Multi-agent oriented learning task dynamic optimization method, system and medium

Abstract

The application provides a learning task dynamic optimization method and system for multi-subject sharing and a medium, relates to the technical field of task scheduling optimization, and the method comprises the following steps: when a current learning interaction event is detected, a first subject using a learning machine is acquired; a silence period timing value of each knowledge point associated is called; a first task scheduling instruction is generated based on the silence period timing value of each knowledge point; a short-time interaction record of the first subject is acquired, and a long-time mastering state file of the first subject is combined; a second task scheduling instruction is generated according to the difference between the short-time performance and the long-time state. Through the application, the technical problem that, in the prior art, due to the lack of dynamic perception of the differentiated cognitive state of multiple subjects, task pushing only depends on static statistical indicators, and individual real-time forgetting rhythm and ability fluctuation cannot be matched is solved; the real mastering state and forgetting process of learners are accurately perceived, and knowledge transition or abnormal state is timely identified for targeted intervention.

Description

Technical Field

[0001] This application relates to the field of task scheduling optimization technology, specifically to a dynamic optimization method, system, and medium for learning tasks shared by multiple stakeholders. Background Technology

[0002] Currently, most existing learning task scheduling methods for multi-student sharing rely on static statistical indicators such as learners' historical answer accuracy and task completion rates, allocating learning tasks and resources to each subject through periodic updates. While these methods are simple to implement, they fail to dynamically perceive each learner's forgetting process and real-time changes in cognitive state regarding specific knowledge points. Furthermore, in scenarios where multiple subjects share the same learning terminal, it is difficult to accurately maintain each individual's independent cognitive baseline, further exacerbating the mismatch between task assignments and individual needs, leading to a double decline in both learning resource allocation efficiency and individual learning efficiency.

[0003] In summary, existing technologies suffer from a lack of dynamic perception of the differentiated cognitive states of multiple subjects, resulting in task delivery relying solely on static statistical indicators. This fails to match the real-time forgetting rhythm and ability fluctuations of individuals, further impacting the efficiency of learning resource allocation and learning efficiency. Summary of the Invention

[0004] The purpose of this application is to provide a method, system, and medium for dynamic optimization of learning tasks shared by multiple subjects, in order to solve the technical problem in the prior art that, due to the lack of dynamic perception of the differentiated cognitive states of multiple subjects, task push relies only on static statistical indicators, which cannot match the real-time forgetting rhythm and ability fluctuations of individuals, and further affects the efficiency of learning resource allocation and learning efficiency.

[0005] To achieve the above objectives, this application provides a method, system, and medium for dynamic optimization of learning tasks shared by multiple stakeholders.

[0006] Firstly, this application provides a method for dynamically optimizing learning tasks shared by multiple subjects. This method is implemented through a system for dynamically optimizing learning tasks shared by multiple subjects. The method includes: when a current learning interaction event is detected, obtaining the first subject currently using the learning machine; retrieving the silent period time value of each knowledge point associated with the first subject, where the silent period time value represents the duration of the corresponding knowledge point from the end of the last correctly mastered interaction to the current moment; generating a first task scheduling instruction based on the silent period time value of each knowledge point, where the first task scheduling instruction is used to determine the task push priority and / or initial task difficulty of the learning task to be pushed; executing the first task scheduling instruction, obtaining the short-term interaction record of the first subject, and combining it with the long-term mastery status file of the first subject, generating a second task scheduling instruction based on the difference between short-term performance and long-term status, whereby the second task scheduling instruction is used to instruct dynamic adjustment of the task content.

[0007] Optionally, the silent period time values ​​of all knowledge points to be reviewed associated with the first subject are obtained; each silent period time value is compared with a preset first duration threshold and a second duration threshold, and each knowledge point is divided into an active zone, a consolidation golden zone, and a cooling zone, wherein the first duration threshold is less than the second duration threshold; for knowledge points in the consolidation golden zone, a highest priority push instruction and an adjustment instruction to increase the task difficulty are generated; for knowledge points in the active zone, a push priority reduction instruction and an adjustment instruction to maintain the baseline difficulty are generated; for knowledge points in the cooling zone, a task reordering instruction is generated to postpone the push of the current knowledge point practice task and prioritize the push of the review task of the preceding dependent knowledge points.

[0008] Optionally, knowledge points whose silent period time value is less than the first duration threshold are classified into the active zone; knowledge points whose silent period time value is between the first duration threshold and the second duration threshold are classified into the consolidation golden zone; and knowledge points whose silent period time value is greater than the second duration threshold are classified into the cooling zone.

[0009] Optionally, a preset knowledge graph is queried to obtain at least one prerequisite knowledge point for a knowledge point in a cooldown state; a task rearrangement instruction is generated to insert the review task of the prerequisite knowledge point into the queue to be pushed, and the difficulty of subsequent tasks for the knowledge point in a cooldown state is preset to the lowest level; after the review task of the prerequisite knowledge point is detected to be completed correctly, the practice tasks for the knowledge point in a cooldown state are gradually released from the lowest difficulty.

[0010] Optionally, after executing the first task scheduling instruction, the real-time interactive behavior of the first subject is obtained and added to the short-term interaction record; the long-term mastery status file of the first subject is collected; the short-term mastery degree of any knowledge point in the short-term interaction record is compared with the stable mastery degree of the long-term mastery status file, and based on the comparison result, a second task adjustment instruction for differentiated push of subsequent tasks is generated.

[0011] Optionally, the long-term mastery status file includes the learning records of the first subject for each knowledge point from the first time they used the learning machine.

[0012] Optionally, the sequence of correct and incorrect results for any knowledge point in the short-term interaction record is obtained from the long-term mastery status file; after removing the sequence segments with fluctuations greater than a preset fluctuation threshold from the sequence of correct and incorrect results, the long-term accuracy is calculated to obtain the stable mastery.

[0013] Optionally, if the short-term mastery exceeds the stable mastery in a preset number of consecutive interactions, a knowledge leap is determined to have occurred, and a leap-type task acceleration instruction is generated as the second task scheduling instruction; if the short-term mastery is lower than the stable mastery in a preset number of consecutive interactions, the current state is determined to be abnormal, and an associated activation task insertion instruction is generated as the second task scheduling instruction.

[0014] Secondly, this application also provides a dynamic optimization system for learning tasks shared by multiple subjects, used to execute the dynamic optimization method for learning tasks shared by multiple subjects as described in the first aspect. The dynamic optimization system for learning tasks shared by multiple subjects includes: a subject identification module, used to acquire the first subject currently using the learning machine when a current learning interaction event is detected; a silent period timing value acquisition module, used to retrieve the silent period timing values ​​of each knowledge point associated with the first subject, wherein the silent period timing value represents the duration of the corresponding knowledge point from the end of the last correctly mastered interaction to the current moment; a scheduling instruction generation module, used to generate a first task scheduling instruction based on the silent period timing values ​​of each knowledge point, wherein the first task scheduling instruction is used to determine the task push priority and / or initial task difficulty of the learning task to be pushed; and a task adjustment module, used to execute the first task scheduling instruction, acquire the short-term interaction record of the first subject, and, in conjunction with the long-term mastery status file of the first subject, generate a second task scheduling instruction based on the difference between the short-term performance and the long-term status, wherein the second task scheduling instruction is used to instruct dynamic adjustment of the task content.

[0015] Thirdly, a computer-readable storage medium storing a computer program that, when executed, implements the steps of the dynamic optimization method for a multi-subject shared learning task as described in any of the first aspects above.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: When a current learning interaction event is detected, the system obtains the first subject currently using the learning machine; retrieves the silent period time value of each knowledge point associated with the first subject, where the silent period time value represents the duration of the corresponding knowledge point from the end of the last correctly mastered interaction to the current moment; based on the silent period time value of each knowledge point, generates a first task scheduling instruction, which is used to determine the task push priority and / or initial task difficulty of the learning task to be pushed; executes the first task scheduling instruction, obtains the short-term interaction record of the first subject, and combines it with the long-term mastery status file of the first subject, generating a second task scheduling instruction based on the difference between the short-term performance and the long-term status, which is used to instruct the dynamic adjustment of the task content. In other words, by introducing a silent period timer value to quantify the forgetting pattern for a specific user of the learning machine, a first task scheduling instruction is generated. Then, by comparing the differences between short-term interaction records and long-term mastery status files, a second task scheduling instruction is generated. This forms a two-layer dynamic optimization based on open-loop pre-scheduling based on forgetting patterns and closed-loop feedback correction based on differences between short and long-term data. This accurately perceives the learner's true mastery status and forgetting process, enables adaptive dynamic adjustment of task push priority and difficulty, and can promptly identify knowledge leaps or abnormal states for targeted intervention, improving the efficiency of learning resource allocation and thus significantly enhancing learning outcomes.

[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the dynamic optimization method for learning tasks shared by multiple stakeholders, as described in this application.

[0020] Figure 2 This is a schematic diagram of the structure of the dynamic optimization system for learning tasks shared by multiple entities, as described in this application.

[0021] Figure labeling: Subject identification module 11, silent period time value acquisition module 12, scheduling instruction generation module 13, task adjustment module 14. Detailed Implementation

[0022] This application addresses the technical problem in existing technologies where the lack of dynamic perception of the differentiated cognitive states of multiple learners leads to task assignments relying solely on static statistical indicators, failing to match individual real-time forgetting rhythms and ability fluctuations, thus further impacting the efficiency of learning resource allocation and learning efficiency. By introducing a silent period timer to quantify forgetting patterns for a specific learner using the learning machine, a second task scheduling instruction is generated by comparing short-term interaction records with long-term mastery status profiles. This forms a two-layer dynamic optimization: open-loop pre-scheduling based on forgetting patterns and closed-loop feedback correction based on long- and short-term differences. This accurately perceives the learner's true mastery status and forgetting process, enabling adaptive dynamic adjustment of task assignment priority and difficulty. Furthermore, it can promptly identify knowledge leaps or abnormal states for targeted intervention, improving the efficiency of learning resource allocation and significantly enhancing learning outcomes.

[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0024] Example 1, please refer to the appendix. Figure 1 This application provides a method for dynamic optimization of learning tasks for multi-agent sharing, wherein the method is applied to a system for dynamic optimization of learning tasks for multi-agent sharing, and the method specifically includes the following steps: When a current learning interaction event is detected, the first subject of the learning machine currently in use is obtained.

[0025] Specifically, during operation, the system continuously monitors the learning machine's interface. When the learning machine receives any behavioral signal triggered by the user that belongs to a predefined interaction type, it determines that behavior as a learning interaction event. Other non-critical tasks are then paused, and the system proceeds to the subject recognition process.

[0026] Check if the learning machine is currently logged in. If the learning machine supports persistent account login and there is already an account with an active session, use the unique entity identifier associated with that account as the primary entity. If the learning machine is not logged in or is in shared guest mode, activate the user selection interface, requiring the current user to select their identity from a pre-registered entity list, or automatically obtain the entity identity through a user-authorized biometric recognition module. After obtaining the entity identifier, mark that entity as the primary entity for the current learning interaction event and record the timestamp of this event. All operations related to this interaction event, such as task scheduling, knowledge status updates, and silent period timing, will be performed based on the independent data space of this primary entity until the next interaction event triggers re-identification.

[0027] To avoid performance overhead caused by frequent subject switching, a short-term caching mechanism is set up. If an interaction event is detected again within a preset time window after the most recent subject recognition, and no explicit subject switching signal is received, the first subject identified last time will be used by default, and there is no need to repeat the recognition process.

[0028] In scenarios where multiple entities share the same physical device, identifying the primary entity avoids serious problems such as incorrect task assignment and incorrect status updates caused by entity confusion.

[0029] Retrieve the silent period time value of each knowledge point associated with the first subject. The silent period time value represents the duration of the corresponding knowledge point from the last time the interaction ended when it was correctly mastered to the current time.

[0030] Specifically, the learning status database of the first subject is read, and a status record is maintained for each knowledge point using the subject identifier as the primary key. This record includes the unique identifier of the knowledge point, the timestamp of the last correct mastery interaction, and the silent period timer value of the knowledge point.

[0031] The system iterates through the knowledge point list of the primary subject, typically predefined according to the course structure, containing all knowledge points that should be learned or may be reviewed. For each knowledge point in the list, the timestamp of the last correctly mastered interaction is retrieved. The current system real-time time is obtained, and the historical timestamp is subtracted from the current time to obtain the time difference, which is the silent period timer value for that knowledge point. The silent period timer value is a numerical value in units of time, used to quantify the length of time that has elapsed since the subject last correctly mastered a knowledge point, reflecting the degree of forgetting of that knowledge point at the current moment; the longer the time, the greater the possibility of forgetting.

[0032] If a subject has never correctly mastered a certain knowledge point, meaning there is no previous interaction point where the knowledge point was correctly mastered, then the silent period timer value for that knowledge point is set to a special preset value, such as directly marking it as unlearned. Unlearned knowledge points can be categorized into a cooling-off zone or a special processing zone, thereby triggering a prerequisite review or a minimum difficulty start strategy.

[0033] The silent period timer is calculated only when each interactive event is triggered, and there is no need to update the timers for all knowledge points periodically. This is because the silent period timer is essentially the difference between the current moment and a historical moment, and the subtraction operation is only required when it is used. Otherwise, only the historical timestamp needs to be stored.

[0034] Based on the silent period time value of each knowledge point, a first task scheduling instruction is generated. The first task scheduling instruction is used to determine the task push priority and / or initial task difficulty of the learning task to be pushed.

[0035] Furthermore, this application also includes the following steps: obtaining the silent period time values ​​of all knowledge points to be reviewed associated with the first subject; comparing each silent period time value with a preset first duration threshold and a second duration threshold, and dividing each knowledge point into an active zone, a consolidation golden zone, and a cooling zone, wherein the first duration threshold is less than the second duration threshold; generating a highest priority push instruction and an adjustment instruction to increase the task difficulty for knowledge points in the consolidation golden zone; generating a push priority reduction instruction and an adjustment instruction to maintain the baseline difficulty for knowledge points in the active zone; and generating a task reordering instruction to temporarily postpone the push of the current knowledge point practice task and prioritize the push of the review task of the preceding dependent knowledge points for knowledge points in the cooling zone.

[0036] Furthermore, this application also includes the following steps: classifying knowledge points whose silent period time value is less than the first duration threshold into the active zone; classifying knowledge points whose silent period time value is between the first duration threshold and the second duration threshold into the consolidation golden zone; and classifying knowledge points whose silent period time value is greater than the second duration threshold into the cooling zone.

[0037] Furthermore, this application also includes the following steps: querying a preset knowledge graph to obtain at least one prerequisite knowledge point for a knowledge point in a cooldown state; generating a task rearrangement instruction to insert the review task of the prerequisite knowledge point into the queue to be pushed, and preset the difficulty of subsequent tasks for the knowledge point in a cooldown state to the lowest level; after detecting that the review task of the prerequisite knowledge point has been correctly completed, gradually releasing the practice tasks for the knowledge point in a cooldown state from the lowest difficulty.

[0038] Specifically, it obtains the silent period time value of all knowledge points to be reviewed associated with the first subject, that is, the duration of each knowledge point from the last time it was correctly mastered to the current time. A first duration threshold and a second duration threshold are preset, with the first duration threshold being less than the second duration threshold. The first duration threshold is generally one day, and the second duration threshold is generally seven days.

[0039] Create three empty lists to store knowledge point identifiers belonging to the active zone, the consolidation zone, and the cooldown zone, respectively, along with their corresponding silent period time values.

[0040] Iterate through all the knowledge points to be reviewed and evaluate the silent period timer value for each knowledge point. If the silent period timer value is less than the first duration threshold, the corresponding knowledge point is placed in the active zone, and its identifier is added to the active zone list. If the silent period timer value is between the first and second duration thresholds, the corresponding knowledge point is placed in the consolidation golden zone. If the silent period timer value is greater than the second duration threshold, the corresponding knowledge point is placed in the cooldown zone.

[0041] Based on the attributes of each zone, corresponding task scheduling instructions are generated. For knowledge points in the consolidation "Golden Zone," a highest-priority push instruction is generated, requiring the task scheduler to place the practice tasks for these knowledge points at the forefront of the push queue. An adjustment instruction to increase the task difficulty is also attached for each knowledge point, instructing the task generation engine to add one difficulty level to the baseline difficulty. For knowledge points in the "Active Zone," a lower-priority push instruction is generated, requiring the tasks for these knowledge points to be placed in the latter half of the queue, after all knowledge points in the consolidation "Golden Zone," while an adjustment instruction to maintain the baseline difficulty is attached, without making any further adjustments to the difficulty.

[0042] For knowledge points in the cooling-off zone, generate a task reordering instruction to suspend the generation of any practice tasks for the current knowledge points in the cooling-off zone; then query the pre-built knowledge graph to find the direct prerequisite knowledge points of each knowledge point in the cooling-off zone; insert the review tasks of these prerequisite knowledge points at the front of the task queue, and give them higher priority than the tasks in the consolidation of the golden zone.

[0043] Specifically, for knowledge points in the cooling-off period, a pre-stored knowledge graph structure is accessed in the database. The knowledge graph is organized as a graph database, with each knowledge point node storing a list of identifiers for all its direct predecessor knowledge points. Each node represents a knowledge point, and directed edges represent dependencies. If learning or understanding knowledge point B requires knowledge point A as a foundation, then there exists an edge pointing from A to B. The knowledge graph is constructed and stored before system operation, based on the course syllabus, expert rules, or mined from historical learning data. Using the identifier of the currently cooling-off knowledge point as the query key, a search is performed in the knowledge graph to obtain the set of direct predecessor knowledge points for that knowledge point. If the query result is empty, meaning the knowledge point has no predecessor dependencies, it is considered an isolated knowledge point, and an alternative remedial strategy is adopted, such as directly pushing the lowest difficulty practice for that knowledge point. Typically, a cooling-off knowledge point has at least one predecessor dependency. A task reordering instruction is generated. For each retrieved predecessor dependency knowledge point, a corresponding review task is generated, rather than a new practice task, and these review tasks are inserted at the front of the current queue of tasks to be pushed. If multiple prerequisite knowledge points exist, they are inserted according to their own quiet period time values. After insertion, the tasks at the head of the original queue are automatically moved to the back. The planned practice tasks for knowledge points in the cooldown state are temporarily removed from the queue and do not participate in the current push order. The preset difficulty value of subsequent tasks for knowledge points in the cooldown state is modified to the lowest level defined by the system, and this preset value is stored in the task configuration corresponding to that knowledge point for future use. After the task queue is rearranged, user interaction results are continuously monitored, especially for review tasks inserted at the front of the queue. Whenever a user completes a review task and submits an answer, it is determined whether the task corresponds to a specific prerequisite knowledge point and whether it was completed correctly. When a review task for a prerequisite knowledge point is detected to be completed correctly, a "prerequisite fixed" mark is added to the status record of that knowledge point.

[0044] Once all direct prerequisite knowledge points for a knowledge point in a cooldown state are marked as repaired, practice tasks for that knowledge point are gradually released. The release process doesn't involve pushing out all practice questions of full difficulty at once; instead, it generates the first practice question according to a preset difficulty tier, starting from the lowest level. After the user completes the question, if the answer is correct, the difficulty level is increased by one level, and the next practice question is generated; if the answer is incorrect, the difficulty is not increased or slightly decreased, and an additional hint or explanation may be inserted. This process repeats until the difficulty returns to the original baseline difficulty of the knowledge point or reaches the system's set upper limit. Throughout the gradual release, practice tasks for that knowledge point are inserted into the appropriate position in the push queue with normal priority, usually after prerequisite review tasks, but can be preempted by higher-priority new tasks.

[0045] All generated instructions are encapsulated into a structured scheduling command object, namely the first task scheduling instruction. This instruction determines the task push priority and / or initial task difficulty of the learning tasks to be pushed, and decides which knowledge point to push first and at what difficulty based on the forgetting curve. Task push priority is an indicator that determines the order in which multiple learning tasks are pushed; higher-priority tasks are placed at the front of the push queue and are executed by the user first. Initial task difficulty is the difficulty level set when generating the first learning task for a certain knowledge point. It is usually a numerical scale, such as level 1 to 5. The initial task difficulty may be adjusted upwards, maintained, or downwards based on the silent period timer value.

[0046] By prioritizing and increasing the difficulty of knowledge points in the "golden zone" (the area of ​​high learning intensity), learners are ensured to receive content precisely at the point of forgetting within the optimal memory reinforcement window. This maximizes long-term memory gain with minimal review, improving learning efficiency. Knowledge points in the "active zone" are given lower priority while maintaining the same difficulty, preventing wasted learning time on new content that hasn't yet begun to be forgotten. This allows limited learning resources to be focused on more urgent review tasks. Knowledge points in the "cooling-off zone" are inserted for prior review through task rearrangement and preset with a minimum difficulty level to prevent the frustration of directly challenging forgotten knowledge.

[0047] Execute the first task scheduling instruction, obtain the short-term interaction record of the first subject, and combine it with the long-term mastery status file of the first subject. Based on the difference between the short-term performance and the long-term status, generate a second task scheduling instruction. The second task scheduling instruction is used to instruct the dynamic adjustment of the task content.

[0048] Furthermore, this application also includes the following steps: after executing the first task scheduling instruction, acquiring the real-time interactive behavior of the first subject and adding it to the short-term interaction record; collecting the long-term mastery status file of the first subject; comparing the short-term mastery degree of any knowledge point in the short-term interaction record with the stable mastery degree of the long-term mastery status file, and generating a second task adjustment instruction for differentiated push of subsequent tasks based on the comparison result.

[0049] Furthermore, this application also includes the following steps: the long-term mastery status file includes the learning records of the first subject for each knowledge point from the first time using the learning machine.

[0050] Furthermore, this application also includes the following steps: obtaining the sequence of correct and incorrect results for any knowledge point in the short-term interaction record from the long-term mastery status file; calculating the long-term accuracy to obtain the stable mastery after removing the sequence segments with fluctuations greater than a preset fluctuation threshold from the sequence of correct and incorrect results.

[0051] Furthermore, this application also includes the following steps: if the short-term mastery exceeds the stable mastery in a preset number of consecutive interactions, a knowledge leap is determined to have occurred, and a leap-type task acceleration instruction is generated as the second task scheduling instruction; if the short-term mastery is lower than the stable mastery in a preset number of consecutive interactions, the current state is determined to be abnormal, and an associated activation task insertion instruction is generated as the second task scheduling instruction.

[0052] Specifically, after the tasks assigned by the first task scheduling instruction are executed, the real-time interactive behavior of the primary subject during the execution of these tasks is recorded. That is, when a user submits an answer to a question, the timestamp of that interaction, the identifier of the knowledge point to which the question belongs, whether the user's answer is correct, the number of milliseconds taken to answer the question, whether hints were used or explanations were viewed, etc., are captured, encapsulated into an interaction record object, and appended to the short-term interaction record list maintained in memory for this learning session. The short-term interaction record is set to retain all records from the beginning to the end of the current learning event. The short-term interaction record is a temporary storage structure used to store real-time interactive behavior data generated in the current learning event; its capacity is usually limited, such as storing a maximum of the most recent 30 records for comparison with long-term archives. After the first task scheduling instruction is executed, the real-time behavior of that interaction is recorded and appended to the short-term interaction record.

[0053] Collect a long-term mastery status profile of the first user, including learning records for each knowledge point from the first time the user used the learning machine. Each record should include at least a knowledge point identifier, an interaction timestamp, a boolean value indicating whether the answer is correct, the difficulty level of the question, and the reaction time. Extract the sequence of correct and incorrect results for all knowledge points from the short-term interaction records in the long-term mastery status profile, and sort them from oldest to newest. This represents all records of the first user for all knowledge points in the first task scheduling instruction, starting from the first time the user used the learning machine.

[0054] A volatility analysis is performed on the sequence of correct and incorrect results to eliminate segments considered to be affected by random factors. A fixed-length window is set, such as five consecutive interactions, and the variance of the accuracy or the number of changes between adjacent results within this window is calculated. If the volatility of a window exceeds a preset threshold, such as a variance greater than 0.25, the segment within that window is marked as a high-volatility segment and removed from the original sequence. This removal operation should be cautious; typically, only short sequences that clearly belong to abnormal states are removed, rather than large segments of stable records. After completing the removal of volatile segments, a filtered sequence segment is obtained. The long-term accuracy of the sequence segment is calculated, which is the total number of correct results divided by the total length of the sequence; this represents the stable mastery of the knowledge point.

[0055] Based on recent performance on a specific knowledge point in short-term interaction records—that is, the proportion of correct answers in the most recent interactions—it reflects the learner's instantaneous or short-term performance level at the current moment. If the number of short-term interaction records for that knowledge point is less than the preset number, such as less than 3 times, then the comparison will not be triggered temporarily.

[0056] Short-term mastery is compared with stable mastery. To prevent misjudgment due to single, accidental fluctuations, a second instruction is not generated based on a single comparison, but rather requires conditions to be met in multiple consecutive interactions. A counter is maintained for each knowledge point to record the number of consecutive exceedances, while another counter is maintained to record the number of consecutive falls below. After each interaction for that knowledge point is completed and a new short-term mastery is calculated, this short-term mastery is compared with the pre-stored stable mastery.

[0057] If short-term mastery exceeds stable mastery, the consecutive exceedance counter is incremented by 1, while the consecutive fall below counter is reset to zero. The consecutive exceedance counter value is checked to see if it has reached a preset threshold of 3 consecutive occurrences. If the preset threshold is reached, it is determined that a knowledge leap has occurred, meaning the learner's actual mastery level has substantially improved, rather than experiencing short-term random fluctuations. At this point, a leap-type task acceleration instruction is generated, explicitly specifying the difficulty of all subsequent related tasks for this knowledge point, including subsequent exercises, quizzes, or review tasks, increasing it by one level from the original settings. After increasing the level, the difficulty parameter for this knowledge point in the task configuration is updated to match the learner's improved ability level. After the instruction is generated, the consecutive exceedance counter is reset to zero to avoid repeated triggering.

[0058] If short-term mastery is less than stable mastery, the "consecutive below" counter is incremented by 1, while the "consecutive above" counter is reset to zero. When the value of the "consecutive below" counter reaches a preset threshold, the current state is deemed abnormal, meaning the learner's performance is significantly and consistently below their expected stable level. Instead of simply reducing the difficulty or skipping the knowledge point, a related activation task insertion instruction is generated, pausing the currently executing learning path—that is, temporarily delaying the push of the next task from the original queue. A pre-built knowledge graph is queried, starting from the current knowledge point, to identify all related knowledge points with a preset positive transfer relationship to the current knowledge point and marked as stable mastery in the subject's long-term mastery status file. For each found related knowledge point, one or more short warm-up exercises are generated, typically basic and quick to complete, and these warm-up tasks are inserted at the front of the current task queue with the highest priority. Existing tasks are moved to the back. After insertion, task execution resumes, allowing the learner to complete these warm-up exercises first. After completing the warm-up exercises, the original normal learning path continues. Simultaneously, the "consecutive below" counter is reset to zero to avoid repeated insertions within a short period.

[0059] Throughout the process, if the short-term mastery is equal to the stable mastery, both counters remain unchanged, neither increasing nor decreasing, and no judgment is triggered, thus avoiding frequent false triggers in boundary conditions.

[0060] Based on the judgment result, a corresponding second task adjustment instruction is generated to dynamically adjust the task content determined by the first task scheduling instruction. The second task adjustment instruction does not change the task push order, but directly modifies the content attributes of the task itself, such as increasing or decreasing the difficulty level of subsequent tasks, inserting additional warm-up exercises, or activating tasks.

[0061] By combining short-term interaction records with long-term mastery status profiles and employing a mechanism of continuous, multiple comparisons, the system accurately distinguishes between learners' knowledge leaps and abnormal states, avoiding misjudgments caused by single, accidental correct or incorrect responses. Secondly, task scheduling instructions directly affect the dynamic adjustment of task content, such as modifying difficulty or inserting warm-up tasks, rather than simply reordering tasks. This truly matches the learner's real-time cognitive needs: when learners progress, the challenge difficulty is increased in a timely manner to maintain learning efficiency; when learners are in an abnormal state, relevant knowledge already acquired is activated to help them recover, rather than abruptly reducing difficulty or skipping content.

[0062] In summary, the dynamic optimization method for multi-subject shared learning tasks provided in this application has the following technical effects: when a current learning interaction event is detected, the first subject currently using the learning machine is obtained; the silent period time value of each knowledge point associated with the first subject is retrieved, wherein the silent period time value represents the duration of the corresponding knowledge point from the end of the last correctly mastered interaction to the current moment; based on the silent period time value of each knowledge point, a first task scheduling instruction is generated, wherein the first task scheduling instruction is used to determine the task push priority and / or initial task difficulty of the learning task to be pushed; the first task scheduling instruction is executed to obtain the short-term interaction record of the first subject, and combined with the long-term mastery status file of the first subject, a second task scheduling instruction is generated based on the difference between the short-term performance and the long-term status, wherein the second task scheduling instruction is used to instruct the dynamic adjustment of the task content. In other words, by introducing a silent period timer value to quantify the forgetting pattern for a specific user of the learning machine, a first task scheduling instruction is generated. Then, by comparing the differences between short-term interaction records and long-term mastery status files, a second task scheduling instruction is generated. This forms a two-layer dynamic optimization based on open-loop pre-scheduling based on forgetting patterns and closed-loop feedback correction based on differences between short and long-term data. This accurately perceives the learner's true mastery status and forgetting process, enables adaptive dynamic adjustment of task push priority and difficulty, and can promptly identify knowledge leaps or abnormal states for targeted intervention, improving the efficiency of learning resource allocation and thus significantly enhancing learning outcomes.

[0063] Example 2: Based on the same inventive concept as the dynamic optimization method for multi-subject shared learning tasks in Example 1, this application also provides a dynamic optimization system for multi-subject shared learning tasks. Please refer to the appendix. Figure 2 The dynamic optimization system for learning tasks shared by multiple entities includes: The subject identification module 11 is used to obtain the first subject currently using the learning machine when a current learning interaction event is detected; the silent period timing value acquisition module 12 is used to retrieve the silent period timing value of each knowledge point associated with the first subject, wherein the silent period timing value represents the duration of the corresponding knowledge point from the end of the last correctly mastered interaction to the current moment; the scheduling instruction generation module 13 is used to generate a first task scheduling instruction based on the silent period timing value of each knowledge point, wherein the first task scheduling instruction is used to determine the task push priority and / or initial task difficulty of the learning task to be pushed; the task adjustment module 14 is used to execute the first task scheduling instruction, obtain the short-term interaction record of the first subject, and combine it with the long-term mastery status file of the first subject, and generate a second task scheduling instruction based on the difference between the short-term performance and the long-term status, wherein the second task scheduling instruction is used to instruct the dynamic adjustment of the task content.

[0064] Furthermore, the scheduling instruction generation module 13 in the dynamic optimization system for learning tasks shared by multiple subjects is also used to: obtain the silent period time value of all knowledge points to be reviewed associated with the first subject; compare each silent period time value with a preset first duration threshold and a second duration threshold respectively, and divide each knowledge point into an active zone, a consolidation golden zone, and a cooling zone, wherein the first duration threshold is less than the second duration threshold; generate a highest priority push instruction and an adjustment instruction to increase the task difficulty for knowledge points in the consolidation golden zone; generate a push priority reduction instruction and an adjustment instruction to maintain the baseline difficulty for knowledge points in the active zone; and generate a task reordering instruction to postpone the push of the current knowledge point practice task and prioritize the push of the review task of the preceding dependent knowledge points for knowledge points in the cooling zone.

[0065] Furthermore, the scheduling instruction generation module 13 in the learning task dynamic optimization system for multi-subject sharing is also used to: classify knowledge points whose silent period time value is less than the first duration threshold into the active zone; classify knowledge points whose silent period time value is between the first duration threshold and the second duration threshold into the consolidation golden zone; and classify knowledge points whose silent period time value is greater than the second duration threshold into the cooling zone.

[0066] Furthermore, the scheduling instruction generation module 13 in the learning task dynamic optimization system for multi-subject sharing is also used to: query a preset knowledge graph to obtain at least one prerequisite knowledge point for a knowledge point in a cooldown state; generate a task rearrangement instruction to insert the review task of the prerequisite knowledge point into the queue to be pushed, and preset the difficulty of subsequent tasks for the knowledge point in a cooldown state to the lowest level; after detecting that the review task of the prerequisite knowledge point has been correctly completed, gradually release the practice task of the knowledge point in a cooldown state from the lowest difficulty.

[0067] Furthermore, the task adjustment module 14 in the learning task dynamic optimization system for multi-subject sharing is also used to: after executing the first task scheduling instruction, obtain the real-time interaction behavior of the first subject and add it to the short-term interaction record; collect the long-term mastery status file of the first subject; compare the short-term mastery degree of any knowledge point in the short-term interaction record with the stable mastery degree of the long-term mastery status file, and generate the second task adjustment instruction for differentiated push of subsequent tasks based on the comparison result.

[0068] Furthermore, the task adjustment module 14 in the learning task dynamic optimization system for multi-subject sharing is also used for: the long-term mastery status file includes the learning records of the first subject for each knowledge point from the first time using the learning machine.

[0069] Furthermore, the task adjustment module 14 in the learning task dynamic optimization system for multi-subject sharing is also used to: obtain the sequence of correct and incorrect results for any knowledge point in the short-term interaction record from the long-term mastery status file; and calculate the long-term accuracy to obtain the stable mastery after removing the sequence segments with fluctuations greater than a preset fluctuation threshold from the sequence of correct and incorrect results.

[0070] Furthermore, the task adjustment module 14 in the learning task dynamic optimization system for multi-subject sharing is also used to: if the short-term mastery exceeds the stable mastery in a preset number of consecutive interactions, then a knowledge leap is determined to have occurred, and a leap-type task acceleration instruction is generated as the second task scheduling instruction; if the short-term mastery is lower than the stable mastery in a preset number of consecutive interactions, then the current state is determined to be abnormal, and an associated activation task insertion instruction is generated as the second task scheduling instruction.

[0071] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The dynamic optimization method and specific examples for multi-subject shared learning tasks in the foregoing Embodiment 1 are also applicable to the dynamic optimization system for multi-subject shared learning tasks in this embodiment. Through the foregoing detailed description of the dynamic optimization method for multi-subject shared learning tasks, those skilled in the art can clearly understand the dynamic optimization system for multi-subject shared learning tasks in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0072] In embodiment three, based on the same inventive concept as the dynamic optimization method for learning tasks shared by multiple subjects in embodiment one, this application also provides a computer-readable storage medium storing a computer program, which, when executed, implements the steps of the dynamic optimization method for learning tasks shared by multiple subjects in any one of embodiments one above.

[0073] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0074] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A dynamic optimization method for learning tasks shared by multiple stakeholders, characterized in that, include: When a current learning interaction event is detected, obtain the first subject currently using the learning machine; Retrieve the silent period time value of each knowledge point associated with the first subject. The silent period time value represents the duration of the corresponding knowledge point from the last time the interaction ended when it was correctly mastered to the current time. Based on the silent period time value of each knowledge point, a first task scheduling instruction is generated. The first task scheduling instruction is used to determine the task push priority and / or initial task difficulty of the learning task to be pushed. Execute the first task scheduling instruction, obtain the short-term interaction record of the first subject, and combine it with the long-term mastery status file of the first subject. Based on the difference between the short-term performance and the long-term status, generate a second task scheduling instruction. The second task scheduling instruction is used to instruct the dynamic adjustment of the task content.

2. The dynamic optimization method for multi-agent shared learning tasks as described in claim 1, characterized in that, Based on the silent period time value of each knowledge point, the first task scheduling instruction is generated, including: Obtain the silent period timing values ​​of all knowledge points to be reviewed associated with the first subject; The timing values ​​of each silent period are compared with the preset first duration threshold and second duration threshold respectively, and each knowledge point is divided into an active zone, a consolidation golden zone and a cooling zone, wherein the first duration threshold is less than the second duration threshold. For knowledge points that are in the consolidation golden zone, generate the highest priority push command and the adjustment command to increase the task difficulty; For knowledge points in the active zone, generate instructions to reduce push priority and adjust the difficulty to maintain the baseline. For knowledge points in the cooling-off zone, generate task reordering instructions to temporarily postpone the current knowledge point practice tasks and prioritize the push of prerequisite dependent knowledge point review tasks.

3. The dynamic optimization method for multi-agent shared learning tasks as described in claim 2, characterized in that, The timing values ​​for each silent period are compared with preset first and second duration thresholds, and each knowledge point is divided into an active zone, a consolidation golden zone, and a cooling-off zone, including: Knowledge points whose silent period time value is less than the first duration threshold are classified into the active area; Knowledge points whose silent period time value falls between the first duration threshold and the second duration threshold are classified into the consolidation golden zone; Knowledge points whose silent period time value is greater than the second duration threshold are classified into the cooling zone.

4. The dynamic optimization method for multi-agent shared learning tasks as described in claim 2, characterized in that, For knowledge points in the cooling-off zone, generate task reordering instructions to temporarily postpone pushing current knowledge point practice tasks and prioritize pushing review tasks for prerequisite dependent knowledge points, including: Query the preset knowledge graph to obtain at least one prerequisite knowledge point for the knowledge point in the cooldown state; Generate a task reordering instruction, insert the review task of the aforementioned prerequisite knowledge points into the queue to be pushed, and preset the difficulty of subsequent tasks for knowledge points in the cooldown state to the lowest level. Once the review tasks for the prerequisite knowledge points are detected to have been completed correctly, the practice tasks for the knowledge points in the cooling-off state will be gradually released from the lowest difficulty level.

5. The dynamic optimization method for multi-agent shared learning tasks as described in claim 1, characterized in that, Execute the first task scheduling instruction, obtain the short-term interaction record of the first subject, and combine it with the long-term mastery status profile of the first subject. Based on the difference between the short-term performance and the long-term status, generate a second task scheduling instruction, including: After executing the first task scheduling instruction, the real-time interaction behavior of the first subject is obtained and added to the short-term interaction record; Collect long-term status records of the first subject; By comparing the short-term mastery of any knowledge point in the short-term interaction record with the stable mastery of the long-term mastery status file, and based on the comparison results, a second task adjustment instruction is generated to provide differentiated push for subsequent tasks.

6. The dynamic optimization method for multi-agent shared learning tasks as described in claim 5, characterized in that, The long-term mastery status file includes the learning records of the first subject for each knowledge point from the first time they used the learning machine.

7. The dynamic optimization method for multi-agent shared learning tasks as described in claim 5, characterized in that, The calculation steps for the stability of mastery in the long-term mastery status record include: Obtain the sequence of correct and incorrect results for any knowledge point in the short-term interaction record from the long-term mastery status archive; After removing sequence segments with fluctuations greater than a preset fluctuation threshold from the correct and incorrect result sequence, the long-term accuracy is calculated to obtain the stability mastery.

8. The dynamic optimization method for multi-agent shared learning tasks as described in claim 5, characterized in that, Compare the short-term mastery level of any knowledge point in the short-term interaction record with the stable mastery level in the long-term mastery status file. Based on the comparison result, generate a second task adjustment instruction for differentiated push of subsequent tasks, including: If the short-term mastery exceeds the stable mastery in a preset number of consecutive interactions, a knowledge leap is determined to have occurred, and a leap-type task acceleration instruction is generated as the second task scheduling instruction. If the short-term mastery level is lower than the stable mastery level in a preset number of consecutive interactions, the current state is determined to be abnormal, and an associated activation task insertion instruction is generated as the second task scheduling instruction.

9. A dynamic optimization system for learning tasks shared by multiple stakeholders, characterized in that: The steps for implementing the dynamic optimization method for multi-agent shared learning tasks according to any one of claims 1 to 8, wherein the dynamic optimization system for multi-agent shared learning tasks comprises: The subject recognition module is used to obtain the first subject currently using the learning machine when a current learning interaction event is detected; The silent period timing value acquisition module is used to retrieve the silent period timing value of each knowledge point associated with the first subject. The silent period timing value represents the duration of the corresponding knowledge point from the end of the last interaction when it was correctly mastered to the current time. The scheduling instruction generation module is used to generate a first task scheduling instruction based on the silent period time value of each knowledge point. The first task scheduling instruction is used to determine the task push priority and / or initial task difficulty of the learning task to be pushed. The task adjustment module is used to execute the first task scheduling instruction, obtain the short-term interaction record of the first subject, and combine it with the long-term mastery status file of the first subject. Based on the difference between the short-term performance and the long-term status, it generates a second task scheduling instruction, which is used to instruct the task content to be dynamically adjusted.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the steps of the dynamic optimization method for multi-subject shared learning tasks as described in any one of claims 1 to 8.

Images