Learning task generation method based on historical answer and real-time evaluation data

Abstract

The invention provides a learning task generation method based on historical answering and real-time evaluation data, and the method comprises the steps: firstly constructing an initial state vector of a student, extracting a knowledge point range, building an association network, and then evaluating a current mastering state of the student through a probability graph model; for weak knowledge points, a greedy strategy optimization algorithm is adopted to generate a task sequence, and grading is performed according to difficulty and content depth; dynamically adjusting task difficulty and a content presentation sequence in combination with learning preferences of students, and generating a customized task sequence; through a real-time feedback mechanism, student state vectors are continuously updated, the mastery degree is re-evaluated, and a task sequence is optimized; finally, a personalized task set is generated by considering time constraints. According to the invention, accurate evaluation of the learning state of the student and dynamic optimization of the task sequence are realized, and the effect of personalized learning is effectively improved.

Description

Technical Field

[0001] The present invention relates to a learning task generation method based on historical question answering and real-time evaluation data. Background Art

[0002] Accurately estimating the mastery of knowledge points by primary and secondary school students is crucial in the field of educational technology, directly impacting the effectiveness of personalized instruction and improving student learning efficiency. By accurately assessing students' knowledge status, educational systems can tailor learning paths for each student, significantly improving learning outcomes. Currently, many educational platforms rely on traditional testing and statistical analysis to assess student ability. However, these methods are often too simplistic to fully capture students' mastery of complex knowledge systems, ignoring the interdependencies between knowledge points and the dynamic changes in student learning. This results in inaccurate assessment results, making it difficult to effectively predict students' future learning performance.

[0003] In the process of achieving accurate predictions, a core challenge is how to build a model that can dynamically reflect the correlation between knowledge points. Knowledge points do not exist in isolation, but are interconnected through logic and hierarchy to form a complex network structure. If the model cannot accurately characterize this correlation, it will be difficult to quantify the student's mastery of a certain knowledge point, let alone predict their potential performance on related knowledge points. Due to the lack of dynamic correlation modeling, another derivative problem is the inability to generate adaptive task sequences based on students' historical learning data and real-time assessment results. The design of task sequences needs to comprehensively consider students' knowledge blind spots, learning progress, and the logical progressive relationship between knowledge points. This requires the system to efficiently screen the optimal path from a massive combination of knowledge points. Traditional machine learning methods often face a contradiction between computational complexity and accuracy when dealing with such multi-dimensional, highly dynamic optimization problems. Summary of the Invention

[0004] The present invention proposes a learning task generation method based on historical answering and real-time assessment data, which can dynamically capture the model of knowledge point correlation and generate the optimal task sequence based on students' historical data and real-time assessment results, thereby realizing accurate estimation of knowledge points for primary and secondary school students.

[0005] The technical solution of the present invention is implemented as follows: a method for generating learning tasks based on historical answer data and real-time assessment data, specifically comprising: S1, obtains the correct answer rate, answer time and error type distribution from the student's historical answer records and real-time assessment data, constructs the student's initial state vector, which contains the mastery data of each knowledge point and learning preference information, and generates the student's initial state description; S2, based on the knowledge point mastery data contained in the student's initial state description, determines the scope of the knowledge points involved. Based on the prerequisite relationships and logical dependencies between knowledge points, the edge weights are calculated using the co-occurrence frequency in the historical answer data, and a knowledge point association network is constructed to generate a knowledge point association strength matrix. S3: Based on the knowledge point association strength matrix and the student's initial state description, a probabilistic graphical model is used to calculate the posterior probability distribution of the mastery of each knowledge point, and the student's current knowledge point state vector is updated and determined. This vector contains the probability value of each knowledge point, and generates a description of the student's current mastery state. Based on the posterior probability distribution in the student's current mastery state description, it is determined whether there are knowledge points with a mastery level below the preset threshold. If so, the subnetwork directly related to the weak knowledge points is extracted from the knowledge point association strength matrix to generate a weak knowledge point association submatrix. S4, based on the weak knowledge point correlation submatrix, combined with the preset task sequence length limit, adopts a greedy strategy-based optimization algorithm and a scoring mechanism with the goal of improving mastery to generate a preliminary task sequence candidate set, which contains multiple task combinations; S5: Based on the preliminary task sequence candidate set, combined with the preset task difficulty grading standards and content depth stratification rules, the tasks are divided into three levels: basic, advanced, and challenging. The content is divided into three depth levels: conceptual understanding, application practice, and comprehensive expansion. This generates an optimized task sequence. S6, for optimizing task sequences, combines the learning preferences and historical learning speed data in the student's initial state description to dynamically adjust the task difficulty increment rhythm and content presentation order, ensuring that task types are divided into explanation, practice, and assessment types, and generating customized task sequences; S7, based on the customized task sequence and real-time assessment data, builds a student state transition probability model based on historical answer records, collects student answer feedback through the feedback response mechanism, dynamically adjusts the task difficulty and content depth, recalculates the posterior probability distribution, and generates an updated student state vector; S8, if the updated student state vector shows that the posterior probability distribution of some knowledge points is still lower than the preset mastery threshold, then based on the error frequency distribution in the historical feedback data, combined with the error tolerance range and knowledge point coverage requirements, the weak knowledge point correlation submatrix is ​​re-extracted, the task difficulty and correlation density are adjusted, and an improved task sequence is generated; S9, in order to improve the task sequence, combined with the students' learning time constraints and the time allocation ratio for task completion, through difficulty balance adjustment and content presentation order optimization, generates the final personalized task set including knowledge point exercises and assessments for the implementation of students' learning practice.

[0006] The beneficial effects of the present invention are as follows: the method first constructs the student's initial state vector, extracts the knowledge point range and establishes an association network, and then uses a probabilistic graphical model to assess the student's current mastery status; for weak knowledge points, the present invention uses a greedy strategy optimization algorithm to generate task sequences and classify them according to difficulty and content depth; based on student learning preferences, the task difficulty and content presentation order are dynamically adjusted to generate customized task sequences; through a real-time feedback mechanism, the present invention continuously updates the student's state vector, reassesses mastery, and optimizes the task sequence; finally, a personalized task set is generated considering time constraints. The present invention achieves accurate assessment of student learning status and dynamic optimization of task sequences, effectively improving the effectiveness of personalized learning. DETAILED DESCRIPTION

[0007] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the embodiments of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0008] A learning task generation method based on historical answer data and real-time assessment data, comprising: S1, obtains the correct answer rate, answer time and error type distribution from the student's historical answer records and real-time assessment data, constructs the student's initial state vector, which contains the mastery data of each knowledge point and learning preference information, and generates a description of the student's initial state. Specifically, it includes: Extract the correct answer rate, answer time, and error type distribution from students' historical answer records and real-time assessment data. Use data cleaning tools to denoise and unify the data to obtain a collated answer performance dataset. Based on the collated answer performance dataset, use statistical tools to classify and summarize the correct answer rate and error type distribution for each knowledge point. Combined with the answer time data, if the correct answer rate is higher than a preset threshold and the answer time is lower than the average, it is determined that the mastery level is high, and the mastery score data for each knowledge point is obtained. For the mastery score data, the distribution of students' answering time and error types for different question types and difficulty levels is obtained, and the data is grouped and analyzed using a clustering tool to determine the students' learning preference characteristics; by integrating the mastery score data with the learning preference characteristics, a data fusion tool is used to generate the student's initial state vector, which is used for subsequent knowledge point range extraction.

[0009] S2, based on the knowledge point mastery data contained in the student's initial state description, determines the scope of the knowledge points involved. Based on the prerequisite relationships and logical dependencies between knowledge points, the edge weights are calculated using the co-occurrence frequency in the historical answer data, and a knowledge point association network is constructed to generate a knowledge point association strength matrix. This includes: Based on the data in the initial state description, the student's mastery of each knowledge point is obtained, the mastery information is classified and sorted, and matched with a pre-established knowledge point classification library to obtain a preliminary knowledge point range division result. Combined with the preset prerequisite relationship network and logical dependency relationship table, the dependency path between the knowledge points is obtained, and the dependency strength between the knowledge points is calculated using a path analysis tool to determine the association hierarchical structure between the knowledge points; Through the records in the historical answer data, the co-occurrence frequency statistics are performed on the knowledge point combinations in the associated hierarchical structure to obtain the number of occurrences of the knowledge point combinations in the answers, and the frequency calculation tool is used to accumulate the numbers to determine the edge weight values ​​between the knowledge points; represents the comprehensive edge weight value between knowledge point i and knowledge point j, Indicates the co-occurrence frequency of two knowledge points, represents the point mutual information value of two knowledge points, α and β are weight coefficients used to balance the influence of frequency and relevance in the final weight; If the edge weight value is higher than a preset threshold, the corresponding knowledge point pair is included in the association strength matrix construction process, and the final matrix data is obtained for subsequent mastery status evaluation to obtain the dynamic evaluation basis of the student on the knowledge point.

[0010] S3: Based on the knowledge point association strength matrix and the student's initial state description, a probabilistic graphical model is used to calculate the posterior probability distribution of the mastery of each knowledge point, update and determine the student's current knowledge point state vector, which contains the probability value of each knowledge point, and generate the student's current mastery state description; based on the posterior probability distribution in the student's current mastery state description, determine whether there are knowledge points below the preset mastery threshold. If so, extract the subnetwork directly related to the weak knowledge points from the knowledge point association strength matrix to generate the weak knowledge point association submatrix. Specifically, it includes: Based on the knowledge point association strength matrix, the association weight values ​​of each knowledge point and other knowledge points are obtained. By constructing a probability graph structure, the association weight values ​​are converted into conditional probability distribution relationships. The initial state data of the knowledge points are integrated to obtain preliminary probability distribution data. The Bayesian inference method is used to obtain the conditional probability value of each knowledge point from the preliminary probability distribution data. When calculating the posterior probability distribution, if the conditional probability value is lower than a preset threshold, the corresponding knowledge point is marked as a node to be updated, and the posterior probability adjustment range of the node to be updated is determined. By iteratively adjusting the probability value of the node to be updated, obtaining the updated probability value from the posterior probability adjustment range, and performing normalization processing on the final probability value of the knowledge point, the student's current knowledge point state vector is obtained, from which the mastery probability value of each knowledge point is obtained. If the mastery probability value is lower than a preset threshold, it is marked as a weak point, and the student's knowledge mastery status is judged; The student's posterior probability data is obtained from the state description, and the probability value of each knowledge point is compared with a preset mastery threshold. If the probability value of a knowledge point is lower than the preset mastery threshold, it is marked as a weak knowledge point, thereby obtaining a preliminary screened set of weak knowledge points; the associated data directly related to the marked weak knowledge points are extracted from a pre-established strength matrix, and a corresponding sub-network structure is generated using a graph traversal tool to determine the range of knowledge points closely connected to the weak knowledge points; Construct a correlation sub-matrix for the weak knowledge points, obtain the correlation strength values ​​between the knowledge points in the sub-matrix, and if there are knowledge point pairs with high-intensity correlation, increase their priority to obtain an optimized correlation sub-matrix. Combined with the task sequence design requirements, obtain the learning path data corresponding to the high-intensity correlation knowledge points, generate a personalized task sequence through the path planning tool, and determine the final task sequence plan.

[0011] S4, based on the weak knowledge point correlation submatrix and the preset task sequence length limit, uses a greedy strategy-based optimization algorithm and a scoring mechanism with the goal of improving mastery to generate a preliminary task sequence candidate set, which contains multiple task combinations. Specifically, it includes: According to the preset task sequence length limit, at least one task combination is extracted from the weak knowledge point association submatrix, a mastery improvement score is calculated for the task combination, and a preliminary task sequence candidate set is obtained by ranking the scores to obtain an initial set containing multiple task combinations; For the task combinations obtained from the initial set, a pre-established difficulty grading rule library is used to determine the difficulty coefficient of the task combination. If the difficulty coefficient exceeds a preset threshold, it is marked as a high-difficulty task combination, and the task set that needs to be split is determined; According to the high-difficulty task set, the subtask units within the set are obtained, the subtask units are recombined using a logical splitting tool, and it is determined whether the subtask units meet a preset difficulty threshold, thereby obtaining an adjusted task sequence; For the adjusted task sequence, a data comparison tool is used to verify the score difference with the initial set. If the difference is less than a preset threshold, the task sequence is included in the final candidate set to determine the final combination for difficulty grading.

[0012] This formula represents the score difference between the adjusted task sequence and the initial set, where represents the score of the adjusted task sequence, represents the score of the initial set, · represents the absolute value, and ΔS represents the difference in scores between the two.

[0013] S5, based on the preliminary task sequence candidate set, combined with the preset task difficulty grading standards and content depth stratification rules, divide the tasks into three levels: basic, advanced, and challenging, and divide the content into three depth levels: conceptual understanding, application practice, and comprehensive expansion, to generate an optimized task sequence. Specifically, it includes: According to the preset task difficulty grading standard, the difficulty score data of each task is extracted from the task sequence candidate set, and the score data is classified and processed. The score data is compared with the preset threshold value using the interval matching method. If the score data is lower than the first threshold, it is classified as the basic level; if the score data is between the first threshold and the second threshold, it is classified as the advanced level; if the score data is higher than the second threshold, it is classified as the challenging level, and a preliminary stratification result is obtained; By deeply analyzing the task content in the preliminary stratification results, the knowledge points involved in the task content are obtained, and the knowledge points are matched with the content depth level rules. If the task content only involves the explanation of a single knowledge point, it is classified as concept understanding; if the task content involves the combination of knowledge points and operations, it is classified as application practice, and the depth level classification is determined; According to the depth level classification and combined with the difficulty level, a two-dimensional matrix distribution of the task sequence is generated, and the matrix distribution is optimized and adjusted. The task sequence is rearranged using a sorting tool to obtain an optimized sequence; the rhythm distribution of the optimized sequence is calculated to obtain the distribution density of each level. If the distribution density is higher than a preset threshold, the splitting tool is used to adjust it to determine whether the final sequence meets the rhythm requirements.

[0014] S6, for optimizing task sequences, combines the learning preferences and historical learning speed data in the student's initial state description to dynamically adjust the task difficulty increment rhythm and content presentation order, ensure that task types are divided into explanation type, practice type and assessment type, and generate customized task sequences. Specifically, it includes: Obtaining the learning preferences and historical speed data from a student database, classifying each student's learning ability level using pre-established classification rules based on their personalized information, and using comparative analysis tools to determine the degree of match between the level and task difficulty to obtain a preliminary task difficulty benchmark value; Based on the preliminary task difficulty benchmark value and the preset threshold of the incremental pace, a logical calculation tool is used to adjust the increase in task difficulty. If the increase exceeds the preset threshold, the incremental pace is reduced; otherwise, the original pace is maintained, thereby determining a final task difficulty increment plan. Obtaining a task type classification rule through the final task difficulty increasing scheme, matching the explanation content, the practice intensity, and the assessment criteria with the task difficulty, using a sorting tool to adjust the content sequence, determining whether the sequence is consistent with the adaptability of the student's state, and obtaining an optimized task presentation sequence; For the optimized task presentation sequence, combined with a dynamic adjustment mechanism, real-time feedback data during the student's learning process is obtained, and a conditional judgment tool is used. If the feedback data indicates a change in the student's status, the content proportion and task type distribution of the customized sequence are adjusted to determine the final personalized learning task sequence.

[0015] S7, based on customized task sequences and real-time assessment data, builds a student state transition probability model based on historical answer records, collects student answer feedback through the feedback response mechanism, dynamically adjusts task difficulty and content depth, recalculates the posterior probability distribution, and generates an updated student state vector. Specifically, it includes: Obtaining student answering behavior data from a historical answering record database, classifying the accuracy and time-consuming information in the answering behavior data, and determining the state transition probability distribution of students at different task difficulties using a pre-established probability calculation tool to obtain an initial student state vector; If the probability value in the initial student state vector is lower than a preset threshold, the student's answer feedback data is collected through a real-time assessment tool, and the task difficulty parameter is dynamically updated based on the feedback data and the preset content depth adjustment rules. It is determined whether the updated task difficulty parameter matches the student's ability, and an adjusted task allocation plan is obtained; Adopting the adjusted task allocation scheme, updating the probability distribution of the student state vector through a posterior probability calculation tool, obtaining adjusted state vector data, and using a data comparison tool to determine the student's mastery trend based on the adjusted state vector data to obtain a mastery assessment result; According to the mastery assessment results, the student state vector data is updated to the database through a data storage tool to obtain the latest student ability profile. Based on the changes in the student ability profile, a logical judgment tool is used to determine the need for task content adjustment to obtain the final task optimization plan.

[0016] S8: If the updated student state vector shows that the posterior probability distribution of some knowledge points is still lower than the preset mastery threshold, then based on the error frequency distribution in the historical feedback data, combined with the error tolerance range and knowledge point coverage requirements, the weak knowledge point association submatrix is ​​re-extracted, the task difficulty and association density are adjusted, and an improved task sequence is generated. Specifically, this includes: Obtaining the posterior probability distribution data from the student status record, comparing the probability value of each knowledge point with a preset mastery threshold, and marking at least one knowledge point as a weak knowledge point if the posterior probability is lower than the threshold, thereby obtaining a preliminarily screened set of weak knowledge points; Extract error frequency distribution data related to the weak knowledge point set based on historical feedback records, and determine whether the error frequency of the weak knowledge point exceeds the tolerance range in combination with the preset error tolerance range. If so, mark it as the high-risk knowledge point and determine a list of high-risk knowledge points; For the list of high-risk knowledge points, obtain the relevant correlation submatrix data based on the knowledge point coverage requirements, use a database query tool to extract the correlation density value in the submatrix, and if the correlation density is lower than a preset threshold, adjust the task difficulty parameter to obtain an updated task difficulty configuration; According to the task difficulty configuration, combined with the associated density value and the time allocation constraint, the task sequence is optimized by a task sequence generation tool to determine the final task sequence content.

[0017] S9, for improving task sequences, combines students’ learning time constraints and task completion time allocation, adjusts difficulty balance and optimizes content presentation order to generate a final personalized task set containing knowledge point exercises and assessments for students’ learning practice implementation. Specifically, it includes: Based on the student's learning time constraints and the time allocation for task completion, a task set that matches the student's available time period is obtained from a pre-established task database. The task set is preliminarily screened based on its difficulty and knowledge point coverage to obtain an initial task list that meets the time constraints. By balancing and adjusting the difficulty values ​​of the tasks in the initial task list, the tasks are grouped using a preset difficulty threshold range. If the average difficulty value of a group of tasks in the group exceeds the upper threshold, the difficulty weights of some tasks in the group are reduced to obtain a task combination with a balanced difficulty distribution; Based on the task combination with a balanced difficulty distribution, the content presentation order of the task combination is obtained, and the order is adjusted according to the logical progressive relationship of the knowledge points. If the preceding content of a knowledge point in the task combination is not covered, relevant practice tasks are supplemented from the database to determine the final task presentation sequence; Through the final task presentation sequence, combined with the student's learning progress record, the assessment task content corresponding to the sequence is obtained, and dynamic matching is performed based on the mastery level of the knowledge points. If the mastery level is lower than the preset threshold, intensive practice tasks for the relevant knowledge points are added to obtain a personalized task set that includes practice and assessment.

[0018] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A learning task generation method based on historical answering and real-time assessment data, characterized in that: include: S1, obtains the correct answer rate, answer time and error type distribution from the student's historical answer records and real-time assessment data, constructs the student's initial state vector, which contains the mastery data of each knowledge point and learning preference information, and generates the student's initial state description; S2, based on the knowledge point mastery data contained in the student's initial state description, determines the scope of the knowledge points involved. Based on the prerequisite relationships and logical dependencies between knowledge points, the edge weights are calculated using the co-occurrence frequency in the historical answer data, and a knowledge point association network is constructed to generate a knowledge point association strength matrix. S3: Based on the knowledge point association strength matrix and the student's initial state description, a probabilistic graphical model is used to calculate the posterior probability distribution of the mastery of each knowledge point, and the student's current knowledge point state vector is updated and determined. This vector contains the probability value of each knowledge point, and generates a description of the student's current mastery state. Based on the posterior probability distribution in the student's current mastery state description, it is determined whether there are knowledge points with a mastery level below the preset threshold. If so, the subnetwork directly related to the weak knowledge points is extracted from the knowledge point association strength matrix to generate a weak knowledge point association submatrix. S4, based on the weak knowledge point correlation submatrix, combined with the preset task sequence length limit, adopts a greedy strategy-based optimization algorithm and a scoring mechanism with the goal of improving mastery to generate a preliminary task sequence candidate set, which contains multiple task combinations; S5: Based on the preliminary task sequence candidate set, combined with the preset task difficulty grading standards and content depth stratification rules, the tasks are divided into three levels: basic, advanced, and challenging. The content is divided into three depth levels: conceptual understanding, application practice, and comprehensive expansion. This generates an optimized task sequence. S6, for optimizing task sequences, combines the learning preferences and historical learning speed data in the student's initial state description to dynamically adjust the task difficulty increment rhythm and content presentation order, ensuring that task types are divided into explanation, practice, and assessment types, and generating customized task sequences; S7, based on the customized task sequence and real-time assessment data, builds a student state transition probability model based on historical answer records, collects student answer feedback through the feedback response mechanism, dynamically adjusts the task difficulty and content depth, recalculates the posterior probability distribution, and generates an updated student state vector; S8, if the updated student state vector shows that the posterior probability distribution of some knowledge points is still lower than the preset mastery threshold, then based on the error frequency distribution in the historical feedback data, combined with the error tolerance range and knowledge point coverage requirements, the weak knowledge point correlation submatrix is ​​re-extracted, the task difficulty and correlation density are adjusted, and an improved task sequence is generated; S9, in order to improve the task sequence, combined with the students' learning time constraints and the time allocation ratio for task completion, through difficulty balance adjustment and content presentation order optimization, generates the final personalized task set including knowledge point exercises and assessments for the implementation of students' learning practice.

2. The method according to claim 1, wherein Step S1 specifically includes: Extract the correct answer rate, answer time, and error type distribution from students' historical answer records and real-time assessment data. Use data cleaning tools to denoise and unify the data to obtain a collated answer performance dataset. Based on the collated answer performance dataset, use statistical tools to classify and summarize the correct answer rate and error type distribution for each knowledge point. Combined with the answer time data, if the correct answer rate is higher than a preset threshold and the answer time is lower than the average, it is determined that the mastery level is high, and the mastery score data for each knowledge point is obtained. For the mastery score data, the distribution of students' answering time and error types for different question types and difficulty levels is obtained, and the data is grouped and analyzed using a clustering tool to determine the students' learning preference characteristics; by integrating the mastery score data with the learning preference characteristics, a data fusion tool is used to generate the student's initial state vector, which is used for subsequent knowledge point range extraction.

3. The method for generating learning tasks based on historical answering and real-time assessment data according to claim 1, characterized in that: Step S2 specifically includes: Based on the data in the initial state description, the student's mastery of each knowledge point is obtained, the mastery information is classified and sorted, and matched with a pre-established knowledge point classification library to obtain a preliminary knowledge point range division result. Combined with the preset prerequisite relationship network and logical dependency relationship table, the dependency path between the knowledge points is obtained, and the dependency strength between the knowledge points is calculated using a path analysis tool to determine the association hierarchical structure between the knowledge points; Through the records in the historical answer data, the co-occurrence frequency statistics are performed on the knowledge point combinations in the associated hierarchical structure to obtain the number of occurrences of the knowledge point combinations in the answers, and the frequency calculation tool is used to accumulate the numbers to determine the edge weight values ​​between the knowledge points; represents the comprehensive edge weight value between knowledge point i and knowledge point j, Indicates the co-occurrence frequency of two knowledge points, represents the point mutual information value of two knowledge points, α and β are weight coefficients used to balance the influence of frequency and relevance in the final weight; If the edge weight value is higher than a preset threshold, the corresponding knowledge point pair is included in the association strength matrix construction process, and the final matrix data is obtained for subsequent mastery status evaluation to obtain the dynamic evaluation basis of the student on the knowledge point.

4. The method for generating learning tasks based on historical answering and real-time assessment data according to claim 1, characterized in that: Step S3 specifically includes: Based on the knowledge point association strength matrix, the association weight values ​​of each knowledge point and other knowledge points are obtained. By constructing a probability graph structure, the association weight values ​​are converted into conditional probability distribution relationships. The initial state data of the knowledge points are integrated to obtain preliminary probability distribution data. The Bayesian inference method is used to obtain the conditional probability value of each knowledge point from the preliminary probability distribution data. When calculating the posterior probability distribution, if the conditional probability value is lower than a preset threshold, the corresponding knowledge point is marked as a node to be updated, and the posterior probability adjustment range of the node to be updated is determined. By iteratively adjusting the probability value of the node to be updated, obtaining the updated probability value from the posterior probability adjustment range, and performing normalization processing on the final probability value of the knowledge point, the student's current knowledge point state vector is obtained, from which the mastery probability value of each knowledge point is obtained. If the mastery probability value is lower than a preset threshold, it is marked as a weak point, and the student's knowledge mastery status is judged; The student's posterior probability data is obtained from the state description, and the probability value of each knowledge point is compared with a preset mastery threshold. If the probability value of a knowledge point is lower than the preset mastery threshold, it is marked as a weak knowledge point, thereby obtaining a preliminary screened set of weak knowledge points; the associated data directly related to the marked weak knowledge points are extracted from a pre-established strength matrix, and a corresponding sub-network structure is generated using a graph traversal tool to determine the range of knowledge points closely connected to the weak knowledge points; Construct a correlation sub-matrix for the weak knowledge points, obtain the correlation strength values ​​between the knowledge points in the sub-matrix, and if there are knowledge point pairs with high-intensity correlation, increase their priority to obtain an optimized correlation sub-matrix. Combined with the task sequence design requirements, obtain the learning path data corresponding to the high-intensity correlation knowledge points, generate a personalized task sequence through the path planning tool, and determine the final task sequence plan.

5. The method for generating learning tasks based on historical answering and real-time assessment data according to claim 1, characterized in that: Step S4 specifically includes: According to the preset task sequence length limit, at least one task combination is extracted from the weak knowledge point association submatrix, a mastery improvement score is calculated for the task combination, and a preliminary task sequence candidate set is obtained by ranking the scores to obtain an initial set containing multiple task combinations; For the task combinations obtained from the initial set, a pre-established difficulty grading rule library is used to determine the difficulty coefficient of the task combination. If the difficulty coefficient exceeds a preset threshold, it is marked as a high-difficulty task combination, and the task set that needs to be split is determined; According to the high-difficulty task set, the subtask units within the set are obtained, the subtask units are recombined using a logical splitting tool, and it is determined whether the subtask units meet a preset difficulty threshold, thereby obtaining an adjusted task sequence; For the adjusted task sequence, a data comparison tool is used to verify the score difference with the initial set. If the difference is less than a preset threshold, the task sequence is included in the final candidate set to determine the final combination for difficulty grading. This formula represents the score difference between the adjusted task sequence and the initial set, where represents the score of the adjusted task sequence, represents the score of the initial set, · represents the absolute value, and ΔS represents the difference in scores between the two.

6. The method for generating learning tasks based on historical answering and real-time assessment data according to claim 1, characterized in that: Step S5 specifically includes: According to the preset task difficulty grading standard, the difficulty score data of each task is extracted from the task sequence candidate set, and the score data is classified and processed. The score data is compared with the preset threshold value using the interval matching method. If the score data is lower than the first threshold, it is classified as the basic level; if the score data is between the first threshold and the second threshold, it is classified as the advanced level; if the score data is higher than the second threshold, it is classified as the challenging level, and a preliminary stratification result is obtained; By deeply analyzing the task content in the preliminary stratification results, the knowledge points involved in the task content are obtained, and the knowledge points are matched with the content depth level rules. If the task content only involves the explanation of a single knowledge point, it is classified as concept understanding; if the task content involves the combination of knowledge points and operations, it is classified as application practice, and the depth level classification is determined; According to the depth level classification and combined with the difficulty level, a two-dimensional matrix distribution of the task sequence is generated, and the matrix distribution is optimized and adjusted. The task sequence is rearranged using a sorting tool to obtain an optimized sequence; the rhythm distribution of the optimized sequence is calculated to obtain the distribution density of each level. If the distribution density is higher than a preset threshold, the splitting tool is used to adjust it to determine whether the final sequence meets the rhythm requirements.

7. The method for generating learning tasks based on historical answering and real-time assessment data according to claim 1, characterized in that: Step S6 specifically includes: Obtaining the learning preferences and historical speed data from a student database, classifying each student's learning ability level using pre-established classification rules based on their personalized information, and using comparative analysis tools to determine the degree of match between the level and task difficulty to obtain a preliminary task difficulty benchmark value; Based on the preliminary task difficulty benchmark value and the preset threshold of the incremental pace, a logical calculation tool is used to adjust the increase in task difficulty. If the increase exceeds the preset threshold, the incremental pace is reduced; otherwise, the original pace is maintained, thereby determining a final task difficulty increment plan. Obtaining a task type classification rule through the final task difficulty increasing scheme, matching the explanation content, the practice intensity, and the assessment criteria with the task difficulty, using a sorting tool to adjust the content sequence, determining whether the sequence is consistent with the adaptability of the student's state, and obtaining an optimized task presentation sequence; For the optimized task presentation sequence, combined with a dynamic adjustment mechanism, real-time feedback data during the student's learning process is obtained, and a conditional judgment tool is used. If the feedback data indicates a change in the student's status, the content proportion and task type distribution of the customized sequence are adjusted to determine the final personalized learning task sequence.

8. The method for generating learning tasks based on historical answering and real-time assessment data according to claim 1, characterized in that: Step S7 specifically includes: Obtaining student answering behavior data from a historical answering record database, classifying the accuracy and time-consuming information in the answering behavior data, and determining the state transition probability distribution of students at different task difficulties using a pre-established probability calculation tool to obtain an initial student state vector; If the probability value in the initial student state vector is lower than a preset threshold, the student's answer feedback data is collected through a real-time assessment tool, and the task difficulty parameter is dynamically updated based on the feedback data and the preset content depth adjustment rules. It is determined whether the updated task difficulty parameter matches the student's ability, and an adjusted task allocation plan is obtained; Adopting the adjusted task allocation scheme, updating the probability distribution of the student state vector through a posterior probability calculation tool, obtaining adjusted state vector data, and using a data comparison tool to determine the student's mastery trend based on the adjusted state vector data to obtain a mastery assessment result; According to the mastery assessment results, the student state vector data is updated to the database through a data storage tool to obtain the latest student ability profile. Based on the changes in the student ability profile, a logical judgment tool is used to determine the need for task content adjustment to obtain the final task optimization plan.

9. The method for generating learning tasks based on historical answering and real-time assessment data according to claim 1, characterized in that: Step S8 specifically includes: Obtaining the posterior probability distribution data from the student status record, comparing the probability value of each knowledge point with a preset mastery threshold, and marking at least one knowledge point as a weak knowledge point if the posterior probability is lower than the threshold, thereby obtaining a preliminarily screened set of weak knowledge points; Extract error frequency distribution data related to the weak knowledge point set based on historical feedback records, and determine whether the error frequency of the weak knowledge point exceeds the tolerance range in combination with the preset error tolerance range. If so, mark it as the high-risk knowledge point and determine a list of high-risk knowledge points; For the list of high-risk knowledge points, obtain the relevant correlation submatrix data based on the knowledge point coverage requirements, use a database query tool to extract the correlation density value in the submatrix, and if the correlation density is lower than a preset threshold, adjust the task difficulty parameter to obtain an updated task difficulty configuration; According to the task difficulty configuration, combined with the associated density value and the time allocation constraint, the task sequence is optimized by a task sequence generation tool to determine the final task sequence content.

10. The method for generating learning tasks based on historical answering and real-time assessment data according to claim 1, characterized in that: Step S9 specifically includes: Based on the student's learning time constraints and the time allocation for task completion, a task set that matches the student's available time period is obtained from a pre-established task database. The task set is preliminarily screened based on its difficulty and knowledge point coverage to obtain an initial task list that meets the time constraints. By balancing and adjusting the difficulty values ​​of the tasks in the initial task list, the tasks are grouped using a preset difficulty threshold range. If the average difficulty value of a group of tasks in the group exceeds the upper threshold, the difficulty weights of some tasks in the group are reduced to obtain a task combination with a balanced difficulty distribution; Based on the task combination with a balanced difficulty distribution, the content presentation order of the task combination is obtained, and the order is adjusted according to the logical progressive relationship of the knowledge points. If the preceding content of a knowledge point in the task combination is not covered, relevant practice tasks are supplemented from the database to determine the final task presentation sequence; Through the final task presentation sequence, combined with the student's learning progress record, the assessment task content corresponding to the sequence is obtained, and dynamic matching is performed based on the mastery level of the knowledge points. If the mastery level is lower than the preset threshold, intensive practice tasks for the relevant knowledge points are added to obtain a personalized task set that includes practice and assessment.

Images