Personalized learning recommendation method based on long time series knowledge mastery degree dynamic modeling

Abstract

The present application relates to the field of intelligent education and education data processing, and particularly relates to a personalized learning recommendation method based on long-time knowledge mastery dynamic modeling. The method comprises: obtaining a learner-related learning log file, cognitive interaction data and a preset knowledge graph, generating learning behavior records and knowledge graph data through data preprocessing; constructing a knowledge point network structure, generating a knowledge point set and a knowledge point association relationship, and forming a multi-dimensional historical performance feature vector of the knowledge mastery level of each knowledge point; identifying a target knowledge point when new cognitive interaction data arrives, updating the feature vector set of the target knowledge point and associated knowledge points, and then generating the knowledge mastery level at a future target time, outputting a priority review order, a learning path and a resource sequence. The present application effectively improves the matching of learning recommendation and long-time knowledge mastery state and the accuracy of resource sequence generation.

Description

Technical Field

[0001] This invention relates to the fields of smart education and educational data processing, and in particular to a personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery. Background Technology

[0002] In the fields of smart education and educational data processing, existing personalized learning recommendation solutions typically acquire learners' relevant learning log files, cognitive interaction data, and pre-set knowledge graphs. After data preprocessing to form learning behavior records, learner models or learning paths are constructed based on knowledge points, teaching interaction content, and learner feedback. However, these solutions suffer from limitations such as a static representation of knowledge mastery levels, knowledge point relationships primarily used for path organization, and a lack of continuous generation of knowledge mastery levels at future target times. Existing methods often rely on past quiz results, error rates, repetition counts, learning gains, or learner feedback to determine the current learning status, and then generate recommended content based on target knowledge points, prerequisite nodes, or matching practice question lists. This approach suffers from the problem of difficulty in synchronously applying newly added cognitive interaction data to related knowledge points.

[0003] In scenarios where long-term learning behavior records continuously increase, learners' new cognitive interaction data on target knowledge points typically only update that target knowledge point or the learner's overall status. It's difficult to write the error rate, repetition count, learning gains, learner feedback, and current knowledge mastery level of the target knowledge point into related knowledge points according to their pre- and post-requirement relationships, prior nodes, necessary nodes, parallel nodes, coverage relationships, and / or relationships. This easily leads to situations where the status of related knowledge points lags behind and the priority review order deviates from the learning behavior records, making it difficult to meet the requirement of stable generation of learning paths and resource sequences.

[0004] For the joint processing of target knowledge points, knowledge point relationships, and updated feature vector sets, existing technologies generally lack a continuous process from adding new cognitive interaction data to updating related knowledge points, from the updated feature vector set to the knowledge mastery level at the future target time, and then to the priority review order and resource sequence. Existing solutions struggle to form a consistent chain of collection, alignment, updating, generation, and recording in personalized learning recommendation scenarios, making it difficult for the knowledge mastery level at the future target time to jointly support the subsequent learning path and resource sequence with the knowledge point relationships. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery, comprising:

[0006] S100: Obtain learner-related learning log files, cognitive interaction data, and preset knowledge graphs; generate learning behavior records and knowledge graph data through data preprocessing.

[0007] S200: Construct a knowledge point network structure based on the learning behavior records and knowledge graph data, and generate a knowledge point set and knowledge point association relationships;

[0008] S300. Based on the knowledge point set, the knowledge point relationship, and learning behavior records, generate a multi-dimensional historical performance feature vector of the knowledge mastery level of each knowledge point.

[0009] S400: Receive newly added cognitive interaction data, identify target knowledge points from the knowledge point set, and update the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge points.

[0010] The newly added cognitive interaction data refers to new records generated by learners after completing tasks, learning interactive teaching content, matching practice question lists, learning resource sequences, correcting mistakes, or receiving learner feedback. These records include interactive teaching content, past task results, interaction time, error rate, number of repetitions, learning benefits, and learner feedback.

[0011] S500. Based on the target knowledge points, the relationships between knowledge points, and the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge points, update the multi-dimensional historical performance feature vector of the knowledge mastery level of the related knowledge points, and generate an updated feature vector set.

[0012] S600. Based on the updated feature vector set and learning behavior records, generate the knowledge mastery level for the future target time.

[0013] S700: Generate a priority review order based on the level of knowledge mastery and the relationship between knowledge points at the future target time, and generate a learning path and resource sequence based on the priority review order.

[0014] Furthermore, the learning behavior record includes the results of each answer, teaching interaction content, interaction time, error rate, number of repetitions, learning benefits, learner feedback, semantic features of questions, basic attribute features of students, matching exercise question list record, and resource sequence record.

[0015] Furthermore, the knowledge point network structure consists of first-level branches, second-level branches, knowledge points, and interactive teaching content; the knowledge point relationships include pre-order and post-order relationships, prior nodes, necessary nodes, parallel nodes, coverage relationships, and OR relationships.

[0016] Furthermore, the multi-dimensional historical performance feature vector of the knowledge mastery level includes short-term learning gains, medium-term learning gains, long-term learning gains, error rate, number of repetitions, learner feedback, current knowledge mastery level, historical information transmission status, historical information forgetting status, influence status of prior knowledge points, influence status of subsequent knowledge points, and response status of matching practice questions.

[0017] Furthermore, the short-term learning gain is generated based on the most recent 10 answer records, the medium-term learning gain is generated based on the middle 50 answer records, and the long-term learning gain is generated based on all historical answer records; the short-term learning gain, the medium-term learning gain, and the long-term learning gain are normalized and then concatenated to form a cross-scale historical performance feature vector.

[0018] Furthermore, in S400, updating the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge point includes: identifying the target knowledge point based on the teaching interaction content in the newly added cognitive interaction data; updating the current knowledge mastery level of the target knowledge point based on the answer results, error rate, and number of repetitions; updating the historical information forgetting status of the target knowledge point based on the interaction time; and updating the response status of the matching practice questions of the target knowledge point based on learner feedback.

[0019] Furthermore, in S500, updating the multi-dimensional historical performance feature vector of the knowledge mastery level of related knowledge points includes: when the related knowledge point is a preorder node of the target knowledge point, updating the influence status of the preorder knowledge point of the related knowledge point according to the error rate and repetition count of the target knowledge point; when the related knowledge point is a subsequent knowledge point of the target knowledge point, updating the influence status of the subsequent knowledge point of the related knowledge point according to the current knowledge mastery level of the target knowledge point; when the related knowledge point and the target knowledge point have parallel nodes, coverage relationships, or OR relationships, updating the historical information transmission status of the related knowledge point according to learning benefits and learner feedback.

[0020] Furthermore, in S600, generating the knowledge mastery level for the future target time includes: reading the short-term learning gain, medium-term learning gain, long-term learning gain, interaction time, historical information transmission status, and historical information forgetting status from the updated feature vector set; obtaining the time interval corresponding to the future target time based on the interaction time; and correcting the current knowledge mastery level based on the time interval and the historical information forgetting status to generate the knowledge mastery level for the future target time.

[0021] Furthermore, in S700, generating a priority review order includes: generating knowledge point ranking data based on the current knowledge mastery level, the knowledge mastery level at the future target time, and the learning objectives; and modifying the knowledge point ranking data based on the prior nodes, necessary nodes, parallel nodes, coverage relationships, and / or relationships in the knowledge point association relationship to generate a priority review order.

[0022] Furthermore, in S700, generating learning paths and resource sequences includes: when a knowledge point in the priority review order has a preorder node, placing the preorder node before the corresponding knowledge point; when a knowledge point in the priority review order has parallel nodes, coverage relationships, or relationships, selecting teaching interaction content based on learner feedback; matching practice question lists and learning resources for each knowledge point according to the priority review order, generating a resource sequence, and returning the learner feedback corresponding to the resource sequence as new cognitive interaction data to S400.

[0023] The key innovations of this invention include:

[0024] (1) Based on the target knowledge points, the knowledge point relationships and the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge points, update the multi-dimensional historical performance feature vector of the knowledge mastery level of the related knowledge points, and generate the updated feature vector set, so that the knowledge point relationships participate in the updating process of the multi-dimensional historical performance feature vector of the knowledge mastery level.

[0025] (2) Based on the updated feature vector set and learning behavior records, generate the knowledge mastery level for the future target time, so that the knowledge mastery level for the future target time is formed by the target knowledge points, related knowledge points and learning behavior records.

[0026] (3) Generate a priority review order based on the knowledge mastery level and knowledge point correlation of the future target time, and generate a learning path and resource sequence based on the priority review order, so that the learning path and resource sequence can inherit the processing results of the updated feature vector set.

[0027] The following are its main beneficial effects:

[0028] (1) In view of the problem that newly added cognitive interaction data is difficult to be applied to related knowledge points in a synchronous manner, this invention updates the multi-dimensional historical performance feature vector of the knowledge mastery level of related knowledge points through the target knowledge point, the knowledge point relationship and the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge point, so that the learning behavior record of the target knowledge point is entered into the status record of the related knowledge point, reducing the situation of the status lag of the related knowledge point.

[0029] (2) In response to the problem of the lack of continuous generation of knowledge mastery level for future target time, this invention generates knowledge mastery level for future target time through updated feature vector set and learning behavior record, so that short-term learning benefits, medium-term learning benefits, long-term learning benefits, historical information transmission status and historical information forgetting status enter the same processing link to form knowledge mastery level for future target time.

[0030] (3) In response to the problem that the priority review order deviates from the learning behavior record, the present invention generates the priority review order based on the knowledge mastery level and knowledge point correlation at the future target time, so that the current knowledge mastery level, the knowledge mastery level at the future target time and the knowledge point correlation together affect the knowledge point ranking data, forming a priority review order corresponding to the learning behavior record.

[0031] (4) In view of the problem that the expression of knowledge mastery level in the existing scheme is too static, the present invention generates a multi-dimensional historical performance feature vector of knowledge mastery level of each knowledge point through learning behavior records, knowledge point set and knowledge point association relationship, and updates the target knowledge point and related knowledge point after the arrival of new cognitive interaction data, so that the knowledge mastery level changes continuously with the learning behavior record.

[0032] (5) To address the problem of the lack of continuous input links in learning paths and resource sequences, this invention sequentially transmits learning behavior records, knowledge graph data, knowledge point network structure, multi-dimensional historical performance feature vectors of knowledge mastery level, updated feature vector set, and knowledge mastery level at future target time to the priority review order generation process, so that learning paths and resource sequences are built on continuous processing results. Attached Figure Description

[0033] Figure 1 A flowchart illustrating the personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery provided in this application embodiment;

[0034] Figure 2 The diagram shows the structure of the personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery provided in this application embodiment. Detailed Implementation

[0035] Example 1: Refer to Figure 1 This is a flowchart illustrating a personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery provided in an embodiment of the present invention. The process may include at least steps S100-S700:

[0036] S100: Obtain learner-related learning log files, cognitive interaction data, and preset knowledge graphs; generate learning behavior records and knowledge graph data through data preprocessing.

[0037] S200: Construct a knowledge point network structure based on learning behavior records and knowledge graph data to generate a set of knowledge points and the relationships between them.

[0038] S300. Based on the knowledge point set, the knowledge point relationship, and learning behavior records, generate a multi-dimensional historical performance feature vector of the knowledge mastery level of each knowledge point.

[0039] S400: Receive new cognitive interaction data, identify target knowledge points from the knowledge point set, and update the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge points;

[0040] S500. Based on the target knowledge points, the relationships between knowledge points, and the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge points, update the multi-dimensional historical performance feature vector of the knowledge mastery level of the related knowledge points, and generate an updated feature vector set.

[0041] S600. Based on the updated feature vector set and learning behavior records, generate the knowledge mastery level for the future target time.

[0042] S700: Generate a priority review order based on the level of knowledge mastery and the relationship between knowledge points at the future target time, and generate a learning path and resource sequence based on the priority review order.

[0043] S100: Obtain learner-related learning log files, cognitive interaction data, and preset knowledge graphs; generate learning behavior records and knowledge graph data through data preprocessing.

[0044] In step S100, the data preprocessing module of the personalized learning recommendation platform receives learner-related learning log files, cognitive interaction data, and a preset knowledge graph, using these three types of inputs as the data source for this step. The learner-related learning log files are learning records formed by learners on the online learning platform, the answer page corresponding to the question network graph, the teaching interaction content page, the matching practice question list page, and the resource sequence page. These records include historical course registration records, course scores, course-related attribute data, past answer results, resource sequence records, and learner feedback. The cognitive interaction data is the interaction data generated by learners when learning the teaching interaction content corresponding to knowledge points. This data includes teaching interaction content, interaction time, error rate, repetition count, learning gains, question semantic features, and student basic attribute features. The preset knowledge graph is graph data composed of knowledge points, teaching interaction content, questions, course-related attribute data, and knowledge point relationships. It includes first-level branches, second-level branches, knowledge points, pre- and post-relationships, preorder nodes, necessary nodes, parallel nodes, coverage relationships, and / or relationships.

[0045] Specifically, the data preprocessing module receives the learner's relevant learning log files and cognitive interaction data when the learner logs into the online learning platform and performs actions such as answering, browsing, submitting, correcting, reviewing, clicking resource sequences, or providing feedback. The data preprocessing module reads the learner identifier, course identifier, question identifier, teaching interaction content identifier, knowledge point identifier, and interaction time, and groups the previous answers for the same learner, the same course, and the same knowledge point into the same learning record. The interaction time uses a timestamp generated by the online learning platform. The previous answers are recorded as correct, incorrect, and incomplete. The error rate is generated by the number of incorrect answers for the same knowledge point corresponding to the total number of answers. The repetition count is generated by the number of times the same teaching interaction content or the same question appears under the same learner's name. The learning gains are generated jointly by the previous answers, course score, and completion status of the teaching interaction content. The learner feedback is recorded through feedback entry points on the resource sequence page, the matching exercise list page, and the teaching interaction content page.

[0046] Furthermore, the data preprocessing module performs field completion, duplicate record merging, anomaly record marking, and chronological sorting on the learner-related learning log files and the cognitive interaction data. Field completion maps missing knowledge point identifiers for question records under the same learner to knowledge points in a pre-defined knowledge graph based on the question's semantic features and the teaching interaction content. Duplicate record merging retains only one record among multiple records with the same learner identifier, question identifier, and interaction time, and writes the merging process into the data sample update record. Anomaly record marking adds an anomaly marker to records lacking learner identifiers, interaction times, question identifiers, or teaching interaction content identifiers, and retains the anomaly-marked records in the preprocessing record of the learner-related learning log files. Chronological sorting arranges the answer results, error rate, number of repetitions, learning gains, and learner feedback into a long-term time-series record according to the interaction time, for subsequent use in short-term, medium-term, and long-term learning gains.

[0047] Furthermore, after receiving the preset knowledge graph, the data preprocessing module reads its first-level branches, second-level branches, knowledge points, teaching interaction content, question semantic features, and knowledge point relationships, and assigns node numbers, relationship numbers, and version records to the preset knowledge graph. The node numbers record the positions of the first-level branches, second-level branches, knowledge points, and teaching interaction content within the preset knowledge graph. The relationship numbers record pre- and post-relationships, prior nodes, necessary nodes, parallel nodes, coverage relationships, and / or relationships. The version records record the source, receiving time, and update batch of the preset knowledge graph. When the online learning platform imports new course-related attribute data or new teaching interaction content, the data preprocessing module generates a new version record for the new content and retains the old version of the knowledge graph data in the data sample update record. When a node number conflict occurs, the data preprocessing module matches the nodes according to the order of course identifier, knowledge point identifier, and teaching interaction content identifier, and writes the conflict record to the exception record.

[0048] Understandably, the learning behavior record is formed by learner-related learning log files and cognitive interaction data after field completion, duplicate record merging, abnormal record marking, and chronological arrangement. The learning behavior record includes each answer result, teaching interaction content, interaction time, error rate, number of repetitions, learning gains, learner feedback, question semantic features, student basic attribute features, matching practice question list records, and resource sequence records. The knowledge graph data is formed by a pre-defined knowledge graph after node numbering, relationship numbering, and version recording. The knowledge graph data includes first-level branches, second-level branches, knowledge points, teaching interaction content, pre- and post-relationships, prior nodes, necessary nodes, parallel nodes, coverage relationships, and / or relationships. After this step, the learning behavior record and the knowledge graph data are passed to S200, which constructs a knowledge point network structure based on the learning behavior record and knowledge graph data, and generates a knowledge point set and knowledge point relationships.

[0049] In one engineering embodiment, the online learning platform receives daily learners' past quiz results, course scores, completion status of teaching interaction content, and learner feedback for a specific course. The data preprocessing module automatically reads the newly added cognitive interaction data after the daily learning data is entered into the database and writes it into a long-time sequence record according to learner identification and interaction time. When a learner completes 3 quiz responses, 1 teaching interaction, and 1 matching practice question list feedback for the same knowledge point, the data preprocessing module generates past quiz results from the 3 quiz responses, generates an error rate from the error status and total number of quiz records, generates a repetition count from the number of times the same teaching interaction was accessed repeatedly, generates learning gains from the course score and completion status, and generates learner feedback from the feedback entry content. Subsequently, the data preprocessing module maps the above records to knowledge points in a preset knowledge graph according to the semantic features of the questions, and writes the first-level branch, second-level branch, pre- and post-relationships, and preorder nodes of the knowledge point into the knowledge graph data. If two records of the same question appear at the same interaction time, the data preprocessing module retains the record written to the database earlier and writes the other record into the exception record. If the newly added teaching interaction content does not appear in the preset knowledge graph, the data preprocessing module generates a new version record and temporarily stores the teaching interaction content in the record to be mapped. After the course identifier, knowledge point identifier and question semantic features are matched, it is written into the knowledge graph data.

[0050] In summary, this step achieves the following technical results: By uniformly receiving learner-related learning log files, cognitive interaction data, and pre-set knowledge graphs, it generates learning behavior records and knowledge graph data that can be directly accessed by the S200. This step organizes past quiz results, error rates, repetition counts, learning gains, and learner feedback into long-term time-series records, providing a data source for subsequent multi-dimensional historical performance feature vectors of knowledge mastery levels. This step preserves the source and batches of knowledge graph data during the update process through node numbers, relationship numbers, and version records.

[0051] S200: Construct a knowledge point network structure based on learning behavior records and knowledge graph data to generate a set of knowledge points and the relationships between them.

[0052] In S200, the knowledge point network structure construction unit receives the learning behavior records and knowledge graph data generated in S100, and processes the data from the learning behavior records—including past answer results, teaching interaction content, interaction time, error rate, repetition count, learning benefits, learner feedback, question semantic features, student basic attribute features, matching exercise list records, and resource sequence records—to correspond with the first-level branches, second-level branches, knowledge points, teaching interaction content, pre- and post-relationships, prior nodes, necessary nodes, parallel nodes, coverage relationships, and / or relationships in the knowledge graph data. The knowledge point network structure is a directed structure composed of first-level branches, second-level branches, knowledge points, and teaching interaction content. The first-level branch represents the upper-level content scope of a course or subject. The second-level branch represents the chapter content scope under the first-level branch. The knowledge point represents the learning unit that both the past answer results and teaching interaction content in the learning behavior records point to. The teaching interaction content represents the content corresponding to the knowledge point in the questions, exercises, course resources, and matching exercise list.

[0053] Specifically, the knowledge point network structure construction unit first reads the first-level and second-level branches in the knowledge graph data and establishes branch affiliation relationships according to course identifiers, chapter identifiers, and knowledge point identifiers. These branch affiliation relationships assign each knowledge point to its corresponding second-level branch, and each second-level branch to its corresponding first-level branch. When the same knowledge point appears in multiple teaching interaction content, the knowledge point network structure construction unit reads the question semantic features, teaching interaction content identifiers, and knowledge point identifiers, merges duplicate knowledge points, and retains the corresponding teaching interaction content. For records in the learning behavior record that have a question identifier but lack a knowledge point identifier, the knowledge point network structure construction unit searches for the corresponding knowledge point in the knowledge graph data based on the question semantic features and teaching interaction content. If the search is successful, the record is assigned to the corresponding knowledge point. If the search fails, the record is written to the unmapped record, and the course identifier, question identifier, teaching interaction content identifier, and interaction time are recorded.

[0054] Furthermore, the knowledge point network structure construction unit generates edge relationships between knowledge points based on the pre- and post-relationships, prior nodes, necessary nodes, parallel nodes, coverage relationships, and OR relationships in the knowledge graph data. The pre- and post-relationships indicate that one knowledge point precedes or follows another in the learning order. The prior nodes represent knowledge points that need to be read before the knowledge point corresponding to the learning target. The necessary nodes represent knowledge points that need to be retained in the learning path. The parallel nodes represent knowledge points that are under the same secondary branch and can appear in the same learning stage. The coverage relationship indicates that one knowledge point covers the associated content of another knowledge point. The OR relationship indicates that there are replaceable learning contents among multiple knowledge points. The knowledge point network structure construction unit writes the above edge relationships into the knowledge point association relationships and records the relationship type, starting knowledge point, ending knowledge point, first-level branch, second-level branch, and corresponding teaching interaction content for each knowledge point association relationship.

[0055] Furthermore, the knowledge point network structure construction unit writes learning behavior records into the knowledge point nodes of the knowledge point network structure. Specifically, the knowledge point network structure construction unit reads the results of each answer according to the knowledge point identifier, and writes the error rate, number of repetitions, and learning gains into the corresponding knowledge point node; it reads the resource sequence records and matching exercise list records according to the teaching interaction content identifier, and writes the learner feedback into the corresponding teaching interaction content; it arranges the learning behavior records under the same knowledge point according to the interaction time, generating a long-time sequence record for that knowledge point. For multiple records with the same interaction time under the same knowledge point, the knowledge point network structure construction unit reads the retained records according to the abnormal record processing results in S100, and writes the unused records into the data sample update record. In cases where the knowledge graph data undergoes version changes, the knowledge point network structure construction unit reads the version record and saves the branch affiliation relationships of the same knowledge point in different versions, with the branch affiliation relationships in the current version entering the knowledge point network structure.

[0056] Understandably, the knowledge point set consists of knowledge points after branch affiliation processing, merging of duplicate knowledge points, isolation of records to be mapped, and writing of learning behavior records. Each knowledge point in the knowledge point set has a knowledge point identifier, its first-level branch, its second-level branch, corresponding teaching interaction content, past answer results, error rate, number of repetitions, learning benefits, and learner feedback. The knowledge point association relationship consists of pre- and post-relationships, prior nodes, necessary nodes, parallel nodes, coverage relationships, and / or relationships after the relationship type records are completed. The knowledge point set and the knowledge point association relationship together constitute a knowledge point network structure and serve as input for S300 to generate a multi-dimensional historical performance feature vector of knowledge mastery level for each knowledge point. Simultaneously, the knowledge point association relationship is used for updating associated knowledge points in S500 and for generating priority review order, learning path, and resource sequence in S700.

[0057] In one engineering embodiment, after the online learning platform receives the knowledge graph data of a course, the knowledge point network structure construction unit reads the first-level branch "Function", the second-level branch "Linear Function", the knowledge point "Function Graph", and the teaching interaction content "Function Graph Exercises" from the course. When the learning behavior record generated by S100 shows the learner's three consecutive answers to the "Function Graph Exercises", the knowledge point network structure construction unit categorizes the three consecutive answers into the knowledge point "Function Graph" according to the teaching interaction content identifier, and writes the error rate, number of repetitions, learning gains, and learner feedback. If the knowledge graph data records "Function Concept" as a preorder node of "Function Graph", and "Function Graph" as a prerequisite knowledge point of "Function Application", then the knowledge point network structure construction unit connects "Function Concept", "Function Graph", and "Function Application" into a knowledge point association relationship. If "Function Graph" and "Function Table" are parallel nodes, then they are recorded as parallel node relationships. After the above processing is completed, knowledge points such as "function concept", "function graph", "function application" and "function table" are formed in the knowledge point set. Preorder nodes, predecessor and successor relationships and parallel nodes are formed in the knowledge point association relationship and enter S300.

[0058] In summary, this step embeds the learning behavior records generated by S100 into the knowledge graph data, forming a network structure of knowledge points with each answer result, error rate, repetition count, learning gains, and learner feedback. This step organizes pre- and post-relationships, prerequisite nodes, necessary nodes, parallel nodes, coverage relationships, and / or relationships into knowledge point associations, providing a structural foundation for the hierarchical updating of subsequent target knowledge points and related knowledge points. The knowledge point set and knowledge point associations output by this step enable S300 to generate multi-dimensional historical performance feature vectors of knowledge mastery levels based on knowledge points.

[0059] S300. Based on the knowledge point set, the knowledge point relationship, and learning behavior records, generate a multi-dimensional historical performance feature vector of the knowledge mastery level of each knowledge point.

[0060] In S300, the multi-dimensional historical performance feature vector generation module receives the knowledge point set and knowledge point relationships generated by S200, and calls the learning behavior records generated by S100. The knowledge point set includes knowledge point identifiers, their primary branches, their secondary branches, and corresponding teaching interaction content. The knowledge point relationships include pre- and post-relationships, prior nodes, necessary nodes, parallel nodes, coverage relationships, and OR relationships. The learning behavior records include past answer results, teaching interaction content, interaction time, error rate, number of repetitions, learning gains, learner feedback, question semantic features, student basic attribute features, matching practice question list records, and resource sequence records. The multi-dimensional historical performance feature vector generation module consists of a knowledge point record reading unit, a time window division unit, a learning gains generation unit, an association state writing unit, and a vector recording unit. The knowledge point record reading unit reads each knowledge point in the knowledge point set. The time window division unit organizes the learning behavior records according to the interaction time. The learning gains generation unit generates short-term learning gains, medium-term learning gains, and long-term learning gains. The association state writing unit reads the knowledge point relationships. The vector recording unit generates a multi-dimensional historical performance feature vector of the knowledge mastery level of each knowledge point.

[0061] Specifically, the knowledge point record reading unit takes the set of knowledge points as input, reads the corresponding teaching interaction content one by one according to the knowledge point identifier, and extracts the answer results, interaction time, error rate, repetition count, learning benefits, and learner feedback that are identical to the corresponding teaching interaction content from the learning behavior record. If the same knowledge point corresponds to multiple teaching interaction contents, the knowledge point record reading unit will group the learning behavior records of multiple teaching interaction contents into the same knowledge point. If the same teaching interaction content corresponds to multiple knowledge points, the knowledge point record reading unit will write the learning behavior records into the record area of ​​the corresponding knowledge point according to the semantic features of the question and the relationship between knowledge points. If the learning behavior record is missing answer results or interaction time, the knowledge point record reading unit will read the abnormal record marker in S100 and write the learning behavior record to be completed. The record to be completed is not included in this vector generation.

[0062] Furthermore, the time window segmentation unit arranges the learning behavior records under the same knowledge point into long-term time-series records according to the interaction time, and divides them into short-term time windows, medium-term time windows, and long-term time windows. The short-term time window corresponds to the most recent 10 answer records. The medium-term time window corresponds to the middle 50 answer records. The long-term time window corresponds to all historical answer records. If there are fewer than 10 answer records for a certain knowledge point, the time window segmentation unit assigns all existing answer records to the short-term time window, and records the medium-term and long-term time windows as existing historical answer records under the same knowledge point. If there are more than 50 answer records for a certain knowledge point, the time window segmentation unit extracts the middle 50 answer records according to the interaction time order to generate a medium-term time window. The long-term time window retains all historical answer records, matching exercise list records, and resource sequence records for that knowledge point.

[0063] Furthermore, the learning benefit generation unit reads past quiz results, course scores, completion status of teaching interaction content, and learner feedback according to short-term, medium-term, and long-term time windows, respectively, to generate short-term, medium-term, and long-term learning benefits. The short-term learning benefit represents the learning benefit corresponding to the most recent 10 quiz records. The medium-term learning benefit represents the learning benefit corresponding to the middle 50 quiz records. The long-term learning benefit represents the learning benefit corresponding to all historical quiz records. The learning benefit generation unit normalizes the short-term, medium-term, and long-term learning benefits. Normalization adjusts the values ​​of learning benefits under different knowledge points to a unified recording range while retaining the original record source. After normalization, the learning benefit generation unit concatenates the short-term, medium-term, and long-term learning benefits into a cross-scale historical performance feature vector. This cross-scale historical performance feature vector is written into the vector recording area of ​​the corresponding knowledge point.

[0064] Further, the association state writing unit reads the pre- and post-relationships, prior nodes, necessary nodes, parallel nodes, coverage relationships, and OR relationships of the current knowledge point according to the knowledge point association relationships, and writes the reading results into the association state of the current knowledge point. When the current knowledge point has a prior node, the association state writing unit generates the pre-knowledge point influence state of the current knowledge point based on the error rate, repetition count, and learning benefit of the prior node. When the current knowledge point has a post-knowledge point, the association state writing unit generates the post-knowledge point influence state of the current knowledge point based on the current knowledge mastery level of the current knowledge point. When the current knowledge point has parallel nodes, coverage relationships, and OR relationships, the association state writing unit generates the historical information transmission state based on the learning benefits and learner feedback corresponding to the parallel nodes, coverage relationships, and OR relationships. The historical information forgetting state is generated by the interaction time and the interval between the learning behavior records of the current knowledge point. The matching exercise response state is generated by the matching exercise list records and learner feedback.

[0065] Furthermore, the vector recording unit writes short-term learning gains, medium-term learning gains, long-term learning gains, error rate, number of repetitions, learner feedback, current knowledge mastery level, historical information transmission status, historical information forgetting status, influence status of prior knowledge points, influence status of subsequent knowledge points, and response status of matching practice questions into a multi-dimensional historical performance feature vector of the knowledge mastery level for the same knowledge point. The current knowledge mastery level is generated from the previous answer results, error rate, number of repetitions, and learning gains for that knowledge point. The historical information transmission status is generated from knowledge point associations and learning gains. The historical information forgetting status is generated from interaction time and long-term time-series records. The influence status of prior knowledge points is generated from preorder nodes and prior / subsequent relationships. The influence status of subsequent knowledge points is generated from subsequent knowledge points and the current knowledge mastery level. The response status of matching practice questions is generated from the matching practice question list record and learner feedback. After each knowledge point is written, the vector recording unit generates a vector version record for that knowledge point. The vector version record includes the knowledge point identifier, vector generation time, learning behavior record source, and knowledge graph data version.

[0066] In one engineering embodiment, a learner on an online learning platform has 68 historical answer records, 3 records of completed teaching interactions, 2 records of matching practice question lists, and 1 record of a resource sequence under the knowledge point "Function Graph". The time window division unit selects the 10 most recent answer records to generate a short-term time window based on the interaction time, selects the middle 50 answer records to generate a medium-term time window, and writes all 68 historical answer records into the long-term time window. The learning benefit generation unit generates short-term learning benefits based on the 10 most recent answer records, medium-term learning benefits based on the middle 50 answer records, and long-term learning benefits based on the 68 historical answer records. The association state writing unit reads the knowledge point association relationship generated in S200, and obtains that "Function Concept" is the preorder node of "Function Graph", "Function Application" is the postorder knowledge point of "Function Graph", and "Function Table" is the parallel node of "Function Graph". Subsequently, the vector recording unit writes the short-term learning gains, medium-term learning gains, long-term learning gains, error rate, number of repetitions, learner feedback, current knowledge mastery level, historical information transmission status, historical information forgetting status, influence status of prior knowledge points, influence status of subsequent knowledge points, and response status of matching exercises into the knowledge mastery level multi-dimensional historical performance feature vector of the "function graph", and writes this vector into the feature vector library.

[0067] Understandably, the multi-dimensional historical performance feature vectors of knowledge mastery level for each knowledge point output in this step correspond one-to-one with the knowledge points in the knowledge point set, and maintain correspondence with the pre- and post-relationships, prior nodes, necessary nodes, parallel nodes, coverage relationships, and / or relationships in the knowledge point association relationships. These multi-dimensional historical performance feature vectors of knowledge mastery level serve as input for updating the target knowledge point in S400, as input for updating associated knowledge points in S500, and as input for generating the knowledge mastery level for the future target time in S600. The vector version record is written into the data sample update record and is updated as new content is added to the learning behavior record.

[0068] In summary, this step transforms the knowledge point set, its relationships, and learning behavior records into a multi-dimensional historical performance feature vector of knowledge mastery level for each knowledge point. This step also writes short-term, medium-term, and long-term learning gains, error rate, repetition count, and learner feedback into the vector recording area for the same knowledge point. Furthermore, this step writes the influence status of prior knowledge points, the influence status of subsequent knowledge points, the historical information transmission status, and the historical information forgetting status into the vector structure, providing input for the update processes of S400 and S500.

[0069] S400: Receive new cognitive interaction data, identify target knowledge points from the knowledge point set, and update the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge points;

[0070] In S400, the target knowledge point update module receives newly added cognitive interaction data and calls the knowledge point set generated in S200 and the multi-dimensional historical performance feature vector of knowledge mastery level of each knowledge point generated in S300. The newly added cognitive interaction data is a new record generated by the learner after completing question answering, learning interactive teaching content, matching practice question list, learning resource sequence, correcting mistakes, or receiving learner feedback. It includes teaching interaction content, previous answer results, interaction time, error rate, number of repetitions, learning benefits, and learner feedback. The target knowledge point is the knowledge point corresponding to the newly added cognitive interaction data in the knowledge point set. The multi-dimensional historical performance feature vector of knowledge mastery level of the target knowledge point is a vector record generated by S300 for the target knowledge point, including short-term learning benefits, medium-term learning benefits, long-term learning benefits, error rate, number of repetitions, learner feedback, current knowledge mastery level, historical information transmission status, historical information forgetting status, influence status of preceding knowledge points, influence status of subsequent knowledge points, and response status of matching practice questions.

[0071] Specifically, when the online learning platform receives new question submission records, teaching interaction content completion records, matching exercise list records, resource sequence records, or learner feedback, the target knowledge point update module reads the learner identifier, course identifier, question identifier, teaching interaction content identifier, interaction time, and previous answer results from the newly added cognitive interaction data, and writes the newly added cognitive interaction data into the corresponding learner's learning behavior record. The target knowledge point update module reads the knowledge point identifier, its primary branch, its secondary branch, and the corresponding teaching interaction content from the knowledge point set, and matches the teaching interaction content identifier in the newly added cognitive interaction data with the corresponding teaching interaction content in the knowledge point set. If the match is successful, the knowledge point to which the corresponding teaching interaction content belongs is recorded as the target knowledge point. If the match fails, the target knowledge point update module reads the question semantic features and searches for the knowledge point in the knowledge point set that corresponds to the question semantic features. Newly added cognitive interaction data that fails to find its match is entered into a pending mapping record, and the interaction time, question identifier, teaching interaction content identifier, and learner feedback are written into it.

[0072] Furthermore, the target knowledge point update module reads the multi-dimensional historical performance feature vector of the current knowledge mastery level of the target knowledge point, and writes and replaces the corresponding dimensions according to the newly added cognitive interaction data. When the previous answer results are correct, the target knowledge point update module writes the record into the previous answer results of the target knowledge point and updates the current knowledge mastery level. When the previous answer results are incorrect, the target knowledge point update module writes the record into the error status record, and generates a new error rate based on the number of error status records and the total number of answer records under the same target knowledge point. The number of repetitions is generated by the number of times the same learner enters the same teaching interaction content or the same question under the same target knowledge point. The learning gains are jointly written by the previous answer results, course score, teaching interaction content completion status, and learner feedback in the newly added cognitive interaction data. After completing the above writing, the target knowledge point update module writes the new error rate, number of repetitions, and learning gains into the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge point.

[0073] Furthermore, the target knowledge point update module processes the historical information forgetting state of the target knowledge point according to the interaction time. The interaction time represents the submission time of the new cognitive interaction data in the online learning platform. The target knowledge point update module reads the last interaction time of the target knowledge point and records the interval between the interaction time of the new cognitive interaction data and the last interaction time as the interaction interval. The interaction interval enters the historical information forgetting state. If the target knowledge point does not have a previous interaction time, the target knowledge point update module records the interaction time of the new cognitive interaction data as the initial interaction time and writes the historical information forgetting state into the initial state. If the interaction time of the new cognitive interaction data is earlier than the existing interaction time of the target knowledge point, the target knowledge point update module rearranges the previous answer results, error rate, repetition count, learning benefits, and learner feedback under the target knowledge point in chronological order, and writes the rearrangement process into the data sample update record.

[0074] Furthermore, the target knowledge point update module updates the response status of matching practice questions based on learner feedback. This learner feedback comes from the teaching interaction content page, the matching practice question list page, and the resource sequence page, and includes completed, incomplete, corrected, relearned, and feedback text. The target knowledge point update module reads the learner feedback and categorizes it into the matching practice question response status for the target knowledge point. If the learner feedback corresponds to a matching practice question list record, the target knowledge point update module writes that record along with the corresponding question identifier into the matching practice question response status. If the learner feedback corresponds to a resource sequence record, the target knowledge point update module writes that record along with the resource sequence record into the learner feedback. If the newly added cognitive interaction data lacks learner feedback, the target knowledge point update module retains the original matching practice question response status and writes the newly added cognitive interaction data lacking learner feedback into an exception record.

[0075] Furthermore, the target knowledge point update module synchronously maintains short-term, medium-term, and long-term learning gains. The short-term learning gains are generated based on the 10 most recent answer records for the target knowledge point. The medium-term learning gains are generated based on the 50 most recent answer records for the target knowledge point. The long-term learning gains are generated based on all historical answer records for the target knowledge point. After the new cognitive interaction data is written, the target knowledge point update module rereads the 10 most recent answer records, the 50 most recent answer records, and all historical answer records according to the interaction time, and writes the generated short-term, medium-term, and long-term learning gains into the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge point. If the number of answer records for the target knowledge point is less than 10, the target knowledge point update module writes the existing answer records into the short-term learning gains. If the number of answer records is less than 50, the target knowledge point update module writes the existing answer records into the medium-term learning gains. The long-term learning gains retain all historical answer records for the target knowledge point.

[0076] Understandably, after completing the above processing, the target knowledge point update module generates an updated multi-dimensional historical performance feature vector of the target knowledge point's knowledge mastery level and generates a target knowledge point update record. The target knowledge point update record includes the target knowledge point identifier, learner identifier, interaction time, previous answer results, error rate, number of repetitions, learning gains, learner feedback, the vector version before the update, and the vector version after the update. The updated multi-dimensional historical performance feature vector of the target knowledge point's knowledge mastery level is entered into S500, and in S500, combined with the target knowledge point and its association with other knowledge points, it serves as input for updating the multi-dimensional historical performance feature vector of the associated knowledge point's knowledge mastery level. The target knowledge point update record is entered into the data sample update record and is called up with each new piece of cognitive interaction data.

[0077] In one engineering embodiment, after a learner completes the "Function Graph Practice" questions on an online learning platform, the platform generates new cognitive interaction data. This new cognitive interaction data includes the teaching interaction content "Function Graph Practice," interaction time, previous answer results, error rate, number of repetitions, learning gains, and learner feedback. The target knowledge point update module reads the teaching interaction content "Function Graph Practice" from the knowledge point set and identifies the corresponding knowledge point "Function Graph" as the target knowledge point. Subsequently, the target knowledge point update module reads the multi-dimensional historical performance feature vector of the knowledge mastery level of "Function Graph," writes the new previous answer results into the current knowledge mastery level of this vector, writes the new error status into the error rate, writes the number of times the learner re-entered "Function Graph Practice" into the number of repetitions, writes the interaction time into the historical information forgetting status, and writes the learner feedback into the matching practice question response status. After writing is complete, the target knowledge point update module generates an updated vector version of "Function Graph" and submits this updated vector version to S500.

[0078] In summary, this step directly writes the newly added cognitive interaction data into the multi-dimensional historical performance feature vector of the target knowledge point's knowledge mastery level. This step updates the corresponding dimensions according to the teaching interaction content, past answer results, error rate, number of repetitions, interaction time, learning benefits, and learner feedback. This step outputs the updated multi-dimensional historical performance feature vector of the target knowledge point's knowledge mastery level, which is then used by the S500 to update related knowledge points based on their association relationships.

[0079] S500. Based on the target knowledge points, the relationships between knowledge points, and the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge points, update the multi-dimensional historical performance feature vector of the knowledge mastery level of the related knowledge points, and generate an updated feature vector set.

[0080] In S500, the related knowledge point update module receives the target knowledge point, target knowledge point update record, and multi-dimensional historical performance feature vector of the target knowledge point's knowledge mastery level generated in S400, and calls the knowledge point association relationship generated in S200 and the multi-dimensional historical performance feature vector of each knowledge point's knowledge mastery level generated in S300. The related knowledge point is a knowledge point that has a predecessor / follower relationship, a preorder node, a necessary node, a parallel node, a coverage relationship, or a OR relationship with the target knowledge point. The multi-dimensional historical performance feature vector of the related knowledge point's knowledge mastery level is a vector record formed in S300 for the related knowledge point, including short-term learning gain, medium-term learning gain, long-term learning gain, error rate, number of repetitions, learner feedback, current knowledge mastery level, historical information transmission status, historical information forgetting status, influence status of predecessor knowledge points, influence status of successor knowledge points, and response status of matching practice questions. The related knowledge point update module consists of a relationship reading unit, a related knowledge point filtering unit, a vector reading unit, a relationship writing unit, and a feature vector set generation unit.

[0081] Specifically, the relationship reading unit reads the knowledge point identifier of the target knowledge point and searches for relationship records in the knowledge point association relationships where the starting or ending knowledge point is the same as the target knowledge point. The relationship record includes the relationship type, starting knowledge point, ending knowledge point, first-level branch, second-level branch, and corresponding teaching interaction content. The relationship type can be a pre-or post-relationship, a preorder node, a necessary node, a parallel node, a covering relationship, or a OR relationship. The associated knowledge point filtering unit extracts associated knowledge points based on the relationship records and merges duplicate associated knowledge points into a single update record. If an associated knowledge point in a relationship record is not in the knowledge point set, the associated knowledge point filtering unit writes the relationship record into the unmapped record. If the relationship type of the relationship record is missing, the associated knowledge point filtering unit reads the data sample update record from S200 and writes the relationship record into the exception record.

[0082] Further, the vector reading unit reads the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge point, and extracts the error rate, repetition count, learning gain, learner feedback, current knowledge mastery level, and historical information transmission status. The error rate and repetition count serve as inputs for updating the preorder node. The current knowledge mastery level serves as input for updating the subsequent knowledge point. The learning gain and learner feedback serve as inputs for updating parallel nodes, covering relationships, and / or relationships. The historical information transmission status serves as input for updating necessary nodes. The vector reading unit synchronously reads the original multi-dimensional historical performance feature vector of the knowledge mastery level of the associated knowledge point, and records the vector version before the update, the associated knowledge point identifier, the relationship type, and the target knowledge point identifier.

[0083] Furthermore, when the associated knowledge point is a preorder node of the target knowledge point, the relationship writing unit updates the influence status of the preceding knowledge point of the associated knowledge point based on the error rate and repetition count of the target knowledge point. Specifically, the relationship writing unit reads the error rate change record and repetition count change record of the target knowledge point and writes these records into the influence status of the preceding knowledge point of the associated knowledge point. If the target knowledge point has consecutive error state records, the relationship writing unit writes the interaction time corresponding to the consecutive error state records into the historical information transmission status of the associated knowledge point. If the target knowledge point does not have any error state records, the relationship writing unit retains the original influence status of the preceding knowledge point of the associated knowledge point and records the result of this relationship reading.

[0084] Furthermore, when the related knowledge point is a subsequent knowledge point of the target knowledge point, the relationship writing unit updates the subsequent knowledge point influence status of the related knowledge point according to the current knowledge mastery level of the target knowledge point. Specifically, the relationship writing unit reads the current knowledge mastery level, short-term learning benefits, medium-term learning benefits, and long-term learning benefits of the target knowledge point, and writes this data into the subsequent knowledge point influence status of the related knowledge point. If the related knowledge point has multiple preceding knowledge points, the relationship writing unit reads the current knowledge mastery level corresponding to each preceding knowledge point according to the interaction time, and appends the record corresponding to the target knowledge point to the subsequent knowledge point influence status of the related knowledge point. If the target knowledge point and the related knowledge point belong to different first-level branches, the relationship writing unit reads the corresponding second-level branches and teaching interaction content, and records the branch source in the subsequent knowledge point influence status.

[0085] Furthermore, when a related knowledge point and a target knowledge point have parallel nodes, coverage relationships, or OR relationships, the relationship writing unit updates the historical information transmission status of the related knowledge point based on learning gains and learner feedback. Specifically, the relationship writing unit reads the learning gains, learner feedback, matching exercise response status, and resource sequence records of the target knowledge point, and writes the learning gains and learner feedback into the historical information transmission status of the related knowledge point. Parallel nodes correspond to knowledge points under the same secondary branch, coverage relationships correspond to knowledge points covering the same teaching interaction content, and OR relationships correspond to knowledge points with replaceable learning content. If learner feedback is missing, the relationship writing unit retains the original historical information transmission status of the related knowledge point and writes the missing feedback record into the exception record. If the same related knowledge point has both parallel nodes and coverage relationships, the relationship writing unit writes the historical information transmission status according to the relationship number order in the knowledge point association relationship.

[0086] Furthermore, when the associated knowledge point is a necessary node, the relationship writing unit reads the historical information transmission status, historical information forgetting status, and current knowledge mastery level of the target knowledge point, and writes the reading results into the historical information transmission status of the associated knowledge point. The necessary node retains its original position in the knowledge point association relationship. If the necessary node is also a preorder node, the relationship writing unit first updates the influence status of the preceding knowledge point, and then updates the historical information transmission status. If the necessary node is also a subsequent knowledge point, the relationship writing unit first updates the influence status of the subsequent knowledge point, and then updates the historical information transmission status. All of the above update processes generate relationship update records. The relationship update record includes the target knowledge point identifier, the associated knowledge point identifier, the relationship type, the update field, the vector version before the update, the vector version after the update, and the interaction time.

[0087] Understandably, the feature vector set generation unit merges the updated multi-dimensional historical performance feature vectors of the knowledge mastery level of the associated knowledge points, the multi-dimensional historical performance feature vectors of the knowledge mastery level of the target knowledge points generated in S400, and the multi-dimensional historical performance feature vectors of the knowledge mastery level of the knowledge points that have not been updated, to generate an updated feature vector set. In the updated feature vector set, each knowledge point retains its knowledge point identifier, its first-level branch, its second-level branch, its corresponding teaching interaction content, short-term learning benefits, medium-term learning benefits, long-term learning benefits, error rate, number of repetitions, learner feedback, current knowledge mastery level, historical information transmission status, historical information forgetting status, influence status of preceding knowledge points, influence status of subsequent knowledge points, and response status of matching practice questions. The updated feature vector set is then fed into S600 as input for generating the knowledge mastery level for the future target time. The relationship update record is written into the data sample update record and maintains the same interaction time as the target knowledge point update record in S400.

[0088] In one engineering embodiment, S400 identifies the "function image" as the target knowledge point and generates an updated vector version of the "function image". The associated knowledge point update module reads the knowledge point association relationship in S200, obtaining that "function concept" is the preorder node of "function image", "function application" is the postorder knowledge point of "function image", and "function table" is the parallel node of "function image". The relationship writing unit reads the error rate and repetition count of "function image" and updates the influence status of the preorder knowledge point of "function concept". The relationship writing unit reads the current knowledge mastery level of "function image" and updates the influence status of the postorder knowledge point of "function application". The relationship writing unit reads the learning benefits and learner feedback of "function image" and updates the historical information transmission status of "function table". After processing, the feature vector set generation unit writes the updated vector versions of "function image", "function concept", "function application" and "function table" into the updated feature vector set and submits it to S600.

[0089] In summary, this step updates the target knowledge points and writes them into the multi-dimensional historical performance feature vector of the knowledge mastery level of the related knowledge points according to their relationships. This step writes the preorder nodes, subsequent knowledge points, parallel nodes, coverage relationships, OR relationships, and necessary nodes into their respective state fields. This step outputs the updated feature vector set, which is then used by the S600 to generate the knowledge mastery level for the future target time.

[0090] S600. Based on the updated feature vector set and learning behavior records, generate the knowledge mastery level for the future target time.

[0091] In S600, the future target time generation module receives the updated feature vector set generated by S500 and calls the learning behavior record generated by S100. The updated feature vector set includes multi-dimensional historical performance feature vectors of knowledge mastery level for each knowledge point. These multi-dimensional historical performance feature vectors include short-term learning gains, medium-term learning gains, long-term learning gains, error rate, number of repetitions, learner feedback, current knowledge mastery level, historical information transmission status, historical information forgetting status, influence status of prior knowledge points, influence status of subsequent knowledge points, and response status of matching practice questions. The learning behavior record includes past answer results, teaching interaction content, interaction time, learning gains, learner feedback, matching practice question list record, and resource sequence record. The future target time represents the time point at which the knowledge mastery level needs to be read for the learning objective, learning path, or resource sequence. The future target time generation module consists of a vector reading unit, a time interval acquisition unit, a historical information processing unit, and a knowledge mastery level generation unit.

[0092] Specifically, the vector reading unit reads the multi-dimensional historical performance feature vectors of knowledge mastery level one by one from the updated feature vector set according to the knowledge point identifier, and reads the short-term learning gains, medium-term learning gains, long-term learning gains, interaction time, historical information transmission status, and historical information forgetting status. The short-term learning gains correspond to the most recent 10 answer records. The medium-term learning gains correspond to the middle 50 answer records. The long-term learning gains correspond to all historical answer records. The interaction time is the submission time of each answer result, teaching interaction content, matching exercise list record, and resource sequence record in the learning behavior record. The historical information transmission status comes from the update process of related knowledge points in S500. The historical information forgetting status comes from the update process of target knowledge points in S400. After the vector reading unit completes the reading, it writes the read fields into the future target time generated record.

[0093] Furthermore, the time interval acquisition unit reads the interaction time from the learning behavior record and reads the future target time corresponding to the learning objective. The time interval acquisition unit compares the future target time with the most recent interaction time of each knowledge point to obtain the time interval corresponding to each knowledge point. If a knowledge point has multiple interaction times, the time interval acquisition unit reads the last interaction time in the time sequence. If a knowledge point has no interaction time, the time interval acquisition unit reads the most recent interaction time of the teaching interaction content under the secondary branch to which the knowledge point belongs and writes this record to the exception record. If the future target time is earlier than the most recent interaction time, the time interval acquisition unit uses the most recent interaction time as the current reading time and writes the knowledge point to the verification record. The time interval generated by the time interval acquisition unit is written into the future target time generation record of the corresponding knowledge point.

[0094] Furthermore, the historical information processing unit corrects the current knowledge mastery level of each knowledge point based on time intervals, historical information forgetting status, and historical information transmission status. Specifically, the historical information processing unit reads the current knowledge mastery level and the short-term, medium-term, and long-term learning gains for the same knowledge point. The short-term learning gains represent changes in learning gains corresponding to recent answer records. The medium-term learning gains represent changes in learning gains corresponding to a stage of answer records. The long-term learning gains represent changes in learning gains corresponding to all historical answer records. The historical information processing unit writes the time interval into the historical information forgetting status and corrects the current knowledge mastery level based on this status. The historical information processing unit writes the influence status of preceding knowledge points, the influence status of subsequent knowledge points, and the historical information transmission status into the status record of the current knowledge point, and retains both the current knowledge mastery level before and after correction in the future target time generation record for the same knowledge point.

[0095] Furthermore, the knowledge mastery level generation unit generates the knowledge mastery level for the future target time based on the corrected current knowledge mastery level, short-term learning gains, medium-term learning gains, long-term learning gains, historical information transmission status, and historical information forgetting status. The knowledge mastery level for the future target time corresponds one-to-one with the knowledge point identifier and records the corresponding first-level branch, second-level branch, corresponding teaching interaction content, and future target time. If the short-term, medium-term, and long-term learning gains of a certain knowledge point are missing, the knowledge mastery level generation unit reads the previous answer results and learner feedback from the learning behavior record to supplement them, and writes the supplementary source into the future target time generation record. If the historical information transmission status of a certain knowledge point comes from multiple related knowledge points, the knowledge mastery level generation unit reads the corresponding status according to the interaction time order of the relationship update record generated by S500, and writes the reading result into the future target time generation record of that knowledge point.

[0096] Understandably, after processing each knowledge point, the future target time generation module generates the knowledge mastery level for the future target time. This future target time knowledge mastery level includes the knowledge point identifier, current knowledge mastery level, future target time, time interval, short-term learning gains, medium-term learning gains, long-term learning gains, historical information transmission status, historical information forgetting status, and learner feedback. The future target time generation record includes the feature vector version, learning behavior record source, future target time, time interval, and abnormal records. The future target time knowledge mastery level is input to S700, which generates a priority review order based on the future target time knowledge mastery level and the knowledge point correlation. The future target time generation record is written to the data sample update record and continues to be read with the next updated feature vector set.

[0097] In one engineering embodiment, after the learner forms an updated set of feature vectors under the knowledge point "function image", the future target time generation module reads the short-term learning gains, medium-term learning gains, long-term learning gains, interaction time, historical information transmission status, and historical information forgetting status of the "function image". If the future target time corresponding to the learning target is the date of the next test, the time interval acquisition unit reads the most recent interaction time of the "function image" and generates the corresponding time interval. The historical information processing unit corrects the current knowledge mastery level of the "function image" according to the time interval and the historical information forgetting status, and reads the influence status of the preceding knowledge points of the "function concept" transmitted to the "function image". The knowledge mastery level generation unit generates the knowledge mastery level of the "function image" at the future target time according to the corrected current knowledge mastery level, short-term learning gains, medium-term learning gains, long-term learning gains, and historical information transmission status, and submits the result to S700.

[0098] In summary, this step generates the knowledge mastery level for each knowledge point at the future target time based on the updated feature vector set and learning behavior records. This step incorporates time intervals, historical information transmission status, and historical information forgetting status into the same generation process, ensuring a correspondence between the current knowledge mastery level and long-term time-series records. The knowledge mastery level at the future target time output by this step is entered into S700 as input for priority review order, learning path, and resource sequence.

[0099] S700: Generate a priority review order based on the level of knowledge mastery and the relationship between knowledge points at the future target time, and generate a learning path and resource sequence based on the priority review order.

[0100] In S700, the priority review order generation module receives the knowledge mastery level at the future target time generated by S600, and calls the knowledge point associations generated by S200, the multi-dimensional historical performance feature vector of the knowledge mastery level generated by S300, and the learning behavior records generated by S100. The knowledge mastery level at the future target time includes knowledge point identifiers, current knowledge mastery level, future target time, time interval, short-term learning gains, medium-term learning gains, long-term learning gains, historical information transmission status, historical information forgetting status, and learner feedback. The knowledge point associations include pre- and post-relationships, prior nodes, necessary nodes, parallel nodes, coverage relationships, and / or relationships. The priority review order represents the review arrangement result of each knowledge point before the future target time. The learning path represents a sequence of knowledge points arranged according to the knowledge point associations. The resource sequence represents the teaching interaction content, matching exercise list, and learning resources corresponding to each knowledge point in the learning path.

[0101] Specifically, the priority review order generation module reads the current knowledge mastery level, the knowledge mastery level at the future target time, and the learning objective for each knowledge point, and generates knowledge point ranking data. The learning objective comes from the course identifier, target knowledge point, and resource sequence record in the learning behavior record. The knowledge point ranking data includes the knowledge point identifier, the knowledge mastery level at the future target time, the current knowledge mastery level, short-term learning benefits, medium-term learning benefits, long-term learning benefits, learner feedback, and interaction time. The priority review order generation module first maps the knowledge mastery level at the future target time to the current knowledge mastery level, then reads the target knowledge point corresponding to the learning objective, and writes the knowledge points related to the learning objective into the candidate knowledge point record. If a knowledge point lacks a knowledge mastery level at the future target time, the priority review order generation module reads the future target time generation record in S600. If no corresponding record is found, the knowledge point is written into the verification record.

[0102] Furthermore, the priority review order generation module modifies the knowledge point sorting data based on the knowledge point relationships. When a current knowledge point has a pre-order node, the priority review order generation module reads the knowledge mastery level of the pre-order node at the future target time and arranges the pre-order node before the current knowledge point. When a current knowledge point has a necessary node, the priority review order generation module retains the necessary node in the priority review order and records its relationship type with the current knowledge point. When a current knowledge point has a parallel node, the priority review order generation module reads the learner feedback, short-term learning benefits, and medium-term learning benefits of the parallel node and writes the parallel node and the current knowledge point into the same review stage. When a current knowledge point has a coverage relationship, the priority review order generation module reads the teaching interaction content corresponding to the coverage relationship and records the covered knowledge point range in the knowledge point sorting data. When a current knowledge point has an OR relationship, the priority review order generation module reads the learner feedback and matching exercise list records of each knowledge point in the OR relationship and writes the knowledge points with complete feedback records into the priority review order.

[0103] Furthermore, the learning path generation module reads the priority review order and the relationship between knowledge points, and generates a learning path. Starting with the first knowledge point in the priority review order, the module writes the learning path in the order of pre-order nodes, necessary nodes, current knowledge point, and subsequent knowledge points. If multiple pre-order nodes exist for a knowledge point in the priority review order, the module reads the most recent learning record of each pre-order node in the learning behavior record according to the interaction time, and places the pre-order node with the earlier interaction time and still within the learning target range at the beginning. If parallel nodes exist for a knowledge point in the priority review order, the module writes the parallel nodes into the same learning stage and records the corresponding teaching interaction content. If a coverage relationship exists for a knowledge point in the priority review order, the module reads the teaching interaction content with the more complete coverage in the coverage relationship and writes the covered knowledge points into the same learning path node. If an OR relationship exists for a knowledge point in the priority review order, the module selects the corresponding teaching interaction content based on learner feedback and the matching exercise list record, and writes the selected record into the path generation record.

[0104] Further, the resource sequence generation module generates a resource sequence based on the learning path. This module reads the knowledge point identifiers, their primary branches, their secondary branches, corresponding teaching interaction content, and matching exercise list records from the learning path, and generates a correspondence between knowledge points and learning resources in the resource sequence records. The learning resources include teaching interaction content, the matching exercise list, course resources, and existing records in the resource sequence. For preorder nodes in the learning path, the resource sequence generation module prioritizes reading their corresponding teaching interaction content and matching exercise list. For necessary nodes in the learning path, the resource sequence generation module retains their corresponding learning resources. For parallel nodes, the resource sequence generation module selects teaching interaction content from the same learning stage based on learner feedback. For coverage relationships, the resource sequence generation module reads the teaching interaction content corresponding to the coverage area. For OR relationships, the resource sequence generation module reads the complete teaching interaction content recorded in the feedback record. After generating the resource sequence, the resource sequence generation module records the resource sequence identifier, knowledge point identifier, learning path position, teaching interaction content, matching exercise list, and learning resources.

[0105] Understandably, after the resource sequence enters the online learning platform, when learners complete the teaching interaction content, match the exercise question list, or access learning resources, the online learning platform generates new learner feedback. This learner feedback includes completion status, incomplete status, correction status, relearning status, and feedback text. The resource sequence generation module writes the learner feedback corresponding to the resource sequence into new cognitive interaction data and returns the new cognitive interaction data to S400. This new cognitive interaction data includes the teaching interaction content, previous answer results, interaction time, error rate, number of repetitions, learning gains, and learner feedback. After the new cognitive interaction data enters S400, the target knowledge point is identified from the knowledge point set, and the multi-dimensional historical performance feature vector of the target knowledge point's knowledge mastery level is updated.

[0106] In one engineering embodiment, S600 generates the learner's knowledge mastery level of "function concept," "function graph," "function application," and "function table" at a future target time. The priority review order generation module reads the target knowledge point "function application" from the learning objectives and reads the knowledge point association relationship, obtaining that "function concept" is the pre-order node of "function graph," "function graph" is the prerequisite knowledge point of "function application," and "function table" is the parallel node of "function graph." The priority review order generation module generates the priority review order of "function concept," "function graph," "function table," and "function application" based on the knowledge mastery level and knowledge point association relationship at the future target time. The learning path generation module places "function concept" before "function graph," writes "function table" into the same learning stage as "function graph," and places "function application" in a later position. The resource sequence generation module reads the teaching interaction content and matching exercise list corresponding to each knowledge point, generating a resource sequence containing "function concept practice," "function graph practice," "function table practice," and "function application practice." After the learner completes the resource sequence, the online learning platform writes the completion status and learner feedback into new cognitive interaction data and returns S400.

[0107] In summary, this step transforms the level of knowledge mastery and the relationships between knowledge points at the future target time into a priority review order. Based on this priority review order, a learning path is generated, incorporating preorder nodes, necessary nodes, parallel nodes, coverage relationships, and / or relationships into the path generation process. Next, a resource sequence is generated based on the learning path, and the learner feedback corresponding to this resource sequence is returned to S400, forming the input for the next update of the target knowledge points.

[0108] Example 2: Figure 2 A structural block diagram of a personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery according to an embodiment of the present invention is shown. Figure 2 As shown, the structure may include:

[0109] The data preprocessing module 01 receives learner-related learning log files, cognitive interaction data, and preset knowledge graphs. It generates learning behavior records and knowledge graph data through data preprocessing and sends these data to the knowledge point network structure construction module. Specifically, the data preprocessing module includes a log receiving interface, an interaction data receiving interface, a knowledge graph receiving interface, and a preprocessing record area. The log receiving interface receives learner-related learning log files, the interaction data receiving interface receives cognitive interaction data, and the knowledge graph receiving interface receives preset knowledge graphs. The data preprocessing module reads past answer results, teaching interaction content, interaction time, error rate, repetition count, learning gains, learner feedback, question semantic features, student basic attribute features, matching practice question list records, and resource sequence records. It then performs field completion, duplicate record merging, abnormal record marking, and chronological ordering according to learner identifier, course identifier, question identifier, and teaching interaction content identifier. The data preprocessing module reads first-level branches, second-level branches, knowledge points, teaching interaction content, pre- and post-relationships, prior nodes, necessary nodes, parallel nodes, coverage relationships, and / or relationships from the preset knowledge graph, and records node numbers, relationship numbers, and version numbers. Records with missing fields are written as exception records, and teaching interaction content that cannot be mapped is written as unmapped records. The data preprocessing module records the completed answer results, teaching interaction content, interaction time, error rate, number of repetitions, learning benefits, and learner feedback as learning behavior records, and records the completed first-level branches, second-level branches, knowledge points, and knowledge point relationships as knowledge graph data, and provides the learning behavior records and knowledge graph data to the knowledge point network structure construction module.

[0110] The knowledge point network structure construction module 02 is used to receive learning behavior records and knowledge graph data, construct a knowledge point network structure, generate a set of knowledge points and their relationships, and send the set of knowledge points and their relationships to the multi-dimensional historical performance feature vector generation module. Specifically, the knowledge point network structure construction module is equipped with a branch attribution processing unit, a knowledge point merging unit, and a relationship writing unit. The branch attribution processing unit receives learning behavior records and knowledge graph data from the data preprocessing module, reads first-level branches, second-level branches, knowledge points, and teaching interaction content, and assigns each knowledge point to its corresponding second-level branch, and each second-level branch to its corresponding first-level branch. The knowledge point merging unit reads the semantic features of the questions, teaching interaction content, and knowledge point identifiers, merges repeated knowledge points into the same knowledge point record, and retains the corresponding teaching interaction content. The relationship writing unit reads the pre- and post-relationships, prior nodes, necessary nodes, parallel nodes, coverage relationships, and / or relationships, and records the starting knowledge point, ending knowledge point, the first-level branch to which it belongs, the second-level branch to which it belongs, and the corresponding teaching interaction content for each relationship. Learning behavior records that cannot be categorized into the knowledge point set are written to the pending mapping record. Records with missing relationship types are written to the exception record. The knowledge point set generated by the knowledge point network structure construction module includes the knowledge point identifier, its first-level branch, its second-level branch, and the corresponding teaching interaction content. The generated knowledge point relationships include pre- and post-relationships, prior nodes, necessary nodes, parallel nodes, coverage relationships, and OR relationships. The knowledge point set and knowledge point relationships are provided to the multi-dimensional historical performance feature vector generation module.

[0111] The multi-dimensional historical performance feature vector generation module 03 receives a set of knowledge points, knowledge point relationships, and learning behavior records, generates a multi-dimensional historical performance feature vector of knowledge mastery level for each knowledge point, and sends the multi-dimensional historical performance feature vector of knowledge mastery level for each knowledge point to the target knowledge point update module. Specifically, the multi-dimensional historical performance feature vector generation module is equipped with a knowledge point record reading unit, a time window division unit, a learning benefit generation unit, a relationship status writing unit, and a vector recording unit. The knowledge point record reading unit receives a set of knowledge points and knowledge point relationships from the knowledge point network structure construction module, and calls the learning behavior records from the data preprocessing module. The time window division unit arranges the learning behavior records under the same knowledge point into long-term time-series records according to the interaction time, and divides them into short-term time windows, medium-term time windows, and long-term time windows. The short-term time window corresponds to the most recent 10 answer records, the medium-term time window corresponds to the middle 50 answer records, and the long-term time window corresponds to all historical answer records. The learning benefit generation unit generates short-term, medium-term, and long-term learning benefits, and concatenates these three types of learning benefits after normalization into a cross-scale historical performance feature vector. The association state writing unit reads the knowledge point association relationships and writes the historical information transmission state, historical information forgetting state, the influence state of preceding knowledge points, and the influence state of subsequent knowledge points. The vector recording unit writes the short-term learning benefits, medium-term learning benefits, long-term learning benefits, error rate, number of repetitions, learner feedback, current knowledge mastery level, historical information transmission state, historical information forgetting state, the influence state of preceding knowledge points, the influence state of subsequent knowledge points, and the response state of matching practice questions into the multi-dimensional historical performance feature vector of knowledge mastery level for each knowledge point, and provides the multi-dimensional historical performance feature vector of knowledge mastery level for each knowledge point to the target knowledge point update module.

[0112] The target knowledge point update module 04 is used to receive newly added cognitive interaction data, identify target knowledge points from the knowledge point set, update the multi-dimensional historical performance feature vector of the target knowledge point's knowledge mastery level, and send the target knowledge point and the multi-dimensional historical performance feature vector of the target knowledge point's knowledge mastery level to the associated knowledge point update module. Specifically, the target knowledge point update module is equipped with a new interaction receiving unit, a target knowledge point identification unit, and a target vector update unit. The new interaction receiving unit receives newly added cognitive interaction data after the learner completes answering questions, learning interactive teaching content, matching practice question lists, learning resource sequences, correcting wrong questions, or receiving feedback. The newly added cognitive interaction data includes interactive teaching content, previous answer results, interaction time, error rate, number of repetitions, learning benefits, and learner feedback. The target knowledge point identification unit reads the knowledge point identifier and corresponding interactive teaching content from the knowledge point set, matches the interactive teaching content in the newly added cognitive interaction data with the corresponding interactive teaching content in the knowledge point set, and records the matched knowledge point as the target knowledge point. New cognitive interaction data that fails to match is written to the pending mapping record. The target vector update unit reads the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge point, updates the current knowledge mastery level based on previous answer results, error rate, and repetition count, updates the historical information forgetting status based on interaction time, updates the response status of matching practice questions based on learner feedback, and simultaneously updates short-term, medium-term, and long-term learning gains. The target knowledge point update module generates target knowledge point update records and provides the target knowledge point and the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge point to the associated knowledge point update module.

[0113] The related knowledge point update module 05 is used to receive target knowledge points, knowledge point relationships, and multi-dimensional historical performance feature vectors of the target knowledge point's knowledge mastery level, update the multi-dimensional historical performance feature vectors of the related knowledge points' knowledge mastery level, generate an updated feature vector set, and send the updated feature vector set to the future target time generation module. Specifically, the related knowledge point update module is equipped with a relationship reading unit, a related knowledge point filtering unit, a vector reading unit, a relationship writing unit, and a feature vector set generation unit. The relationship reading unit receives the target knowledge point and the multi-dimensional historical performance feature vectors of the target knowledge point's knowledge mastery level from the target knowledge point update module, and calls the knowledge point relationships from the knowledge point network structure construction module. The related knowledge point filtering unit reads knowledge points from the knowledge point relationships that have a predecessor / follower relationship, a preorder node, a necessary node, a parallel node, a coverage relationship, or a OR relationship with the target knowledge point, and records them as related knowledge points. The vector reading unit reads the error rate, repetition count, learning benefit, learner feedback, current knowledge mastery level, and historical information transmission status of the target knowledge point, and reads the original multi-dimensional historical performance feature vector of the knowledge mastery level of the related knowledge point. The relationship writing unit updates the influence status of the preceding knowledge point of the related knowledge point based on the error rate and repetition count of the target knowledge point when the related knowledge point is a pre-order node of the target knowledge point; updates the influence status of the following knowledge point of the related knowledge point based on the current knowledge mastery level of the target knowledge point when the related knowledge point is a subsequent knowledge point of the target knowledge point; and updates the historical information transmission status of the related knowledge point based on the learning benefit and learner feedback when the related knowledge point and the target knowledge point have parallel nodes, a covering relationship, or a OR relationship. The feature vector set generation unit merges the multi-dimensional historical performance feature vectors of the knowledge mastery level of the target knowledge point, related knowledge points, and knowledge points that have not been updated, generates an updated feature vector set, and provides the updated feature vector set to the future target time generation module.

[0114] The future target time generation module 06 receives the updated feature vector set and learning behavior records, generates the knowledge mastery level for the future target time, and sends the knowledge mastery level for the future target time to the priority review order generation module. Specifically, the future target time generation module is equipped with a vector reading unit, a time interval acquisition unit, a historical information processing unit, and a knowledge mastery level generation unit. The vector reading unit receives the updated feature vector set from the associated knowledge point update module and calls the learning behavior records from the data preprocessing module. The vector reading unit reads the short-term learning benefit, medium-term learning benefit, long-term learning benefit, interaction time, historical information transmission status, and historical information forgetting status of each knowledge point one by one. The time interval acquisition unit reads the future target time corresponding to the learning target and compares the future target time with the most recent interaction time of each knowledge point to obtain the time interval. Knowledge points without interaction time are written into the exception record. The historical information processing unit corrects the current knowledge mastery level according to the time interval and historical information forgetting status, and writes the historical information transmission status into the generation record of the same knowledge point. The knowledge mastery level generation unit generates the knowledge mastery level for the future target time based on the corrected current knowledge mastery level, short-term learning benefits, medium-term learning benefits, long-term learning benefits, historical information transmission status, and historical information forgetting status, and provides the knowledge mastery level for the future target time to the priority review order generation module.

[0115] The priority review order generation module 07 receives the knowledge mastery level and knowledge point relationships at a future target time, generates a priority review order, and sends the priority review order to the learning path and resource sequence generation module. Specifically, the priority review order generation module sets up a knowledge point sorting unit and a sorting correction unit. The knowledge point sorting unit receives the knowledge mastery level at the future target time from the future target time generation module and calls the knowledge point relationships from the knowledge point network structure construction module. The knowledge point sorting unit reads the current knowledge mastery level, the knowledge mastery level at the future target time, and the learning objective to generate knowledge point sorting data. The sorting correction unit reads the preorder nodes, necessary nodes, parallel nodes, coverage relationships, and / or relationships in the knowledge point relationships and corrects the knowledge point sorting data. When a current knowledge point has a preorder node, the sorting correction unit arranges the preorder node before the current knowledge point. When a current knowledge point has a necessary node, the sorting correction unit retains the necessary node in the priority review order. When parallel nodes, coverage relationships, or other relationships exist for the current knowledge point, the sorting correction unit reads the learner feedback and matching exercise list records, and writes them into the same review stage or the corresponding replacement record. The priority review order generation module generates a priority review order and provides the priority review order to the learning path and resource sequence generation module.

[0116] The learning path and resource sequence generation module 08 receives the priority review order, generates a learning path and resource sequence, and sends the learner feedback corresponding to the resource sequence as new cognitive interaction data to the target knowledge point update module. Specifically, the learning path and resource sequence generation module includes a learning path generation unit, a resource sequence generation unit, and a feedback write-back unit. The learning path generation unit receives the priority review order from the priority review order generation module and invokes the knowledge point association relationships. The learning path generation unit writes the learning path in the order of preorder nodes, necessary nodes, current knowledge points, and subsequent knowledge points. When there are parallel nodes among the knowledge points in the priority review order, the learning path generation unit writes the parallel nodes into the same learning stage. When there are coverage relationships or OR relationships among the knowledge points in the priority review order, the learning path generation unit selects the corresponding teaching interaction content based on the learner feedback. The resource sequence generation unit reads the knowledge point identifiers, corresponding teaching interaction content, and matching exercise list records from the learning path and generates a resource sequence. The feedback write-back unit receives learner feedback generated after the learner completes the resource sequence, writes the learner feedback, teaching interaction content, previous answer results, interaction time, error rate, number of repetitions, and learning benefits into new cognitive interaction data, and sends the new cognitive interaction data to the target knowledge point update module.

Claims

1. A personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery, characterized in that, include: S100: Obtain learner-related learning log files, cognitive interaction data, and preset knowledge graphs; generate learning behavior records and knowledge graph data through data preprocessing. S200: Construct a knowledge point network structure based on the learning behavior records and knowledge graph data, and generate a knowledge point set and knowledge point association relationships; S300. Based on the knowledge point set, the knowledge point relationship, and learning behavior records, generate a multi-dimensional historical performance feature vector of the knowledge mastery level of each knowledge point. S400: Receive newly added cognitive interaction data, identify target knowledge points from the knowledge point set, and update the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge points. The newly added cognitive interaction data refers to new records generated by learners after completing tasks, learning interactive teaching content, matching practice question lists, learning resource sequences, correcting mistakes, or receiving learner feedback. These records include interactive teaching content, past task results, interaction time, error rate, number of repetitions, learning benefits, and learner feedback. S500. Based on the target knowledge points, the relationships between knowledge points, and the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge points, update the multi-dimensional historical performance feature vector of the knowledge mastery level of the related knowledge points, and generate an updated feature vector set. S600. Based on the updated feature vector set and learning behavior records, generate the knowledge mastery level for the future target time. S700: Generate a priority review order based on the level of knowledge mastery and the relationship between knowledge points at the future target time, and generate a learning path and resource sequence based on the priority review order.

2. The personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery as described in claim 1, characterized in that, The learning behavior records include the results of each quiz, teaching interaction content, interaction time, error rate, number of repetitions, learning benefits, learner feedback, semantic features of questions, basic attribute features of students, records of matching practice question lists, and records of resource sequences.

3. The personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery as described in claim 1, characterized in that, The knowledge point network structure consists of first-level branches, second-level branches, knowledge points, and interactive teaching content; the knowledge point relationships include pre-order and post-order relationships, prior nodes, necessary nodes, parallel nodes, coverage relationships, and OR relationships.

4. The personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery as described in claim 1, characterized in that, The multi-dimensional historical performance feature vector of knowledge mastery level includes short-term learning gains, medium-term learning gains, long-term learning gains, error rate, number of repetitions, learner feedback, current knowledge mastery level, historical information transmission status, historical information forgetting status, influence status of prior knowledge points, influence status of subsequent knowledge points, and response status of matching practice questions.

5. The personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery as described in claim 4, characterized in that, The short-term learning gain is generated based on the most recent 10 answer records, the medium-term learning gain is generated based on the middle 50 answer records, and the long-term learning gain is generated based on all historical answer records; the short-term learning gain, the medium-term learning gain, and the long-term learning gain are normalized and then concatenated to form a cross-scale historical performance feature vector.

6. The personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery as described in claim 1, characterized in that, In S400, updating the multi-dimensional historical performance feature vector of the knowledge mastery level of the target knowledge point includes: identifying the target knowledge point based on the teaching interaction content in the newly added cognitive interaction data; updating the current knowledge mastery level of the target knowledge point based on the answer results, error rate, and number of repetitions; updating the historical information forgetting status of the target knowledge point based on the interaction time; and updating the response status of the matching practice questions of the target knowledge point based on learner feedback.

7. The personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery as described in claim 1, characterized in that, In S500, updating the multi-dimensional historical performance feature vector of knowledge mastery level of related knowledge points includes: when the related knowledge point is a preorder node of the target knowledge point, updating the influence status of the preceding knowledge point of the related knowledge point according to the error rate and repetition frequency of the target knowledge point; when the related knowledge point is a subsequent knowledge point of the target knowledge point, updating the influence status of the subsequent knowledge point of the related knowledge point according to the current knowledge mastery level of the target knowledge point; when the related knowledge point and the target knowledge point have parallel nodes, coverage relationship, or OR relationship, updating the historical information transmission status of the related knowledge point according to learning benefits and learner feedback.

8. The personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery as described in claim 1, characterized in that, In S600, generating the knowledge mastery level for the future target time includes: reading the short-term learning gain, medium-term learning gain, long-term learning gain, interaction time, historical information transmission status, and historical information forgetting status from the updated feature vector set; obtaining the time interval corresponding to the future target time based on the interaction time; and correcting the current knowledge mastery level based on the time interval and the historical information forgetting status to generate the knowledge mastery level for the future target time.

9. The personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery as described in claim 1, characterized in that, In S700, generating a priority review order includes: generating knowledge point ranking data based on the current knowledge mastery level, the knowledge mastery level at the future target time, and the learning objectives; and modifying the knowledge point ranking data based on the prior nodes, necessary nodes, parallel nodes, coverage relationships, and / or relationships in the knowledge point association relationship to generate a priority review order.

10. The personalized learning recommendation method based on dynamic modeling of long-term knowledge mastery as described in claim 1, characterized in that, In S700, generating learning paths and resource sequences includes: when a knowledge point in the priority review order has a preorder node, placing the preorder node before the corresponding knowledge point; when a knowledge point in the priority review order has parallel nodes, coverage relationships, or OR relationships, selecting teaching interaction content based on learner feedback; matching practice question lists and learning resources for each knowledge point according to the priority review order, generating resource sequences, and returning learner feedback corresponding to the resource sequences as new cognitive interaction data to S400.

Images