, etc. These features are then input into the cognitive load prediction model to obtain the learner's estimated cognitive load on the current knowledge unit. CL t Maintain a sequence of cognitive loads from the most recent units within the current learning session to determine the trend of cognitive load changes. The cognitive load prediction model can be a supervised learning model or a two-stream neural network model. Sub-step S115: Based on the learner profile vector, ability estimation results, knowledge mastery state vector, and cognitive load sequence, construct the learner state vector.

[0043] In this embodiment, the learner profile vector LP, the ability estimation results (overall ability θ and knowledge point ability), and the knowledge mastery state vector are used. K t and cognitive load estimates CL t The data is then fused to form the learner's state vector. s t It is used to drive adaptive decisions on curriculum structure.

[0044] Step S120: Based on the learner's state vector and the course graph of the target course, the next learning unit and its corresponding course structure parameter configuration are determined from the candidate knowledge unit set through a multi-objective optimization decision model; the course structure parameters include at least the support strength parameters.

[0045] When making adaptive decisions about the curriculum structure, for the current set of learnable candidate knowledge units... U (Its prerequisites are) K t (The above is basically satisfied), considering different course structure parameter configurations. p u Define candidate actions ( u , p u The optimal action is selected based on a multi-objective optimization function to evaluate the impact of each candidate action on expected knowledge acquisition gain, cognitive load control, and time cost, thus deriving the next learning unit. u *and its corresponding course structure parameter configuration p u* This includes unit granularity, practice ratio, number of examples, support strength, and review schedule.

[0046] In some possible embodiments, step S120 above may include: determining a set of candidate knowledge units that can be learned currently based on the knowledge mastery state vector in the learner's state vector and the course graph; generating different course structure parameter configurations for each candidate knowledge unit in the candidate knowledge unit set as a set of candidate actions; evaluating the impact of each candidate action in the candidate action set on the expected knowledge gain, expected cognitive load, and time cost through a multi-objective optimization function according to the learner's state vector, and selecting the optimal candidate action from the candidate action set as the next learning unit and its course structure parameter configuration.

[0047] In one possible implementation, the above multi-objective optimization function can be: ; in, Learner state vector s t Candidate knowledge units u Course structure parameter configuration p u The objective function value is below. For the expected knowledge gain, For the expected cognitive load, CLtarget The target cognitive load level is preset. C ( p u (Time cost) α , β , γ These are the preset weighting coefficients.

[0048] Step S130: Based on the knowledge mastery state vector and cognitive load estimate in the learner's state vector, dynamically adjust the support strength parameter for the next learning unit.

[0049] This embodiment incorporates dynamic support and fading strategies, which can adjust the support intensity based on the learner's mastery probability and cognitive load for each knowledge point. h sup In the initial stage, or when mastery is low and cognitive load is high, a higher support intensity (providing complete examples and detailed hints) is adopted. As mastery improves and cognitive load stabilizes within an appropriate range, the support intensity is gradually reduced, and examples and hints are decreased to achieve fading. When mastery is high and cognitive load is relatively low, the support intensity is further reduced and the proportion of independent practice is increased. The updated support intensity is then combined with other course structure parameters to form the final course structure parameter configuration for the current knowledge unit (i.e., the next learning unit).

[0050] In some possible embodiments, step S130 above may include: The support strength parameter is defined as multiple discrete support strength levels; If the mastery probability of the first knowledge point is higher than the preset first threshold for N consecutive times and the cognitive load estimate is within the preset appropriate range, then the support strength level will be reduced in the next learning unit related to the first knowledge point; where N is a preset positive integer. If the mastery probability of the second knowledge point is lower than the preset second threshold and the cognitive load estimate is higher than the preset third threshold, then the support strength level will be maintained or increased in the next learning unit related to the second knowledge point. If the mastery probability of the third knowledge point is higher than the first threshold and the estimated cognitive load is lower than the preset fourth threshold, then the support strength level will be reduced and the proportion of self-practice will be increased in the next learning unit related to the third knowledge point.

[0051] The first threshold is greater than the second threshold, and the third threshold is greater than the fourth threshold. A suitable range is the area between the fourth and third thresholds. The first, second, third, and fourth thresholds can all be set according to actual needs; no specific restrictions are imposed here.

[0052] Step S140: Generate and present the learning content of the next learning unit based on the adjusted course structure parameter configuration of the next learning unit.

[0053] When generating and presenting adaptive learning content, the knowledge units obtained in step S120 can be used as a basis. u *and the adjusted course structure parameters in step S130 p u*The system constructs a content generation template, specifying the target knowledge points, difficulty level, expected explanation length, number of examples / exercises, and prompt types. It then calls the content generation module (which can be implemented through the Large Language Model LLM interface) to generate or rewrite the explanation text, example questions, exercises, and analysis content for that knowledge unit. The system performs basic verification on the generated target learning content (such as whether the knowledge point annotations are consistent and whether the difficulty is within the predetermined range). After the verification is passed, the content is presented to the learner on the learning terminal.

[0054] Based on this, in some possible embodiments, step S140 above may include: constructing a content generation template according to the next learning unit and its adjusted course structure parameter configuration; calling the content generation module to generate or rewrite the corresponding learning content according to the content generation template to obtain the target learning content; and presenting the target learning content to the learner after quality verification.

[0055] Step S150: Based on the learner's newly generated behavioral data in the next learning unit, update the learner's state vector, and re-execute the course graph based on the learner's state vector and the target course. Through a multi-objective optimization decision model, determine the next learning unit and its corresponding course structure parameter configuration from the candidate knowledge unit set until the learning objective is achieved, so as to realize closed-loop adaptive optimization.

[0056] This step involves state updating and iterative execution. After the learner completes the learning of a knowledge unit, their behavioral data and response records are recorded, and the following are updated accordingly: the ability estimate in the IRT model and the knowledge mastery status in the knowledge tracing model. K t+1 Cognitive load estimation CL t+1 ; thereby updating the learner's state vector s t+1 Then repeat steps S120 to S140 until the entire course is completed or the predetermined learning objective is achieved.

[0057] The adaptive optimization method for course structure provided in this invention, based on the existing course structure, combines learner profiles, ability estimation, and cognitive load estimation to perform real-time adaptive control of the course structure for individual learners, realizing fine-grained personalized control of the course structure, and systematizing the dynamic support / fading strategy into a computable support strength parameter, thereby realizing dynamic control of cognitive load.

[0058] Corresponding to the above-described adaptive optimization method for curriculum structure, this embodiment of the invention also provides an adaptive optimization system for curriculum structure. This system can be deployed on a server or cloud platform and is used to implement the aforementioned adaptive optimization method for curriculum structure. See also... Figure 2The diagram shows a structural schematic of a curriculum structure adaptive optimization system, which includes: The learner state construction module 201 is used to construct a learner state vector. The learner state vector integrates at least the learner profile vector constructed based on the learner's static profile data and historical behavior data, the learner's ability estimation result and knowledge mastery state vector obtained based on IRT / 3PL, and the cognitive load estimate obtained based on learning behavior log prediction. The adaptive curriculum structure engine module 202 is used to determine the next learning unit and its corresponding curriculum structure parameter configuration from the set of candidate knowledge units based on the learner's state vector and the curriculum graph of the target curriculum through a multi-objective optimization decision model; the curriculum structure parameters include at least the support strength parameter; and the support strength parameter of the next learning unit is dynamically adjusted according to the knowledge mastery state vector and the cognitive load estimate in the learner's state vector. The content generation and presentation module 203 is used to generate and present the learning content of the next learning unit based on the adjusted course structure parameter configuration of the next learning unit. The adaptive optimization module 204 is used to update the learner's state vector based on the new behavioral data generated by the learner in the next learning unit, and to call the course structure adaptive engine module again until the learning objective is achieved, so as to realize closed-loop adaptive optimization.

[0059] The adaptive optimization system for course structure provided in this invention, based on the existing course structure, combines learner profiles, ability estimation, and cognitive load estimation to perform real-time adaptive control of the course structure for individual learners, realizing fine-grained personalized control of the course structure, and systematizing the dynamic support / fading strategy into calculable support strength parameters, thereby realizing dynamic control of cognitive load.

[0060] Furthermore, the learner state construction module 201 is specifically used for: acquiring learner static profile data and historical behavior data to construct a learner profile vector; wherein, the static profile data includes basic attribute information, learning objectives, and time constraint information; acquiring the course diagram and course structure parameter definitions of the target course; wherein, the course structure parameters of each knowledge unit in the course diagram include at least granularity parameters, practice ratio parameters, example ratio parameters, support strength parameters, review parameters, and difficulty level parameters; using IRT / 3PL to calibrate the parameters of questions related to the target course in the preset question bank, and combining the historical answer records in the historical behavior data to estimate the learner's overall ability parameters and knowledge point-level local ability parameters to obtain ability estimation results, and determining the knowledge mastery state vector of each knowledge point at each time step through the knowledge tracing model; determining the cognitive load sequence through the cognitive load prediction model based on the learning behavior log in the historical behavior data, the cognitive load sequence including the cognitive load estimates of the most recent preset number of knowledge units; and constructing the learner state vector based on the learner profile vector, ability estimation results, knowledge mastery state vector, and cognitive load sequence.

[0061] Furthermore, the cognitive load prediction model is a supervised learning model or a two-stream neural network model; the learner state construction module 201 is also used to: input the answer sequence corresponding to the historical answer record in time order into the knowledge tracking model to obtain the knowledge mastery state vector corresponding to each knowledge point; wherein, the knowledge tracking model is a deep knowledge tracking model based on a recurrent neural network or Transformer architecture.

[0062] Furthermore, the aforementioned adaptive curriculum structure engine module 202 is specifically used for: determining the set of candidate knowledge units that can be learned at present based on the knowledge mastery state vector in the learner's state vector and the curriculum graph; generating different curriculum structure parameter configurations for each candidate knowledge unit in the candidate knowledge unit set as a set of candidate actions; evaluating the impact of each candidate action in the candidate action set on the expected knowledge gain, expected cognitive load, and time cost through a multi-objective optimization function according to the learner's state vector, and selecting the optimal candidate action from the candidate action set as the next learning unit and its curriculum structure parameter configuration.

[0063] Furthermore, the above multi-objective optimization function can be: ; in, Learner state vector s t Candidate knowledge units u Course structure parameter configuration p u The objective function value is below. For the expected knowledge gain, For the expected cognitive load, CLtarget The target cognitive load level is preset. C ( p u (Time cost) α , β , γ These are the preset weighting coefficients.

[0064] Furthermore, the aforementioned curriculum structure adaptive engine module 202 is also used to: define the support strength parameter as multiple discrete support strength levels; if the mastery probability of the first knowledge point is higher than a preset first threshold for N consecutive times and the cognitive load estimate is within a preset appropriate range, then the support strength level is reduced in the next learning unit related to the first knowledge point; where N is a preset positive integer; if the mastery probability of the second knowledge point is lower than a preset second threshold and the cognitive load estimate is higher than a preset third threshold, then the support strength level is maintained or increased in the next learning unit related to the second knowledge point; if the mastery probability of the third knowledge point is higher than the first threshold and the cognitive load estimate is lower than a preset fourth threshold, then the support strength level is reduced and the proportion of independent practice is increased in the next learning unit related to the third knowledge point.

[0065] Furthermore, the aforementioned content generation and presentation module 203 is specifically used for: constructing a content generation template based on the next learning unit and its adjusted course structure parameter configuration; calling the content generation module to generate or rewrite the corresponding learning content according to the content generation template to obtain the target learning content; and presenting the target learning content to learners after quality verification.

[0066] This invention also provides another structure for the above-mentioned adaptive optimization system for course structure, including: 1. Learner profile building module, used to collect learner basic information, learning goals and learning behavior data, and build and update learner profile vectors.

[0067] 2. The course structure management module is used to obtain course diagrams from external course design systems or manual configurations, and to define and store adjustable course structure parameters for each knowledge unit.

[0068] 3. The Ability Estimation and Knowledge Tracking module is used to calibrate the 3PL parameters of the questions and estimate the ability parameters and the mastery probability of each knowledge point based on the learner's answer records.

[0069] 4. Cognitive load estimation module, used to extract cognitive load features from learning behavior and output the cognitive load values ​​of the current and most recent knowledge units through the cognitive load prediction model.

[0070] 5. Learner State Fusion Module, which integrates learner profile vectors, ability estimation results, knowledge mastery status, and cognitive load status into a unified learner state vector.

[0071] 6. The adaptive course structure engine module is used to evaluate and select candidate knowledge units and their course structure parameters based on the current learner status and course graph, determine the next learning unit and course structure parameter configuration, and execute dynamic support / fading strategies.

[0072] 7. Content generation and presentation module, used to generate or rewrite course content based on the decisions of the structure adaptive engine, and display it on the learning terminal, while collecting new learning behavior data.

[0073] 8. Data storage and model service module, used to store learner profile data, course graph and course structure parameter data, question bank data, learning behavior log data, and provide online inference services for IRT model, knowledge tracing model, cognitive load model and adaptive decision strategy model.

[0074] The above modules can be implemented as software modules or as a combination of hardware circuits and software. The modules interact with each other via a bus or network.

[0075] The course structure adaptive optimization system provided in this embodiment has the same implementation principle and technical effects as the aforementioned course structure adaptive optimization method embodiment. For the sake of brevity, any parts not mentioned in the course structure adaptive optimization system embodiment can be referred to the corresponding content in the aforementioned course structure adaptive optimization method embodiment.

[0076] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: 1. Focusing on real-time individual adaptation in the "learning process stage": The embodiments of the present invention do not change the way the initial course outline is generated. Instead, given the course diagram, the structure is adapted to the real-time state of an individual learner, which makes up for the lack of program-controlled adaptation in the existing technology at the learning process level.

[0077] 2. Unifying and integrating learner profiles, ability status, and cognitive load status: This embodiment of the invention unifies explicit profiles, implicit ability status (IRT / knowledge tracking), and cognitive load status into a learner state vector through a learner profile construction module, an ability estimation and knowledge tracking module, and a cognitive load estimation module, providing a comprehensive and detailed basis for subsequent adaptive decision-making.

[0078] 3. Fine-grained adjustments at the course structure level, rather than just adjusting the difficulty of questions: This embodiment of the invention introduces adjustable parameters in multiple dimensions such as unit granularity, practice ratio, example ratio, support strength, and review frequency at the course structure parameter level, so as to achieve fine-grained personalized control of the course structure, rather than simply changing the difficulty or order of questions.

[0079] 4. Systematizing the dynamic support / fading strategy into a computable support strength parameter: This embodiment of the invention abstracts the worked examples + fading strategy from teaching practice into support strength parameters. h sup It automatically adjusts the support level based on the learner's mastery and cognitive load status, achieving dynamic control of cognitive load and avoiding long-term high support leading to learning dependence or premature removal of support leading to cognitive overload.

[0080] For ease of understanding, please refer to the following: Figure 3 The adaptive optimization method for the above-mentioned course structure is described in detail. For example... Figure 3 As shown, another course structure adaptive optimization method provided by this embodiment of the invention includes the following steps: Step S310: Learner profile construction.

[0081] 1. Static image collection.

[0082] The system collects the following information when a learner registers or uses the system for the first time: Basic attributes: age, education level, years of work experience, current position or educational level, etc. Learning objectives: Target certificate, target competency level, or job requirements; Time constraints: available study time and target completion time; Learning preferences: Do you prefer videos, text and images, fast-paced or slow-paced learning, etc.?

[0083] 2. Behavioral profiling collection and statistics.

[0084] During learners' use of the platform, the system continuously collects their learning behavior data, including: Course access count, chapter completion rate, and exit / interruption status; Time spent on each knowledge unit, number of times skipped, and number of times rewatched; Records of answers to questions (question number, answer time, correctness, and any retakes).

[0085] 3. Image vector construction and updating.

[0086] The system encodes static profile features and behavioral features (such as one-hot encoding, numerical normalization, and embedding mapping) to form learner profile vectors (LPs), and updates them when changes in behavioral data exceed a preset threshold.

[0087] Step S320: Obtain course diagram and course structure parameters.

[0088] 1. Obtaining course images.

[0089] The system obtains the course graph of the target course from an external course design subsystem or through manual import. This course graph consists of multiple knowledge unit nodes and their prerequisite relationship edges. The generation process of the course graph can be handled by other systems, and this embodiment of the invention does not limit the process.

[0090] 2. Definition of course structure parameters.

[0091] For each knowledge unit u The system's predefined set of structural parameters: ;in: g Granularity parameter, describing the number of knowledge points contained in the unit, the estimated duration, etc.; r pr Practice ratio parameter, which is the ratio of the number of practice questions to the amount of content explained in this unit; r ex Example ratio parameter, i.e., the number of worked examples; h sup Support strength parameter, with values ​​ranging from 0 to 3; f rev : Review parameters, used to control the insertion frequency or insertion strategy of review units; lvl Difficulty level parameter, which can be estimated based on the 3PL parameters of the questions / knowledge points involved.

[0092] Step S330: Capacity estimation and knowledge tracking.

[0093] 1. 3PL parameter calibration.

[0094] For questions related to the course in the question bank q Perform project response theory modeling and estimate the discrimination index. a q Difficulty b q Guessing parameters c q This makes it possible to achieve a certain level of ability. θAnswer the questions correctly q The probability satisfies: .

[0095] 2. Overall capability estimation.

[0096] For learners u The answer record {( q i , y i Constructing the likelihood function L ( θ The overall capability parameters are obtained using maximum likelihood estimation or Bayesian estimation. θ u It can also be updated incrementally as new responses arrive.

[0097] 3. Knowledge-point level capabilities and knowledge tracking.

[0098] The question set is divided according to knowledge points, and each knowledge point... k Estimating local capabilities on the corresponding problem subset θ u,k ; Input learners' time-series response data into the deep knowledge tracing model, and output each time step. t Knowledge mastery state vector: .

[0099] Step S340: Cognitive load estimation.

[0100] 1. Feature extraction.

[0101] For each learned knowledge unit, the system extracts: The overall difficulty of the unit (estimated from the difficulty of the questions and knowledge points it contains); The ratio of learning time spent to system-estimated time; The number and frequency of pause, rewind, and jump actions; Time deviation in answering questions and number of consecutive errors; Mouse or keyboard operation frequency, etc.

[0102] 2. Cognitive load prediction model.

[0103] Constructing a cognitive load prediction model may include: Neural networks used to process task complexity and behavioral intensity characteristics; Or it can be used for joint modeling of knowledge state and load state in a two-stream neural network.

[0104] The cognitive load prediction model outputs a certain time step t Cognitive load estimates CLt It can be a continuous value [0,1] or a discrete level {low, medium, high}.

[0105] like Figure 4 As shown, in a two-stream neural network, the knowledge state stream above takes the learner's temporal interaction sequence (questions and answers) as input, and after embedding and encoding by an RNN / Transformer model, outputs a dynamically updated knowledge mastery state vector. K t The cognitive load state stream below simultaneously processes the current task and fine-grained behavioral features, obtaining a real-time cognitive load estimate through feature encoding by a neural network (NN / RNN). CL t These two core state vectors are fused into a unified joint state representation. z t This joint state representation is ultimately input into a prediction head, which simultaneously outputs key decision information: the expected knowledge gain (Δ). K ), future trends in cognitive load, and the probability of answering the next question correctly.

[0106] 3. Load status maintenance. The system caches the most recent data in the current learning session. L The load value of each unit { CL t-L+1 ,..., CL t This is used to determine whether the load is showing a continuous upward, downward, or stable trend.

[0107] Step S350: Learner state vector construction.

[0108] The system will generate learner profile vectors LP u Overall capabilities θ u Knowledge mastery status K t With cognitive load CL t The learner's state vector is composed of the following: This state is used to drive subsequent adaptive decisions regarding the course structure.

[0109] Step S360: Adaptive decision-making for curriculum structure.

[0110] 1. The set of candidate knowledge units is determined.

[0111] The system uses the prerequisite relationships in the course graph and the current knowledge mastery state vector. K t Determine the set of candidate knowledge units that can be learned. UFor example, unlocking is only possible when the mastery probability of all the previous units is higher than a certain threshold.

[0112] 2. Generation of candidate course structure parameters.

[0113] For each candidate knowledge unit u ∈ U Several feasible course structure parameter configurations are generated within the structure parameter space. p u =( g , r pr , r ex , h sup , f rev , lvl ), forming a set of candidate actions A t .

[0114] 3. Calculation of the objective function.

[0115] For each candidate action ( u , p u System estimation: Expected knowledge gain Prediction can be made based on knowledge tracing models or empirical rules; Expected cognitive load Based on cognitive load model and curriculum structure parameter estimation; Time cost C ( p u Based on granularity parameters g Estimate the learning time.

[0116] Define a multi-objective optimization function: .

[0117] in α , β , γ These are the weighting coefficients. CLtarget The target cognitive load level is (generally moderate load).

[0118] 4. Optimal structure decision.

[0119] The system selects the action that maximizes R from the candidate action set. u *, p u* () serves as the next learning unit and its course structure parameter.

[0120] 5. Supplement with heuristic strategies.

[0121] In conjunction with the objective function (i.e., the multi-objective optimization function), this embodiment can add the following heuristic rules: like CL t If the probability of mastering certain key knowledge points is low, prioritize units with smaller granularity, higher example ratio, and stronger support, and appropriately insert review units. like CL t If the probability of mastering most of the relevant knowledge points is high, then choose units with higher difficulty and a larger proportion of practice, and reduce the support intensity.

[0122] Step S370: Dynamic support and fading strategy.

[0123] 1. Definition of support strength level.

[0124] Define for each knowledge unit h sup ∈0,1,2,3: 3. High level of support, providing complete worked examples and detailed step-by-step explanations; 2: Medium support, providing examples and key step tips; 1: Low support, only a few hints provided; 0: No support provided, only the task and basic instructions are given.

[0125] 2. Fade-out rule.

[0126] For a certain knowledge point k or unit u After learners complete several learning sessions, the system will assess their mastery level. p ( K k,t ) and cognitive load CL t Adjust support strength: like p ( K k,t (Multiple consecutive times exceeding the threshold) K high and CL t In [ CL low , CL high If the support strength of the next related unit is reduced by one level, then the support strength of the next related unit will be reduced by one level. like p ( K k,t (lower than) K low and CLt Higher than CL high If so, then maintain or increase the support strength; like p ( K k,t (Higher than) K high and CL t Below CL low This can further reduce the support intensity and increase the proportion of practice.

[0127] 3. Combined with structural parameters.

[0128] Updated h sup As mentioned above p u The other parameters are combined to form the final course structure configuration, which guides content generation and presentation.

[0129] like Figure 5 As shown, the support intensity changes with mastery and cognitive load as follows: Support starts at level 3 (high-intensity support providing "complete examples + detailed explanations"). When the system detects that the learner's "mastery is increasing and cognitive load is moderate," it automatically lowers the support level to level 2 ("examples + key step hints"). If mastery continues to improve and the load remains good, it further decreases to level 1 ("a few hints or ideas"), until finally reaching level 0 ("no support, only tasks"), to achieve fully autonomous learning. Conversely, if it detects "decreasing mastery or increasing errors, and high cognitive load," the system automatically increases the support level in the opposite direction (e.g., from level 2 to level 3) to provide more comprehensive assistance and prevent the learner from getting stuck. This entire process forms a closed-loop adjustment system that uses the learner's real-time status as feedback. Its goal is to maintain the learner's cognitive load within the optimal range while ensuring learning effectiveness, thereby truly achieving personalized and adaptive learning support.

[0130] Step S380: Adaptive content generation and presentation.

[0131] 1. Prompt template construction.

[0132] According to the established ( u *, p u* ), construct content to generate templates, including: List of knowledge points and difficulty level; Expected length and structure of the explanation; The number and type of examples to be generated; The number and types of practice questions to be generated; The level of detail in the hints / explanation (corresponding to) h sup ).

[0133] 2. Call the content generation module.

[0134] The system calls the content generation module (which can be implemented through a large language model) to generate or rewrite the explanatory text, examples, and exercises for this unit according to the template.

[0135] 3. Quality verification and presentation.

[0136] Perform basic validation on the generated content, such as: Are the annotations of the knowledge points consistent? Is the difficulty level within the expected range? Does the structure meet the parameter requirements?

[0137] After verification, the results are presented to the learner on the learning terminal (Web, App, etc.) and their interactive behavior and answers are continuously recorded during the learning process.

[0138] Step S390: State update and loop.

[0139] 1. Data collection.

[0140] After learners complete the current unit, the system collects: Record of answers to new questions; New learning behavior log (duration, pauses, replays, error patterns, etc.).

[0141] 2. Incremental model update.

[0142] System for: IRT capability parameters θ u Perform incremental updates; Knowledge Tracking Status K t+1 Update; Cognitive load CL t+1 Update.

[0143] 3. Re-decision.

[0144] In the updated state s t+1 As input, repeat steps S360 to S380 until the course is completed or the learning objective is achieved.

[0145] like Figure 6 As shown in the figure, the system architecture of a curriculum structure adaptive optimization system provided by an embodiment of the present invention is as follows: I. Front-end interaction layer.

[0146] Web / App front-end: Located on the learning terminal, it provides the user learning interface, displays adaptive learning content, and collects user behavior and response data.

[0147] API Gateway: Serving as a unified entry point for communication between the front-end and back-end, it forwards requests via HTTP / HTTPS protocols, ensuring communication security and traffic control.

[0148] II. Backend Services and Model Layer.

[0149] The core functional modules include: 1. Learner profile building module, including: The static profile submodule is used to collect and store learners' basic attributes, learning goals, and preferences; The behavior profiling submodule is used to collect and analyze learning time, chapter completion status, question answering records, etc. The profile update submodule is used to encode the above information into learner profile vectors and update them when behavior changes.

[0150] 2. Course structure management module, including: The course map acquisition submodule is used to acquire course maps from external course design systems or manual presets; The structural parameter definition submodule is used to define structural parameters such as granularity, practice ratio, example ratio, support strength, review parameters, and difficulty level for each knowledge unit.

[0151] 3. Capability estimation and knowledge tracking module, including: The IRT parameter calibration submodule is used to estimate the 3PL parameters of the problem; The ability estimation submodule is used to estimate learners’ overall ability and knowledge-point level ability based on their response records. The knowledge tracking submodule is used to output a knowledge mastery state vector based on a sequence model. K t .

[0152] 4. Cognitive load estimation module, including: The feature extraction submodule is used to extract features such as task difficulty, duration, number of pauses / replays, and error patterns from the learning behavior log; The load forecasting submodule is used to output the cognitive load estimate for each knowledge unit. CL t ; The load status caching submodule is used to cache the cognitive load sequence of the most recent units.

[0153] 5. Learner State Fusion Module: This module integrates learner profiles, ability estimates, knowledge mastery status, and cognitive load status into a learner state vector. s t .

[0154] 6. The course structure adaptive engine module includes: The candidate set generation submodule is used to determine the set of currently learnable knowledge units based on the course graph and knowledge mastery status; The objective function evaluation submodule is used to calculate the expected return for each candidate action. R ; The structural decision submodule is used to select the optimal candidate action as the next learning unit and its structural parameters. The dynamic support / fading-out submodule is used to adjust the support strength based on knowledge mastery and cognitive load estimates. h sup .

[0155] 7. Content generation and presentation module, including: The template construction submodule is used to generate templates based on structural parameters. The content generation submodule is used to call the large language model or other generative models to generate / rewrite course content; The quality verification submodule is used to check the consistency of knowledge points and the matching degree of difficulty of the generated content; The terminal presentation submodule is used to display adaptive course content and collect behavioral data on user terminals.

[0156] III. Data Storage and Model Service Layer.

[0157] The data storage and model service module includes: a user profile library, a course graph and structure parameter library, a question bank, and a behavior log library; it also includes: IRT model service, knowledge tracing model service, cognitive load model service, and adaptive decision strategy service, which are used to provide online inference for the IRT model, knowledge tracing model, cognitive load model, and adaptive decision strategy model, respectively.

[0158] like Figure 7 As shown, an electronic device 700 provided in this embodiment of the invention includes: a processor 701, a memory 702 and a bus. The memory 702 stores a computer program that can run on the processor 701. When the electronic device 700 is running, the processor 701 and the memory 702 communicate through the bus, and the processor 701 executes the computer program to implement the above-mentioned adaptive optimization method for course structure.

[0159] Specifically, the memory 702 and processor 701 mentioned above can be general-purpose memory and processor, without any specific limitations here.

[0160] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the adaptive optimization method for course structure described in the preceding method embodiments. The computer-readable storage medium includes various media capable of storing program code, such as a USB flash drive, external hard drive, read-only memory (ROM), RAM, magnetic disk, or optical disk.

[0161] In this document, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0162] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limitations; therefore, other examples of exemplary embodiments may have different values.

[0163] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0164] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interface, and the indirect coupling or communication connection between modules may be electrical, mechanical, or other forms.

[0165] The modules described as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0166] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0167] 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. A method for adaptive optimization of curriculum structure, characterized in that, include: Construct a learner state vector, which at least integrates a learner profile vector constructed based on learner static profile data and historical behavior data, the learner's ability estimation result and knowledge mastery state vector obtained based on IRT / 3PL, and the cognitive load estimate obtained based on learning behavior log prediction. Based on the learner state vector and the course graph of the target course, a multi-objective optimization decision model is used to determine the next learning unit and its corresponding course structure parameter configuration from the candidate knowledge unit set; the course structure parameters include at least a support strength parameter; wherein, the multi-objective optimization decision model performs collaborative optimization based on the expected knowledge gain, the deviation between the expected cognitive load and the preset target cognitive load level, and time cost; the multi-objective optimization function is: ; Learner state vector s t Candidate knowledge units u Course structure parameter configuration p u The objective function value is below. For the expected knowledge gain, For the expected cognitive load, CLtarget The target cognitive load level is preset. C ( p u (Time cost) α , β , γ These are preset weighting coefficients; Based on the knowledge mastery state vector and cognitive load estimate in the learner's state vector, the support strength parameter of the next learning unit is dynamically adjusted, including: defining the support strength parameter as multiple discrete support strength levels; if the mastery probability of the first knowledge point is higher than a preset first threshold for N consecutive times and the cognitive load estimate is within a preset suitable range, then the support strength level is reduced in the next learning unit related to the first knowledge point; where N is a preset positive integer; if the mastery probability of the second knowledge point is lower than a preset second threshold and the cognitive load estimate is higher than a preset third threshold, then the support strength level is maintained or increased in the next learning unit related to the second knowledge point; if the mastery probability of the third knowledge point is higher than the first threshold and the cognitive load estimate is lower than a preset fourth threshold, then the support strength level is reduced and the proportion of self-practice is increased in the next learning unit related to the third knowledge point. Based on the adjusted course structure parameter configuration of the next learning unit, the learning content of the next learning unit is generated and presented, including: constructing a content generation template based on the next learning unit and its adjusted course structure parameter configuration; calling the content generation module to generate or rewrite the corresponding learning content according to the content generation template to obtain the target learning content; and presenting the target learning content to the learner after quality verification. Based on the learner's newly generated behavioral data in the next learning unit, the learner's state vector is updated, and the process of determining the next learning unit and its corresponding course structure parameter configuration from the candidate knowledge unit set through a multi-objective optimization decision model is re-executed until the learning objective is achieved, so as to realize closed-loop adaptive optimization.

2. The adaptive optimization method for curriculum structure according to claim 1, characterized in that, The construction of the learner state vector includes: Acquire static profile data and historical behavior data of learners to construct learner profile vectors; wherein, the static profile data includes basic attribute information, learning objectives and time constraint information; Obtain the course diagram and course structure parameter definition of the target course; wherein, the course structure parameters of each knowledge unit in the course diagram include at least the granularity parameter, practice ratio parameter, example ratio parameter, support strength parameter, review parameter, and difficulty level parameter; The IRT / 3PL is used to calibrate the parameters of questions related to the target course in the preset question bank, and the overall ability parameters and local ability parameters at the knowledge point level of the learner are estimated by combining the historical answer records in the historical behavior data to obtain the ability estimation results. The knowledge mastery state vector of each knowledge point at each time step is determined by the knowledge tracing model. Based on the learning behavior logs in the historical behavior data, a cognitive load sequence is determined by a cognitive load prediction model. The cognitive load sequence includes the cognitive load estimates of the most recent preset number of knowledge units. The learner state vector is constructed based on the learner profile vector, the ability estimation result, the knowledge mastery state vector, and the cognitive load sequence.

3. The adaptive optimization method for curriculum structure according to claim 2, characterized in that, The cognitive load prediction model is a supervised learning model or a two-stream neural network model; The process of determining the knowledge mastery state vector for each knowledge point at each time step using a knowledge tracing model includes: The answer sequence, ordered by time, corresponding to the historical answer records is input into the knowledge tracing model to obtain the knowledge mastery state vector corresponding to each knowledge point; wherein, the knowledge tracing model is a deep knowledge tracing model based on a recurrent neural network or Transformer architecture.

4. The adaptive optimization method for curriculum structure according to claim 1, characterized in that, The course graph based on the learner's state vector and the target course, through a multi-objective optimization decision model, determines the next learning unit and its corresponding course structure parameter configuration from the candidate knowledge unit set, including: Based on the knowledge mastery state vector in the learner's state vector and the course graph, determine the current set of candidate knowledge units that can be learned. For each candidate knowledge unit in the candidate knowledge unit set, different course structure parameter configurations are generated as a candidate action set; Based on the learner's state vector, the impact of each candidate action in the candidate action set on the expected knowledge gain, expected cognitive load, and time cost is evaluated through a multi-objective optimization function, and the optimal candidate action is selected from the candidate action set as the next learning unit and its course structure parameter configuration.

5. A curriculum structure adaptive optimization system, characterized in that, include: The learner state construction module is used to construct a learner state vector. The learner state vector is at least a fusion of a learner profile vector constructed based on learner static profile data and historical behavior data, a learner ability estimation result and knowledge mastery state vector obtained based on IRT / 3PL, and a cognitive load estimation value obtained based on learning behavior log prediction. The course structure adaptive engine module is used to determine the next learning unit and its corresponding course structure parameter configuration from the candidate knowledge unit set based on the learner state vector and the course graph of the target course, through a multi-objective optimization decision model; the course structure parameters include at least a support strength parameter; wherein, the multi-objective optimization decision model performs collaborative optimization based on the expected knowledge gain, the deviation between the expected cognitive load and the preset target cognitive load level, and time cost; the multi-objective optimization function is: ; Learner state vector s t Candidate knowledge units u Course structure parameter configuration p u The objective function value is below. For the expected knowledge gain, For the expected cognitive load, CLtarget The target cognitive load level is preset. C ( p u (Time cost) α , β , γ The preset weighting coefficients are used; the support strength parameters of the next learning unit are dynamically adjusted based on the knowledge mastery state vector and cognitive load estimate in the learner's state vector. The course structure adaptive engine module is further configured to: define the support strength parameter as multiple discrete support strength levels; if the mastery probability of the first knowledge point is higher than a preset first threshold for N consecutive times and the cognitive load estimate is within a preset appropriate range, then the support strength level is reduced in the next learning unit related to the first knowledge point; where N is a preset positive integer; if the mastery probability of the second knowledge point is lower than a preset second threshold and the cognitive load estimate is higher than a preset third threshold, then the support strength level is maintained or increased in the next learning unit related to the second knowledge point; if the mastery probability of the third knowledge point is higher than the first threshold and the cognitive load estimate is lower than a preset fourth threshold, then the support strength level is reduced and the proportion of self-practice is increased in the next learning unit related to the third knowledge point; The content generation and presentation module is used to generate and present the learning content of the next learning unit according to the adjusted course structure parameter configuration of the next learning unit. The content generation and presentation module is specifically used for: constructing a content generation template based on the next learning unit and its adjusted course structure parameter configuration; calling the content generation module to generate or rewrite the corresponding learning content according to the content generation template to obtain the target learning content; and presenting the target learning content to the learner after quality verification. The adaptive optimization module is used to update the learner's state vector based on the new behavioral data generated by the learner in the next learning unit, and to re-call the course structure adaptive engine module until the learning objective is achieved, so as to realize closed-loop adaptive optimization.

6. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the course structure adaptive optimization method as described in any one of claims 1-4.

7. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program is executed by the processor to perform the curriculum structure adaptive optimization method as described in any one of claims 1-4.

Images