Assignment arrangement method and system based on learning information

Abstract

The application provides a homework arrangement method and system based on learning information, in the method, multi-source learning information is collected; class students are grouped based on an improved clustering algorithm, and a knowledge point topology graph is constructed by using an effective learning vector set and a conduction weight is calculated, a time sequence weighted weak frequency is calculated in combination with a dynamic mastery sequence, a weak combination with high frequency and high conduction is screened, and a center vector cluster is generated; a multi-objective optimization function is constructed, a maximum profile coefficient, a minimum group knowledge point overlap rate and a maximum teaching constraint satisfaction degree are taken as targets, an optimal clustering number is solved and output; similarity of the effective learning vector and the center vector cluster is calculated based on Gower distance, and the clustering center is iteratively updated until convergence; a hybrid recommendation algorithm of content recommendation and collaborative filtering is adopted, and candidate homework questions are matched according to the characteristic labels of each group. The embodiment solves the problems that the existing homework arrangement method cannot fully utilize learning information and homework arrangement lacks pertinence.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence, mainly involving the field of smart education, and more specifically, to a method and system for assigning homework based on learning information. Background Technology

[0002] Homework is an effective way for teachers to check students' mastery of knowledge and provide feedback on teaching effectiveness. It is also an important way for students to acquire knowledge and develop learning habits. Therefore, assigning homework scientifically and reasonably is a key link in achieving a closed teaching loop and improving teaching quality. However, there are still significant shortcomings in the current homework assignment field: on the one hand, mainstream homework assignment methods emphasize commonality over individuality. Usually, all students in a class are assigned the same homework, making it difficult for teachers to accurately understand each student's learning status through homework. On the other hand, although some personalized teaching methods on the market can distribute homework randomly, their main purpose in teaching is to prevent students from copying homework, and they are not deeply integrated with students' actual learning situation. As a result, it is difficult for students' learning process to achieve true personalization, the intelligence in personalized education is weak, and the actual teaching effect is poor.

[0003] Currently, there are still significant shortcomings in the field of homework assignment. Although existing technologies have established a preliminary process for collecting learning data, basic clustering and grouping, and homework delivery, the learning analysis is based only on numerical data such as the degree of mastery of knowledge points. This cannot make full use of learning information, resulting in insufficient targeting of subsequent homework delivery. Furthermore, teachers need to manually match strategies, which is inefficient and cannot adjust homework content according to students' actual learning situation, making it difficult to achieve truly personalized learning support. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for assigning homework based on learning information, so as to at least solve the technical problem that existing homework assignment methods cannot make full use of learning information, resulting in insufficient targeting of subsequent homework assignments.

[0005] To achieve the above objectives, the embodiments of this application provide the following technical solutions.

[0006] According to one embodiment of this application, a method for assigning homework based on learning information is provided, comprising the following steps:

[0007] Step S101: Collect multi-source learning information; use the Kalman filter algorithm to smooth the structured data of the multi-source learning information to obtain a dynamic mastery sequence; use the BERT fine-tuning model to perform entity recognition and feature extraction on the unstructured data of the multi-source learning information, and convert it into quantitative labels of error type and severity; convert the scenario-based data of the multi-source learning information into teaching scenario coefficients.

[0008] Step S102: Group students in the class based on the improved clustering algorithm, construct a knowledge point topology map using the effective learning vector set and calculate the transmission weight, calculate the time-weighted weak frequency in combination with the dynamic mastery degree sequence, screen the weak combinations with high frequency and high transmission, and generate the central vector cluster;

[0009] Step S103: Construct a multi-objective optimization function with the objectives of maximizing the silhouette coefficient, minimizing the overlap rate of knowledge points between groups, and maximizing the satisfaction of teaching constraints. Solve and output the optimal number of clusters.

[0010] Step S104: Calculate the similarity between the effective learning vector and the center vector cluster based on Gower distance, iteratively update the cluster centers until convergence, and output the affiliation relationship between students and groups and the feature labels of each group;

[0011] Step S105: Use a hybrid recommendation algorithm that combines content recommendation and collaborative filtering to match candidate assignment questions based on the feature labels of each group; solve the optimal combination of assignments for assignment placement using a constraint satisfaction problem algorithm.

[0012] Furthermore, in step S101, the smoothing process using the Kalman filter algorithm includes:

[0013] The degree of students' mastery of knowledge points is taken as the state of a dynamic system, and a single mastery level is treated as a noisy observation. Data smoothing is achieved through iterative prediction and updating using Kalman filtering. The state equation is expressed as:

[0014]

[0015] In the formula, A represents the true confidence level of the k-th observation; A represents the state transition coefficient. Indicates process noise; This indicates the true confidence level of the (k-1)th observation;

[0016] The observation equation is expressed as: ,in, H represents the original knowledge of the k-th observation, and H represents the observation coefficient. Indicates observation noise;

[0017] For each knowledge point of each student, a dynamic mastery sequence is generated.

[0018] Furthermore, in step S102, the effective learning information vector set is obtained through the following steps:

[0019] Step S201: Based on multiple weak knowledge points covered by the class as the basic dimensions, each basic dimension corresponds to the latest value of the dynamic mastery sequence; add error type labels, error severity quantification values, and teaching scenario coefficients as extended dimensions;

[0020] Step S202: Based on the dynamic weights assigned to each dimension by the self-attention mechanism, the basic and extended dimensions are integrated to generate a weighted learning information vector set; based on the teaching scenario coefficient, the original abnormal scores of the isolated forest algorithm are corrected, abnormal students are identified, and after classifying and matching the corresponding intervention strategies, an effective learning information vector set is obtained.

[0021] Furthermore, in step S102, the step of calculating the time-weighted weak frequencies includes:

[0022] The dynamic mastery sequence is weighted according to time distance, and the number of weak points for each knowledge point of each student is accumulated after time-weighted summation; a high-frequency weak point set H is extracted, and the transmission correlation between any two knowledge points i and j in set H is calculated. Based on transmission correlation Select a combination of high conductivity and weak materials;

[0023] For each high-transmission weak combination, extract the effective learning information vectors of all students in that combination, and calculate the mean of the vectors as the initial center vector; if the combination contains k knowledge points, generate k center vectors to form a center vector cluster, and match the number of vectors in the cluster with the candidate value of the optimal clustering number K.

[0024] Furthermore, in step S103, the silhouette coefficient is maximized to quantify the similarity of students within a group and the differences between students between groups. The silhouette coefficient ranges from [-1, 1] and is obtained in the following way:

[0025] The average distance within a group is calculated as the average Gower distance between student i and all other students in the same group, expressed as:

[0026]

[0027] in, This indicates the group that student i belongs to. Representation group The number of people, This represents the Gower distance between student i and student j;

[0028] Calculate student i and all other non-students. The minimum of the average distances between groups is expressed as:

[0029]

[0030] in, Indicates except Any group outside of, Representation group The number of people, This represents the Gower distance between student i and student j;

[0031] Calculate the overall profile coefficient of the class , represented as:

[0032]

[0033]

[0034] In the formula, N represents the total number of students in the class. This represents the profile coefficient of a single student.

[0035] Furthermore, in step S103, minimizing the overlap rate of knowledge points between groups is used to represent the proportion of the intersection of the core weak knowledge points of any two groups of students to the union of the core weak knowledge points of the two groups.

[0036] Wherein, the overlap rate for any two sets of k and m , represented as:

[0037]

[0038] In the formula, This represents the set of core weak knowledge points for group k. This represents the set of core weak knowledge points of group m; This represents the number of intersections between two sets of knowledge points. This represents the number of items in the union of two sets of knowledge points.

[0039] The average overlap rate of all pairs is expressed as: In the formula, Let K represent the overlap rate between any two groups k and m, and let K represent the total number of groups.

[0040] Furthermore, in step S103, the step of solving for and outputting the optimal number of clusters includes:

[0041] The range of K values ​​is initially selected based on teaching constraints; the upper limit of K in the initial selection is no less than 2, and the lower limit is the minimum value among the number of class hours per week, the number of homework templates, and the lower limit of group size.

[0042] Initialize the NSGA-II population and set its parameters;

[0043] For each candidate K value, calculate the score, expressed as: In the formula, Represents the overall profile coefficient of the class. This represents the average overlap rate across all pairs. Indicates the degree of satisfaction of teaching constraints;

[0044] In the non-dominated sorting and crowding calculation, individuals in the population are stratified according to their dominance relationship. The first layer is the Pareto optimal layer where no individual dominates. The crowding distance of each individual in its layer is calculated, and individuals with large crowding distances are retained. The crowding distance is used to measure the degree of dispersion of individuals in the target space.

[0045] In the iterative optimization and convergence determination, the next generation of population is generated through roulette wheel selection, single-point crossover and gene position flipping mutation. When the total objective score of the Pareto optimal layer of the population fluctuates less than the threshold for several consecutive generations, the iteration stops and the K value with the highest total objective score in the Pareto optimal layer is taken as the optimal solution.

[0046] Furthermore, in step S104, the similarity between the effective learning vector and the central vector cluster is calculated based on the Gower distance, and is expressed as:

[0047]

[0048] In the formula, This represents the effective learning information vector of student i and the center vector of the kth group. The smaller the distance value, the higher the similarity; F represents the total dimension of the vectors; This represents the dynamic weight of the f-th dimension; Represents the local distance in the f-th dimension;

[0049] In the local distance In the context of continuous data, For discrete data, 0 or 1, where 0 indicates the same label and 1 indicates different labels;

[0050] In the iterative update of cluster centers, based on the cluster of center vectors generated by high-frequency, high-transmission, and weak combinations, for each student i, the relationship between the cluster and all center vectors is calculated. The dynamic weights are calculated using the Gower distance, and the dynamic weights are assigned to the group with the smallest distance.

[0051] Furthermore, in step S105, dynamic weights of group features are introduced to adjust the ratio of content recommendation to collaborative filtering in the hybrid recommendation algorithm, and the results are calculated based on the comprehensive score. Sort the questions from highest to lowest to determine the candidate pool and filter out questions with duplicate knowledge points; construct an optimization model with educational scenario constraints and use a weighted backtracking algorithm to solve for the optimal combination;

[0052] The step of using a weighted backtracking algorithm to solve for the optimal combination includes:

[0053] Step S301: Sort the constraints by weight and prioritize satisfying the constraints with higher weights;

[0054] Step S302: Each time a candidate topic is selected, calculate the degree of satisfaction of the current combination with each constraint and remove branches that do not satisfy the high-weight constraints;

[0055] Step S303: Output the combination of jobs that satisfies all constraints and has the highest overall score.

[0056] According to another embodiment of this application, a homework assignment system based on learning information is provided, including the following modules:

[0057] The learning information preprocessing module is used to collect learning information from multiple sources. The structured data of the multi-source learning information is smoothed using the Kalman filter algorithm to obtain a dynamic mastery sequence. The unstructured data of the multi-source learning information is used to perform entity recognition and feature extraction using the BERT fine-tuning model, which is converted into quantitative labels of error type and severity. The scenario-based data of the multi-source learning information is converted into teaching scenario coefficients.

[0058] The knowledge point screening module is used to group students in a class based on an improved clustering algorithm, construct a knowledge point topology map using an effective learning vector set and calculate the transmission weight, and combine the dynamic mastery degree sequence to calculate the time-weighted weak frequency, screen the high-frequency and highly transmissible weak combinations, and generate a central vector cluster.

[0059] The optimization and solution module is used to construct a multi-objective optimization function with the objectives of maximizing the profile coefficient, minimizing the overlap rate of knowledge points between groups, and maximizing the satisfaction of teaching constraints. It solves and outputs the optimal number of clusters.

[0060] The iterative update module is used to calculate the similarity between the effective learning vector and the center vector cluster based on the Gower distance, iteratively update the cluster centers until convergence, and output the affiliation relationship between students and groups and the feature labels of each group.

[0061] The assignment combination recommendation module uses a hybrid recommendation algorithm that combines content recommendation and collaborative filtering to match candidate assignment questions based on the feature labels of each group; and solves the optimal assignment combination for assignment placement through a constraint satisfaction problem algorithm.

[0062] Compared with the prior art, the beneficial effects of the embodiments of this application are:

[0063] This embodiment uses Kalman filtering to smooth structured data, which can eliminate random errors in a single test; it uses the BERT fine-tuning model to process unstructured data, which improves the accuracy of entity recognition and can also convert textual evaluations into quantitative labels; it also converts contextual data into unified coefficients; and through collaborative processing of multi-source data, it avoids misjudgment of single-dimensional data, making the learning data more consistent with the students' true level.

[0064] This embodiment utilizes an improved K-means clustering approach to introduce a topological map of knowledge points and time-weighted weakness frequencies, making the initial centers more closely reflect the actual weakness distribution in the class. It employs multi-objective optimization, including silhouette coefficient, inter-group overlap rate, and teaching constraints, to determine the K value, thereby improving the accuracy of the clustering grouping results. The biconditional convergence provided in this embodiment ensures that students within a group have similar levels of weakness, resulting in grouping results that are both accurate and tailored to the teacher's tutoring workload and teaching progress.

[0065] In the hybrid recommendation algorithm provided in this embodiment, the ratio of content recommendation and collaborative filtering is adjusted based on dynamic weights to ensure that candidate questions match the core weak knowledge points and high-frequency errors within the group. In addition, the weighted backtracking algorithm proposed in this embodiment optimizes the homework combination, and based on the constraints of the educational scenario, it ensures that the homework is both targeted at filling gaps in knowledge and meets the actual needs of class teaching stage, completion time, and other factors. Attached Figure Description

[0066] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0067] In the attached diagram:

[0068] Figure 1 This is a flowchart illustrating the implementation of the homework assignment method based on student learning information in an embodiment of this application.

[0069] Figure 2 This is a sub-flowchart of the homework assignment method based on learning information in an embodiment of this application;

[0070] Figure 3 This is another sub-flowchart of the homework assignment method based on learning information in this application embodiment;

[0071] Figure 4 This is a structural block diagram of the homework assignment system based on learning information, as described in an embodiment of this application.

[0072] Figure 5 This is a hardware structure block diagram of an electronic terminal for a homework assignment method based on learning information according to an embodiment of this application. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0074] In some of the processes described in the specification, claims and accompanying drawings of this invention, there are multiple operations that appear in a specific order. However, it should be clearly understood that these operations may not be performed in the order they appear herein or may be performed in parallel. The sequence numbers of the operations, such as 101, 102, etc., are merely used to distinguish the different operations and do not represent any execution order.

[0075] Furthermore, these processes can include more or fewer operations, and these operations can be executed sequentially or in parallel. It should be noted that the terms "first," "second," etc., used in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0076] The technical solutions in the exemplary embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described exemplary embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0077] The technical solutions provided in the embodiments of this application are described in detail below with reference to the accompanying drawings.

[0078] Figure 1 A flowchart illustrating the implementation of a homework assignment method based on learning information, as provided in an embodiment of the present invention.

[0079] like Figure 1 As shown, the homework assignment method based on learning information provided in this embodiment includes the following steps:

[0080] Step S101: Collect multi-source learning information; use the Kalman filter algorithm to smooth the structured data of the multi-source learning information to obtain a dynamic mastery sequence; use the BERT fine-tuning model to perform entity recognition and feature extraction on the unstructured data of the multi-source learning information, and convert it into quantitative labels of error type and severity; convert the scenario-based data of the multi-source learning information into teaching scenario coefficients.

[0081] In this embodiment, the structured data of multi-source learning information includes quantified values ​​of students' knowledge point mastery in each test, such as calculations based on the correct answer rate of each knowledge point in the test, using a 0-1 interval for quantification; it also includes answering time, such as the total time for students to complete a single test / assignment, and the average time to complete questions corresponding to a single knowledge point; and it also includes the number of incorrect answers, such as the cumulative number of incorrect answers in each test / assignment counted by knowledge point dimension. Regarding the collection method of structured data, this embodiment can automatically synchronize the above structured data through the API interface of the school's teaching management system (such as the academic affairs system or online testing platform); for historical data without system records (such as paper tests), the structured conversion is completed through a manual input module by the teacher.

[0082] Furthermore, in the preprocessing of structured data, to eliminate random errors and obtain a dynamic mastery sequence that reflects the true learning trend of students, this embodiment uses the Kalman filter algorithm for smoothing, including:

[0083] The degree of students' mastery of a certain knowledge point is taken as the state of the dynamic system, and the mastery of a single test / assignment is regarded as a noisy observation. Data smoothing is achieved through iterative prediction and updating using Kalman filtering. The state equation is expressed as:

[0084]

[0085] In the formula, denoted by , indicating the true confidence level of the k-th observation; A represents the state transition coefficient, with a value of 0.7; Indicates process noise; This indicates the true confidence level of the (k-1)th observation;

[0086] The observation equation is expressed as: ,in, H represents the original knowledge of the k-th observation, and H represents the observation coefficient. Indicates observation noise;

[0087] For each knowledge point of each student, a dynamic mastery sequence is generated. For example, the initial mastery sequence of a certain knowledge point based on the last 5 observations is: {0.5, 0.55, 0.3, 0.68, 0.72}. The mastery sequence after smoothing with the Kalman filter algorithm is: {0.5, 0.55, 0.62, 0.68, 0.72}. This eliminates the 0.3 mastery level, which is the error in the initial mastery sequence. After eliminating random errors, the fluctuation of the mastery level is no greater than 0.15.

[0088] Therefore, this embodiment uses the Kalman filter algorithm to process structured data, which can avoid the interference of abnormal word data on the judgment of learning progress.

[0089] Furthermore, in this embodiment, the unstructured data for multi-source learning information refers to qualitative learning information that cannot be directly quantified, such as teachers' subjective evaluation texts of students and students' notes on incorrect answers. The unstructured data is collected through text input via the teacher's learning evaluation module and the student's incorrect answer note module. For paper comments or notes, OCR text recognition methods can be used to identify the comments or notes in the paper images, converting them into electronic text. This electronic text is then verified by the teacher / student, and the verified electronic text can be used for unstructured data.

[0090] In this embodiment, unstructured data of multi-source learning information is processed using a BERT fine-tuning model for entity recognition and feature extraction. In the step of converting this data into quantitative labels for error type and severity, a BERT-base model fine-tuned based on educational domain corpus is used for entity recognition and feature quantification. The educational domain corpus consists of three years of teacher comments from within the school, student error notes, and the publicly available educational text dataset Knewton. The fine-tuning parameters are set as follows: batch size is 32, initial learning rate is 2e-5, number of iterations is 5, the optimizer is AdamW, the optimizer weight decay coefficient is 0.01, and the loss function is cross-entropy loss.

[0091] Specifically, in entity recognition, knowledge point entities and error type entities are extracted from the text;

[0092] In feature quantization, the severity of error types is labeled, and the severity label is quantized into a specific value in the range of 0-1; finally, an unstructured data quantization label set is generated.

[0093] In this embodiment, the contextualized data of multi-source learning information is used to characterize the teaching background that generates learning and the environmental information that affects the interpretation of learning, such as the teaching stage of knowledge points, the overall mastery of the class, and the curriculum standard level of knowledge points; wherein, the curriculum standard level in this embodiment is divided into understanding level, comprehension level, mastery level and application level.

[0094] In the scenario-based data collection method, the teaching stage and curriculum standard level are preset by the teacher in the knowledge point management module; the overall class mastery is calculated based on the dynamic mastery sequence of individual students. Specifically, the overall class mastery is the average of the sum of the latest dynamic mastery values ​​of all valid students.

[0095] In this embodiment, the update frequency of the contextual data is synchronized with the update frequency of the structured data.

[0096] Furthermore, in the step of transforming scenario-based data of multi-source learning information into teaching scenario coefficients, the labels / quantified values ​​of the scenario-based data are converted into unified teaching scenario coefficients. Specifically, the conversion is achieved through label mapping and numerical normalization: for the teaching stage coefficient, the smaller the coefficient, the higher the requirement for high mastery; for the overall class mastery coefficient, the original quantified value (0-1) is directly used to judge the deviation between individual student mastery and the group; for the curriculum standard level coefficient, the larger the coefficient, the higher the importance of the knowledge point.

[0097] The final set of teaching scenario coefficients is generated, which includes the overall class mastery coefficient, teaching stage coefficient, and curriculum standard level coefficient.

[0098] In step S101 of this embodiment, Kalman filtering is used to smooth structured data, which can eliminate random errors in a single test; BERT fine-tuning model is used to process unstructured data, which improves the accuracy of entity recognition and can also convert text evaluations into quantitative labels; and scenario-based data is converted into unified coefficients. Through the above processing, multi-source data can be processed collaboratively, avoiding misjudgment of single-dimensional data and making learning data more consistent with students' true level.

[0099] Please continue to refer to Figure 1 The homework assignment method based on learning information provided in this application embodiment further includes the following steps:

[0100] Step S102: Group students in the class based on the improved clustering algorithm, construct a knowledge point topology map using the effective learning vector set and calculate the transmission weight, calculate the time-weighted weak frequency in combination with the dynamic mastery degree sequence, screen the weak combinations with high frequency and high transmission, and generate the central vector cluster;

[0101] The clustering algorithm used in this embodiment is the K-means clustering algorithm.

[0102] Specifically, such as Figure 2 As shown, in step S102, the effective learning information vector set is obtained through the following steps:

[0103] Step S201: Based on multiple weak knowledge points covered by the class as the basic dimensions, each basic dimension corresponds to the latest value of the dynamic mastery sequence; add error type labels, error severity quantification values, and teaching scenario coefficients as extended dimensions;

[0104] Step S202: Based on the dynamic weights assigned to each dimension by the self-attention mechanism, the basic and extended dimensions are integrated to generate a weighted learning information vector set; based on the teaching scenario coefficient, the original abnormal scores of the isolated forest algorithm are corrected, abnormal students are identified, and after classifying and matching the corresponding intervention strategies, an effective learning information vector set is obtained.

[0105] Furthermore, in the step of assigning dynamic weights to each dimension based on the self-attention mechanism, the calculation of the attention weights is expressed as follows:

[0106]

[0107] In the formula, This represents the attention score for the f-th dimension; if the dimension is a core knowledge point, ,in, This represents the curriculum standard grade coefficient for the f-th dimension;

[0108] If the dimension is an error type, then:

[0109]

[0110] in, Indicates the frequency of error types. This indicates the total frequency of errors in the class.

[0111] If the dimension is another dimension, then , This represents the numerical value of the corresponding dimension.

[0112] Furthermore, in the step of assigning dynamic weights to each dimension based on the self-attention mechanism, the curriculum standard level coefficient is used to determine the weight of the knowledge point dimension. For example, the curriculum standard level coefficient is set to 1.5 for the knowledge point dimension at the mastery level, while the curriculum standard level coefficient is set to 1.0 for the knowledge point dimension at the understanding level, so as to avoid secondary knowledge points from affecting the clustering results.

[0113] Similarly, in this embodiment, the teaching stage coefficient is used to assist in adjusting the error type dimension weight: if the teaching stage coefficient is for a new lesson, the error dimension weight for the concept understanding category is 1.2, prioritizing the identification of concept errors caused by the initial learning; if the teaching stage coefficient is for a review lesson, the error dimension weight for the calculation error category is set to 1.2, prioritizing the identification of calculation problems that still exist after review; this embodiment allows the effective learning vector to focus on the teaching focus, such as core knowledge points and typical errors in the current teaching stage, so that they can be identified first in subsequent clustering processing, avoiding secondary features dominating the subsequent grouping process.

[0114] In the step of correcting the original abnormal scores of the isolated forest algorithm based on the teaching scenario coefficient, the isolated forest algorithm is used to calculate the original abnormal scores of student learning data. For example, if a student's knowledge point mastery is 0.3, the original score is high and easily judged as abnormal. Therefore, this embodiment further adjusts the score based on the teaching scenario coefficient: if the teaching stage coefficient is 0.8 for a new lesson and the overall class mastery coefficient is 0.35, then the student's mastery of 0.3 is within the normal level of a new lesson, and the abnormal score is reduced after correction. At this time, it is not judged as abnormal. If the teaching stage coefficient is 0.4 for a review lesson, then the mastery of 0.3 will be judged as a systematic weakness abnormality.

[0115] Furthermore, in the step of classifying and matching corresponding intervention strategies, the priority of abnormalities is distinguished by the curriculum standard level coefficient: if the curriculum standard level coefficient of the weak knowledge point of the abnormal student is mastery level, then the knowledge point breakthrough path is matched first; if the curriculum standard level coefficient is understanding level, then the basic consolidation task is matched to avoid excessive intervention on low-priority knowledge points; the intervention strategy of this embodiment can ensure that the abnormality judgment is consistent with the actual teaching situation.

[0116] Furthermore, in this embodiment, the steps of constructing a knowledge point topology map using an effective learning vector set and calculating the transmission weights include:

[0117] Based on the subject curriculum standards and target teaching materials, a directed acyclic topological graph is constructed. The nodes of the graph are knowledge points, and the directed edges of the graph represent the preceding and following relationships of knowledge points. For example, there is a directed edge between an algebraic expression and a linear equation in one variable.

[0118] In this embodiment, the pre- and post-knowledge relationships represent the dependencies between knowledge points. For example, when learning a post-knowledge point, it is necessary to master the pre-knowledge points as a prerequisite, and the pre-knowledge points are the foundation for learning the post-knowledge points. When one post-knowledge point corresponds to multiple pre-knowledge points, they are sorted by their dependency (dependency = the proportion of content in the post-knowledge point that uses the pre-knowledge points). Pre-knowledge points with a dependency ≥ 50% are marked as core pre-knowledge points, and those with a dependency < 50% are marked as auxiliary pre-knowledge points. When one pre-knowledge point corresponds to multiple post-knowledge points, directed edges are marked sequentially according to the chapter order of the target teaching materials.

[0119] Additionally, it should be noted that the target teaching aids in this embodiment are versions that are adapted to the subject curriculum standards. When the order of knowledge points in the teaching aids conflicts with the curriculum standards, the logical order specified in the curriculum standards shall prevail.

[0120] When calculating the transmission weights, each directed edge is assigned a transmission weight W. For example, the weight of a direct preceding relationship is set to 0.9, the weight of an indirect preceding relationship is set to 0.5, and the weight of a connection with at least two knowledge points in between is set to 0.3. An N×N transmission weight matrix is ​​generated, with matrix elements... This represents the transmission weight of knowledge point i to knowledge point j.

[0121] When calculating the time-weighted frequency of weaknesses, the dynamic mastery sequence is weighted according to time distance, and the time-weighted number of weaknesses for each knowledge point of each student is accumulated; a high-frequency weakness set H is extracted, and the transmission correlation between any two knowledge points i and j in set H is calculated. Select The combination of these is considered a high-conductivity, low-conductivity combination;

[0122] For each high-transmission weak combination, extract the effective learning information vectors of all students in that combination, and calculate the mean of the vectors as the initial center vector; if the combination contains k knowledge points, generate k center vectors to form a center vector cluster, and match the number of vectors in the cluster with the candidate value of the optimal clustering number K.

[0123] In step S102 of this embodiment, an improved K-means clustering method is used to introduce a knowledge point topology map and time-weighted weak frequency, and the initial center is more in line with the actual weak distribution of the class; the K value is determined by multi-objective optimization including silhouette coefficient, inter-group overlap rate and teaching constraints, which can improve the accuracy of clustering grouping results.

[0124] In addition, the dual-condition convergence provided in this embodiment can ensure that students in the group have similar levels of weakness, and the grouping results are both accurate and suitable for the teacher's tutoring workload and teaching progress.

[0125] Please continue to refer to Figure 1 The homework assignment method based on learning information provided in this application embodiment further includes the following steps:

[0126] Step S103: Construct a multi-objective optimization function with the objectives of maximizing the silhouette coefficient, minimizing the overlap rate of knowledge points between groups, and maximizing the satisfaction of teaching constraints. Solve and output the optimal number of clusters.

[0127] In this embodiment, the silhouette coefficient is maximized to quantify the similarity of students within a group and the differences between students in different groups. The silhouette coefficient ranges from -1 to 1. The closer the silhouette coefficient is to 1, the better the grouping effect, that is, the more similar the students are within the group and the more different the students are between the groups. In specific implementation, the average distance within the group is calculated first. For example, the average Gower distance between student i and all other students in the same group is calculated as follows:

[0128]

[0129] in, This indicates the group that student i belongs to. Representation group The number of people, This represents the Gower distance between student i and student j;

[0130] Further calculate student i and all other non- The minimum of the average distances between groups is expressed as:

[0131]

[0132] in, Indicates except Any group outside of, Representation group The number of people, This represents the Gower distance between student i and student j;

[0133] Calculate the overall profile coefficient of the class , represented as:

[0134]

[0135]

[0136] In the formula, N represents the total number of students in the class. This represents the profile coefficient of a single student.

[0137] In this embodiment, the minimized inter-group knowledge point overlap rate is used to represent the proportion of the intersection of the core weak knowledge points of any two groups of students to the union of the core weak knowledge points of the two groups. The lower the proportion, the greater the difference in the weak characteristics of the two groups, and the more targeted the grouping results are. Specifically, for any two groups k and m, the overlap rate... , represented as:

[0138]

[0139] In the formula, This represents the set of core weak knowledge points for group k. This represents the set of core weak knowledge points of group m; This represents the number of intersections between two sets of knowledge points. This represents the number of items in the union of two sets of knowledge points.

[0140] The average overlap rate of all pairs is then expressed as:

[0141]

[0142] In the formula, Let K represent the overlap rate between any two groups k and m, and let K represent the total number of groups.

[0143] In maximizing the satisfaction of teaching constraints in this embodiment, the teaching constraints are broken down into quantifiable sub-constraints, including group size constraints. Class time adaptation constraints Resource matching constraints The group size constraint is that each group has no less than 5 people and the number of people in each group is no more than 20% of the class size. For example, for a class of 40 people, the number of people in each group is between [5, 8]. In the class time adaptation constraint, the number of clusters K is no more than the number of weekly subject tutoring hours. In the resource matching constraint, the number of clusters K is no more than the number of existing group assignment templates.

[0144] Furthermore, the degree of satisfaction of teaching constraints, T, is represented as: The values ​​of the three sub-constraints are all 0 or 1, with 1 being the value when the sub-constraint is satisfied and 0 being the value when it is not satisfied.

[0145] Furthermore, the specific steps in solving and outputting the optimal number of clusters include:

[0146] The range of K values ​​is initially selected based on teaching constraints to avoid invalid calculations; the upper limit of K in the initial selection is no less than 2, and the lower limit is the minimum value among the number of weekly class hours, the number of homework templates, and the lower limit of group size.

[0147] Initialize the NSGA-II population and set the following parameters:

[0148] The population size is set to 20, the encoding method is integer encoding, and the gene locus of each individual is a K value; for each K value candidate, a score is calculated, such as clustering based on the K value, calculating the silhouette coefficient, inter-group overlap rate, and teaching constraint satisfaction corresponding to the K value, and substituting them into the total target score formula to obtain the score of the K value;

[0149] In this embodiment, the scoring formula is expressed as: ;

[0150] In the formula, Represents the overall profile coefficient of the class. This represents the average overlap rate across all pairs. Indicates the degree of satisfaction of teaching constraints;

[0151] In the non-dominated ranking and crowding calculation, individuals in the population are stratified according to their dominance relationship. If all target scores of individual A are not lower than those of individual B, and at least one target score is higher than that of B, then A dominates B. The first layer is the Pareto optimal layer where no individual dominates. The crowding distance of each individual in its layer is calculated, and individuals with large crowding distances are retained to avoid the optimal solution being concentrated in local areas. The crowding distance is used to measure the degree of dispersion of individuals in the target space.

[0152] In the iterative optimization and convergence determination, the next generation population is generated through roulette wheel selection, single-point crossover, and gene locus flipping mutation. The crossover probability of the genetic operation parameters is set to 0.8, and the mutation probability is set to 0.02.

[0153] When the total objective score of the Pareto optimal layer fluctuates by ≤0.05 for 5 consecutive generations of the population, the iteration stops, and the K value with the highest total objective score in the Pareto optimal layer is taken as the optimal solution.

[0154] Please continue to refer to Figure 1 The homework assignment method based on learning information provided in this application embodiment further includes the following steps:

[0155] Step S104: Calculate the similarity between the effective learning vector and the center vector cluster based on Gower distance, iteratively update the cluster centers until convergence, and output the affiliation relationship between students and groups and the feature labels of each group;

[0156] In this embodiment, the similarity between the effective learning vector and the central vector cluster calculated based on Gower distance is represented as follows:

[0157]

[0158] In the formula, This represents the effective learning information vector of student i and the center vector of the kth group. The smaller the distance value, the higher the similarity; F represents the total dimension of the vectors; This represents the dynamic weight of the f-th dimension; Represents the local distance in the f-th dimension;

[0159] In this embodiment, the dynamic weights In the context of core knowledge points, the weights are... The value is set to 1.5 to prioritize the clustering accuracy of core knowledge points; for the high-frequency error type dimension, the weight is... The value is set to 1.2 to enhance the differentiation of error causes; for other dimensions, the weights are... The value in the region is 1.0.

[0160] Furthermore, in the local distance In the context of continuous data, For discrete data, 0 or 1, where 0 indicates the same label and 1 indicates different labels.

[0161] In the iterative update of cluster centers, based on the cluster of center vectors generated by high-frequency, high-transmission, and weak combinations, for each student i, the relationship between the cluster and all center vectors is calculated. The dynamic weights are calculated using Gower distance, and the dynamic weights are assigned to the groups with the smallest distances. The center vectors are then updated. During the update, for each group k, the mean of all student vectors in that group is recalculated as the new center. , represented as: In the formula, This represents the number of people in group k. This represents the value of student i in the f-th dimension;

[0162] In this embodiment, a teaching constraint is introduced as the condition for stopping the iteration. In specific implementation, the algorithm convergence condition and the teaching constraint condition must be met simultaneously. The algorithm convergence condition is that the distance between the new and old center vectors of all groups is not greater than 0.05. The teaching constraint condition is that the standard deviation of the students' mastery of the core knowledge points in each group is not greater than 0.15. The teaching constraint condition is used to ensure that the students in the group have similar levels of weakness, which facilitates unified tutoring.

[0163] Preferably, the maximum number of iterations in this embodiment is set to 50 to avoid infinite iteration loops.

[0164] Furthermore, in the step of outputting the affiliation between students and groups and the characteristic tags of each group, a structured output containing the following tags is generated, such as core weak knowledge point tags, error reason tags, and teaching adaptation tags; wherein, the core weak knowledge point tags include knowledge points that are weak for no less than 60% of students in the group, sorted by the frequency of weakness; the error reason tags include high-frequency error types and percentages in the group; and the teaching adaptation tags include recommended tutoring duration and suitable course type.

[0165] Please continue to refer to Figure 1 The homework assignment method based on learning information provided in this application embodiment further includes the following steps:

[0166] Step S105: Use a hybrid recommendation algorithm that combines content recommendation and collaborative filtering to match candidate assignment questions based on the feature labels of each group; solve the optimal combination of assignments for assignment placement using a constraint satisfaction problem algorithm.

[0167] Specifically, this application's embodiments introduce dynamic weights for group features to adjust the ratio of content recommendation to collaborative filtering in the hybrid recommendation algorithm, expressed as follows:

[0168]

[0169] In the formula, This represents the overall score of the candidate topic q; a higher value indicates a higher degree of matching. Indicates dynamic weights. If the core weak knowledge points within the group are clearly identified, then Prioritize content recommendations to enhance the relevance of knowledge points. If the reasons for errors within a group are scattered, then... To increase the collaborative filtering weight;

[0170] This represents the content recommendation score, which is calculated based on the matching degree between the title and the group feature tags, and is expressed as:

[0171]

[0172] In the formula, This indicates the degree of match between question q and the core weak knowledge points of the group. This indicates that question q is specific to the high-frequency error types within the group. This indicates the degree to which the difficulty level of the questions matches the average level of the students in the group;

[0173] The collaborative filtering score is calculated based on the accuracy of historical similar groups on question q, and is expressed as:

[0174]

[0175] In the formula, S represents the set of historical groups whose similarity to the current group k is greater than a threshold (e.g., 0.8); group similarity The calculation is performed using the Gower distance inverse transformation, and is expressed as follows: ; This represents the dynamic weighted Gower distance between group k and historical group s. This represents the maximum Gower distance among all pairs in the historical set S. This represents the accuracy rate of history group s on question q;

[0176] Based on the overall score Sort the questions from highest to lowest and select the top 50 as a candidate pool to filter out questions with duplicate knowledge points and avoid content redundancy.

[0177] Furthermore, an optimization model incorporating educational scenario constraints is constructed, and a weighted backtracking algorithm is used to solve for the optimal combination. The educational scenario constraints include knowledge point coverage, difficulty gradient, completion time, question quantity control, and error type matching, with the weights of the five educational scenario constraints decreasing sequentially.

[0178] In the educational scenario of this embodiment, whether the homework covers weak knowledge points is the main indicator for judging the value of the homework. If the knowledge points are not covered enough, even if the difficulty, number of questions and time are appropriate, the homework cannot solve the students' main problems. Therefore, the coverage of knowledge points is set as the highest weight to ensure that the relevance of the homework takes precedence over the adaptability of the form, which is in line with the goal of personalized homework assignment.

[0179] like Figure 3 As shown, the steps of solving the optimal combination using a weighted backtracking algorithm in this embodiment include:

[0180] Step S301: Sort the constraints by weight and prioritize satisfying the constraints with higher weights;

[0181] Step S302: Each time a candidate topic is selected, the degree of satisfaction of the current combination with each constraint is calculated, branches that do not satisfy the high-weight constraints are removed to avoid wasting computing power in directions where it is impossible to satisfy the high-weight constraints, while ensuring that each step of topic selection can move closer to the goal of satisfying all constraints.

[0182] Step S303: Output the combination of jobs that satisfies all constraints and has the highest overall score.

[0183] In the hybrid recommendation algorithm provided in this embodiment, the ratio of content recommendation and collaborative filtering is adjusted based on dynamic weights to ensure that candidate questions match the core weak knowledge points and high-frequency errors within the group. In addition, the weighted backtracking algorithm proposed in this embodiment optimizes the homework combination, and the constraints of 5 types of educational scenarios ensure that the homework is both targeted at filling gaps and meets the actual needs of class teaching stage, completion time and other factors.

[0184] Please refer to Figure 4 According to another embodiment of this application, a homework assignment system based on learning information is provided, including the following modules:

[0185] The learning information preprocessing module 401 is used to collect multi-source learning information, smooth the structured data of the multi-source learning information using the Kalman filter algorithm to obtain a dynamic mastery sequence; perform entity recognition and feature extraction using the BERT fine-tuning model for the unstructured data of the multi-source learning information, and convert it into quantitative labels of error type and severity; and convert the scenario-based data of the multi-source learning information into teaching scenario coefficients.

[0186] The knowledge point screening module 402 is used to group students in a class based on an improved clustering algorithm, construct a knowledge point topology map using an effective learning vector set and calculate the transmission weight, and combine the dynamic mastery degree sequence to calculate the time-weighted weak frequency, screen the weak combinations with high frequency and high transmission, and generate a central vector cluster.

[0187] The optimization and solution module 403 is used to construct a multi-objective optimization function with the objectives of maximizing the profile coefficient, minimizing the overlap rate of knowledge points between groups, and maximizing the satisfaction of teaching constraints, and solves and outputs the optimal number of clusters.

[0188] The iterative update module 404 is used to calculate the similarity between the effective learning vector and the center vector cluster based on the Gower distance, iteratively update the cluster centers until convergence, and output the affiliation relationship between students and groups and the feature labels of each group.

[0189] The assignment combination recommendation module 405 is used to match candidate assignment questions based on the feature labels of each group using a hybrid recommendation algorithm of content recommendation and collaborative filtering; and to solve the optimal assignment combination for assignment placement through a constraint satisfaction problem algorithm.

[0190] like Figure 5 As shown, in this embodiment of the invention, an electronic terminal is provided. At the hardware level, the terminal includes a processor and optionally also includes an internal bus, a network interface, and a memory.

[0191] The memory may include main memory, such as high-speed random access memory, or non-volatile memory, such as at least one disk drive. Of course, the electronic device may also include other hardware required for its operations.

[0192] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA bus, PCI bus, or EISA bus, etc.

[0193] The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0194] Memory is used to store programs. Specifically, programs may include program code, which includes computer operation instructions. Memory may include main memory and non-volatile memory, and provides instructions and data to the processor.

[0195] The processor reads the corresponding computer program from non-volatile memory into memory and then runs it, forming a homework assignment system based on learning information at the logical level.

[0196] The processor executes the program stored in the memory, and specifically performs the steps of the homework assignment method based on learning information provided in the above embodiments.

[0197] The above is as stated in this application. Figure 1 The homework assignment method based on learning information disclosed in the illustrated embodiment can be applied to a processor, or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities.

[0198] In implementation, each step of the above method can be completed by integrated logic circuits in the processor hardware or by instructions in software. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor, etc.; it can also be a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor.

[0199] The steps of the method disclosed in the embodiments of this application can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor.

[0200] The software module can reside in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0201] This application also proposes a computer-readable storage medium that stores one or more programs, the programs including instructions that, when executed by an electronic device including multiple applications, enable the electronic device to perform... Figure 1 The illustrated embodiment shows a homework assignment method based on student learning information.

[0202] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media containing computer-usable program code.

[0203] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0204] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0205] These computer program instructions can also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0206] In a typical configuration, a computing device includes one or more processors, input / output interfaces, network interfaces, and memory.

[0207] Memory may include non-persistent storage in computer-readable media, random access memory, and / or non-volatile memory, such as read-only memory or flash memory. Memory is an example of computer-readable media.

[0208] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data.

[0209] Examples of computer storage media include, but are not limited to, phase-change memory, static random access memory, dynamic random access memory, other types of random access memory, read-only memory, electrically erasable programmable read-only memory, flash memory or other memory technologies, optical disc read-only memory, digital versatile optical disc or other optical storage, magnetic cassette tape, magnetic tape disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media do not include temporary computer-readable media such as modulated data signals and carrier waves.

[0210] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0211] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for assigning homework based on student learning information, characterized in that, Includes the following steps: Step S101: Collect multi-source learning information. For the structured data of the multi-source learning information, use the Kalman filter algorithm for smoothing. Specifically, the student's mastery of knowledge points is treated as the dynamic system state, and a single mastery level is considered a noisy observation. Data smoothing is achieved through iterative prediction and update using Kalman filtering. For each knowledge point of each student, a dynamic mastery sequence is generated. For the unstructured data of the multi-source learning information, use the BERT fine-tuning model for entity recognition and feature extraction, converting it into quantitative labels for error type and severity. Finally, convert the scenario-based data of the multi-source learning information into teaching scenario coefficients. Step S102: Group students in the class based on the improved clustering algorithm, construct a knowledge point topology map using the effective learning vector set, calculate the transmission weight, and calculate the time-weighted weak frequency in combination with the dynamic mastery sequence. The dynamic mastery sequence is weighted according to time distance, and the time-weighted weak frequency is accumulated for each knowledge point of each student. High-frequency and highly transmissible weak combinations are selected to generate central vector clusters. The effective learning vector set is obtained through the following steps: using multiple weak knowledge points covered by the class as the basic dimension, each basic dimension corresponding to the latest value of the dynamic mastery sequence; adding error type labels, error severity quantification values, and teaching scenario coefficients as extended dimensions; using the self-attention mechanism to assign dynamic weights to each dimension, merging the basic and extended dimensions to generate a weighted learning vector set; correcting the original abnormal scores of the isolated forest algorithm based on the teaching scenario coefficient, identifying abnormal students, classifying and matching corresponding intervention strategies, and obtaining the effective learning vector set. Step S103: Construct a multi-objective optimization function with the objectives of maximizing the silhouette coefficient, minimizing the overlap rate of knowledge points between groups, and maximizing the satisfaction of teaching constraints. Solve and output the optimal number of clusters. The silhouette coefficient is used to quantify the similarity of students within a group and the differences between students between groups. Step S104: Calculate the similarity between the effective learning vector and the center vector cluster based on Gower distance, iteratively update the cluster centers until convergence, and output the affiliation relationship between students and groups and the feature labels of each group; Step S105: Use a hybrid recommendation algorithm that combines content recommendation and collaborative filtering to match candidate assignment questions based on the feature labels of each group; solve the optimal combination of assignments for assignment placement using a constraint satisfaction problem algorithm.

2. The homework assignment method based on learning information according to claim 1, characterized in that, In step S101, where the student's mastery of the knowledge points is taken as the state of the dynamic system, the state equation is expressed as: In the formula, A represents the true confidence level of the k-th observation; A represents the state transition coefficient. Indicates process noise; This indicates the true confidence level of the (k-1)th observation; In the step of treating a single mastery level as a noisy observation, the observation equation is expressed as: ,in, H represents the original knowledge of the k-th observation, and H represents the observation coefficient. This indicates observation noise.

3. The homework assignment method based on learning information according to claim 2, characterized in that, In step S102, the step of screening for weak combinations with high frequency and high conductivity to generate a central vector cluster includes: Extract the high-frequency weak set H, and calculate the transmission correlation between any two knowledge points i and j in set H. Based on transmission correlation Select a combination of high conductivity and weak materials; For each high-transmission weak combination, extract the effective learning information vectors of all students in that combination, and calculate the mean of the vectors as the initial center vector; if the combination contains k knowledge points, generate k center vectors to form a center vector cluster, and match the number of vectors in the cluster with the candidate value of the optimal clustering number K.

4. The homework assignment method based on learning information according to claim 3, characterized in that, In step S103, the profile coefficient ranges from [-1, 1] and is obtained as follows: The average distance within a group is calculated as the average Gower distance between student i and all other students in the same group, expressed as: in, This indicates the group that student i belongs to. Representation group The number of people, This represents the Gower distance between student i and student j; Calculate student i and all other non-students. The minimum of the average distances between groups is expressed as: in, Indicates except Any group outside of, Representation group The number of people, This represents the Gower distance between student i and student j; Calculate the overall profile coefficient of the class , represented as: In the formula, N represents the total number of students in the class. This represents the profile coefficient of a single student.

5. The homework assignment method based on learning information according to claim 4, characterized in that, In step S103, the inter-group knowledge point overlap rate is used to represent the proportion of the intersection of the core weak knowledge points of any two groups of students to the union of the core weak knowledge points of the two groups. Wherein, the overlap rate for any two sets of k and m , represented as: In the formula, This represents the set of core weak knowledge points for group k. This represents the set of core weak knowledge points of group m; This represents the number of intersections between two sets of knowledge points. This represents the number of items in the union of two sets of knowledge points. The average overlap rate of all pairs is expressed as: In the formula, Let K represent the overlap rate between any two groups k and m, and let K represent the total number of groups.

6. The homework assignment method based on learning information according to claim 5, characterized in that, Step S103, the step of solving and outputting the optimal number of clusters, includes: The range of K values ​​is initially selected based on teaching constraints; the upper limit of K in the initial selection is no less than 2, and the lower limit is the minimum value among the number of weekly class hours, the number of homework templates, and the lower limit of group size. Initialize the NSGA-II population and set its parameters; For each candidate K value, calculate the score, expressed as: In the formula, Represents the overall profile coefficient of the class. This represents the average overlap rate across all pairs, and T represents the degree of satisfaction of teaching constraints. In the non-dominated sorting and crowding calculation, individuals in the population are stratified according to their dominance relationship. The first layer is the Pareto optimal layer where no individual dominates. The crowding distance of each individual in its layer is calculated, and individuals with large crowding distances are retained. The crowding distance is used to measure the degree of dispersion of individuals in the target space. In the iterative optimization and convergence determination, the next generation of population is generated through roulette wheel selection, single-point crossover and gene position flipping mutation. When the total objective score of the Pareto optimal layer of the population fluctuates less than the threshold for several consecutive generations, the iteration stops and the K value with the highest total objective score in the Pareto optimal layer is taken as the optimal solution.

7. The homework assignment method based on learning information according to claim 6, characterized in that, In step S104, the similarity between the effective learning vector and the central vector cluster is calculated based on the Gower distance, and is expressed as: In the formula, This represents the effective learning information vector of student i and the center vector of the kth group. The smaller the distance value, the higher the similarity; F represents the total dimension of the vector; This represents the dynamic weight of the f-th dimension; Represents the local distance in the f-th dimension; In the local distance In the context of continuous data, ; For discrete data, 0 or 1, where 0 indicates the same label and 1 indicates different labels; In the iterative update of cluster centers, based on the cluster of center vectors generated by high-frequency, high-transmission, and weak combinations, for each student i, the relationship between the cluster and all center vectors is calculated. The dynamic weights are calculated using the Gower distance, and the dynamic weights are assigned to the group with the smallest distance.

8. The homework assignment method based on learning information according to claim 7, characterized in that, In step S105, dynamic weights of group features are introduced to adjust the ratio of content recommendation to collaborative filtering in the hybrid recommendation algorithm, and the results are calculated based on the comprehensive score. Sort the questions from highest to lowest to determine the candidate pool, and filter out questions with duplicate knowledge points; An optimization model incorporating educational scenario constraints is constructed, and a weighted backtracking algorithm is used to solve for the optimal combination. The step of using a weighted backtracking algorithm to solve for the optimal combination includes: Step S301: Sort the constraints by weight and prioritize satisfying the constraints with higher weights; Step S302: Each time a candidate topic is selected, calculate the degree of satisfaction of the current combination with each constraint and remove branches that do not satisfy the high-weight constraints; Step S303: Output the combination of jobs that satisfies all constraints and has the highest overall score.

9. A system for implementing the homework assignment method based on learning information as described in any one of claims 1 to 8, characterized in that, Includes the following modules: The learning information preprocessing module is used to collect learning information from multiple sources. The structured data from these sources is smoothed using a Kalman filter algorithm. Specifically, the student's mastery of knowledge points is treated as a dynamic system state, and a single mastery level is considered a noisy observation. Data smoothing is achieved through iterative prediction and updates using Kalman filtering, generating a dynamic mastery sequence for each knowledge point for each student. For the unstructured data from these multiple sources, a BERT fine-tuning model is used for entity recognition and feature extraction, transforming it into quantitative labels for error type and severity. Finally, the contextualized data from these multiple sources is converted into teaching scenario coefficients. The knowledge point screening module is used to group students in a class based on an improved clustering algorithm, construct a knowledge point topology map using an effective learning vector set, calculate transmission weights, and calculate time-weighted weakness frequencies by combining the dynamic mastery sequence. Specifically, the dynamic mastery sequence is weighted according to time distance, and the time-weighted weakness frequency is accumulated for each knowledge point of each student. High-frequency and highly transmissible weakness combinations are screened to generate central vector clusters. The effective learning vector set is obtained through the following steps: using multiple weak knowledge points covered by the class as basic dimensions, each basic dimension corresponding to the latest value of the dynamic mastery sequence; adding error type labels, error severity quantification values, and teaching scenario coefficients as extended dimensions; using a self-attention mechanism to assign dynamic weights to each dimension, merging the basic and extended dimensions to generate a weighted learning vector set; and correcting the original abnormal scores of the isolated forest algorithm based on the teaching scenario coefficient, identifying abnormal students, classifying and matching corresponding intervention strategies to obtain the effective learning vector set. The optimization solution module is used to construct a multi-objective optimization function with the objectives of maximizing the silhouette coefficient, minimizing the overlap rate of knowledge points between groups, and maximizing the satisfaction of teaching constraints. It solves and outputs the optimal number of clusters. Among them, the silhouette coefficient is used to quantify the similarity of students within a group and the differences between students between groups. The iterative update module is used to calculate the similarity between the effective learning vector and the center vector cluster based on the Gower distance, iteratively update the cluster centers until convergence, and output the affiliation relationship between students and groups and the feature labels of each group. The assignment combination recommendation module uses a hybrid recommendation algorithm combining content recommendation and collaborative filtering to match candidate assignment questions based on the feature labels of each group; and solves the optimal assignment combination for assignment placement through a constraint satisfaction problem algorithm.

Images