Knowledge graph-based full-scene learning navigation system

Abstract

The application discloses a full-scene learning navigation system based on a knowledge graph and belongs to the field of educational data processing systems. The system comprises a time sequence feature extraction module, a mastery level diagnosis module, a urgency determination module, a branch generation module and a path merging module. The mastery probability is corrected by the time interval of the answer sequence and the correctness state, the mastery state of the hierarchical diagnosis concept cognition, principle understanding and application transfer level is diagnosed, the learning trigger gear is determined by combining the urgency calculation and the threshold comparison, the complete branch, the delayed branch or the inserted node and other learning forms are dynamically generated, and the personalized learning path is formed by merging the main path. The application realizes the real-time linkage of the cognitive state and the learning path, and improves the adaptability and intervention accuracy of the learning navigation.

Description

Technical Field

[0001] This invention relates to the field of educational data processing systems, and more particularly to a full-scenario learning navigation system based on knowledge graphs. Background Technology

[0002] The full-scenario learning navigation system covers the collection, governance, storage, and analysis of multi-source learning behavior data. Its core lies in leveraging distributed architecture, natural language processing, and graph neural networks to achieve standardized integration of multimodal learning resources and dynamic correlation analysis of cognitive paths, providing decision support for personalized learning recommendations and adaptive navigation.

[0003] Among them, the full-scenario learning navigation system based on knowledge graphs and cognitive tracking refers to a mechanism for joint analysis and dynamic path planning of multi-source heterogeneous data in an online-offline integrated learning environment. Targeting all learning scenarios, including pre-class preparation, classroom interaction, post-class exercises, and practical training, the system constructs a subject knowledge ontology using algorithms such as named entity recognition, extracts structured features from multimedia resources using convolutional neural networks, integrates multi-dimensional features such as learning behavior sequences, cognitive state evolution, and interactive feedback, and maps them to a knowledge graph. Through joint reasoning using graph neural networks and cognitive diagnostic models, it outputs personalized learning paths or real-time navigation suggestions.

[0004] Existing technologies rely on static knowledge graphs and pre-set learning paths, treating learning behavior and cognitive state as isolated dimensions and ignoring the real-time evolution of cognitive load and knowledge mastery during the learning process. Feature alignment lacks a cognitive feedback mechanism, making it difficult to capture the coupling relationship between the dynamic association of knowledge points and the learner's personalized cognitive evolution. Fixed navigation paths cannot adapt to instantaneous changes in learning states, resulting in lagging navigation decisions and making it difficult to provide real-time and accurate learning path optimization and intervention for learners with different cognitive levels. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a knowledge graph-based full-scene learning navigation system.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A knowledge graph-based full-scenario learning navigation system includes the following steps: Based on the answer sequence of the same knowledge point, the correctness status and time interval of adjacent answers are extracted. According to the comparison result of the time interval with the preset value and the matching result of the correctness status with the time sequence pattern, the time sequence correction mastery probability is obtained. The mastery rate of each cognitive level is determined based on the time-series corrected mastery probability, and compared with the corresponding benchmark value to generate the current mastery level identifier; By combining the time-series correction of mastery probability with the current mastery level identifier, and based on the comparison results of mastery probability and trigger reference value, as well as the correspondence between target level and current level, the urgency of remedial learning and the corresponding remedial learning trigger level are determined. Based on the supplementary learning trigger level, retrieve the corresponding supplementary learning resource identifier to form a continuous supplementary learning branch containing supplementary learning nodes or insertion nodes; Connect or embed supplementary learning nodes with or into the main path nodes to form a complete learning path queue.

[0007] Preferably, the method for obtaining the timing correction mastery probability is as follows: Obtain the timestamp matrix of the answer sequence. Each row records the question identifier, answer time, and time consumption value. Arrange the answer records under the same knowledge point in ascending order according to the answer time to form the answer sequence. Extract the correctness results of two consecutive answers, the first and second answers, and the time interval between the two answers; Compare the time interval with the short-term memory threshold; The matching factor is combined based on the two correctness results: the incremental factor is obtained after the correct result, the inhibition factor is obtained after the correct result or the correct result after the error, and the zero-value factor is obtained after the error. The current mastery probability is added to the increment factor or subtracted from the inhibition factor to obtain the initial correction value. Then, it is compared with the upper and lower limits of the mastery probability to obtain the boundary value, thus generating the time-series corrected mastery probability.

[0008] The preferred method currently known for generating hierarchical identifiers is: The cognitive levels are divided into the conceptual cognition level, the principle understanding level, and the application transfer level; Calculate the ratio of the number of correct test questions to the total number of test questions in each level to obtain the mastery rate of concept cognition, mastery rate of principle understanding, and mastery rate of application and transfer. Compare the mastery rate of each level with the preset corresponding level's reference value, and write the identifier of the level that meets the standard into the depth layer identifier field to form the current mastery level identifier.

[0009] The preferred method for determining the trigger level for supplementary learning is as follows: If the time-corrected mastery probability is less than the preset mastery probability trigger threshold, then the urgency calculation for supplementary learning will be triggered. Subtract the required mastery level from the highest level identifier in the current mastery level identifier to obtain the level difference, and then multiply it by the preset importance weight coefficient to obtain the urgency of remedial learning; The urgency of supplementary learning is compared with the high, medium, and low thresholds in sequence to obtain the high, medium, or low level as the supplementary learning trigger level.

[0010] The preferred method for generating the continuous form of supplementary learning branches is as follows: When the supplementary learning trigger level is a high level, retrieve the application migration level test question set identifier and the principle understanding level explanation content identifier, write them into the supplementary learning branch node set, and generate a complete branch; When the supplementary learning trigger level is the middle level, the principle understanding level explanation content identifier is retrieved and written into the supplementary learning branch node set, and the delay duration value is written into the starting node to generate a delay branch; When the supplementary learning trigger level is low, the concept cognition level review content identifier is retrieved and written into the main path insertion node set to generate the insertion node.

[0011] Preferably, path merging includes: The complete learning path queue is generated as follows: When the supplementary learning branch continues to be a complete branch, all node identifiers in the supplementary learning branch node set are merged with the main path node queue to generate a complete learning path queue. When the supplementary learning branch continues to be a delayed branch, the supplementary learning branch node queue of the delayed branch will be merged with the main path node queue after the delay duration value ends, to generate a complete learning path queue. When the supplementary learning branch continues to be an insertion node, the retrospective node identifier of the insertion node is written to the corresponding position in the main path node queue according to the insertion position offset value, and a complete learning path queue is generated.

[0012] Preferably, the complete branch is formed by sequentially connecting the test node corresponding to the application migration level test question set identifier and the explanation node corresponding to the principle understanding level explanation content identifier; the delayed branch is formed by the explanation node corresponding to the principle understanding level explanation content identifier and the starting waiting node associated with the delay duration value; the inserted node is formed by the review node corresponding to the concept cognition level review content identifier and is embedded in the specified position of the main path.

[0013] Preferably, when merging complete branches, all node identifiers in the supplementary learning branch node set are appended to the end of the main path node queue in the order of node topology; when merging delayed branches, the delayed branch node queue is inserted after the current execution position of the main path node queue after the delay duration value ends; when merging inserted nodes, the review node identifier is written into the corresponding index position in the main path node queue according to the insertion position offset value.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: 1. This invention constructs a timestamp matrix from the answer sequence and arranges it in ascending order according to the same knowledge point. It extracts the correctness state combinations and time intervals of adjacent answers, compares the time intervals with short-term memory thresholds, and matches the correctness state combinations with a temporal pattern mapping table. Incremental or inhibitory factors are used to correct the current mastery probability, achieving dynamic adjustment of knowledge mastery over time and eliminating the neglect of memory decay and consolidation patterns by static knowledge representation. Based on the time-corrected mastery probability, the mastery rate of different cognitive levels, such as concept cognition, principle understanding, and application transfer, is calculated and compared with the level attainment threshold to generate a deep hierarchical identification field. This refines cognitive diagnosis from a single correctness rate to multi-level mastery states, clearly identifying the learner's cognitive stage.

[0015] 2. By comparing the mastery probability with the mastery probability trigger threshold through time-series correction, and combining the difference between the required mastery level value and the current mastery level identifier, the urgency of remedial learning is calculated. Based on the comparison results of the urgency with high, medium, and low thresholds, the remedial learning trigger level is determined, realizing the quantitative classification of the urgency of learning needs, so that the timing and degree of intervention are accurately matched with actual needs. According to the remedial learning trigger level, the corresponding remedial learning resource identifier is retrieved. At the high level, the application transfer level test question set and the principle understanding level explanation content are retrieved to construct a complete branch. At the medium level, the principle understanding level explanation content is retrieved and a delay duration value is written to construct a delayed branch. At the low level, the concept cognition level review content is retrieved to construct an insertion node, so that the form of remedial learning content and urgency are dynamically adapted, avoiding resource waste and cognitive overload.

[0016] 3. Merge or insert the set of supplementary learning branch nodes or the set of inserted nodes with the main path node queue. Control the timing of merging in the delayed branch and locate the inserted node based on the offset value to realize dynamic reconstruction of the learning path. Ensure that the cognitive state and path planning are linked in real time, so that the learning navigation can respond to changes in actual situations such as cognitive load and time pressure. Attached Figure Description

[0017] Figure 1 This is a system flowchart of the present invention; Detailed Implementation

[0018] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0019] like Figure 1 As shown, the knowledge graph-based full-scenario learning navigation system includes the following modules: Temporal feature extraction module: Based on the answer sequence of the same knowledge point, extract the correctness status and time interval of adjacent answers, and obtain the temporal correction mastery probability through comparison analysis of time intervals and matching analysis of correctness status and temporal pattern; The process of obtaining the timing-corrected mastery probability includes: Retrieves all answer records for a specified learner and a specified knowledge point from the learning behavior database. Each record includes: question identifier, answer time, time taken, and correctness result (correct = 1, incorrect = 0). All answer records for the same knowledge point are arranged in ascending order of the time of answering, forming a chronological answer sequence; Each element in the sequence is recorded as a record, numbered sequentially as the 1st, 2nd, ..., nth answer; Starting from the first record in the sequence, take two adjacent records in sequence (the i-th time and the (i+1)-th time) to form a pair of answers; For each answer pair, three features are extracted, including: The result of the first answer (denoted as R1, with a value of either correct or incorrect); The correctness result of the second answer (denoted as R2); The time interval between two answers (denoted as Δt, in seconds) is calculated as: the time of the second answer minus the time of the first answer; The preset short-term memory threshold (e.g., 86,400 seconds, or 24 hours) can be dynamically adjusted according to the characteristics of the subject, the difficulty of the knowledge points, or the learner's historical forgetting curve. For each correct answer, compare the time interval Δt between the two answers with the short-term memory threshold: If the time interval Δt between two responses is less than or equal to the threshold, then the two responses are considered to have occurred within the effective short-term memory range, and temporal correction is required. If the time interval Δt between two responses is greater than the threshold, then the time interval between the two responses is considered too long, and memory decline has exceeded the range of short-term effects. In this case, the response will not be included in the correction (or a zero-factor correction will be used). Based on the combination of two correctness results, a preset correction factor is matched. There are four combinations: Correct-correct combination: This indicates that the learner answered correctly consecutively within a short period of time, indicating that the level of mastery is stable or improving, and is matched with an incremental factor (e.g., +0.1). Combination correct-incorrect: indicates a shift from correct to incorrect, which may be due to memory lapse or instability, and is matched with an inhibitory factor (e.g., -0.1). Combination error-correct: Indicates a shift from error to correctness, possibly due to momentary recall or guesswork, matching the inhibitory factor (e.g., -0.1). Combination error - error: indicates consecutive errors, indicating no change in mastery level, and matches the zero-value factor (0); Get the current mastery probability of the current knowledge point (a floating-point number between 0 and 1; the initial value can be set according to the correctness of the first answer, such as 0.6 for the first correct answer and 0.3 for the first incorrect answer). Perform calculations based on the matched factor type: If it is an incremental factor: Initial correction value = Current mastery probability of the current knowledge point + Incremental factor value; If it is a suppression factor: Initial correction value = current mastery probability of the current knowledge point - absolute value of the suppression factor (i.e., subtract the suppression factor); If it is a zero-value factor: the initial correction value = the current mastery probability of the current knowledge point; The upper limit (e.g., 0.95) and lower limit (e.g., 0.05) of the probability of mastery are preset. The upper limit prevents over-optimism, and the lower limit retains the minimum probability of mastery. Compare the initial correction value with the upper limit: if it is greater than the upper limit, take the upper limit value; then compare the result with the lower limit: if it is less than the lower limit, take the lower limit value; finally, output the timing correction mastery probability. For example, learner Zhang San answered questions on the topic of quadratic equations in one variable a total of 4 times, and the answer records are as follows (in chronological order): |Question Number|Question Identifier|Answer Time|Correctness|Time Spent (seconds)|; |1|Q01|2025-01-01 09:00:00|Correct|45|; |2|Q02|2025-01-01 09:05:00||Correct|38|; |3|Q03|2025-01-01 09:30:00|Error|52|; |4|Q04|2025-01-02 10:00:00|Correct|40|; System parameters: Short-term memory threshold = 86400 seconds (24 hours), initial mastery probability = 0.5, increment factor = +0.1, inhibition factor = -0.1, upper limit of mastery probability = 0.95, lower limit = 0.05; Calculation steps: Answering questions for the first and second time (numbered 1-2); R1 = Correct, R2 = Correct, combine "Correct-Correct"; Δt = 09:05:00 - 09:00:00 = 300 seconds ≤ 86400 seconds, participates in the correction; Matching factor: Increment factor +0.1; The current probability of knowledge is P_current = 0.5; The initial correction value = 0.5 + 0.1 = 0.6; Boundary constraints: 0.6 is within [0.05, 0.95], the timing correction mastery probability = 0.6; Answer the questions for the second and third time (numbered 2-3); R1 = Correct, R2 = Incorrect, combine "Correct - Incorrect"; Δt = 09:30:00 - 09:05:00 = 1500 seconds ≤ 86400 seconds, participates in the correction; Matching factor: Inhibition factor -0.1; P_current=0.6; The initial correction value is 0.6 - 0.1 = 0.5. Boundary constraint: 0.5 within the interval, the probability of mastering the time series correction is 0.5; Answer the questions for the 3rd and 4th time (numbered 3-4); R1 = Error, R2 = Correct, combine "Error-Correct"; Δt = 2025-01-02 10:00:00 - 2025-01-01 09:30:00 = 88200 seconds ≈ 24.5 hours > 86400 seconds, exceeding the short-term memory threshold, and will not be included in the correction (using zero factor). Matching factor: Zero factor 0; P_current=0.5; Initial correction value = 0.5 + 0 = 0.5; The probability of mastering the timing correction is 0.5. Understandably, the logic of time-series correction is as follows: if a learner answers the same knowledge point questions correctly in a short period of time, it indicates that their mastery has improved, and the probability of mastery should be increased; if they go from correct to incorrect or from incorrect to correct in a short period of time, it indicates that their mastery is unstable, and the probability of mastery should be decreased; if they make consecutive mistakes, the probability of mastery remains unchanged. It should be noted that the role of temporal feature extraction is to introduce the time dimension into cognitive diagnosis, reflect the patterns of memory decline and consolidation, and dynamically adapt the mastery probability to the learner's real-time state. Mastery level diagnosis module: Based on time-series corrected mastery probability, determine the mastery rate of each cognitive level, compare and analyze it with the corresponding benchmark value, and generate the current mastery level identifier; The process of generating the currently known hierarchical identifier includes: According to Bloom's Taxonomy of Educational Objectives, cognitive levels are defined as follows: Conceptual cognition level: focuses on the ability to remember and recognize facts, terms, definitions, classifications, etc. Level of understanding of principles: Focuses on the ability to explain, interpret, and summarize the relationships between concepts, rules, and principles; Application transfer level: focuses on applying knowledge in new contexts, solving unstructured problems, and creatively transferring abilities; For the current knowledge point, all answered questions under that knowledge point are selected from the learning behavior data; According to the question-level mapping table, each question is assigned to the corresponding level. The question-level mapping table is pre-marked by subject matter experts, and each question corresponds to a cognitive level (concept / principle / application). The total number of test questions and the number of correct test questions under each level are counted separately. The ratio of the number of correct test questions to the total number of test questions in each level is calculated to obtain the concept cognition mastery rate, principle understanding mastery rate, and application transfer mastery rate. Compare the mastery rate of each level with the preset corresponding level achievement reference value (such as the concept cognition achievement threshold of 0.6, the principle understanding achievement threshold of 0.7, and the application transfer achievement threshold of 0.8), and write the achievement level identifier into the depth layer identifier field to form the current mastery level identifier; For example, under the same knowledge point of quadratic equations, learner Zhang San has completed 6 test questions. The cognitive level and correctness of each question are as follows: |Question ID|Cognitive Level|Correctness|; |T01|Conceptual Understanding|Correct|; |T02|Conceptual Understanding|Correct|; |T03|Conceptual Understanding|Errors|; |T04|Principle Understanding|Correct|; |T05|Principle Understanding|Error|; |T06|Application Migration|Error|; Reference values ​​for achieving the following standards: Conceptual cognition threshold = 0.6, Principle understanding threshold = 0.7, Application and transfer threshold = 0.8; Calculation steps: Calculate the total number and correct number for each level: Conceptual understanding: Total number of test questions = 3, number of correct answers = 2, mastery rate = 2 / 3 ≈ 0.667; Understanding the principle: Total number of questions = 2, number of correct answers = 1, mastery rate = 1 / 2 = 0.5; Application and transfer: Total number of test questions = 1, number of correct answers = 0, mastery rate = 0 / 1 = 0; Compared with the target threshold: Conceptual understanding mastery rate 0.667 ≥ 0.6, meeting the standard; The understanding and mastery rate of the principles was 0.5 < 0.7, which did not meet the standard. Application migration mastery rate 0 < 0.8, not meeting the standard; Understandably, the logic of tiered diagnosis is that different cognitive levels correspond to different ability requirements. Simply counting the overall accuracy rate cannot pinpoint weak areas. Tiered diagnosis can clarify the learner's specific mastery status at each stage of concept memorization, principle understanding, and application transfer. It should be noted that the role of mastering hierarchical diagnosis is to refine cognitive states into structured identifiers, providing accurate input for subsequent urgency calculation and remedial learning strategy selection; Urgency determination module: Combines the mastery probability with the current mastery level identifier, and analyzes the comparison of mastery probabilities to determine whether supplementary learning is needed. If so, it determines the urgency of supplementary learning and the corresponding supplementary learning trigger level based on the correspondence between the target level and the current level. The process of determining whether remedial learning is needed includes: The time-corrected mastery probability is compared with the preset mastery probability trigger threshold. If the time-corrected mastery probability is greater than or equal to the preset mastery probability trigger threshold, it is considered that the current mastery level has met the basic requirements and no supplementary learning is triggered. Otherwise, supplementary learning is required. The process of determining the trigger level for supplementary learning includes: If the time-corrected mastery probability is less than the preset mastery probability trigger threshold (e.g., 0.5), then the urgency calculation for supplementary learning will be triggered. Subtract the required mastery level (e.g., the expected application transfer level, which is 3) from the highest level identifier in the current mastery level identifier (e.g., the principle understanding level, which is 2) to obtain the level difference (1), and then multiply it by the preset importance weight coefficient (e.g., 0.3) to obtain the urgency of supplementary learning (0.3). The urgency of remedial learning was compared with the high threshold (0.7), median threshold (0.4), and low threshold (0.1) in that order: If the urgency of remedial learning is greater than or equal to the high threshold, it is judged as a high level; If the median threshold ≤ the urgency of remedial learning ≤ the high threshold, it is judged as the median level; If the low threshold is less than or equal to the urgency of remedial learning and less than or equal to the middle threshold, it is judged as a low level. If the urgency of supplementary learning is less than the low threshold, supplementary learning will not be triggered (or it will be considered the lowest priority). Ultimately, a high, medium, or low gear is selected as the trigger gear for supplementary learning; Understandably, the logic for determining urgency is as follows: when the probability of mastering the material is low and there is a large gap between the current level and the target level, the need for remedial learning is urgent and should be prioritized; otherwise, it should be dealt with more slowly. It should be noted that the purpose of urgency assessment is to quantify the priority of remedial learning, avoid using the same intervention strategy for all unmastered knowledge points, and achieve precise allocation of resources. Branch generation module: Based on the supplementary learning trigger level, retrieve the corresponding supplementary learning resource identifier to form a continuous supplementary learning branch form containing supplementary learning nodes and insertion nodes; The generation process of the supplementary learning branch persistence form includes: Retrieve the resource mapping table, which stores the identifiers of supplementary learning resources at each level corresponding to each knowledge point, including: Conceptual cognition level review content identifiers (e.g., micro-lesson video ID, knowledge point card ID); Identifiers for hierarchical explanations of principles (e.g., explanation animation ID, interactive explanation module ID); Application transfer level test question set identifier (e.g., a set of application question IDs); The first point to clarify is that if it is a high gear, then: Retrieve the application migration level test question set identifier and the principle understanding level explanation content identifier of the current knowledge point from the resource mapping table, and write them into a supplementary learning branch node set in the following order: first explanation (principle understanding), then test (application migration). Generate a complete branch structure, where each branch represents the learner's need to fully experience both the understanding of the principle and the transfer and application of that principle before returning to the main path. Secondly, it should be noted that if it is in the middle gear position, then: Retrieve the principle understanding level explanation content identifier of the current knowledge point from the resource mapping table, and write the identifier into the supplementary learning branch node set; Append a delay duration value (e.g., 3600 seconds, or 1 hour) to the starting node of the branch, indicating that the branch will not be executed immediately, but will wait for the specified duration before it can be executed; Generate delayed branching patterns for situations where understanding is required but the urgency is moderate, giving learners some time to digest the information independently. Thirdly, it should be noted that if it is a low gear, then: Retrieve the concept cognition level review content identifier of the current knowledge point from the resource mapping table and write it into the main path insertion node set; Generate the insertion node form as a lightweight review, which is directly embedded into the current main learning path without generating branch jumps; Understandably, the logic behind branch generation is as follows: high urgency requires complete supplementary learning of understanding and transfer capabilities, medium urgency allows for delayed supplementary learning of the understanding layer, and low urgency only requires a light review to avoid interrupting the main path; It should be noted that the purpose of branch generation is to dynamically adapt the form of supplementary learning content to the urgency level, thereby avoiding cognitive overload or waste of resources. Path merging module: Connects and embeds supplementary learning nodes and insertion nodes with main path nodes to form a complete learning path queue; The process of generating the complete learning path queue includes: If it is a complete branch merge, then: Obtain the set of supplementary learning branch nodes, where the nodes are arranged in the topological order of explanation followed by testing; Append all node identifiers in the supplementary learning branch node set to the end of the main path node queue in their original order; A new, complete learning path queue is generated. This means that learners must complete all nodes on the current main path before sequentially executing the supplementary learning branches. If it is a delayed branch merge, then: Get the set of supplementary learning branch nodes (containing only nodes for explaining the principles) and their delay duration values; Start a timer (or record a timestamp) to trigger the merge action after the delay period; When the delay period expires, obtain the current execution position index of the current main path node queue (i.e., the node position that the learner is currently executing or is about to execute). Insert the delayed branch node queue as a whole after the current execution position of the main path node queue (i.e., immediately after the current activity, without interrupting the current activity). Finally, an updated complete learning path queue is generated; If the merging is for inserted nodes, then: Get the set of nodes to be inserted on the main path (including the identifiers of the concept review nodes) and their insertion position offset values ​​(e.g., offset value = 3, which means that it is inserted after the 3rd node on the main path). Based on the offset value, the retrospective node identifier is written to the corresponding index position in the main path node queue; If the offset value exceeds the queue length, it is appended to the end; Finally, an updated and complete learning path queue is generated, and review nodes will be triggered naturally when the main path reaches a specified position. For example, the current main path node queue (learning activity sequence) is as follows: |Location Index|Activity Identifier|; |1|video_intro|; |2|exercise_basic_01|; |3|video_advanced|; |4|quiz_final|; The learner is executing position 2 (exercise_basic_01).

[0020] The branch generation module outputs a delayed branch, containing the node interactive_principle_quad, with a delay of 3600 seconds, and inserted at the position "after the current execution position".

[0021] Merging process: Delayed branch merging steps: Get the supplementary learning branch node queue = [interactive_principle_quad], delay = 3600 seconds; Start a timer; the merge will be triggered after 3600 seconds. When the delay period expires, get the current execution position index = 2 (because the learner may have already completed position 2 and is currently or about to enter position 3). Insert the delayed branch node queue after index 2 of the main path node queue, that is, before position 3.

[0022] The new main path queue becomes: |New Location Index|Activity Identifier|: |1|video_intro|; |2|exercise_basic_01|; |3|interactive_principle_quad|, the inserted branch node |4|video_advanced|; |5|quiz_final|; The learning engine executes according to the new queue order, completing the nodes at the original position 3 and beyond first, but the branch nodes have already been inserted into the appropriate positions.

[0023] Module output: Complete learning path queue = [video_intro, exercise_basic_01, interactive_principle_quad, video_advanced, quiz_final] Understandably, the logic of path merging is as follows: complete branches are appended to the end, delayed branches are inserted after the timer is completed and the inserted node is directly embedded at the specified position to ensure the continuity of the learning process. It should be noted that the purpose of path merging is to enable real-time linkage between cognitive state and learning path, so that the navigation system can respond to changes in actual situations such as cognitive load and time constraints.

[0024] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A knowledge graph-based full-scene learning navigation system, characterized in that: Includes the following steps: Based on the answer sequence of the same knowledge point, the correctness status and time interval of adjacent answers are extracted. According to the comparison results of the time interval and the preset value, and the matching results of the correctness status and the time sequence pattern, the time sequence correction mastery probability is obtained. The mastery rate of each cognitive level is determined based on the time-series corrected mastery probability, and compared with the corresponding benchmark value to generate the current mastery level identifier; By combining the time-series correction of mastery probability with the current mastery level identifier, and based on the comparison results of mastery probability and trigger reference value, as well as the correspondence between target level and current level, the urgency of remedial learning and the corresponding remedial learning trigger level are determined. Based on the supplementary learning trigger level, retrieve the corresponding supplementary learning resource identifier to form a continuous supplementary learning branch containing supplementary learning nodes or insertion nodes; Connect or embed supplementary learning nodes with or into the main path nodes to form a complete learning path queue.

2. The knowledge graph-based full-scene learning navigation system according to claim 1, characterized in that: The method for obtaining the timing correction mastery probability is as follows: Obtain the timestamp matrix of the answer sequence. Each row records the question identifier, answer time, and time consumption value. Arrange the answer records under the same knowledge point in ascending order according to the answer time to form the answer sequence. Extract the correctness results of two consecutive answers, the first and second answers, and the time interval between the two answers; Compare the time interval with the short-term memory threshold; The matching factor is combined based on the two correctness results: the incremental factor is obtained after the correct result, the inhibition factor is obtained after the correct result or the correct result after the error, and the zero-value factor is obtained after the error. The current mastery probability is added to the increment factor or subtracted from the inhibition factor to obtain the initial correction value. Then, it is compared with the upper and lower limits of the mastery probability to obtain the boundary value, thus generating the time-series corrected mastery probability.

3. The knowledge graph-based full-scene learning navigation system according to claim 2, characterized in that: The method of combining matching factors based on the two correctness results is as follows: If the result is correct, the increment factor is obtained; if the result is incorrect, the inhibition factor is obtained; if the result is incorrect, the zero factor is obtained.

4. The knowledge graph-based full-scene learning navigation system according to claim 1, characterized in that: The current method for generating the hierarchical identifier is as follows: The cognitive levels are divided into the conceptual cognition level, the principle understanding level, and the application transfer level; Calculate the ratio of the number of correct test questions to the total number of test questions in each level to obtain the mastery rate of concept cognition, mastery rate of principle understanding, and mastery rate of application and transfer. Compare the mastery rate of each level with the preset corresponding level's reference value, and write the identifier of the level that meets the standard into the depth layer identifier field to form the current mastery level identifier.

5. A knowledge graph-based full-scene learning navigation system according to claim 1, characterized in that: The method for determining the trigger level for supplementary learning is as follows: If the time-corrected mastery probability is less than the preset mastery probability trigger threshold, then the urgency calculation for supplementary learning will be triggered. Subtract the required mastery level from the highest level identifier in the current mastery level identifier to obtain the level difference, and then multiply it by the preset importance weight coefficient to obtain the urgency of remedial learning; The urgency of supplementary learning is compared with the high, medium, and low thresholds in sequence to obtain the high, medium, or low level as the supplementary learning trigger level.

6. A knowledge graph-based full-scene learning navigation system according to claim 5, characterized in that: The method for calculating the urgency of remedial learning is as follows: The required mastery level is subtracted from the highest level identifier in the current mastery level identifier to obtain the level difference value, which is then multiplied by the preset importance weight coefficient to obtain the urgency of remedial learning.

7. A knowledge graph-based full-scene learning navigation system according to claim 1, characterized in that: The method for generating the continuous form of the supplementary learning branch is as follows: When the supplementary learning trigger level is a high level, retrieve the application migration level test question set identifier and the principle understanding level explanation content identifier, write them into the supplementary learning branch node set, and generate a complete branch; When the supplementary learning trigger level is the middle level, the principle understanding level explanation content identifier is retrieved and written into the supplementary learning branch node set, and the delay duration value is written into the starting node to generate a delay branch; When the supplementary learning trigger level is low, the concept cognition level review content identifier is retrieved and written into the main path insertion node set to generate the insertion node.

8. A knowledge graph-based full-scene learning navigation system according to claim 7, characterized in that: The complete learning path queue is generated as follows: When the supplementary learning branch continues to be a complete branch, all node identifiers in the supplementary learning branch node set are merged with the main path node queue to generate a complete learning path queue. When the supplementary learning branch continues to be a delayed branch, the supplementary learning branch node queue of the delayed branch will be merged with the main path node queue after the delay duration value ends, to generate a complete learning path queue. When the supplementary learning branch continues to be an insertion node, the retrospective node identifier of the insertion node is written to the corresponding position in the main path node queue according to the insertion position offset value, and a complete learning path queue is generated.

9. A knowledge graph-based full-scene learning navigation system according to claim 7, characterized in that: The complete branch is formed by sequentially connecting the test node corresponding to the application migration level test question set identifier and the explanation node corresponding to the principle understanding level explanation content identifier; the delayed branch is formed by the explanation node corresponding to the principle understanding level explanation content identifier and the start waiting node associated with the delay duration value; the inserted node is formed by the review node corresponding to the concept cognition level review content identifier and is embedded in the specified position of the main path.

10. A knowledge graph-based full-scene learning navigation system according to claim 8, characterized in that: When merging complete branches, all node identifiers in the supplementary learning branch node set are appended to the end of the main path node queue in the order of node topology; when merging delayed branches, the delayed branch node queue is inserted after the current execution position of the main path node queue after the delay duration value ends; when merging inserted nodes, the retrospective node identifier is written to the corresponding index position in the main path node queue according to the insertion position offset value.

Images