Search

CN-121980068-A - Learning path recommendation method based on knowledge graph, computer equipment and storage medium

CN121980068ACN 121980068 ACN121980068 ACN 121980068ACN-121980068-A

Abstract

The invention belongs to the technical field of teaching informatization, and relates to a learning path recommending method, computer equipment and a storage medium based on a knowledge graph. The method comprises the steps of generating a basic learning path based on a knowledge map first repair relation, a learning target and knowledge points mastered by a learner, collecting and filtering learning behavior data to obtain effective evidence, updating knowledge point mastering states by adopting a double-threshold hysteresis mechanism, detecting remedy rollback oscillation by combining the mastering states through double windows, accurately positioning bottleneck nodes according to oscillation contribution degree, executing minimum residence control on the bottleneck nodes, generating a minimum remedy packet path only comprising forward gain nodes, and finishing path combination after verifying the mastering states. The invention can effectively solve the problems of rollback oscillation of learning path remedy, unstable state judgment, redundancy of remedy range and the like, can inhibit path oscillation, reduce invalid learning and improve learning path convergence and learning efficiency.

Inventors

  • YOU YUAN
  • CHEN HONGMEI
  • SHI YIFEI

Assignees

  • 四川吉利学院

Dates

Publication Date
20260505
Application Date
20260403

Claims (13)

  1. 1. The learning path recommending method based on the knowledge graph is characterized by comprising the following steps of: Generating an initial basic learning path based on the first repair relation of the knowledge graph, the learning target set and the knowledge point set mastered by the learner; in the basic learning path execution, learning behavior data of a learner are collected and filtered in real time, and an effective evidence set is generated; The method comprises the steps of calculating a grasping degree estimated value of a current knowledge point and a confidence interval based on an effective evidence set, adopting a double-threshold hysteresis mechanism to update the grasping state of the current knowledge point, wherein the double-threshold hysteresis mechanism is used for setting an upper grasping state judging threshold T_up and a lower grasping state judging threshold T_down, judging the grasping state to be grasped only when the lower bound of the confidence interval is more than or equal to T_up, judging the grasping state to be in learning only when the upper bound of the confidence interval is less than or equal to T_down, and keeping the grasping state unchanged when the confidence interval [ ci_low, ci_high ] is intersected with the interval [ T_down, T_up ]; according to the path execution log and the updated mastering state, calculating an oscillation index in the oscillation detection window, and judging to trigger remedial rollback oscillation when the oscillation index meets a judging condition generated based on historical stable learner data; Under the condition of remedied rollback oscillation triggering, calculating the oscillation contribution degree according to the rollback times, the state turnover times and the subsequent failure times of all nodes in an oscillation detection window, and selecting the node with the highest contribution degree as a bottleneck node; Performing minimum residence control on the bottleneck node until a preset minimum residence control release condition is met, and then releasing residence; meanwhile, generating a minimum remedy packet path based on the bottleneck node and a local first-repair structure associated with the bottleneck node, wherein the minimum remedy packet path only comprises nodes with forward gain for improving the subsequent pass rate of the bottleneck node; after the user finishes the minimum remedy packet path learning, checking the current mastering state of the bottle neck node, and merging the minimum remedy packet path into the basic learning path when the mastering state judging condition is met.
  2. 2. The knowledge-based learning path recommendation method of claim 1, wherein generating an initial base learning path comprises the steps of: recursively traversing the first repair relation edges in the knowledge graph until the first repair relation edges reach basic knowledge points without direct first repair nodes, and acquiring first repair closure of each target knowledge point in the learning target set; based on the first repair package and the knowledge point set mastered by the learner, calculating a union set and eliminating mastered nodes to construct a set to be learned; The method comprises the steps of carrying out topological sequencing on knowledge points in a set to be learned by adopting a Kahn algorithm, calculating the initial degree of entry of each knowledge point, adding the knowledge point with the degree of entry of 0 into a queue to be processed, recursively updating the degree of entry of a subsequent node until the queue is empty, and generating a basic learning path meeting the first repair constraint.
  3. 3. The knowledge-based learning path recommendation method as claimed in claim 1, wherein the processing of learning behavior data comprises the steps of: the first layer of filtering, namely performing identical topic deduplication and/or identical template deduplication on the collected original evidence, and only reserving a first answer record in a preset time window; The second layer of filtering, namely counting historical time distribution of learners under the current knowledge point, and eliminating abnormal time records of a lower time limit Q_low and an upper time limit Q_high determined by a quantile method; And (3) carrying out coverage statistics processing, namely extracting skill labels from the effective evidence set formed by the filtering of the first two layers, and maintaining and updating a skill label coverage set of the learner on the current knowledge points.
  4. 4. The learning path recommending method based on the knowledge graph according to claim 3, wherein the quantifying method for the length of the preset time window in the first layer of filtering is as follows: Extracting learners meeting preset stability conditions from historical learning data as target samples, wherein the preset stability conditions are that the rollback frequency is not higher than a first threshold value, the mastering state turnover frequency is not higher than a second threshold value and the one-time passing rate of a subsequent node is not lower than a third threshold value in a preset statistical period, and the first threshold value, the second threshold value and the third threshold value are determined according to the statistical distribution of corresponding indexes of the historical samples; Counting the time interval between adjacent effective learning behaviors of the target sample on the same learning content under the same knowledge point; Taking 75% quantile value of time interval in each group of statistical samples as the preset time window length of the corresponding category.
  5. 5. The learning path recommending method based on the knowledge graph according to claim 3, wherein the same template de-duplication comprises calculating a template-specific identifier and a parameter splicing character string by adopting an MD5 algorithm aiming at a template generation problem to generate a unique characteristic identifier, and judging that a subsequent record is invalid if the same unique characteristic identifier exists in the record of the latest preset times.
  6. 6. The learning path recommendation method based on the knowledge graph according to claim 1, wherein a confidence interval is calculated by adopting a Wilson two-term proportional confidence interval algorithm; The values of the mastering state judgment upper threshold T_up and the mastering state judgment lower threshold T_down are obtained based on mastering degree distribution statistics of a history stable learner group, wherein the history stable learner group is a learner group with rollback frequency not higher than a first threshold, mastering state turnover frequency not higher than a second threshold and one-time passing rate of a subsequent node not lower than a third threshold in a statistics period, T_up takes 80% of mastering degree distribution of a stable learner, and T_down takes 20% of mastering degree distribution of the stable learner.
  7. 7. The knowledge-graph-based learning path recommendation method of claim 1, wherein the oscillation detection window is an intersection of a step number window and a time window; The oscillation indexes comprise the number of rollback actions, the repetition rate and the number of mastering state turnover; The judgment conditions for triggering the remedy rollback oscillation are that the rollback action times are more than or equal to a rollback times judgment threshold, the repetition rate is more than or equal to a repetition rate judgment threshold and the state turnover times are more than or equal to a state turnover times judgment threshold, wherein each judgment threshold is generated based on index statistical distribution of a history stable learner group under the same window setting.
  8. 8. The knowledge-graph-based learning path recommendation method according to claim 1, wherein the formula for calculating the oscillation contribution degree is: Contrib (k) =w back_in ×back_in(k)+w back_out ×back_out(k)+w flip ×flip(k)+w fail_succ ×fail_succ (k), where k is the candidate knowledge node, back_in (k) is the number of times back to candidate knowledge node k in the oscillation detection window, back_out (k) is the number of times back again after advancing from node candidate knowledge node k, flip (k) is the number of times the state of grasp of candidate knowledge node k in the oscillation detection window, fail_succ (k) is the number of times failure after advancing from node candidate knowledge node k to the subsequent node, wback _in is the weight of back_in (k), wback _out is the weight of back_out (k), wflip is the weight of flip (k), wfail _ succ is the weight of fail_succ (k), and each weight is normalized by stabilizing the degree of distinction of learner population and oscillation learner population on the corresponding element.
  9. 9. The knowledge-based learning path recommendation method of claim 1, wherein generating a minimum remedy packet path comprises the steps of: obtaining a local candidate set composed of bottleneck nodes, direct first-repair nodes of the bottleneck nodes and second-order first-repair nodes; Calculating Gain (c) of each node c in the local candidate set, gain (c) =p (subsequent pass |c is mastered) -P (subsequent pass |c is not mastered), wherein P (subsequent pass |c is mastered) represents a probability that a direct subsequent node of the bottleneck node can be passed once under a condition that the node c is mastered; Selecting a node of Gain (c) 0, and intercepting the first q nodes in descending order according to a Gain value to form a core remedying node set, wherein q is a preset positive integer not more than the number of the Gain (c) 0 nodes; And recursively completing the direct repair nodes of all nodes in the core repair node set to form a minimum repair packet path.
  10. 10. The knowledge-graph-based learning route recommendation method according to claim 1, wherein the minimum resident control release condition includes: The accumulated effective evidence number is more than or equal to the minimum evidence quantity generated based on the statistics of the stable learners; The skill label coverage completion rate is more than or equal to the minimum coverage generated based on the statistics of the stable learners; the width of the grasping degree confidence interval is less than or equal to the maximum width generated based on the statistics of the steady learner, wherein the width of the grasping degree confidence interval is the difference between the upper boundary of the grasping degree confidence interval and the lower boundary of the grasping degree confidence interval.
  11. 11. The learning path recommendation method based on a knowledge graph according to claim 7, wherein when two indexes of a rollback action number, a repetition rate and a grasp state turnover number reach corresponding judgment thresholds, and the other index does not reach the corresponding judgment thresholds, an osc_mode field in a path execution log is set to 2 to identify an observed state, and when three indexes do not meet an oscillation triggering condition at the same time, the osc_mode field is set to 0to identify an off state.
  12. 12. A computer device comprising a memory, a processor and a transceiver in communication connection in sequence, wherein the memory is configured to store a computer program, the transceiver is configured to receive and transmit data, and the processor is configured to read the computer program and perform a learning path recommendation method based on a knowledge graph according to any one of claims 1-11.
  13. 13. A computer readable storage medium having instructions stored thereon which, when executed on a computer, perform a knowledge-graph based learning path recommendation method according to any of claims 1-11.

Description

Learning path recommendation method based on knowledge graph, computer equipment and storage medium Technical Field The invention belongs to the technical field of teaching informatization, and particularly relates to a learning path recommending method based on a knowledge graph, computer equipment and a storage medium. Background The prior learning path recommendation scheme based on the knowledge graph takes a 'knowledge point-first repair relation' as a bottom layer structure, and generates a linear or branched learning path sequence meeting the first repair constraint by analyzing the directional dependency relation (such as 'first repair of a unitary one-time equation solution' in the knowledge graph) and combining a learner image (such as age, learning basis and cognition level), learning behavior data (such as correct answer, learning duration and resource access record) or knowledge point mastery estimation model (such as Bayesian estimation and project reaction theory). In the learning process, the system dynamically adjusts the path according to the real-time evaluation result (such as the accuracy rate of the exercise and the small step score) of the learner, wherein the typical adjustment strategy is that when the learner continuously fails (such as the accuracy rate is lower than 30%) in the exercise or evaluation of the current target knowledge point (such as B), the system automatically returns to the direct first-repair knowledge point (such as A) to carry out remedial learning, and after the learner finishes the remedial resource learning of A, the learner is pushed to the original target knowledge point B again. However, in the floor scenario of an actual teaching platform or adaptive learning system, the dynamic adjustment mechanism described above is prone to "remedial rollback oscillations". Taking the first repair relation of knowledge points A, B and C as an example, a learner returns to the knowledge point A for remedying after failing in the practice of the knowledge point B, the system returns to the knowledge point A again after remedying and forms a local cycle of 'A ↔ B', and multi-node cycles of 'A ↔ B ↔ C' can occur in more complex scenes. The oscillation phenomenon can cause three technical defects that firstly, a learning path is not converged, a learner is trapped between a few knowledge points for a long time and cannot advance to a target knowledge point, secondly, ineffective learning time is increased rapidly, contacted knowledge point contents are repeatedly learned, thirdly, mastery degree judgment is distorted, low-quality learning evidences such as brushing questions, second answers, hanging up and the like accumulate in the oscillation process, the mastery degree estimation is interfered by noise, and the path recommendation is further inaccurate, so that a vicious circle of oscillation-misjudgment-more serious oscillation is formed. Disclosure of Invention In order to solve the technical problems, the invention is realized by the following technical scheme: According to the first aspect, a learning path recommending method based on a knowledge graph is provided, and the method comprises the following steps of generating an initial basic learning path based on a first repair relation of the knowledge graph, a learning target set and a knowledge point set mastered by a learner; in the basic learning path execution, learning behavior data of a learner are collected and filtered in real time, and an effective evidence set is generated; the method comprises the steps of calculating a grasping degree estimated value and a confidence interval of a current knowledge point based on an effective evidence set, updating grasping states of the current knowledge point by adopting a double-threshold hysteresis mechanism, wherein the double-threshold hysteresis mechanism comprises the steps of setting an upper grasping state judging threshold T_up and a lower grasping state judging threshold T_down, judging that the grasping states are grasped only when the lower bound of the confidence interval is more than or equal to T_up, judging that the grasping states are learned only when the upper bound of the confidence interval is less than or equal to T_down, keeping the grasping states unchanged when the confidence interval [ ci_low, ci_high ] is intersected with the interval [ T_down, T_up ], establishing an oscillation detection window comprising a step number window and a time window, executing a log and the updated grasping states according to paths, calculating oscillation indexes in the oscillation detection window, judging that remedial rollback is triggered when the oscillation indexes meet judging conditions generated based on historical stable learner data, calculating the oscillation contribution times of all nodes in the oscillation detection window according to the number of rollback, state turnover times and subsequent failure times in the oscillation detection window,