CN120407951B

CN120407951B - A personalized course recommendation method and system based on knowledge graph

Info

Publication number: CN120407951B
Application number: CN202510919033.4A
Authority: CN
Inventors: 贾勇哲; 王鑫; 徐大为
Original assignee: Tianda Zhitu Tianjin Technology Co ltd
Current assignee: Tianda Zhitu Tianjin Technology Co ltd
Priority date: 2025-07-04
Filing date: 2025-07-04
Publication date: 2025-08-26
Anticipated expiration: 2045-07-04
Also published as: CN120407951A

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

Personalized course recommendation method and system based on knowledge graph

Technical Field

The invention relates to the technical field of course recommendation, in particular to a personalized course recommendation method and system based on a knowledge graph.

Background

Along with the wide application of the online education platform, the course recommendation system facing the user plays an important role in improving the learning efficiency and the learning experience. Most of the current recommendation systems mainly rely on behavior data such as click records, browsing frequency, course collection, learning completion rate and the like of users to conduct modeling, and personalized recommendation results are generated through technical means such as collaborative filtering, content matching or knowledge graph reasoning.

For example, the invention with the bulletin number of CN116542731A discloses a knowledge-graph-based exercise planning course recommendation method, and relates to the field of individual recommendation of exercises. A fitness planning course recommendation method based on a knowledge graph comprises the steps of collecting physical quality characteristic conditions and a fitness action course data set of a user, extracting knowledge attributes from motion descriptions of the fitness courses, structuring the knowledge graph, taking a physical quality characteristic u of the user and a course v as input, outputting probability that the physical quality characteristic u of the user can do training course motion v, and adjusting super-parameters of a model.

For example, the invention with publication number CN116501970A provides an online course recommendation method based on a knowledge graph and convolution, a model firstly uses a feature extraction module to extract and convert the user history interaction information and the feature information of the course field into embedded vectors, the vectors are fused into the knowledge graph to generate embedded expressions of users and courses, and finally the information is transmitted to adjacent nodes through a message propagation algorithm to obtain final embedded vectors of the users and the courses for recommendation.

On the other hand, most of the existing recommendation systems use user interests as core driving targets, lack of feedback mechanisms based on capability promotion and learning effects, and easily cause the recommendation systems to fall into behavior-driven recommendation loops.

Therefore, in view of the above problems, there is a need for a personalized course recommendation method and system based on knowledge graph.

Disclosure of Invention

Technical problem to be solved

Aiming at the defects of the prior art, the invention provides a personalized course recommendation method and a personalized course recommendation system based on a knowledge graph, which solve the problem that the prior art is difficult to measure the recommendation cycle that learning effect is easy to fall into behavior driving.

The personalized course recommendation method based on the knowledge graph comprises the following steps of S1, acquiring user learning data through collecting embedded points at the front end of a learning platform and background access log data in real time, preprocessing the user learning data to obtain preprocessed user learning data, S2, constructing a behavior data relation between a user and the knowledge points, carrying out normalization splitting on the preprocessed user learning data, evaluating the mastering condition of the knowledge points by the user, constructing a time sequence knowledge point scoring set, S3, identifying abnormal mastering knowledge points at the current stage of the user based on time sequence knowledge point scoring set, carrying out differentiation analysis by combining the behavior data relation, evaluating the recommendation value of candidate courses, partitioning the candidate courses, generating the recommendation courses, S4, quantifying response difference between the recommendation courses and mastering effects from proportional dimensions, updating recommendation labels and user records, and perfecting recommendation data circulation.

Acquiring user learning data by acquiring the front-end embedded point and the background access log data of the learning platform in real time, wherein the user learning data comprises course codes, course browsing duration, course access times, course historical average browsing duration, video viewing integrity, course scoring data, course scoring full data, course testing full data and course related knowledge points; the method comprises the steps of extracting course codes through course name fields in a course access log, obtaining course browsing time length by recording the difference between the starting time and the exiting time of a user entering a course page each time and summarizing and summing the courses according to the course, obtaining course access times by counting the total times of the user accessing a certain course page, obtaining course historical average browsing time length by dividing the total browsing time length of the user for the same course within one week by the course access times, obtaining video watching integrity by carrying out ratio calculation on the actual playing time length recorded by a video player and the total time length of a course video, obtaining course scoring data and course scoring full score data by extracting scoring fields submitted by the user after the course learning is completed, obtaining course testing score data and course testing full score data by recording scoring results of the user participating in a course embedded test after the course is completed, and obtaining the knowledge point related by analyzing the mapping relation between the courses and knowledge points in a course resource management database.

The method comprises the specific steps of carrying out local and global anomaly detection on user learning data through the combination of a median absolute deviation method of a sliding time window and an isolated forest algorithm, eliminating abnormal user learning data comprising abnormal long-time hanging and class jump brushing progress, carrying out data filling on field deletion problems of the user learning data caused by system delay, buried point loss or terminal fluctuation through a similar user collaborative complementation strategy and an exponential weighted moving average algorithm, carrying out fitting and denoising on the user learning data through a Bayesian dynamic smoothing method and a local weighted regression filtering algorithm, and carrying out transformation and range compression on the user learning data through an exponential transformation combined maximum and minimum normalization method, thereby realizing normalization processing.

Further, a behavioral data relation between a user and knowledge points is built, the preprocessed user learning data is subjected to normalization splitting, the specific steps of evaluating the grasping condition of the user on the knowledge points are as follows, corresponding knowledge point sets are extracted according to knowledge point labels marked in each course, course codes in a user learning record are combined, knowledge point labels contained in the user learned courses are searched one by one and combined in a user dimension to obtain all knowledge point sets for user learning, the user name and the knowledge point labels are used as corresponding items, the associated record between the user and the knowledge points is organized to form a behavioral corresponding relation between the user and the knowledge points in the learning process, the browsing time length, the video watching completeness and the scoring data generated in courses containing the same knowledge points are integrated for each knowledge point, the grasping condition of the user on the knowledge points is evaluated, the browsing time length and the history browsing time length are calculated for a course containing the same, the browsing weight ratio between the user name and the history browsing time length is calculated, the browsing time length is multiplied by the obtained video scoring time length and the obtained video scoring time length is multiplied by the obtained video scoring time length, the obtained video scoring time length is multiplied by the obtained time length product and the obtained time length is multiplied by the score data, the obtained time length is multiplied by the obtained time length of the whole course, and the score is multiplied by the obtained time length of the obtained time length is multiplied by the score data, and recalculating the shared score of all courses containing the knowledge point, and summing the shared scores of all courses to obtain the knowledge point mastering evaluation value.

The method comprises the specific steps of establishing a time sequence knowledge point scoring set, namely, after a knowledge point mastering evaluation value is obtained through calculation, comparing the knowledge point mastering evaluation value with a mastering threshold in real time, marking the knowledge point mastering evaluation value as a normal mastering knowledge point without adjustment when the knowledge point mastering evaluation value is larger than or equal to the mastering threshold, marking the knowledge point mastering evaluation value as an abnormal mastering knowledge point when the knowledge point mastering evaluation value is smaller than the mastering threshold, prompting a current learning path of a user, generating a recommended instruction for adding supplementary courses related to the knowledge point, carrying out time sequence processing on the compared knowledge point mastering evaluation value, namely, recording the knowledge point mastering evaluation value generated by the user in each learning action in time sequence, marking corresponding time stamps, course codes and knowledge point labels, sequencing all knowledge point mastering evaluation values of the same user on the same knowledge point in time sequence, establishing a knowledge point mastering evaluation value sequence of the user on the knowledge point, and carrying out collective arrangement on the knowledge point mastering evaluation value sequence of the user on all knowledge points to form the time sequence knowledge point mastering evaluation value set of the user under the whole knowledge structure.

Further, based on the time sequence knowledge point evaluation set, the abnormal mastering knowledge points of the current stage of the user are identified, and the differentiated analysis is carried out by combining the behavior data relationship; grouping abnormal records according to knowledge point labels, identifying knowledge point labels continuously marked as abnormal in a short time of the same user to form an important supplementary knowledge point set at the current stage of the user, checking back all courses related to all knowledge points on the basis of the identified important supplementary knowledge point set, removing the courses which are completely learned by the user and the courses which are incompletely watched, incorporating the rest courses into a candidate course set, calculating a course history average score for each course in the candidate course set, extracting the superposition quantity of the associated knowledge points and the supplementary knowledge point set, analyzing whether the courses are worth recommending, calculating the square value of the superposition quantity of the associated knowledge points and the supplementary knowledge point set, calculating the product of the knowledge points related to the knowledge points and the quantity marked as abnormal mastery, dividing the square value by the product, taking the ratio as a knowledge point matching score, calculating the course browsing time divided by the history average browsing time, taking the ratio as a learning input score, calculating course score data and taking the score data divided by the sum value and two times of the score data as course data representing score, and taking the point matching score, and multiplying the learning input score and the course score data performance score to obtain a course recommendation strength evaluation value.

The method comprises the specific steps of selecting a candidate course and generating a recommended course, wherein the specific steps are that a multi-dimensional course recommendation heat partition is built on the basis of a course recommendation strength evaluation value, the course recommendation threshold comprises a first-level course recommendation threshold and a second-level course recommendation threshold, when the course recommendation strength evaluation value is larger than or equal to the second-level course recommendation threshold, the high-heat course is judged to be directly included in the current recommended course, the current recommended course is preferentially ordered and pushed, when the course recommendation strength evaluation value is larger than the first-level course recommendation threshold and smaller than the second-level course recommendation threshold, the medium-heat course is judged to enter a to-be-selected area, subject aggregation is conducted through supplementary knowledge point labels covered by the courses, 1-2 courses are recommended according to label distribution in a preferential matching mode according to the historic content of a user, the recommended coverage is expanded, when the course recommendation strength evaluation value is smaller than or equal to the first-level course recommendation threshold, the current-stage recommendation of the user is judged not to be included in the current recommended course, the current-stage recommendation of the user is generated according to the multi-dimensional course recommendation heat partition result, the recommended course is sent to a user learning interface, and the supplementary knowledge point labels covered by each course are marked in a recommendation list.

Further, the specific steps of quantifying response difference between a recommended course and mastering effects from a proportional dimension are as follows, extracting course codes, course browsing time and supplementary knowledge point labels covered by courses related to the recommended course learning after a user finishes the recommended course learning, establishing a content structure of the recommended course learning, synchronously recording course test score data after the recommended course, updating corresponding knowledge point mastering evaluation values in a knowledge point dimension, constructing a knowledge point mastering evaluation value comparison relation before and after recommendation according to the user and the knowledge point, quantifying the response difference between the recommended course and mastering effects according to the knowledge point mastering evaluation value comparison relation before and after recommendation, namely dividing the knowledge point mastering evaluation value of the knowledge point after the recommended course is issued by the knowledge point mastering evaluation value of the knowledge point before the recommended course is issued by the course by the full score data of the course, subtracting the ratio of the test score data to the full score data of the course, taking the absolute value of the difference as a mastering change ratio, adding 1 to the number of the recommended courses related to the knowledge point, taking the obtained logarithm of two as a base, adding 1 to the obtained coverage factor, and multiplying the coverage factor and the response factor to obtain the response change factor.

Further, the specific steps of updating the recommendation tag and the user record to complete the recommendation data loop are as follows, after the feedback response value is obtained by calculation, the feedback response value is compared with the feedback threshold in real time, wherein the mastering feedback threshold comprises a first-level mastering feedback threshold and a second-level mastering feedback threshold, when the mastering feedback response value is greater than or equal to the second-level mastering feedback threshold, the feedback is normal, the covered knowledge points in the current course are marked as recommendation valid tags, the recommendation priority of the current course in the candidate course ranking of the subsequent similar user is improved, when the mastering feedback response value is greater than the first-level mastering feedback threshold, the feedback is lagged, the current course is marked as an intermediate recommendation course, and is taken as a continuous observation object of the delay supplement content, when the mastering feedback response value is less than or equal to the first-level mastering feedback threshold, the feedback abnormality is marked as recommendation offset tag, the current course is marked as recommendation insensitivity, the recommendation structure is marked in the subsequent supplementary teaching course, the recommendation priority is raised as a recommendation valid tag, the current course is generated in different-like course, the current course is generated in the subsequent similar course, the previous course is set-up time, the feedback response value is read-by the comparison result of the first-level, the first-level feedback course is set up time is read, the comparison course is completed, the comparison result is completed, and the three-level of the comparison course is read, and the comparison course is completed, and the first-level comparison course is based on the comparison result is read by the comparison result and the comparison result In the new recommendation generation process, the user recommendation record is read preferentially, the initial recommendation ordering of the courses, the heat distinguishing 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 adaptation adjustment of a recommendation path is completed, so that a recommendation self-closing mechanism based on capability feedback is realized.

The invention provides a personalized course recommendation system based on a knowledge graph, which comprises a user learning data acquisition preprocessing module, a knowledge point mastering and evaluating module, a supplementary course recommendation generating module and a recommendation effect feedback correcting module, wherein the user learning data acquisition preprocessing module is used for acquiring user learning data through real-time acquisition of embedded point at the front end of a learning platform and background access log data and preprocessing the user learning data to obtain preprocessed user learning data, the knowledge point mastering and evaluating module is used for constructing a behavior data relation between a user and the knowledge point, carrying out normalization splitting on the preprocessed user learning data, evaluating the mastering condition of the user on the knowledge point, constructing a time sequence knowledge point scoring set, and the supplementary course recommendation generating module is used for carrying out differentiation analysis by combining the behavior data relation, evaluating the recommendation value of candidate courses, carrying out partition on the candidate courses and generating recommendation courses, and the recommendation effect feedback correcting module is used for quantifying the response difference between the recommendation courses and the success effect from a proportion dimension, updating the recommendation label and the user record, and perfecting the circulation of the recommendation data.

Advantageous effects

The invention has the following beneficial effects:

(1) According to the personalized course recommendation method and system based on the knowledge graph, through integrating the user learning data such as the course browsing time length, the video watching completeness and the course scoring data, an accurate knowledge point mastering and evaluating method is provided, so that evaluation of knowledge point mastering states in a user learning process is more refined and accurate, and the limitation that the traditional system only depends on click rate and completion rate to measure learning results is broken through.

(2) According to the personalized course recommendation method and system based on the knowledge graph, the difference of the evaluation value change and the course test score data is mastered by quantifying the knowledge points before and after recommendation, the feedback of the learning effect is dynamically adjusted by combining the number of recommended courses and the behavioral response of the user, the real influence of the learning behavior on the capacity improvement is reflected, and a real-time feedback basis is provided for optimizing the recommended content.

(3) According to the personalized course recommendation method and system based on the knowledge graph, based on course recommendation intensity evaluation, the partition recommendation strategies of high, medium and low heat are designed, the actual grasping feedback of the recommendation content is combined, the differential pushing strategy is adopted for courses with different intensities, accurate recommendation and exploratory recommendation are balanced, and the adaptability and coverage of the system are improved.

(4) According to the personalized course recommendation method and system based on the knowledge graph, the time sequence mastering evaluation record between the user and the knowledge points is constructed, so that the dynamic tracking and the staged recognition of the knowledge mastering state are realized, the abnormal mastering recognition and the supplementary content generation mechanism are constructed based on the dimension of the knowledge points, and the timeliness and the precision of recommendation decisions are effectively improved.

Of course, it is not necessary for any one product to practice the invention to achieve all of the advantages set forth above at the same time.

Drawings

FIG. 1 is a flow chart of a personalized course recommendation method based on knowledge graph;

FIG. 2 is a block diagram of a personalized course recommendation system based on knowledge graph;

FIG. 3 is a histogram of knowledge point mastery assessment values;

Fig. 4 is a line graph of grasping evaluation values and grasping feedback response values before and after a recommended course.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

Referring to fig. 1-4, an embodiment of the invention provides a personalized course recommendation method based on a knowledge graph, which comprises the following steps of S1, acquiring user learning data through collecting embedded points at the front end of a learning platform and background access log data in real time, preprocessing the user learning data to obtain preprocessed user learning data, S2, constructing a behavior data relation between a user and the knowledge points, carrying out normalization splitting on the preprocessed user learning data, evaluating the mastering condition of the knowledge points by the user, constructing a time sequence knowledge point scoring set, S3, identifying abnormal mastering knowledge points at the current stage of the user based on time sequence knowledge score diversity, carrying out differentiation analysis by combining the behavior data relation, evaluating the recommendation value of candidate courses, partitioning the candidate courses, generating recommended courses, S4, quantifying the response difference between the recommended courses and mastering effects from a proportional dimension, updating the recommended label and the user record, and perfecting the recommended data circulation.

The method comprises the specific steps of acquiring user learning data by acquiring front-end embedded points and background access log data of a learning platform in real time, wherein the user learning data comprises course codes, course browsing duration, course access times, course historical average browsing duration, video viewing integrity, course scoring data, course scoring full data, course testing full data, course related knowledge points; the method comprises the steps of extracting course codes through course name fields in a course access log, obtaining course browsing time length by recording the difference between the starting time and the exiting time of a user entering a course page each time and summarizing and summing the courses according to the course, obtaining course access times by counting the total times of the user accessing a certain course page, obtaining course historical average browsing time length by dividing the total browsing time length of the user for the same course within one week by the course access times, obtaining video watching integrity by carrying out ratio calculation on the actual playing time length recorded by a video player and the total time length of a course video, obtaining course scoring data and course scoring full score data by extracting scoring fields submitted by the user after the course learning is completed, obtaining course testing score data and course testing full score data by recording scoring results of the user participating in a course embedded test after the course is completed, and obtaining the knowledge point related by analyzing the mapping relation between the courses and knowledge points in a course resource management database.

In the embodiment, a multidimensional data base covering the whole process of user behavior is constructed by accurately defining and collecting course codes, course browsing time periods, course access times, course historical average browsing time periods, video viewing completeness, course scoring data, course scoring full score data, course test scoring data, course test full score data and course related knowledge points. The method effectively fuses the front-end embedded point behavior data and the background access log information, realizes high-precision mapping of the user learning behavior and the learning content on the basis of not changing the consistency of the data names, provides accurate, stable and clear-structured data support for grasping and evaluating the follow-up knowledge points, recommending courses and optimizing feedback, and breaks through the bottleneck that the traditional recommendation system does not adequately describe the learning state of the user.

The method comprises the specific steps of preprocessing user learning data to obtain preprocessed user learning data, wherein the method comprises the steps of combining a median absolute deviation method of a sliding time window with an isolated forest algorithm, carrying out local and global anomaly detection on course browsing time, course access times and course scoring data in the user learning data, mainly removing abnormal user learning data which comprises abnormal long-time hanging and course skip progress and violates a normal learning path rule, carrying out intelligent complementation on the course scoring data, course testing scoring data and field deletion problems of video watching integrity caused by system delay, buried point loss and terminal fluctuation through a similar user collaborative complementation strategy and an exponential weighted moving average algorithm, guaranteeing the integrity and continuity of the user learning data, carrying out fitting and denoising on the user learning data through a Bayesian dynamic smoothing method and a local weighted regression filtering algorithm, improving the data stability, carrying out transformation and range compression on the user learning data through an exponential transformation combined maximum and minimum normalization method, and providing a normalization processing data base for consistency of structural specifications for subsequent analysis.

In the embodiment, by systematically preprocessing course browsing duration, course access times, course scoring data, course test scoring data and video watching integrity in the user learning data, abnormal user learning data can be effectively identified and removed, field missing problems are repaired, data noise is eliminated, variable scales are unified, the integrity, stability and comparability of the preprocessed user learning data are remarkably improved, a high-quality and structured data basis is provided for subsequently constructing behavior data relations between users and knowledge points, evaluating knowledge point mastering conditions and generating recommended courses, and the operation accuracy and robustness of a recommendation system are ensured.

The method comprises the specific steps of constructing a behavior data relation between a user and knowledge points, carrying out normalization splitting on preprocessed user learning data, evaluating the mastering condition of the user on the knowledge points, extracting a corresponding knowledge point set according to knowledge point labels marked in each course, combining course codes in a user learning record, searching knowledge point labels contained in the user learned courses one by one, merging in a user dimension to obtain all knowledge point sets of user learning, taking a user name and the knowledge point labels as corresponding items, sorting an associated record between the user and the knowledge points to form a behavior corresponding relation between the user and the knowledge points in the learning process, taking the product of the browsing time length and the video score as the product of the obtained video score, multiplying the product of the obtained product of the browsing time length and the video score as the product of the obtained video score by the obtained video score, multiplying the product of the obtained product of the video score as the obtained product of the video score for each knowledge point by the obtained video score, and multiplying the obtained product of the video score as the obtained product of the video score for each knowledge point by the obtained video score, and recalculating the shared score of all courses containing the knowledge point, and summing the shared scores of all courses to obtain the knowledge point mastering evaluation value. The method comprises the steps of taking course browsing time length, video watching completeness and course scoring data collected in historical user learning data as independent variables, taking a periodic evaluation score as a dependent variable, carrying out parameter fitting by adopting a least square regression algorithm, and obtaining browsing time length weight, video completeness weight and course scoring weight according to interpretation strength of each behavior characteristic to capability grasping effect, wherein the value ranges of the browsing time length weight, the video completeness weight and the course scoring weight are all [0,1].

The specific formula of the knowledge point mastering evaluation value is as follows:

;

In the formula, Indicating that the knowledge point grasps the evaluation value,Representing a course containing this knowledge point,Representing a set of courses containing this knowledge point,Representing the weight of the browsing duration,The duration of course browsing is indicated,Representing a historic average browsing duration of the course,The video integrity weights are represented as such,Indicating the integrity of the video viewing,The course scoring weight is represented as a function of the course scoring weight,The course scoring data is represented as a function of the course scoring data,Representing full-scale data of course scores,Representing courses involves knowledge points.

In this embodiment, table 1 is a knowledge point grasping evaluation value data table, and details of course sets, course scoring full score data, course browsing duration weights, video viewing integrity weights, course scoring weights and finally calculated knowledge point grasping evaluation values of different knowledge points in the evaluation process are recorded, so as to quantify learning grasping conditions of users on various military knowledge points. The method comprises the steps of setting a lesson set corresponding to a knowledge point C1 as B1, setting lesson scoring full-scale data as 10, setting lesson browsing time length weight as 0.3, setting video watching completeness weight as 0.4, setting lesson scoring weight as 0.3, setting knowledge point mastering evaluation value as 0.72, setting a lesson set corresponding to a knowledge point C2 as B2, setting lesson scoring full-scale data as 10, setting lesson browsing time length weight as 0.3, setting knowledge point mastering evaluation value as 0.65, setting lesson set corresponding to a knowledge point C3 as B3, setting lesson scoring full-scale data as 10, setting lesson browsing time length weight as 0.3, setting video watching completeness weight as 0.4, setting lesson scoring weight as 0.3, setting lesson scoring time length weight as 0.78, setting lesson scoring time length corresponding to 0.4, setting lesson scoring weight as 0.4, setting lesson scoring time length weight as 0.5, setting lesson scoring time length as 0.3, setting lesson scoring full-scale data as 0.59, setting lesson scoring time length as 0.5, setting lesson scoring full-scale data as 0.3.

TABLE 1 knowledge point mastering evaluation value data sheet

As shown in fig. 3, which is a visual bar chart of knowledge point grasp evaluation values, it can be seen in combination with table 2 that there is a significant difference in grasp conditions of users at different knowledge points. The highest knowledge point mastering evaluation value of the knowledge point C2 is 0.84, which indicates that the learning effect of the user in the course set corresponding to the knowledge point is optimal, the lowest mastering evaluation value of the knowledge point C3 is only 0.59, which indicates that the course learning behavior under the knowledge point is weak, the mastering evaluation values of the knowledge point C4 and the knowledge point C1 are higher, the learning state is stable, the evaluation value of the knowledge point C5 is 0.65, and the learning level is in a medium lower level. In the whole, the knowledge point mastering evaluation value visualization histogram intuitively reflects the difference of mastering degrees of different knowledge points and can be used as a basis for supplementary course recommendation and capability feedback adjustment.

In the embodiment, through constructing the behavior data relation between the user and the knowledge points, based on course browsing time length, video watching completeness and course scoring data in the preprocessed user learning data, course history average browsing time length, course scoring full score data and course related knowledge points are comprehensively introduced to form a multidimensional scoring system based on browsing time length score, video completeness score and course scoring data score, so that the fine modeling and quantitative evaluation of learning grasping conditions of the user on each knowledge point are realized, the discrimination capability and dynamic expression capability of grasping evaluation values of the knowledge points are effectively enhanced, and a high-reliability evaluation basis is laid for follow-up accurate recommendation and feedback regulation.

The method comprises the specific steps of calculating and obtaining a knowledge point mastering evaluation value, comparing the knowledge point mastering evaluation value with a mastering threshold in real time, marking as a normal mastering knowledge point without adjustment when the knowledge point mastering evaluation value is larger than or equal to the mastering threshold, marking as an abnormal mastering knowledge point when the knowledge point mastering evaluation value is smaller than the mastering threshold, prompting a current learning path of a user, generating a recommended instruction for adding supplementary courses related to the knowledge point, carrying out time serialization processing on the compared knowledge point mastering evaluation value, namely recording the knowledge point mastering evaluation value generated by the user in each learning action in time sequence, marking corresponding time stamps, course codes and knowledge point labels, sorting all knowledge point mastering evaluation values of the same user on the same knowledge point in time sequence, constructing a knowledge point mastering evaluation value sequence of the user on the knowledge point, and carrying out collective arrangement on the knowledge point mastering evaluation value sequence of the user on all knowledge points to form a time sequence knowledge point mastering evaluation value set of the user under the whole knowledge structure.

In the embodiment, through the real-time comparison of the knowledge point mastering evaluation value and the mastering threshold value, the abnormal mastering knowledge point is timely identified, and the supplementary course recommendation instruction is dynamically generated by combining the learning path, the active intervention of the learning blind area is realized, meanwhile, through the labeling and sequencing of the knowledge point mastering evaluation value according to the time sequence, the time sequence knowledge point mastering evaluation value set covering all knowledge points is constructed, the tracking and expression capability of the user knowledge mastering evolution process is effectively improved, and the high-resolution and time sequence data support is provided for the follow-up stage performance evaluation and personalized recommendation strategy.

Specifically, based on time sequence knowledge point evaluation set, identifying abnormal mastering knowledge points of a current stage of a user, carrying out differential analysis in combination with a behavior data relationship, and evaluating recommendation values of candidate courses, wherein the specific steps are that a time sequence knowledge point mastering evaluation value set of the user under the whole knowledge structure is combined, a set marked as the abnormal mastering knowledge points in the current period is extracted, and the number of the knowledge points marked as the abnormal mastering is counted; grouping abnormal records according to knowledge point labels, identifying knowledge point labels continuously marked as abnormal in a short time of the same user to form an important supplementary knowledge point set at the current stage of the user, checking back all courses related to all knowledge points on the basis of the identified important supplementary knowledge point set, removing the courses which are completely learned by the user and the courses which are incompletely watched, incorporating the rest courses into a candidate course set, calculating a course history average score for each course in the candidate course set, extracting the superposition quantity of the associated knowledge points and the supplementary knowledge point set, analyzing whether the courses are worth recommending, calculating the square value of the superposition quantity of the associated knowledge points and the supplementary knowledge point set, calculating the product of the knowledge points related to the knowledge points and the quantity marked as abnormal mastery, dividing the square value by the product, taking the ratio as a knowledge point matching score, calculating the course browsing time divided by the history average browsing time, taking the ratio as a learning input score, calculating course score data and taking the score data divided by the sum value and two times of the score data as course data representing score, and taking the point matching score, and multiplying the learning input score and the course score data performance score to obtain a course recommendation strength evaluation value.

The specific calculation formula of the course recommendation intensity evaluation value is as follows:

;

In the formula, An evaluation value of the intensity of course recommendation is represented,Representing the number of coincidences of course-associated knowledge points with the supplemental knowledge point set,The presentation of the course involves knowledge points,The number of knowledge points marked as abnormally mastered is represented,The duration of course browsing is indicated,Representing a historic average browsing duration of the course,Indicating the integrity of the video viewing,The course scoring data is represented as a function of the course scoring data,Representing a course history average score that is representative of the course,Representing course scoring full score data.

In the embodiment, a time sequence knowledge point mastering evaluation value set is constructed, abnormal mastering knowledge points in the current stage are accurately identified, fine-granularity differential analysis is carried out by combining user behavior data relations to form an important supplementary knowledge point set, the superposition number of course-related knowledge points and the supplementary knowledge point set, course browsing duration, video watching integrity and course scoring data are introduced on the basis, and comprehensive indexes including knowledge point matching scores, learning input scores and course scoring data expression scores are constructed to quantify course recommendation strength evaluation values of candidate courses, so that pertinence and effectiveness of recommended course screening are enhanced.

Based on the course recommendation intensity evaluation value, constructing a course recommendation intensity evaluation index system by analyzing the matching degree of course associated knowledge points and a supplementary knowledge point set, course browsing behavior investment and course scoring data expression, and constructing a multi-dimensional course recommendation heat partition according to a course recommendation threshold; the course recommendation threshold comprises a first course recommendation threshold and a second course recommendation threshold, which are dynamically adjusted according to the fluctuation trend of a course feedback record and knowledge point mastering evaluation value in a history recommendation result, when the course recommendation strength evaluation value is larger than or equal to the second course recommendation threshold, the current recommendation course is judged to be a high-heat course, and the current recommendation course is preferentially ordered and pushed, meanwhile, the follow-up monitoring flow of mastering a feedback response value is triggered in a recommendation strategy module, when the course recommendation strength evaluation value is larger than the first course recommendation threshold and smaller than the second course recommendation threshold, the medium-heat course is judged to enter a to-be-selected area, subject aggregation is carried out through supplementary knowledge point labels covered by courses, and priority matching is carried out by combining with knowledge point labels which are not covered in a user learning record, 1 to 2 courses are recommended according to label distribution so as to expand recommendation coverage and maintain recommendation diversity, when the course recommendation strength evaluation value is smaller than or equal to the first course recommendation threshold, the current recommendation course is not entered, but the current recommendation course is recorded as a retrieval candidate resource when the subsequent feedback is abnormal, the medium-heat course is generated according to the multi-dimensional recommendation strength evaluation value is partitioned, the current course is generated, the supplementary knowledge point labels are covered in a recommendation interface, and identifying whether the course is a non-learned course in the user history learning path so as to assist the user in course selection and path judgment.

In the embodiment, the multi-dimensional hot partition management of candidate courses is realized by introducing the course recommendation intensity evaluation value and constructing the course recommendation threshold interval, so that the matching degree of the course association knowledge points and the supplementary knowledge point set, the course browsing behavior investment and the course scoring data expression are fully considered, and the adaptability and the expansibility of recommended content are enhanced by combining with the knowledge point labels which are not covered in the user learning record. Meanwhile, courses in different heat intervals are endowed with a differential pushing mechanism and are linked with a follow-up feedback response value grasping monitoring flow, a dynamic adjustment and feedback closed-loop mechanism is constructed, and data support and strategy guarantee are provided for accuracy, controllability and dynamic sustainable optimization of recommended results.

The method comprises the specific steps of quantifying response difference between a recommended course and mastering effects from a proportional dimension, wherein after the user finishes the recommended course study, course coding, course browsing time and supplementary knowledge point labels of course coverage, which are related to the recommended course study, are extracted, a content structure of the recommended course study is built, course test score data after the recommended course is synchronously recorded, corresponding knowledge point mastering evaluation values are updated in the knowledge point dimension, a knowledge point mastering evaluation value comparison relation before and after the recommendation is built according to the user and the knowledge points, the response difference between the recommended course and mastering effects is quantified according to the knowledge point mastering evaluation value comparison relation before and after the recommendation is issued, the knowledge point mastering evaluation value of the knowledge point is subtracted from the knowledge point mastering evaluation value of the knowledge point before the recommendation is issued, the ratio is subtracted from the ratio of the knowledge point mastering evaluation value of the knowledge point of the recommended course to the full data, the ratio of the test score data to the course test full data is subtracted, the absolute value of the difference is used as a mastering change ratio, the number of the recommended course is added with 1, the obtained logarithm of the number of the knowledge point is added to the number of the knowledge point, and the obtained by two is used as a coverage factor, and the response factor is multiplied by the mastering change factor.

The specific calculation formula for grasping the feedback response value is as follows:

;

In the formula, Indicating that the feedback response value is grasped,Indicating that the knowledge points grasp the evaluation value after recommendation,Indicating that the pre-recommendation knowledge points grasp the evaluation values,Representing full-scale data of course scores,The course test score data is represented as such,Representing full-scale data of course testing,And representing the number of recommended courses related to the knowledge point in the recommended courses.

In this embodiment, table 2 records evaluation data of 5 knowledge points in order to grasp the feedback response value data table. Each knowledge point corresponds to a series of calculation results, wherein the calculation results comprise knowledge point mastering evaluation values before and after the recommended course and various weight values related to the course. The knowledge point A is characterized in that the knowledge point A before recommendation is used for mastering an evaluation value of 0.42, the knowledge point A after recommendation is used for mastering an evaluation value of 0.68, the full score data of course score is 5.00, the test score data of course is 4.20, the full score data of course test is 5.00, the number of recommended courses is 2, and the final calculated mastering feedback response value is 2.04. Knowledge point B, the knowledge point mastering evaluation value before recommendation is 0.51, the knowledge point mastering evaluation value after recommendation is 0.55, the full score data of course score is 5.00, the test score data of course is 4.00, the full score data of course test is 5.00, the recommended course number is 1, and the mastering feedback response value obtained through final calculation is 1.58. Knowledge point C, the knowledge point mastering evaluation value before recommendation is 0.63, the knowledge point mastering evaluation value after recommendation is 0.76, the course scoring full score data is 5.00, the course test score data is 4.40, the course test full score data is 5.00, the recommended course number is 3, and the mastering feedback response value obtained through final calculation is 2.68. Knowledge point D, the knowledge point mastering evaluation value before recommendation is 0.58, the knowledge point mastering evaluation value after recommendation is 0.60, the course scoring full-scale data is 5.00, the course test score data is 4.30, the course test full-scale data is 5.00, the recommended course number is 2, and the mastering feedback response value obtained through final calculation is 1.95. Knowledge point E, the knowledge point mastering evaluation value before recommendation is 0.49, the knowledge point mastering evaluation value after recommendation is 0.57, the course scoring full-scale data is 5.00, the course test scoring data is 4.10, the course test full-scale data is 5.00, the recommended course number is 1, and the mastering feedback response value obtained through final calculation is 1.61.

Table 2 grasp feedback response value data sheet

As shown in fig. 4, in order to grasp the evaluation value and grasp the feedback response value line graph before and after the recommended course, it can be seen from the combination of table 2 and fig. 4 that five knowledge points grasp the evaluation value and the corresponding grasp the change trend of the feedback response value before and after the recommended course learning. In the knowledge point C, knowledge point grasp evaluation values before and after the recommended course change greatly, and grasp feedback response values are 2.68 at most, which indicates that the recommended course causes obvious grasp deviation. In contrast, knowledge points B grasp the change amplitude of the evaluation value to be minimum at knowledge points before and after recommendation, grasp the feedback response value to be lower than 1.58, and indicate that the recommendation effect is stable. In addition, knowledge points A and D also exhibit moderate response values, while knowledge point E exhibits less variation and lower mastery feedback response values. Overall, the data in the graph clearly reflect the difference of the grasping variation amplitude and the feedback intensity of different knowledge points under the intervention of recommended content.

In the embodiment, the comparison relation of the evaluation values is mastered by constructing knowledge points before and after recommendation, the response difference between the recommended course and mastering effect is quantified, and the diagnostic analysis capability of the recommended content is improved. The change ratio and course coverage factor are mastered, the actual improvement of knowledge point mastering degree and the influence of course distribution on learning effect can be reflected from the proportion dimension, the generation of mastering feedback response values is ensured to have higher accuracy and discrimination, a quantifiable basis is provided for the feedback correction and recommendation strategy adjustment of follow-up recommendation content, and the tracking capability and intervention rationality of the system to the user learning process are enhanced.

The method comprises the specific steps of updating a recommended label and a user record, perfecting a recommended data cycle, namely, after calculating to obtain a mastering feedback response value, comparing the mastering feedback response value with a mastering feedback threshold in real time, wherein the mastering feedback threshold comprises a primary mastering feedback threshold and a secondary mastering feedback threshold; when the mastering feedback response value is larger than or equal to the second-level mastering feedback threshold value, feedback is normal, the covered knowledge points in the current course are marked as recommendation effective labels, the recommendation priority of the current course in the candidate course sequencing in the follow-up similar user is improved, when the mastering feedback response value is larger than the first-level mastering feedback threshold value and smaller than the second-level mastering feedback threshold value, feedback is lagged, the current course is marked as an intermediate transition course and is incorporated in the next recommendation period to be used as a continuous observation object of delay type supplementary content, when the mastering feedback response value is smaller than or equal to the first-level mastering feedback threshold value, feedback abnormality is marked as recommendation offset labels, the current course is marked as a recommendation insensitive course, courses of similar teaching structures are preferentially avoided in the follow-up supplementary course screening, the mastering feedback response values generated by the same knowledge points in different recommendation periods are continuously arranged according to the dimension of the user, the knowledge points corresponding to the first-level feedback response values exist three times and are smaller than the first-level feedback threshold value, the codes corresponding to the knowledge points are extracted, the original course data corresponding to the knowledge points are used as a continuous observation object of delay type supplementary content, the whole-level assessment data is calculated on the basis of the video browsing course, the whole assessment point is used for the previous course, and the result of the learning score is calculated on each course is replaced by the previous course is calculated, and the result is used for the correct the course, in the new recommendation generation process, the user recommendation record is read preferentially, the initial recommendation ordering of the courses, the heat distinguishing 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 adaptation adjustment of a recommendation path is completed, so that a recommendation self-closing mechanism based on capability feedback is realized.

In the embodiment, a multi-level comparison mechanism for grasping a feedback response value and a feedback threshold value is introduced, so that the fine feedback classification and label updating of course recommendation results are realized, the recommendation effectiveness, hysteresis and offset can be dynamically identified, and the course recommendation state is explicitly marked by using a recommendation effective label, an intermediate transition course and a recommendation offset label. Meanwhile, the feedback sequence construction and the stepwise fluctuation monitoring in the user dimension are combined, so that the recognition depth of the system for grasping the change of the knowledge points is improved, and the iterative correction of grasping the evaluation value of the original knowledge points is promoted. By continuously writing course codes, knowledge point labels, course browsing time length, course scoring data, video watching integrity, grasping feedback response values, course test scoring data and evaluation value change results, the system effectively accumulates recommendation record data, dynamically updates course ordering strategies and recommendation thresholds in the next round of recommendation generation, strengthens suitability between candidate courses and key knowledge points, and finally achieves recommendation path self-adaptive adjustment and data closed-loop optimization driven based on grasping feedback response values.

The personalized course recommendation system based on the knowledge graph comprises a user learning data acquisition preprocessing module, a knowledge point mastering and evaluating module, a supplementary course recommendation generation module and a recommendation effect feedback correction module, wherein the user learning data acquisition preprocessing module is used for acquiring user learning data through real-time acquisition of embedded point of the front end of a learning platform and background access log data and preprocessing the user learning data to obtain preprocessed user learning data, the knowledge point mastering and evaluating module is used for constructing a behavior data relation between a user and the knowledge point, carrying out normalization splitting on the preprocessed user learning data, evaluating mastering conditions of the user on the knowledge point and constructing a time sequence knowledge point scoring set, the supplementary course recommendation generation module is used for carrying out differentiation analysis by combining the behavior data relation, evaluating recommendation values of candidate courses, partitioning the candidate courses and generating recommendation courses, and the recommendation effect feedback correction module is used for quantifying response differences between the recommendation courses and mastering effects from a scale dimension, updating labels and user circulation data.

In the embodiment, through the systematic integration of the user learning data acquisition preprocessing module, the knowledge point mastering and evaluation module, the supplementary course recommendation generation module and the recommendation effect feedback correction module, the structural cleaning of the user learning data, the refined evaluation of knowledge point mastering conditions and the dynamic feedback correction of the differentiated recommendation and recommendation effect of the supplementary courses are realized. The system can continuously track the time sequence change of knowledge point mastering state, develop recommendation value analysis and hotness partition of candidate courses in combination with behavior data to realize intelligent generation of course recommendation, simultaneously introduce mastering feedback response mechanism, evaluate recommendation results from proportional dimension and update recommendation labels and user records to form a closed loop flow of data acquisition, analysis, recommendation, feedback and re-optimization, and remarkably improve the matching degree between recommendation content and actual mastering requirements of users.

It is noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.

The preferred embodiments of the invention disclosed above are intended only to assist in the explanation of the invention. The preferred embodiments are not exhaustive or to limit the invention to the precise form disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best understand and utilize the invention. The invention is limited only by the claims and the full scope and equivalents thereof.

Claims

1. The personalized course recommendation method based on the knowledge graph is characterized by comprising the following steps of:

S1, acquiring user learning data by collecting embedded points at the front end of a learning platform and background access log data in real time, and preprocessing the user learning data to obtain preprocessed user learning data;

s2, constructing a behavior data relation between a user and knowledge points, carrying out normalization and splitting on the preprocessed user learning data, evaluating the grasping condition of the user on the knowledge points, and constructing a time sequence knowledge point scoring set;

The method comprises the following specific steps of constructing a behavior data relation between a user and knowledge points, carrying out normalization and splitting on preprocessed user learning data, and evaluating the grasping condition of the user on the knowledge points:

Extracting a corresponding knowledge point set according to the knowledge point labels marked in each course; combining course codes in a user learning record, searching knowledge point labels contained in the learned courses of the user one by one, and combining in the dimension of the user to obtain all knowledge point sets learned by the user;

Based on the established corresponding relation of the behaviors between the user and the knowledge points, evaluating the mastering condition of the user on the knowledge points by integrating the pre-processed user learning data, such as the course browsing time length, the video watching completeness and the course scoring data, generated by the user in the course containing the same knowledge point, for each knowledge point, calculating the ratio between the course browsing time length and the historical average browsing time length of the course, multiplying the ratio by the browsing time length weight, taking the product as the browsing time length score, multiplying the video watching completeness by the video completeness weight, taking the product as the video completeness score, dividing the course scoring data by the course scoring full scoring data, multiplying the ratio by the course scoring weight, taking the product as the scoring data score, adding the browsing time length score, the video point completeness score and the course scoring data score, and multiplying the reciprocal of the course related knowledge, obtaining the spreading score of the knowledge point of the user on the course, recalculating the spreading score of all courses containing the knowledge point, and summing the spreading score of each course to obtain the knowledge point mastering evaluation value;

s3, identifying abnormal mastering knowledge points of the current stage of the user based on time sequence knowledge score aggregation, carrying out differential analysis in combination with behavior data relationship, evaluating recommendation values of candidate courses, partitioning the candidate courses and generating recommendation courses;

s4, quantifying response differences between recommended courses and mastering effects from the proportion dimension, updating recommended labels and user records, and perfecting recommended data circulation.

2. The personalized course recommendation method based on the knowledge graph of claim 1, wherein the specific steps of acquiring the user learning data by collecting the embedded point at the front end of the learning platform and the background access log data in real time are as follows:

Acquiring user learning data through collecting front-end buried points and background access log data of a learning platform in real time, wherein the user learning data comprises course codes, course browsing time length, course access times, course history average browsing time length, video watching completeness, course scoring data, course scoring full-score data, course test scoring data and course related knowledge points, extracting course codes through course name fields in the course access log, obtaining course browsing time length by recording the difference between the starting time and the exiting time of each user entering a course page and summing up according to course summary, obtaining course browsing time length by counting the total times of the user accessing a certain course page, obtaining course access times, obtaining course history average browsing time length by dividing the total browsing time length of the same course in a week by the course access times, obtaining video watching completeness by calculating the ratio of the actual playing time length recorded by a video player to the total time length of the course video, obtaining course scoring data and course scoring data through extracting scoring fields submitted by the user after the completion of the learning, obtaining course testing score data and course full-score data through recording the test results of the user after completion, obtaining course test score and course score data through recording the built-in the course, obtaining the knowledge related knowledge of each course map through the score map of the point in the course map and the time-resource management point.

3. The personalized course recommendation method based on the knowledge graph of claim 1, wherein the specific steps of preprocessing the user learning data to obtain the preprocessed user learning data are as follows:

the method comprises the steps of combining a median absolute deviation method of a sliding time window with an isolated forest algorithm, carrying out local and global anomaly detection on user learning data, removing abnormal user learning data comprising abnormal long-time on-hook and class-jump brushing progress, carrying out data filling on field missing problems of the user learning data caused by system delay, buried point loss or terminal fluctuation through a similar user collaborative complementation strategy and an index weighted moving average algorithm, carrying out fitting and denoising on the user learning data through a Bayesian dynamic smoothing method and a local weighted regression filtering algorithm, and carrying out transformation and range compression on the user learning data through an index transformation combined maximum and minimum normalization method to realize normalization processing.

4. The personalized course recommendation method based on the knowledge graph of claim 1, wherein the specific steps of constructing the time sequence knowledge point scoring set are as follows:

After the knowledge point mastering evaluation value is obtained through calculation, the knowledge point mastering evaluation value and the mastering threshold value are compared in real time, wherein when the knowledge point mastering evaluation value is larger than or equal to the mastering threshold value, the knowledge point is marked as a normal mastering knowledge point without adjustment, and when the knowledge point mastering evaluation value is smaller than the mastering threshold value, the knowledge point is marked as an abnormal mastering knowledge point, the current learning path of a user is prompted, and a recommended instruction for adding supplementary courses related to the knowledge point is generated;

the time series processing is carried out on the knowledge point mastering evaluation values after comparison, namely the knowledge point mastering evaluation values generated by a user in each course learning action are recorded in time sequence, corresponding time stamps, course codes and knowledge point labels are marked, all knowledge point mastering evaluation values of the same user on the same knowledge point are ordered in time sequence, a knowledge point mastering evaluation value sequence of the user on the knowledge point is constructed, and the knowledge point mastering evaluation value sequences of the user on all knowledge points are integrated and tidied to form a time series knowledge point mastering evaluation value set of the user under the whole knowledge structure.

5. The personalized course recommendation method based on the knowledge graph of claim 1, wherein the step of identifying abnormal knowledge points in the current stage of the user based on time series knowledge point evaluation set, performing differential analysis in combination with behavior data relationship, and evaluating recommendation values of candidate courses comprises the following specific steps:

Combining a time sequence knowledge point mastering evaluation value set of a user under the whole knowledge structure, extracting a set marked as an abnormally mastered knowledge point in the current period, and counting the number of the knowledge points marked as abnormally mastered knowledge points;

On the basis of the identified key supplementary knowledge point set, all courses associated with each knowledge point are reviewed, the courses which are learned by the user and the incompletely watched courses are removed, and the rest courses are brought into the candidate course set;

The method comprises the steps of calculating a square value of the superposition quantity of course related knowledge points and a supplementary knowledge point set, calculating the product of the number of knowledge points related to a course and the number of knowledge points marked as abnormal mastery, dividing the square value by the product, taking the ratio as a knowledge point matching score, calculating course browsing time length divided by course historical average browsing time length, taking the ratio plus video watching completeness as a learning input score, calculating course scoring data plus course historical average scoring, taking the sum value divided by twice as course scoring data representation score, and multiplying the knowledge point matching score, the learning input score and the course scoring data representation score to obtain a course recommendation strength evaluation value.

6. The personalized course recommendation method based on the knowledge graph of claim 1, wherein the specific steps of partitioning candidate courses and generating recommended courses are as follows:

The method comprises the steps of establishing a multi-dimensional course recommendation heat partition based on a course recommendation intensity evaluation value, wherein the course recommendation threshold value comprises a primary course recommendation threshold value and a secondary course recommendation threshold value, judging that a high-heat course is directly included in a current recommended course when the course recommendation intensity evaluation value is larger than or equal to the secondary course recommendation threshold value, and carrying out priority sorting pushing;

generating a recommended course of the current stage of the user according to the multi-dimensional course recommendation heat partitioning result, transmitting the recommended course to a user learning interface, and marking supplementary knowledge point labels covered by each course in a recommended course list.

7. The personalized course recommendation method based on the knowledge graph of claim 1, wherein the specific steps of quantifying the response difference between the recommended course and the mastering effect from the scale dimension are as follows:

after the user finishes the recommended course learning, extracting course codes, course browsing time and supplementary knowledge point labels covered by courses related to the recommended course learning, and establishing a content structure of the recommended course learning; synchronously recording course test score data after recommending courses, and updating corresponding knowledge point mastering evaluation values in the dimension of the knowledge points;

According to comparison relation of knowledge point mastering evaluation values before and after recommendation, response difference between a recommended course and mastering effect is quantified, namely knowledge point mastering evaluation values of the knowledge points after the recommendation course is issued are subtracted by knowledge point mastering evaluation values of the knowledge points before the recommendation course is issued and divided by course scoring full-scale data, then the ratio of course testing score data and the ratio of course testing full-scale data is subtracted by the ratio, absolute value of the difference is used as mastering change ratio, the number of recommended courses related to the knowledge points in the recommended course is added by 1 and then logarithm based on two is taken, obtained result is added by 1 and used as course coverage factor, and mastering change ratio and course coverage factor are multiplied to obtain mastering feedback response value.

8. The personalized course recommendation method based on the knowledge graph of claim 1, wherein the specific steps of updating recommendation labels and user records and perfecting recommendation data circulation are as follows:

After the mastering feedback response value is obtained, the mastering feedback response value is compared with the mastering feedback threshold in real time, wherein the mastering feedback threshold comprises a first-level mastering feedback threshold and a second-level mastering feedback threshold, when the mastering feedback response value is larger than or equal to the second-level mastering feedback threshold, feedback is normal, knowledge points covered in a current course are marked as recommended effective labels, the recommended priority of the current course in candidate course sequencing of a subsequent similar user is improved, when the mastering feedback response value is larger than the first-level mastering feedback threshold and smaller than the second-level mastering feedback threshold, feedback is delayed, the current course is marked as an intermediate transition course and is brought into the next recommendation period to be used as a continuous observation object of delay type supplementary content, when the mastering feedback response value is smaller than or equal to the first-level mastering feedback threshold, the feedback abnormality is marked as recommended offset labels, the current course is marked as recommended insensitive courses, and courses of similar teaching structures are preferentially avoided in the subsequent supplementary screening courses;

Continuously arranging mastered feedback response values generated by the same knowledge point in different recommendation periods according to the dimension of a user to construct a mastered feedback response value sequence, extracting course codes and course scoring data corresponding to the knowledge points for three or more times and more than one level of knowledge points with mastered feedback threshold values of the feedback response values, and recalculating knowledge point mastered evaluation values of the user on the course based on the browsing duration of the original course, the video watching integrity and the course scoring data to replace the original result for feedback correction;

Writing course codes, knowledge point labels, course browsing time length, course scoring data, video watching completeness, grasping feedback response values, course test scoring data and evaluation value change results generated by each round of course recommendation learning into user recommendation records, preferentially reading the user recommendation records in a new round of recommendation generation process, dynamically adjusting course initial recommendation ordering, heat distinguishing strategies and recommendation threshold calculation logic, updating binding strength between candidate courses and key knowledge points, and completing active adaptation adjustment of recommendation paths, thereby realizing a recommendation self-closing mechanism based on capability feedback.

9. The personalized course recommendation system based on the knowledge graph is characterized by comprising a user learning data acquisition preprocessing module, a knowledge point mastering and evaluating module, a supplementary course recommendation generating module and a recommendation effect feedback correction module, wherein:

The user learning data acquisition preprocessing module is used for acquiring user learning data through real-time acquisition of embedded points at the front end of the learning platform and background access log data, preprocessing the user learning data and obtaining preprocessed user learning data;

the knowledge point mastering and evaluating module is used for constructing a behavior data relation between a user and knowledge points, evaluating the mastering condition of the user on the knowledge points based on the preprocessed user learning data, and constructing a time sequence knowledge point scoring set;

the supplementary course recommendation generation module is used for identifying abnormal mastering knowledge points at the current stage of a user based on time sequence knowledge critique set, analyzing whether candidate courses are worth recommending, partitioning the candidate courses and generating recommended courses;

the recommending effect feedback correction module is used for quantifying the response difference between the recommending course and mastering effect, updating the recommending label and the user record and perfecting the recommending data circulation.