A personalized course recommendation method and system based on knowledge graph

Abstract

The present invention discloses a personalized course recommendation method and system based on knowledge graph, which relates to the field of course recommendation technology. The personalized course recommendation method and system based on knowledge graph include the following steps: S1, acquiring user learning data by real-time collection of front-end buried points and back-end access log data of the learning platform, and pre-processing the user learning data; S2, constructing a behavioral data relationship between users and knowledge points, evaluating the user's mastery of knowledge points, and constructing a time series knowledge point scoring set; S3, identifying abnormal mastery of knowledge points at the user's current stage based on the time series knowledge point scoring set, analyzing whether the course is worth recommending, partitioning candidate courses and generating recommended courses; S4, quantifying the response difference between recommended courses and mastery results, updating recommendation tags and user records, and improving the recommendation data cycle. It solves the problem that the existing technology is difficult to measure learning effects and is prone to fall into a behavior-driven recommendation cycle.

Description

Technical Field

[0001] The present invention relates to the field of course recommendation technology, and specifically to a personalized course recommendation method and system based on knowledge graph. Background Art

[0002] With the widespread adoption of online education platforms, user-oriented course recommendation systems have played a vital role in improving learning efficiency and experience. Currently, most recommendation systems rely on behavioral data such as user click history, browsing frequency, course collections, and completion rates to generate personalized recommendations. These systems utilize collaborative filtering, content matching, or knowledge graph reasoning to generate personalized recommendations.

[0003] For example, the invention with announcement number CN116542731A discloses a method for recommending fitness plans based on a knowledge graph, which relates to the field of personalized fitness recommendations. This method involves collecting user physical fitness characteristics and a dataset of fitness action courses, extracting knowledge attributes from fitness course action descriptions, and structuring the knowledge graph. The method takes a user physical fitness characteristic u and a course v as input, outputting the probability that a user with physical fitness characteristic u will perform training course action v, and adjusting the model's hyperparameters.

[0004] For example, the invention with publication number CN116501970A provides an online course recommendation method based on knowledge graph and convolution. The model first uses a feature extraction module to extract and convert user historical interaction information and feature information of the course field into embedding vectors, and integrates these vectors into the knowledge graph to generate embedding expressions of users and courses. Finally, this information is passed to adjacent nodes through a message propagation algorithm to obtain the final embedding vectors of users and courses for recommendation.

[0005] However, learning behavior itself is highly goal-oriented and progressive. Simply relying on behavioral indicators such as click-through rate or completion rate is difficult to accurately reflect the learner's true mastery of knowledge content. On the other hand, most existing recommendation systems are driven by user interests and lack feedback mechanisms based on ability improvement and learning outcomes, which can easily lead to the recommendation system falling into a "behavior-driven recommendation cycle."

[0006] Therefore, to address the above issues, there is an urgent need for a personalized course recommendation method and system based on knowledge graphs. Summary of the Invention

[0007] Technical problems solved

[0008] In response to the shortcomings of the existing technology, the present invention provides a personalized course recommendation method and system based on knowledge graph, which solves the problem that the existing technology is difficult to measure learning effects and easily falls into a behavior-driven recommendation cycle.

[0009] Technical solution. To achieve the above objectives, the present invention is implemented through the following technical solution: a personalized course recommendation method based on knowledge graph, comprising the following steps: S1, obtaining user learning data by real-time collection of front-end buried points and back-end access log data of the learning platform, and preprocessing the user learning data to obtain preprocessed user learning data; S2, constructing a behavioral data relationship between users and knowledge points, normalizing and splitting the preprocessed user learning data, evaluating the user's mastery of knowledge points, and constructing a time series knowledge point scoring set; S3, identifying the user's abnormal mastery of knowledge points at the current stage based on the time series knowledge point scoring set, conducting differentiated analysis in combination with the behavioral data relationship, evaluating the recommendation value of candidate courses, partitioning the candidate courses and generating recommended courses; S4, quantifying the response difference between recommended courses and mastery results from the proportion dimension, updating recommendation tags and user records, and improving the recommendation data cycle.

[0010] Furthermore, by collecting the front-end embedded points and back-end access log data of the learning platform in real time, the specific steps for obtaining user learning data are as follows: by collecting the front-end embedded points and back-end access log data of the learning platform in real time, user learning data is obtained, and the user learning data includes course code, course browsing time, number of course visits, historical average browsing time of the course, video viewing completeness, course rating data, course rating full score data, course test score data, course test full score data and number of knowledge points involved in the course; among them, the course code is extracted through the course name field in the course access log; the course browsing time is obtained by recording the difference between the start time and exit time of each user entering the course page, and summing up by course; by counting the number of user visits The total number of times a course page is viewed is used to obtain the number of course visits; the average historical browsing time of the course is obtained by dividing the total browsing time of the same course by the number of course visits in a week; the video viewing completeness is obtained by calculating the ratio of the actual playback time recorded by the video player to the total length of the course video; the course rating data and the full score data of the course rating are obtained by extracting the rating field submitted by the user after completing the course; the course test score data and the full score data of the course test are obtained by recording the rating results of the embedded quiz taken by the user after completing the course; the mapping relationship between courses and knowledge points in the course resource management database is analyzed, and the number of knowledge point tags covered by each course is counted to obtain the number of knowledge points involved in the course.

[0011] Furthermore, the user learning data is preprocessed, and the specific steps to obtain the preprocessed user learning data are as follows: by combining the median absolute deviation method of the sliding time window with the isolation forest algorithm, local and global anomaly detection is performed on the user learning data, and abnormal user learning data including abnormally long idle time and skipping classes to make progress are eliminated; by using the similar user collaborative completion strategy and the exponentially weighted moving average algorithm, data is filled in for missing fields in the user learning data caused by system delays, point loss or terminal fluctuations; by using the Bayesian dynamic smoothing method and the local weighted regression filtering algorithm, the user learning data is fitted and denoised; by using the exponential transformation combined with the maximum and minimum normalization method, the user learning data is transformed and range compressed to achieve normalization processing.

[0012] Furthermore, the behavioral data relationship between users and knowledge points is constructed, the pre-processed user learning data is normalized and split, and the specific steps for evaluating users' mastery of knowledge points are as follows: according to the knowledge point labels marked in each course, the corresponding knowledge point set is extracted; combined with the course code in the user's learning record, the knowledge point labels included in the user's learned courses are searched one by one, and merged in the user dimension to obtain the complete knowledge point set learned by the user; with the user name and the knowledge point label as the corresponding item, the association records between the user and the knowledge point are sorted to form the behavioral correspondence between the user and the knowledge point in the learning process; based on the established behavioral correspondence between users and knowledge points, for each knowledge point, the course browsing time, video viewing completeness and course rating number generated by the user in the course containing the same knowledge point are integrated Based on these preprocessed user learning data, evaluate the user's mastery of this knowledge point: for a course that contains this knowledge point, calculate the ratio of the course browsing time to the historical average browsing time of the course, multiply this ratio by the browsing time weight, and use the product as the browsing time score; multiply the video viewing completeness by the video completeness weight, and use the product as the video completeness score; divide the course rating data by the full score of the course rating, and then multiply the ratio by the course rating weight, and use the product as the course rating data score; add the browsing time score, video completeness score, and course rating data score, and multiply it by the inverse of the number of knowledge points involved in the course to obtain the user's shared score for this knowledge point in the course; recalculate the shared scores for all courses that contain this knowledge point, and sum up the shared scores of each course to obtain the knowledge point mastery assessment value.

[0013] Furthermore, the specific steps for constructing a time series knowledge point scoring set are as follows: after calculating the knowledge point mastery evaluation value, the knowledge point mastery evaluation value and the mastery threshold are compared in real time: when the knowledge point mastery evaluation value is greater than or equal to the mastery threshold, it is marked as a normally mastered knowledge point and no adjustment is required; when the knowledge point mastery evaluation value is less than the mastery threshold, it is marked as an abnormally mastered knowledge point, the user's current learning path is prompted, and a recommendation instruction for adding supplementary courses related to this knowledge point is generated; the compared knowledge point mastery evaluation value is time-serialized: the knowledge point mastery evaluation value generated by the user in each learning behavior is recorded in chronological order, and the corresponding timestamp, course code and knowledge point label are marked; all knowledge point mastery evaluation values ​​of the same user on the same knowledge point are sorted in chronological order to construct a knowledge point mastery evaluation value sequence of the user on this knowledge point; the knowledge point mastery evaluation value sequence of the user on all knowledge points is aggregated and sorted to form a time series knowledge point mastery evaluation value set of the user under the entire knowledge structure.

[0014] Furthermore, based on the time series knowledge point scoring set, the abnormal mastery of knowledge points in the user's current stage is identified, and differentiation analysis is carried out in combination with the behavioral data relationship. The specific steps for evaluating the recommendation value of candidate courses are as follows: Combine the time series knowledge point mastery evaluation value set of the user under the entire knowledge structure, extract the set of knowledge points marked as abnormal mastery in the current period, and count the number of knowledge points marked as abnormal mastery; group the abnormal records according to the knowledge point labels, identify the knowledge point labels that are continuously marked as abnormal within a short period of time within the same user, and form the key supplementary knowledge point set of the user's current stage; based on the identified key supplementary knowledge point set, review all courses associated with each knowledge point, eliminate courses that the user has completed and courses that have not been watched completely, and include the remaining courses in the candidate course set; For each course in the candidate course set, calculate the course's historical average rating, extract the number of overlaps between the course's associated knowledge points and the set of supplementary knowledge points, and analyze whether the course is worthy of recommendation; calculate the square value of the number of overlaps between the course's associated knowledge points and the set of supplementary knowledge points, calculate the product of the number of knowledge points involved in the course and the number of knowledge points marked as abnormally mastered, divide this square value by this product, and use the ratio as the knowledge point matching score; calculate the course browsing time divided by the historical average browsing time of the course, and add this ratio to the video viewing completeness to obtain the learning input score; calculate the course rating data plus the historical average course rating, and divide the sum by twice the full score of the course rating data to obtain the course rating data performance score; multiply the knowledge point matching score, learning input score, and course rating data performance score to obtain the course recommendation strength assessment value.

[0015] Furthermore, the specific steps for partitioning candidate courses and generating recommended courses are as follows: based on the course recommendation strength evaluation value, a multi-dimensional course recommendation heat partition is constructed according to the course recommendation threshold: wherein the course recommendation threshold includes the first-level course recommendation threshold and the second-level course recommendation threshold; when the course recommendation strength evaluation value is greater than or equal to the second-level course recommendation threshold, it is judged as a high-heat course, directly included in the current recommended courses, and prioritized for push; when the course recommendation strength evaluation value is greater than the first-level course recommendation threshold and less than the second-level course recommendation threshold, it is judged as a medium-heat course, enters the selection area, and performs topic aggregation through the supplementary knowledge point tags covered by the course, prioritizes matching according to the user's historical unlearned content, recommends 1 to 2 courses according to the tag distribution, and expands the recommendation coverage; when the course recommendation strength evaluation value is less than or equal to the first-level course recommendation threshold, it is judged as a low-heat course, and does not enter the current recommended courses; generate recommended courses for the user's current stage based on the multi-dimensional course recommendation heat partition results, and send the recommended courses to the user learning interface; mark the supplementary knowledge point tags covered by each course in the recommended course list.

[0016] Furthermore, the specific steps for quantifying the response difference between recommended courses and mastery outcomes from the proportion dimension are as follows: after the user completes the recommended course, the course code, course browsing time, and supplementary knowledge point tags covered by the course are extracted to establish the content structure of the recommended course; the course test score data after the recommended course is recorded simultaneously, and the corresponding knowledge point mastery evaluation value is updated in the knowledge point dimension; a comparison relationship between the knowledge point mastery evaluation values ​​before and after recommendation is established according to the user and the knowledge point; based on the comparison relationship between the knowledge point mastery evaluation values ​​before and after recommendation, the response difference between the recommended course and mastery outcomes is quantified: the knowledge point mastery evaluation value of the knowledge point after the recommended course is issued minus the knowledge point mastery evaluation value of the knowledge point before the recommended course is issued is divided by the full score data of the course rating, and then the ratio of the course test score data to the full score data of the course test is subtracted from this ratio, and the absolute value of the difference is taken as the mastery change ratio; the number of recommended courses involving this knowledge point in the recommended course is added by 1, and the base-two logarithm is taken, and the result is added by 1 as the course coverage factor; the mastery change ratio is multiplied by the course coverage factor to obtain the mastery feedback response value.

[0017] Furthermore, the specific steps for updating the recommendation tags and user records and improving the recommendation data cycle are as follows: after calculating the mastery feedback response value, the mastery feedback response value is compared with the mastery feedback threshold in real time: wherein, the mastery feedback threshold includes the first-level mastery feedback threshold and the second-level mastery feedback threshold; when the mastery feedback response value is greater than or equal to the second-level mastery feedback threshold, the feedback is normal, and the knowledge points covered in the current course are marked as recommendation valid tags, thereby improving the recommendation priority of the current course in the subsequent candidate course ranking among similar users; when the mastery feedback response value is greater than the first-level mastery feedback threshold and less than the second-level mastery feedback threshold, the feedback is delayed, and the current course is marked as an intermediate transition course and included in the next recommendation cycle as a continuous observation object for delayed supplementary content; when the mastery feedback response value is less than or equal to the first-level mastery feedback threshold, the feedback is abnormal, and the knowledge points covered in the current course are marked as recommendation offset tags, marking the current course as a recommendation-insensitive course, and giving priority to avoiding courses with similar teaching structures in subsequent supplementary course screening. ; According to the user dimension, the mastery feedback response values ​​generated by the same knowledge point in different recommendation cycles are arranged continuously to construct a mastery feedback response value sequence; for knowledge points with feedback response values ​​less than the first-level mastery feedback threshold three or more times, the course code and course rating data corresponding to the knowledge point are extracted, and the user's knowledge point mastery evaluation value for this course is recalculated based on the original course browsing time, video viewing completeness and course rating data, replacing the original result for feedback correction; the course code, knowledge point label, course browsing time, course rating data, video viewing completeness, mastery feedback response value, course test score data and evaluation value change results generated by each round of recommended course learning are written into the user recommendation record; in the new round of recommendation generation process, user recommendation records are read first, the initial recommendation ranking of courses, the popularity differentiation strategy and the recommendation threshold calculation logic are dynamically adjusted, the binding strength between candidate courses and key knowledge points is updated, and the active adaptive adjustment of the recommendation path is completed, thereby realizing a recommendation self-closed loop mechanism based on ability feedback.

[0018] The second aspect of the present invention provides a personalized course recommendation system based on knowledge graph, including: a user learning data collection and preprocessing module, a knowledge point mastery evaluation module, a supplementary course recommendation generation module and a recommendation effect feedback correction module, wherein: the user learning data collection and preprocessing module is used to obtain user learning data by real-time collection of front-end buried points and back-end access log data of the learning platform, and preprocess the user learning data to obtain preprocessed user learning data; the knowledge point mastery evaluation module is used to construct a behavioral data relationship between users and knowledge points, normalize and split the preprocessed user learning data, evaluate the user's mastery of knowledge points, and construct a time series knowledge point rating set; the supplementary course recommendation generation module is used to identify the abnormal mastery of knowledge points at the user's current stage based on the time series knowledge point rating set, conduct differential analysis in combination with the behavioral data relationship, evaluate the recommendation value of candidate courses, partition the candidate courses and generate recommended courses; the recommendation effect feedback correction module is used to quantify the response difference between recommended courses and mastery results from the proportion dimension, update recommendation tags and user records, and improve the recommendation data cycle.

[0019] Beneficial effects

[0020] The present invention has the following beneficial effects:

[0021] (1) This personalized course recommendation method and system based on knowledge graph proposes an accurate knowledge point mastery assessment method by integrating user learning data such as course browsing time, video viewing completeness and course rating data, making the assessment of the user's knowledge point mastery status during the learning process more detailed and accurate, breaking through the limitation of traditional systems that only rely on click rate and completion rate to measure learning effectiveness.

[0022] (2) This knowledge graph-based personalized course recommendation method and system quantifies the changes in the knowledge point assessment values ​​before and after recommendation and the differences in course test score data. It combines the number of recommended courses with the user's behavioral response to dynamically adjust the feedback on learning effects, reflecting the real impact of learning behavior on ability improvement, and providing a real-time feedback basis for the optimization of recommended content.

[0023] (3) This knowledge graph-based personalized course recommendation method and system designs a high, medium, and low popularity partition recommendation strategy based on the course recommendation strength evaluation. In addition, it adopts a differentiated push strategy for courses of different intensities in combination with the actual feedback on the recommended content, balances precise recommendations with exploratory recommendations, and improves the adaptability and coverage of the system.

[0024] (4) This personalized course recommendation method and system based on knowledge graph realizes the dynamic tracking and stage-by-stage identification of knowledge mastery status by constructing a time series mastery assessment record between users and knowledge points, and constructs an abnormal mastery identification and supplementary content generation mechanism based on the knowledge point dimension, effectively improving the timeliness and accuracy of recommendation decisions.

[0025] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Flowchart of the personalized course recommendation method based on knowledge graph;

[0027] Figure 2 This is the structure diagram of the personalized course recommendation system based on knowledge graph;

[0028] Figure 3 Visualize the bar chart of knowledge point mastery assessment values;

[0029] Figure 4 It is a line graph of the mastery assessment value and the mastery feedback response value before and after the recommended course. DETAILED DESCRIPTION

[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0031] See also Figure 1-Figure 4 , an embodiment of the present invention provides a technical solution: a personalized course recommendation method based on knowledge graph, comprising the following steps: S1, obtaining user learning data by real-time collection of front-end buried points and back-end access log data of the learning platform, and preprocessing the user learning data to obtain preprocessed user learning data; S2, constructing a behavioral data relationship between users and knowledge points, normalizing and splitting the preprocessed user learning data, evaluating the user's mastery of knowledge points, and constructing a time series knowledge point scoring set; S3, identifying the abnormal mastery of knowledge points at the user's current stage based on the time series knowledge point scoring set, conducting differential analysis in combination with the behavioral data relationship, evaluating the recommendation value of candidate courses, partitioning the candidate courses and generating recommended courses; S4, quantifying the response difference between recommended courses and mastery results from the proportion dimension, updating recommendation tags and user records, and improving the recommendation data cycle.

[0032] Specifically, by collecting the front-end embedded points and back-end access log data of the learning platform in real time, the specific steps for obtaining user learning data are as follows: by collecting the front-end embedded points and back-end access log data of the learning platform in real time, user learning data is obtained, and user learning data includes course code, course browsing time, number of course visits, historical average browsing time of the course, video viewing completeness, course rating data, course rating full score data, course test score data, course test full score data and the number of knowledge points involved in the course; among them, the course code is extracted through the course name field in the course access log; the course browsing time is obtained by recording the difference between the start time and exit time of each user entering the course page, and summing up by course; by counting the number of user visits The total number of times a course page is viewed is used to obtain the number of course visits; the average historical browsing time of the course is obtained by dividing the total browsing time of the same course by the number of course visits in a week; the video viewing completeness is obtained by calculating the ratio of the actual playback time recorded by the video player to the total length of the course video; the course rating data and the full score data of the course rating are obtained by extracting the rating field submitted by the user after completing the course; the course test score data and the full score data of the course test are obtained by recording the rating results of the embedded quiz taken by the user after completing the course; the mapping relationship between courses and knowledge points in the course resource management database is analyzed, and the number of knowledge point tags covered by each course is counted to obtain the number of knowledge points involved in the course.

[0033] In this implementation, a multidimensional data foundation covering the entire user behavior process is constructed by precisely defining and collecting course codes, course browsing time, number of course visits, historical average course browsing time, video viewing completeness, course rating data, full-score course rating data, course test score data, full-score course test data, and the number of knowledge points covered in the course. This method effectively integrates front-end embedded behavior data with back-end access log information, achieving high-precision mapping of user learning behavior and learning content without changing the consistency of data names. This provides accurate, stable, and clearly structured data support for subsequent knowledge point mastery assessment, course recommendations, and feedback optimization, breaking through the bottleneck of traditional recommendation systems' inadequate portrayal of user learning status.

[0034] Specifically, the user learning data is preprocessed to obtain the preprocessed user learning data in the following specific steps: by combining the median absolute deviation method of the sliding time window with the isolation forest algorithm, local and global anomaly detection is performed on the course browsing time, course visit times and course rating data in the user learning data, focusing on eliminating abnormal user learning data that violates the normal learning path rules, including abnormally long idle time and skipping classes to brush up progress; through the similar user collaborative completion strategy and the exponentially weighted moving average algorithm, the field missing problems of course rating data, course test score data and video viewing completeness caused by system delays, point loss and terminal fluctuations are intelligently completed to ensure the integrity and continuity of user learning data; the user learning data is fitted and denoised by the Bayesian dynamic smoothing method and the local weighted regression filtering algorithm to improve data stability; the user learning data is transformed and range compressed by the exponential transformation combined with the maximum and minimum normalization method to achieve normalization processing, providing a structural and consistent preprocessing data foundation for subsequent analysis.

[0035] In this implementation plan, by systematically preprocessing the course browsing time, course visit times, course rating data, course test score data and video viewing completeness in the user learning data, it is possible to effectively identify and eliminate abnormal user learning data, repair field missing problems, eliminate data noise, unify variable scales, and significantly improve the integrity, stability and comparability of the preprocessed user learning data. This provides a high-quality, structured data foundation for the subsequent construction of behavioral data relationships between users and knowledge points, evaluation of knowledge point mastery, and generation of recommended courses, ensuring the accuracy and robustness of the recommendation system.

[0036] Specifically, the behavioral data relationship between users and knowledge points is constructed, the pre-processed user learning data is normalized and split, and the specific steps for evaluating users' mastery of knowledge points are as follows: according to the knowledge point labels marked in each course, the corresponding knowledge point set is extracted; combined with the course code in the user's learning record, the knowledge point labels included in the user's learned courses are searched one by one, and merged in the user dimension to obtain the complete knowledge point set learned by the user; with the user name and the knowledge point label as the corresponding item, the association records between the user and the knowledge point are sorted to form the behavioral correspondence between the user and the knowledge point in the learning process; based on the established behavioral correspondence between users and knowledge points, for each knowledge point, the course browsing time, video viewing completeness and course rating data generated by the user in the course containing the same knowledge point are integrated These preprocessed user learning data are used to evaluate the user's mastery of this knowledge point: for a course that contains this knowledge point, calculate the ratio of the course browsing time to the historical average browsing time of the course, multiply this ratio by the browsing time weight, and use the product as the browsing time score; multiply the video viewing completeness by the video completeness weight, and use the product as the video completeness score; divide the course rating data by the full score of the course rating, and then multiply the ratio by the course rating weight, and use the product as the course rating data score; add the browsing time score, video completeness score, and course rating data score, and multiply it by the inverse of the number of knowledge points involved in the course to obtain the user's shared score for this knowledge point in the course; recalculate the shared scores for all courses that contain this knowledge point, and sum up the shared scores of each course to obtain the knowledge point mastery assessment value. Among them, by taking the course browsing time, video viewing completeness, and course rating data collected from historical user learning data as independent variables and the periodic assessment scores as dependent variables, the least squares regression algorithm is used for parameter fitting. The browsing time weight, video completeness weight, and course rating weight are obtained according to the explanatory strength of each behavioral characteristic on the ability mastery effect. The value range of the browsing time weight, video completeness weight, and course rating weight is [0,1].

[0037] The specific formula for the knowledge point mastery evaluation value is:

[0038] ;

[0039] Where, Indicates the knowledge point mastery assessment value, Indicates the course that contains this knowledge point. Represents the set of courses that contain this knowledge point, Indicates the browsing time weight, Indicates the course browsing time. Indicates the average browsing time of the course history. Indicates the video completeness weight, Indicates the video viewing completeness. Indicates the course scoring weight, Represents course rating data, Indicates the full score data of the course. Indicates the number of knowledge points covered in the course.

[0040] In this embodiment, Table 1 is a knowledge point mastery evaluation value data table, which records in detail the course set, full score data of course rating, course browsing time weight, video viewing completeness weight, course rating weight and the knowledge point mastery evaluation value finally calculated for different knowledge points during the evaluation process, and is used to quantify the user's learning and mastery of various military knowledge points. Among them: the course set corresponding to knowledge point C1 is B1, the full score data of course rating is 10, the course browsing time weight is 0.3, the video viewing completeness weight is 0.4, the course rating weight is 0.3, and the knowledge point mastery evaluation value is 0.72; the course set corresponding to knowledge point C2 is B2, the full score data of course rating is 10, the course browsing time weight is 0.3, the video viewing completeness weight is 0.4, the course rating weight is 0.3, and the knowledge point mastery evaluation value is 0.65; the course set corresponding to knowledge point C3 is B3, the full score data of course rating is 10, the course browsing time weight is 0 .3, the weight of video viewing completeness is 0.4, the weight of course score is 0.3, and the evaluation value of knowledge point mastery is 0.78; the course set corresponding to knowledge point C4 is B4, the full score of course score is 10, the weight of course browsing time is 0.3, the weight of video viewing completeness is 0.4, the weight of course score is 0.3, and the evaluation value of knowledge point mastery is 0.59; the course set corresponding to knowledge point C5 is B5, the full score of course score is 10, the weight of course browsing time is 0.3, the weight of video viewing completeness is 0.4, the weight of course score is 0.3, and the evaluation value of knowledge point mastery is 0.84.

[0041] Table 1 Knowledge point mastery evaluation value data table

[0042]

[0043] like Figure 3The following is a visualization of the knowledge point mastery assessment values. Combined with Table 2, it can be seen that users exhibit significant differences in their mastery of different knowledge points. Knowledge point C2 has the highest mastery assessment value, at 0.84, indicating that users achieve the best learning outcomes within the corresponding course set. Knowledge point C3 has the lowest mastery assessment value, at only 0.59, indicating relatively weak learning behavior within this knowledge point. Knowledge points C4 and C1 have relatively high mastery assessment values, indicating stable learning. Knowledge point C5 has an assessment value of 0.65, placing it at a lower-to-average level. Overall, the visualization of the knowledge point mastery assessment values ​​intuitively reflects the differences in mastery across different knowledge points and can serve as a basis for recommending supplementary courses and for feedback adjustment.

[0044] In this implementation plan, by constructing the behavioral data relationship between users and knowledge points, based on the course browsing time, video viewing completeness and course rating data in the pre-processed user learning data, the historical average browsing time of the course, the full score data of the course rating and the number of knowledge points involved in the course are comprehensively introduced to form a multi-dimensional scoring system based on the browsing time score, video completeness score and course rating data score, which realizes the refined modeling and quantitative evaluation of the user's learning and mastery of each knowledge point, effectively enhancing the discrimination ability and dynamic expression ability of the knowledge point mastery evaluation value, and laying a high-credibility evaluation foundation for subsequent precise recommendation and feedback regulation.

[0045] Specifically, the specific steps for constructing a time series knowledge point scoring set are as follows: after calculating the knowledge point mastery evaluation value, the knowledge point mastery evaluation value and the mastery threshold are compared in real time: when the knowledge point mastery evaluation value is greater than or equal to the mastery threshold, it is marked as a normally mastered knowledge point and no adjustment is required; when the knowledge point mastery evaluation value is less than the mastery threshold, it is marked as an abnormally mastered knowledge point, the user's current learning path is prompted, and a recommendation instruction for adding supplementary courses related to this knowledge point is generated; the compared knowledge point mastery evaluation value is time-serialized: the knowledge point mastery evaluation value generated by the user in each learning behavior is recorded in chronological order, and the corresponding timestamp, course code and knowledge point label are marked; all knowledge point mastery evaluation values ​​of the same user on the same knowledge point are sorted in chronological order to construct a sequence of the user's knowledge point mastery evaluation values ​​on this knowledge point; the user's knowledge point mastery evaluation value sequences on all knowledge points are aggregated and sorted to form a time series knowledge point mastery evaluation value set for the user under the entire knowledge structure.

[0046] In this implementation plan, by comparing the knowledge point mastery assessment value with the mastery threshold in real time, abnormally mastered knowledge points are identified in a timely manner, and supplementary course recommendation instructions are dynamically generated in combination with the learning path, thereby achieving active intervention in learning blind spots; at the same time, by labeling and sorting the knowledge point mastery assessment values ​​in chronological order, a time series knowledge point mastery assessment value set covering all knowledge points is constructed, which effectively improves the ability to track and express the evolution process of user knowledge mastery, and provides high-resolution, time-series data support for subsequent stage-by-stage capability assessment and personalized recommendation strategies.

[0047] Specifically, based on the time series knowledge point scoring set, the abnormal mastery of knowledge points in the user's current stage is identified, and differentiated analysis is carried out in combination with the behavioral data relationship. The specific steps for evaluating the recommendation value of candidate courses are as follows: Combine the time series knowledge point mastery evaluation value set of the user under the entire knowledge structure, extract the set of knowledge points marked as abnormal mastery in the current period, and count the number of knowledge points marked as abnormal mastery; group abnormal records according to knowledge point labels, identify the knowledge point labels that are continuously marked as abnormal within a short period of time within the same user, and form a set of key supplementary knowledge points for the user's current stage; based on the identified set of key supplementary knowledge points, review all courses associated with each knowledge point, eliminate courses that the user has completed and courses that he has not watched completely, and include the remaining courses in the candidate course set; for For each course in the candidate course set, the average historical course rating is calculated, the number of overlaps between the course-related knowledge points and the supplementary knowledge point set is extracted, and the course is analyzed to see if it is worth recommending; the square value of the number of overlaps between the course-related knowledge points and the supplementary knowledge point set is calculated, the product of the number of knowledge points involved in the course and the number of knowledge points marked as abnormally mastered is calculated, this square value is divided by this product, and the ratio is used as the knowledge point matching score; the course browsing time is calculated and divided by the average historical course browsing time, and this ratio is added to the video viewing completeness to obtain the learning input score; the course rating data plus the course historical average rating is calculated, and the sum is divided by twice the full score of the course rating data to obtain the course rating data performance score; the knowledge point matching score, learning input score, and course rating data performance score are multiplied together to obtain the course recommendation strength assessment value.

[0048] The specific calculation formula for the course recommendation strength evaluation value is:

[0049] ;

[0050] Where, Indicates the evaluation value of the course recommendation strength, Indicates the number of overlaps between the course-related knowledge points and the set of supplementary knowledge points. Indicates the number of knowledge points involved in the course, Indicates the number of knowledge points marked as abnormally mastered, Indicates the course browsing time. Indicates the average browsing time of the course history. Indicates the video viewing completeness. Represents course rating data, Indicates the historical average score of the course. Indicates the full score data of the course.

[0051] In this implementation plan, by constructing a time series knowledge point mastery evaluation value set, abnormal mastery knowledge points in the current stage are accurately identified, and fine-grained differentiation analysis is carried out in combination with user behavior data relationships to form a key supplementary knowledge point set; on this basis, the number of overlaps between course-related knowledge points and supplementary knowledge point sets, course browsing time, video viewing completeness and course rating data are introduced to construct a comprehensive indicator including knowledge point matching score, learning investment score and course rating data performance score to quantify the course recommendation strength evaluation value of candidate courses, thereby enhancing the pertinence and effectiveness of recommended course screening.

[0052] Specifically, the specific steps for partitioning candidate courses and generating recommended courses are as follows: based on the course recommendation strength evaluation value, a course recommendation strength evaluation index system is constructed by analyzing the matching degree between the course-related knowledge points and the set of supplementary knowledge points, the course browsing behavior input and the course rating data performance, and a multi-dimensional course recommendation popularity partition is constructed according to the course recommendation threshold; wherein, the course recommendation threshold includes the first-level course recommendation threshold and the second-level course recommendation threshold, which are dynamically adjusted according to the course feedback records in the historical recommendation results and the fluctuation trend of the knowledge point mastery evaluation value; when the course recommendation strength evaluation value is greater than or equal to the second-level course recommendation threshold, it is judged as a high-popularity course, directly included in the current recommended courses, and prioritized for push, and at the same time triggering the subsequent monitoring process of the mastery feedback response value in the recommendation strategy module; when the course recommendation strength evaluation value is greater than the first-level course recommendation threshold, the course recommendation strength evaluation value is greater than or equal to the second-level course recommendation threshold, and the course recommendation strength evaluation value is greater than or equal to the second-level course recommendation threshold. Recommendation threshold and less than the second-level course recommendation threshold, it is determined to be a medium-popularity course and enter the selection area. The subject aggregation is carried out through the supplementary knowledge point tags covered by the course, and priority matching is carried out based on the knowledge point tags not covered in the user's learning record. 1 to 2 courses are recommended according to the tag distribution to expand the recommendation coverage and maintain recommendation diversity; when the course recommendation strength evaluation value is less than or equal to the first-level course recommendation threshold, it is determined to be a low-popularity course and will not enter the current recommended course, but will be recorded as a candidate resource for replenishment in the event of subsequent abnormal mastery feedback; recommended courses for the user's current stage are generated based on the multi-dimensional course recommendation heat partitioning results, and the recommended courses are sent to the user learning interface; the supplementary knowledge point tags covered by each course are marked in the recommended course list, and it is indicated whether it is a course that has not been learned in the user's historical learning path, so as to assist users in course selection and path judgment.

[0053] This implementation, by introducing a course recommendation strength assessment value and establishing a course recommendation threshold range, achieves multi-dimensional popularity zoning management for candidate courses. This not only fully considers the degree of match between the course's associated knowledge points and the set of supplementary knowledge points, course browsing behavior, and course rating data, but also incorporates the tags of uncovered knowledge points in the user's learning history, enhancing the adaptability and scalability of recommended content. At the same time, a differentiated push mechanism is implemented for courses in different popularity ranges, and this is linked to the subsequent mastery feedback response value monitoring process to establish a dynamic adjustment and feedback closed-loop mechanism, providing data support and strategic guarantees for the accuracy, controllability, and dynamic and sustainable optimization of recommendation results.

[0054] Specifically, the specific steps for quantifying the response difference between recommended courses and mastery effectiveness from the proportion dimension are as follows: after the user completes the recommended course, the course code, course browsing time and supplementary knowledge point tags covered by the course involved in this recommended course are extracted to establish the content structure of the recommended course; the course test score data after the recommended course is recorded synchronously, and the corresponding knowledge point mastery evaluation value is updated in the knowledge point dimension; a comparison relationship between the knowledge point mastery evaluation values ​​before and after the recommendation is established according to the user and the knowledge point; based on the comparison relationship between the knowledge point mastery evaluation values ​​before and after the recommendation, the response difference between the recommended course and the mastery effectiveness is quantified: the knowledge point mastery evaluation value of the knowledge point after the recommended course is issued minus the knowledge point mastery evaluation value of the knowledge point before the recommended course is issued is divided by the full score data of the course rating, and then the ratio of the course test score data to the full score data of the course test is subtracted from this ratio, and the absolute value of the difference is taken as the mastery change ratio; the number of recommended courses involving this knowledge point in the recommended course is added by 1, and the base-two logarithm is taken, and the result is added by 1 as the course coverage factor; the mastery change ratio is multiplied by the course coverage factor to obtain the mastery feedback response value.

[0055] Among them, the specific calculation formula for the feedback response value is:

[0056] ;

[0057] Where, Indicates the mastery feedback response value, Indicates the evaluation value of knowledge points after recommendation. Indicates the evaluation value of knowledge points before recommendation. Indicates the full score data of the course. Represents course test score data, Indicates the full score data of the course test. Indicates the number of recommended courses that cover this knowledge point.

[0058] In this embodiment, Table 2 is a mastery feedback response value data table, which records the evaluation data of 5 knowledge points. Each knowledge point corresponds to a series of calculation results, including the knowledge point mastery evaluation value before and after the recommended course and the various weight values ​​related to the course. The specific data are as follows: Knowledge point A: The knowledge point mastery evaluation value before recommendation is 0.42, the knowledge point mastery evaluation value after recommendation is 0.68, the course score full score data is 5.00, the course test score data is 4.20, the course test full score data is 5.00, the number of recommended courses is 2, and the final calculated mastery feedback response value is 2.04. Knowledge point Knowledge point B: The knowledge point mastery evaluation value before recommendation is 0.51, the knowledge point mastery evaluation value after recommendation is 0.55, the course score full score data is 5.00, the course test score data is 4.00, the course test full score data is 5.00, the number of recommended courses is 1, and the final calculated mastery feedback response value is 1.58. Knowledge Point C: The pre-recommendation mastery assessment value was 0.63, and the post-recommendation mastery assessment value was 0.76. The course score was 5.00, the course test score was 4.40, the course test score was 5.00, and the number of recommended courses was 3. The final calculated mastery feedback response value was 2.68. Knowledge Point D: The pre-recommendation mastery assessment value was 0.58, and the post-recommendation mastery assessment value was 0.60. The course score was 5.00, the course test score was 4.30, the course test score was 5.00, and the number of recommended courses was 2. The final calculated mastery feedback response value was 1.95. Knowledge Point E: The pre-recommendation mastery assessment value was 0.49, and the post-recommendation mastery assessment value was 0.57. The course score was 5.00, the course test score was 4.10, the course test score was 5.00, and the number of recommended courses was 1. The final calculated mastery feedback response value was 1.61.

[0059] Table 2 Mastering feedback response value data table

[0060]

[0061] like Figure 4 As shown in Table 2, it is a line graph of the mastery evaluation value and the mastery feedback response value before and after the recommended course. Figure 4The following chart illustrates the changing trends in the mastery assessment values ​​and corresponding mastery feedback response values ​​for the five knowledge points before and after the recommended course. For knowledge point C, the mastery assessment value before and after the recommended course showed a significant change, with the highest mastery feedback response value at 2.68, indicating that the recommended course triggered a significant mastery shift. In contrast, knowledge point B showed the smallest change in mastery assessment value before and after the recommendation, and a relatively low mastery feedback response value of 1.58, indicating a relatively stable recommendation effect. Furthermore, knowledge points A and D also exhibited moderate response values, while knowledge point E showed a smaller change and a lower mastery feedback response value. Overall, the data in the figure clearly demonstrates the differences in the magnitude of mastery change and feedback intensity across different knowledge points under the intervention of recommended content.

[0062] This implementation enhances the diagnostic analysis capabilities of recommended content by establishing a correlation between knowledge point mastery assessment values ​​before and after recommendation, quantifying the difference in responses between recommended courses and mastery outcomes. The introduction of a mastery change ratio and course coverage factor reflects the actual improvement in knowledge point mastery and the impact of course distribution on learning outcomes from a proportional perspective, ensuring more accurate and discriminative mastery feedback response values. This provides a quantifiable basis for subsequent feedback corrections and recommendation strategy adjustments, enhancing the system's ability to track the user's learning process and the rationality of interventions.

[0063] Specifically, the specific steps for updating recommendation tags and user records and improving the recommendation data cycle are as follows: after calculating the mastery feedback response value, the mastery feedback response value is compared with the mastery feedback threshold in real time: wherein the mastery feedback threshold includes the first-level mastery feedback threshold and the second-level mastery feedback threshold; when the mastery feedback response value is greater than or equal to the second-level mastery feedback threshold, the feedback is normal, and the knowledge points covered in the current course are marked as recommendation-effective tags, thereby improving the recommendation priority of the current course in the subsequent candidate course ranking among similar users; when the mastery feedback response value is greater than the first-level mastery feedback threshold and less than the second-level mastery feedback threshold, the feedback is delayed, and the current course is marked as an intermediate transition course and included in the next recommendation cycle as a continuous observation object for delayed supplementary content; when the mastery feedback response value is less than or equal to the first-level mastery feedback threshold, the feedback is abnormal, and the knowledge points covered in the current course are marked as recommendation offset tags, marking the current course as a recommendation-insensitive course, and giving priority to avoiding courses with similar teaching structures in subsequent supplementary course screening; According to the user dimension, the mastery feedback response values ​​generated by the same knowledge point in different recommendation cycles are arranged continuously to construct a mastery feedback response value sequence; for knowledge points with feedback response values ​​less than the first-level mastery feedback threshold three or more times, the course code and course rating data corresponding to the knowledge point are extracted, and the user's knowledge point mastery evaluation value for this course is recalculated based on the original course browsing time, video viewing completeness and course rating data, replacing the original result for feedback correction; the course code, knowledge point label, course browsing time, course rating data, video viewing completeness, mastery feedback response value, course test score data and evaluation value change results generated by each round of recommended course learning are written into the user recommendation record; in the new round of recommendation generation process, user recommendation records are read first, the initial course recommendation ranking, popularity differentiation strategy and recommendation threshold calculation logic are dynamically adjusted, the binding strength between candidate courses and key knowledge points is updated, and the active adaptive adjustment of the recommendation path is completed, thereby realizing a recommendation self-closed loop mechanism based on ability feedback.

[0064] In this implementation plan, by introducing a multi-level comparison mechanism of mastery feedback response values ​​and mastery feedback thresholds, refined feedback classification and label updates for course recommendation results are achieved, which can dynamically identify recommendation effectiveness, lags, and offsets, and clearly mark the course recommendation status with recommendation effective labels, intermediate transition courses, and recommendation offset labels. At the same time, combined with the feedback sequence construction and periodic fluctuation monitoring in the user dimension, the system's recognition depth of changes in knowledge point mastery is improved, and the iterative correction of the original knowledge point mastery evaluation values ​​is promoted. By continuously writing course codes, knowledge point labels, course browsing time, course rating data, video viewing completeness, mastery feedback response values, course test score data, and evaluation value change results, the system effectively accumulates recommendation record data, and dynamically updates the course sorting strategy and recommendation threshold in the next round of recommendation generation, strengthening the adaptability between candidate courses and key knowledge points, and ultimately achieving adaptive adjustment of recommendation paths and data closed-loop optimization driven by mastery feedback response values.

[0065] like Figure 2 As shown, the second aspect of the present invention provides a personalized course recommendation system based on knowledge graph, including: a user learning data collection and preprocessing module, a knowledge point mastery evaluation module, a supplementary course recommendation generation module and a recommendation effect feedback correction module, wherein: the user learning data collection and preprocessing module is used to obtain user learning data by real-time collection of front-end buried points and back-end access log data of the learning platform, and preprocess the user learning data to obtain preprocessed user learning data; the knowledge point mastery evaluation module is used to construct a behavioral data relationship between users and knowledge points, normalize and split the preprocessed user learning data, evaluate the user's mastery of knowledge points, and construct a time series knowledge point rating set; the supplementary course recommendation generation module is used to identify the abnormal mastery of knowledge points at the user's current stage based on the time series knowledge point rating set, conduct differential analysis in combination with the behavioral data relationship, evaluate the recommendation value of candidate courses, partition the candidate courses and generate recommended courses; the recommendation effect feedback correction module is used to quantify the response difference between recommended courses and mastery results from the proportion dimension, update recommendation tags and user records, and improve the recommendation data cycle.

[0066] This implementation method systematically integrates a user learning data collection and preprocessing module, a knowledge point mastery assessment module, a supplementary course recommendation generation module, and a recommendation effect feedback correction module to achieve structured cleaning of user learning data, refined assessment of knowledge point mastery, differentiated recommendation of supplementary courses, and dynamic feedback correction of recommendation effects. The system can continuously track the time series changes in knowledge point mastery status, combine behavioral data to conduct recommendation value analysis and popularity zoning of candidate courses, and realize intelligent generation of course recommendations. At the same time, it introduces a mastery feedback response mechanism to evaluate recommendation effectiveness from a proportional dimension and update recommendation tags and user records, forming a closed-loop process of data collection, analysis, recommendation, feedback, and re-optimization, significantly improving the match between recommended content and users' actual mastery needs.

[0067] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0068] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A personalized course recommendation method based on knowledge graph, characterized by: The following steps are involved: S1, acquires user learning data by real-time collection of front-end tracking data and back-end access log data of the learning platform, and pre-processes the user learning data to obtain pre-processed user learning data; S2, builds the behavioral data relationship between users and knowledge points, normalizes and splits the pre-processed user learning data, evaluates the user's mastery of knowledge points, and constructs a time series knowledge point rating set; The specific steps of constructing the behavioral data relationship between users and knowledge points, normalizing and splitting the pre-processed user learning data, and evaluating the user's mastery of knowledge points are as follows: Extract the corresponding knowledge point set based on the knowledge point tags marked in each course; combine the course codes in the user's learning records, search for the knowledge point tags included in the user's courses one by one, and merge them at the user dimension to obtain the complete knowledge point set learned by the user; With user names and knowledge point labels as corresponding items, the association records between users and knowledge points are sorted out to form the behavioral correspondence between users and knowledge points during the learning process; Based on the established behavioral correspondence between users and knowledge points, for each knowledge point, by integrating the pre-processed user learning data such as course browsing time, video viewing completeness and course rating data generated by users in courses containing the same knowledge point, the user's mastery of this knowledge point is evaluated: for a course containing this knowledge point, calculate the ratio between the course browsing time and the historical average browsing time of the course, multiply this ratio by the browsing time weight, and the resulting product is the browsing time score; multiply the video viewing completeness by the video completeness weight, and the resulting product is the video completeness score; divide the course rating data by the full score of the course rating, and then multiply the ratio by the course rating weight, and the resulting product is the course rating data score; add the browsing time score, video completeness score and course rating data score, and multiply it by the inverse of the number of knowledge points involved in the course to obtain the user's apportioned score for this knowledge point in the course; Recalculate the apportioned scores for all courses that include this knowledge point, and sum up the apportioned scores of each course to obtain the knowledge point mastery assessment value; S3, based on the time series knowledge point score set, identifies the abnormal knowledge points mastered by the user at the current stage, conducts differentiation analysis based on the behavioral data relationship, evaluates the recommendation value of candidate courses, partitions the candidate courses and generates recommended courses; S4, quantify the response difference between recommended courses and mastery outcomes from the proportional dimension, update recommendation tags and user records, and improve the recommendation data cycle.

2. The method for personalized course recommendation based on knowledge graph according to claim 1, characterized in that: The specific steps for obtaining user learning data by real-time collection of front-end tracking data and back-end access log data of the learning platform are as follows: By collecting the front-end buried points and back-end access log data of the learning platform in real time, the user learning data is obtained. The user learning data includes course code, course browsing time, number of course visits, average historical browsing time of the course, video viewing completeness, course rating data, course rating full score data, course test score data, course test full score data and the number of knowledge points involved in the course; among them, the course code is extracted through the course name field in the course access log; the course browsing time is obtained by recording the difference between the start time and exit time of each user entering the course page and summing them up by course; the number of course visits is obtained by counting the total number of times users visit a certain course page; The total browsing time of the same course by users in a week is divided by the number of course visits to obtain the average browsing time of the course history; the video viewing completeness is obtained by calculating the ratio of the actual playback time recorded by the video player to the total duration of the course video; the course rating data and the course rating full score data are obtained by extracting the rating field submitted by the user after completing the course; the course test score data and the course test full score data are obtained by recording the rating results of the embedded quiz taken by the user after completing the course; the mapping relationship between courses and knowledge points in the course resource management database is analyzed, and the number of knowledge point tags covered by each course is counted to obtain the number of knowledge points involved in the course.

3. The method for personalized course recommendation based on knowledge graph according to claim 1, characterized in that: The specific steps of preprocessing the user learning data to obtain the preprocessed user learning data are as follows: By combining the median absolute deviation method of the sliding time window with the isolation forest algorithm, local and global anomaly detection is performed on user learning data to eliminate abnormal user learning data, including abnormally long idle time and skipping classes to make progress; through the similar user collaborative completion strategy and the exponentially weighted moving average algorithm, data is filled in for missing fields in user learning data caused by system delays, lost tracking points or terminal fluctuations; user learning data is fitted and denoised through the Bayesian dynamic smoothing method and the local weighted regression filtering algorithm; user learning data is transformed and range compressed through exponential transformation combined with the maximum and minimum normalization method to achieve normalization processing.

4. The method for personalized course recommendation based on knowledge graph according to claim 1, characterized in that: The specific steps of constructing a time series knowledge point scoring set are as follows: After calculating the knowledge point mastery assessment value, the knowledge point mastery assessment value and the mastery threshold are compared in real time: when the knowledge point mastery assessment value is greater than or equal to the mastery threshold, it is marked as a normal mastery of the knowledge point and no adjustment is required; when the knowledge point mastery assessment value is less than the mastery threshold, it is marked as an abnormal mastery of the knowledge point, and the user's current learning path is prompted, and a recommendation instruction is generated to add supplementary courses related to this knowledge point; The compared knowledge point mastery evaluation values ​​are processed in time series: the knowledge point mastery evaluation values ​​generated by the user in each course learning behavior are recorded in chronological order, and the corresponding timestamps, course codes and knowledge point tags are marked; all knowledge point mastery evaluation values ​​of the same user on the same knowledge point are sorted in chronological order to construct the user's knowledge point mastery evaluation value sequence on this knowledge point; the user's knowledge point mastery evaluation value sequence on all knowledge points is aggregated and organized to form a time series knowledge point mastery evaluation value set of the user in the entire knowledge structure.

5. The method for personalized course recommendation based on knowledge graph according to claim 1, characterized in that: The specific steps of identifying the abnormal knowledge points mastered by the user at the current stage based on the time series knowledge point scoring set, conducting differentiation analysis based on the behavioral data relationship, and evaluating the recommendation value of the candidate courses are as follows: Combine the user's time series knowledge point mastery evaluation value set under the entire knowledge structure, extract the set of knowledge points marked as abnormally mastered in the current period, and count the number of knowledge points marked as abnormally mastered; Group abnormal records by knowledge point tags, identify knowledge point tags that are continuously marked as abnormal within a short period of time within the same user, and form a set of key supplementary knowledge points for the user at the current stage; Based on the identified set of key supplementary knowledge points, we review all courses associated with each knowledge point, remove courses that the user has completed or has not fully watched, and include the remaining courses in the candidate course set. For each course in the candidate course set, we calculate the course's historical average rating, extract the number of overlaps between the course's associated knowledge points and the set of supplementary knowledge points, and analyze whether the course is worth recommending. Calculate the square of the number of overlaps between the course-related knowledge points and the set of supplementary knowledge points. Calculate the product of the number of knowledge points involved in the course and the number of knowledge points marked as abnormally mastered. Divide this square by this product, and use the ratio as the knowledge point matching score. Calculate the course browsing time divided by the average browsing time of the course history, and add this ratio to the video viewing completeness to get the learning investment score; Calculate the course rating data plus the historical average course rating, and divide the sum by twice the full score of the course rating data as the course rating data performance score; multiply the knowledge point matching score, learning input score and course rating data performance score to obtain the course recommendation strength assessment value.

6. The method for personalized course recommendation based on knowledge graph according to claim 1, characterized in that: The specific steps of partitioning candidate courses and generating recommended courses are as follows: Based on the course recommendation strength evaluation value, a multi-dimensional course recommendation popularity partition is constructed according to the course recommendation threshold: the course recommendation threshold includes the first-level course recommendation threshold and the second-level course recommendation threshold; when the course recommendation strength evaluation value is greater than or equal to the second-level course recommendation threshold, it is judged as a high-popularity course, directly included in the current recommended courses, and prioritized for push; when the course recommendation strength evaluation value is greater than the first-level course recommendation threshold and less than the second-level course recommendation threshold, it is judged as a medium-popularity course and enters the selection area, and the topic is aggregated through the supplementary knowledge point tags covered by the course, and priority is given to matching according to the user's historical unlearned content. 1 to 2 courses are recommended according to the tag distribution to expand the recommendation coverage; when the course recommendation strength evaluation value is less than or equal to the first-level course recommendation threshold, it is judged as a low-popularity course and is not included in the current recommended courses; Generate recommended courses for the user's current stage based on the multi-dimensional course recommendation popularity partitioning results, and send the recommended courses to the user's learning interface; mark the supplementary knowledge points covered by each course in the recommended course list.

7. The method for personalized course recommendation based on knowledge graph according to claim 1, characterized in that: The specific steps for quantifying the difference in responses between recommended courses and mastery outcomes from a proportional dimension are as follows: After the user completes the recommended course, the course code, course browsing time, and supplementary knowledge point tags involved in the recommended course are extracted to establish the content structure of the recommended course. The course test score data after the recommended course is simultaneously recorded, and the corresponding knowledge point mastery assessment value is updated in the knowledge point dimension. The comparison relationship between the knowledge point mastery assessment value before and after the recommendation is established according to the user and the knowledge point. Based on the comparison between the knowledge point mastery assessment values ​​before and after recommendation, the response difference between the recommended course and the mastery effect is quantified: the knowledge point mastery assessment value of the knowledge point after the recommended course is issued minus the knowledge point mastery assessment value of the knowledge point before the recommended course is issued, and then divided by the full score data of the course rating. Then, this ratio is subtracted from the ratio of the course test score data to the full score data of the course test, and the absolute value of the difference is used as the mastery change ratio; the number of recommended courses involving this knowledge point in the recommended course is added by 1, and the base 2 logarithm is taken. The result is then added with 1 as the course coverage factor; the mastery change ratio is multiplied by the course coverage factor to obtain the mastery feedback response value.

8. The method for personalized course recommendation based on knowledge graph according to claim 1, characterized in that: The specific steps for updating recommendation tags and user records and improving the recommendation data cycle are as follows: After calculating the mastery feedback response value, the mastery feedback response value is compared with the mastery feedback threshold in real time: the mastery feedback threshold includes the first-level mastery feedback threshold and the second-level mastery feedback threshold; when the mastery feedback response value is greater than or equal to the second-level mastery feedback threshold, the feedback is normal, and the knowledge points covered in the current course are marked as recommendation-valid labels, thereby improving the recommendation priority of the current course in the subsequent candidate course ranking among similar users; when the mastery feedback response value is greater than the first-level mastery feedback threshold and less than the second-level mastery feedback threshold, the feedback is delayed, and the current course is marked as an intermediate transition course and included in the next recommendation cycle as a continuous observation object for delayed supplementary content; when the mastery feedback response value is less than or equal to the first-level mastery feedback threshold, the feedback is abnormal, and the knowledge points covered in the current course are marked as recommendation offset labels, marking the current course as a recommendation-insensitive course, and giving priority to avoiding courses with similar teaching structures in subsequent supplementary course screening; By user dimension, the mastery feedback response values ​​generated by the same knowledge point in different recommendation cycles are continuously arranged to construct a mastery feedback response value sequence. For knowledge points with feedback response values ​​less than the first-level mastery feedback threshold three or more times, the corresponding course code and course rating data are extracted. Based on the original course browsing time, video viewing completeness, and course rating data, the user's knowledge point mastery assessment value for this course is recalculated and replaced with the original result for feedback correction. The course codes, knowledge point tags, course browsing time, course rating data, video viewing completeness, mastery feedback response values, course test score data and evaluation value change results generated in each round of recommended course learning are written into the user recommendation record; in the process of generating a new round of recommendations, user recommendation records are read first, and the initial course recommendation ranking, popularity differentiation strategy and recommendation threshold calculation logic are dynamically adjusted, the binding strength between candidate courses and key knowledge points is updated, and the active adaptive adjustment of the recommendation path is completed, thereby realizing a recommendation self-closed loop mechanism based on ability feedback.

9. A personalized course recommendation system based on knowledge graph, characterized by: include: User learning data collection and preprocessing module, knowledge point mastery assessment module, supplementary course recommendation generation module and recommendation effect feedback correction module, among which: The user learning data collection and preprocessing module is used to obtain user learning data by real-time collection of front-end embedded points and back-end access log data of the learning platform, and preprocess the user learning data to obtain preprocessed user learning data; The knowledge point mastery evaluation module is used to build a behavioral data relationship between users and knowledge points, evaluate users' mastery of knowledge points based on preprocessed user learning data, and construct a time series knowledge point score set; The specific steps of constructing the behavioral data relationship between users and knowledge points, normalizing and splitting the pre-processed user learning data, and evaluating the user's mastery of knowledge points are as follows: Based on the knowledge point labels marked in each course, the corresponding knowledge point set is extracted; combined with the course code in the user's learning record, the knowledge point labels included in the user's courses are searched one by one, and merged at the user dimension to obtain the complete knowledge point set learned by the user; with the user name and knowledge point label as the corresponding item, the association record between the user and the knowledge point is sorted to form the behavioral correspondence between the user and the knowledge point during the learning process; Based on the established behavioral correspondence between users and knowledge points, for each knowledge point, by integrating the pre-processed user learning data such as course browsing time, video viewing completeness and course rating data generated by users in courses containing the same knowledge point, the user's mastery of this knowledge point is evaluated: for a course containing this knowledge point, calculate the ratio between the course browsing time and the historical average browsing time of the course, multiply this ratio by the browsing time weight, and the product obtained is the browsing time score; multiply the video viewing completeness by the video completeness weight, and the product obtained is the video completeness score; divide the course rating data by the full score of the course rating data, and then multiply the ratio by the course rating weight, and the product obtained is the course rating data score; add the browsing time score, video completeness score and course rating data score, and multiply it by the inverse of the number of knowledge points involved in the course to obtain the user's apportioned score for this knowledge point in the course; recalculate the apportioned scores for all courses containing this knowledge point, and sum up the apportioned scores of each course to obtain the knowledge point mastery evaluation value; The supplementary course recommendation generation module is used to identify the abnormal knowledge points mastered by the user at the current stage based on the time series knowledge point rating set, analyze whether the candidate courses are worth recommending, partition the candidate courses and generate recommended courses; The recommendation effect feedback correction module is used to quantify the response difference between recommended courses and mastering results, update recommendation tags and user records, and improve the recommendation data cycle.

Images