Teaching content recommendation method and system driven by online education platform data

Abstract

The invention relates to the technical field of intelligent recommendation, in particular to a teaching content recommendation method and system driven by online education platform data. The invention discloses a teaching content recommendation system driven by online education platform data. The teaching content recommendation system comprises a knowledge point mastering feature data construction module, a mastering degree analysis module, a gap knowledge point determination module and an education content recommendation module. The knowledge point mastering degree of the user is determined by analyzing the evaluation data of the user, the education content knowledge graph is analyzed through the mastering degree analysis model, the dependency relationship between the knowledge points is researched, the deeper knowledge point mastering degree is analyzed, and then more accurate knowledge point mastering degree judgment is obtained. And then determining notched knowledge points with low knowledge point mastering degree of the user based on the knowledge point deep mastering feature data, and determining an educational content learning path needing to be recommended through the notched knowledge points, so that educational content recommendation for checking and repairing the missing of the user can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent recommendation, and in particular to a method and system for recommending teaching content driven by data on an online education platform. Background Art

[0002] With the rapid development and popularization of online education platforms, learning resources have become increasingly rich, and the teaching content available to users has shown an explosive growth. However, the content recommendation mechanisms of existing education platforms mainly recommend based on curriculum integrity, popularity, user historical behavior, or simple knowledge point association relationships, lacking accurate identification and targeted analysis of the actual knowledge mastery status of learners. Such a recommendation method is difficult to meet the personalized learning needs of learners, especially unable to effectively solve the "knowledge gap" problem in the learning process, resulting in low learning efficiency, poor learning experience, and even a series of problems such as a decline in learning motivation and an incomplete construction of the knowledge system. Summary of the Invention

[0003] The present invention determines the degree of knowledge point mastery of a user by analyzing the user's assessment data, and analyzes the educational content knowledge graph through a mastery degree analysis model to study the dependency relationship between knowledge points, analyze the deeper degree of knowledge point mastery, and then obtain a more accurate judgment of the degree of knowledge point mastery. Based on the deep mastery feature data of knowledge points, the gap knowledge points with low user knowledge point mastery are determined, and the learning path of educational content to be recommended is determined through the gap knowledge points, so as to realize the recommendation of educational content for users to fill in the gaps.

[0004] The present invention provides a method for recommending teaching content driven by data on an online education platform, including: Obtaining assessment data for a user, and constructing knowledge point mastery feature data based on the user's assessment data, where the knowledge point mastery feature data includes the feature vector corresponding to each knowledge point and the assessment accuracy rate of the user under the corresponding knowledge point; Sending the knowledge point mastery feature data and the educational content knowledge graph into a mastery degree analysis model for processing, and outputting deep mastery feature data of knowledge points; Determining a set of gap knowledge points based on the deep mastery feature data of knowledge points and the educational content knowledge graph, where the set of gap knowledge points includes several gap knowledge points; Determining a learning path of educational content based on the set of gap knowledge points and the educational content knowledge graph, where the learning path of educational content includes educational content arranged in the learning order, and realizing the recommendation of educational content through the learning path of educational content; The mastery level analysis model includes a graph convolutional layer, a dependency strengthening layer, and an output layer. The graph convolutional layer is used to perform feature fusion on the knowledge point mastery feature data based on the educational content knowledge graph to construct the fused knowledge point mastery features. The dependency strengthening layer is used to strengthen the fused knowledge point mastery features based on the educational content knowledge graph to construct the enhanced knowledge point mastery features. The output layer is used to perform a fully connected operation on the enhanced knowledge point mastery features to obtain the deep knowledge point mastery feature data.

[0005] Preferably, the educational content knowledge graph is used to describe the dependency relationships between all knowledge points in the educational content and is constructed in the following manner: Construct several groups of knowledge point entity - dependency relationship - knowledge point entity triples, and form the educational content knowledge graph from all the knowledge point entity - dependency relationship - knowledge point entity triples. The dependency relationships include pre - dependency relationships, correlation relationships, and auxiliary enhancement relationships. The weight values corresponding to the pre - dependency relationships, correlation relationships, and auxiliary enhancement relationships are the pre - dependency relationship strength, correlation strength, and auxiliary enhancement degree, respectively.

[0006] Preferably, the knowledge point mastery feature data and the educational content knowledge graph are sent into the mastery level analysis model for processing, and the deep knowledge point mastery feature data is output. The specific steps are as follows: Perform feature fusion on the knowledge point mastery feature data based on the educational content knowledge graph through the graph convolutional layer to construct the fused knowledge point mastery features. The specific content is as follows: Obtain the corresponding dependency relationship adjacency matrix based on the educational content knowledge graph. The size of the dependency relationship adjacency matrix is N×N, where N is the total number of knowledge point entities. The data stored in the i - th row and j - th column of the dependency relationship adjacency matrix is the dependency relationship vector between the i - th knowledge point entity and the j - th knowledge point entity, and i, j = 1, 2, 3, …, N. The dependency relationship vector includes the pre - dependency relationship strength, correlation strength, and auxiliary enhancement degree between the i - th knowledge point entity and the j - th knowledge point entity. Perform graph convolutional operations through the following formula: H(t + 1)=σ(H(t)AW(t)), where H(t) is the fused feature map output in the t - th iteration, A is the normalized dependency relationship adjacency matrix, W(t) is the graph convolutional weight matrix corresponding to the t - th iteration, and σ(·) is the sigmoid function. When the number of iterations reaches the iteration upper limit E, the feature fusion is completed, and the fused feature map output in the last iteration is denoted as the fused knowledge point mastery features; The feature strengthening of the mastered features of the integrated knowledge points is carried out based on the educational content knowledge graph through the dependency strengthening layer to construct the strengthened mastered features of the knowledge points, which specifically includes the following content: constructing the corresponding knowledge point feature value vector V and knowledge point feature key vector K based on the mastered features of the integrated knowledge points, constructing the corresponding knowledge point feature query vector Q based on the dependency adjacency matrix, and performing the self-attention mechanism operation through the following formula: G = softmax(QK T / D 0.5 )V, where G is the strengthened mastered feature of the knowledge point, T is the matrix transpose operation, and D is the dimension size of the knowledge point feature key vector K to achieve feature strengthening; Perform a fully connected operation on the strengthened mastered features of the knowledge points to obtain the deep mastered feature data of the knowledge points.

[0007] Preferably, determining the set of missing knowledge points based on the deep mastered feature data of the knowledge points and the educational content knowledge graph specifically includes the following steps: Traverse the knowledge points in the deep mastered feature data of the knowledge points, and perform the following operations on the selected knowledge points: judge whether the corresponding deep mastery degree value of the knowledge point is lower than the first mastery threshold. If the corresponding deep mastery degree value of the knowledge point is lower than the first mastery threshold, record the selected knowledge point as a missing knowledge point. If the corresponding deep mastery degree value of the knowledge point is not lower than the first mastery threshold, do nothing; until all the knowledge points in the deep mastered feature data of the knowledge points are traversed, output all the missing knowledge points; Split the pre-dependency educational content knowledge directed graph and the auxiliary enhancement educational content knowledge directed graph that only have pre-dependency relationships and auxiliary enhancement relationships from the educational content knowledge graph. Traverse all the missing knowledge points, and perform the following operations on each missing knowledge point: in the pre-dependency educational content knowledge directed graph, traverse based on the missing knowledge point against the pre-dependency relationship, record the knowledge point entities, and obtain the pre-dependency path. In the auxiliary enhancement educational content knowledge directed graph, traverse based on the missing knowledge point against the auxiliary enhancement relationship, record the knowledge point entities, and obtain the auxiliary enhancement path; until all the missing knowledge points are traversed, output all the pre-dependency paths and auxiliary enhancement paths. For all the pre-dependency paths, form the first candidate knowledge point set by all the overlapping knowledge point entities of all the pre-dependency paths. Traverse all the knowledge point entities in the first candidate knowledge point set, and record the knowledge points corresponding to the knowledge point entities with the deep mastery degree value lower than the second mastery threshold as missing knowledge points, and the second mastery threshold is higher than the first mastery threshold. For all the auxiliary enhancement paths, form the second candidate knowledge point set by all the overlapping knowledge point entities of all the auxiliary enhancement paths. Traverse all the knowledge point entities in the second candidate knowledge point set, and record the knowledge points corresponding to the knowledge point entities with the deep mastery degree value lower than the second mastery threshold as missing knowledge points; Form the set of missing knowledge points by all the missing knowledge points.

[0008] Preferably, an educational content learning path is determined based on the gap knowledge point set and the educational content knowledge graph, which specifically includes the following steps: map all the gap knowledge points in the gap knowledge point set to the educational content knowledge graph, segment the educational content learning path that retains the prerequisite dependency relationship and the auxiliary enhancement relationship, and associate each knowledge point entity in the educational content learning path with the corresponding educational content.

[0009] Preferably, the mastery analysis model is trained, which specifically includes the following steps: Obtain a number of mastery analysis training samples. The mastery analysis training samples include knowledge point mastery feature data and the educational content knowledge graph. Combine all the mastery analysis training samples to form a mastery analysis training set, and perform unsupervised training on the mastery analysis model through the mastery analysis training set. The training objective is that for all the mastery analysis training samples of the same user in the same period, the knowledge point deep mastery feature data output by the mastery analysis model is consistent.

[0010] The present invention also provides an online education platform data-driven teaching content recommendation system, including: A knowledge point mastery feature data construction module, which is used to obtain the evaluation data of the user and construct the knowledge point mastery feature data based on the user's evaluation data. The knowledge point mastery feature data includes the feature vector corresponding to each knowledge point and the evaluation accuracy rate of the user under the corresponding knowledge point; A mastery analysis module, which is used to send the knowledge point mastery feature data and the educational content knowledge graph into the mastery analysis model for processing and output the knowledge point deep mastery feature data; A gap knowledge point determination module, which is used to determine the gap knowledge point set based on the knowledge point deep mastery feature data and the educational content knowledge graph. The gap knowledge point set includes a number of gap knowledge points; An educational content recommendation module, which is used to determine the educational content learning path based on the gap knowledge point set and the educational content knowledge graph. The educational content learning path includes educational content arranged in the learning order, and realizes the recommendation of educational content through the educational content learning path.

[0011] The present invention has the following advantages: The present invention determines the mastery level of users' knowledge points by analyzing the evaluation data of users, and analyzes the educational content knowledge graph through a mastery level analysis model, studies the dependency relationships between knowledge points, analyzes the mastery level of deeper knowledge points, and then obtains a more accurate judgment of the mastery level of knowledge points. Based on the deep mastery feature data of knowledge points, the missing knowledge points with low mastery level of users' knowledge points are determined, and the learning path of educational content to be recommended is determined through the missing knowledge points, so as to realize the recommendation of educational content for users to fill in the gaps and make up for deficiencies. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic structural diagram of a data-driven teaching content recommendation system for an online education platform adopted in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0013] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0014] Embodiment 1, a data-driven teaching content recommendation method for an online education platform, includes: Obtain the evaluation data for users. During the learning process of users on the online education platform, there will be evaluations in the forms of after-class exercises and comprehensive test papers, etc., to reflect the mastery level of users for the learned content. Here, the evaluation data refers to the evaluation accuracy rate corresponding to each knowledge point under various historical evaluation methods. It should be noted that each exercise in the evaluation corresponds to a knowledge point, and a one-to-one mapping between the exercise and the knowledge point is established. Here, the knowledge points are obtained from the educational content knowledge graph constructed based on the corresponding educational content. For example, the knowledge points corresponding to "High School Mathematics: Functions and Derivatives" include "Function Monotonicity", "Function Value Range", and "Geometric Meaning of Derivatives", etc.; construct knowledge point mastery feature data based on the evaluation data of users. Here, the knowledge point mastery feature data includes the feature vector corresponding to each knowledge point and the evaluation accuracy rate of the user under the corresponding knowledge point. Specifically, during the historical evaluation process of the user, there are several test questions corresponding to each knowledge point. The accuracy rate of all test questions under each knowledge point is statistically calculated, which is the evaluation accuracy rate corresponding to the knowledge point. The evaluation accuracy rate can reflect the user's mastery of the knowledge point. The feature vector corresponding to the knowledge point is specifically the data after word embedding operation; The educational content knowledge graph is used to describe the dependency relationships between all knowledge points in educational content and is constructed as follows: Obtain the textbook content and video explanation materials corresponding to the educational content, and perform named entity recognition in combination with models such as TF-IDF, TextRank, and BERT to obtain corresponding knowledge point entities. Here, each knowledge point entity corresponds to a knowledge point. The dependency relationships between knowledge points are labeled by experts to construct several groups of knowledge point entity - dependency relationship - knowledge point entity triples. It should be noted that the educational content knowledge graph here corresponds to a directed graph. All the knowledge point entity - dependency relationship - knowledge point entity triples are combined to form the educational content knowledge graph; The dependency relationships here include prerequisite dependency relationships, correlation relationships, and auxiliary enhancement relationships. The weight values corresponding to the prerequisite dependency relationships, correlation relationships, and auxiliary enhancement relationships are the prerequisite dependency strength, correlation strength, and auxiliary enhancement degree respectively. The prerequisite dependency strength is the necessity strength of knowledge point A as a prerequisite for knowledge point B calculated through a conditional probability model. The correlation strength is the similarity degree between knowledge point A and knowledge point B calculated through the cosine similarity algorithm. The auxiliary enhancement degree is used to describe the degree of promotion of mastering knowledge point A for learning knowledge point B. It can be labeled through expert experience or obtained through A / B test experiments. The specific calculation method of the auxiliary enhancement degree is as follows: Divide the experimental personnel into an experimental group and a control group. The experimental group first learns knowledge point A and then learns knowledge point B, and the control group learns knowledge point B without learning knowledge point A. Calculate the corresponding learning efficiencies Q1 and Q2 of the experimental group and the control group respectively. The auxiliary enhancement degree is then 1 - (Q2 / Q1). However, the latter A / B test experiment requires a large amount of manpower and material resources and is relatively complicated. First, a small number of experimental samples can be used as labeled data, and then a semi-supervised method can be used to train a model to replace the experiment; The knowledge point mastery feature data and the educational content knowledge graph are sent into the mastery degree analysis model for processing, and the deep mastery feature data of knowledge points is output. The deep mastery feature data of knowledge points includes the feature vector corresponding to each knowledge point and the deep mastery degree value under the corresponding knowledge point. It should be noted that during the user's assessment process, there may be test questions errors caused by the user's own carelessness and other reasons, which affect the judgment of the knowledge point mastery degree. On the other hand, an important aspect of the knowledge point mastery degree is the application in the comprehensive aspect. For example, if knowledge point A is a prerequisite for knowledge point B, the test questions corresponding to knowledge point A may have a high assessment accuracy rate, but the assessment accuracy rate corresponding to knowledge point B is low. That indicates that the comprehensive application mastery degree of knowledge point A may be low, which will affect the judgment of the knowledge point mastery degree of knowledge point A. Therefore, by analyzing the educational content knowledge graph in the mastery degree analysis model, studying the dependency relationships between knowledge points, and then adjusting the judgment of the mastery degree of different knowledge points; Determine the set of knowledge points with gaps based on the deep mastery feature data of knowledge points and the educational content knowledge graph. The set of knowledge points with gaps includes several knowledge points with gaps. Here, a knowledge point with a gap refers to a knowledge point that the user has a low level of mastery, that is, the educational content that the user needs to supplement; Determine the learning path of educational content based on the set of knowledge points with gaps and the educational content knowledge graph. The learning path of educational content includes educational content arranged in the learning order. Here, the educational content includes textbooks, explanatory videos, etc. The recommendation of educational content is realized through the learning path of educational content; This application determines the user's mastery level of knowledge points by analyzing the user's assessment data, and analyzes the educational content knowledge graph through a mastery level analysis model to study the dependence relationship between knowledge points, analyze the deeper mastery level of knowledge points, and then obtain a more accurate judgment of the mastery level of knowledge points. Then, based on the deep mastery feature data of knowledge points, determine the knowledge points with gaps where the user has a low level of mastery, and determine the learning path of educational content to be recommended through the knowledge points with gaps, which can realize the recommendation of educational content for users to check for deficiencies and make up for them.

[0015] Send the knowledge point mastery feature data and the educational content knowledge graph into the mastery level analysis model for processing, and output the deep mastery feature data of knowledge points. The specific steps are as follows: The mastery level analysis model includes a graph convolutional layer, a dependence relationship strengthening layer, and an output layer. Among them, the graph convolutional layer is used to perform feature fusion on the knowledge point mastery feature data based on the educational content knowledge graph to construct a fused knowledge point mastery feature. Here, the feature fusion adjusts the mastery level corresponding to each knowledge point according to the dependence relationship between knowledge points, and weakens the test questions errors caused by the user's own carelessness, etc. For example, the mastery levels corresponding to two knowledge points with a relatively high correlation strength should be similar; the dependence relationship strengthening layer is used to perform feature strengthening on the fused knowledge point mastery feature based on the educational content knowledge graph to construct a strengthened knowledge point mastery feature. Here, the strengthened knowledge point mastery feature strengthens the mastery level corresponding to each knowledge point according to the dependence relationship between knowledge points. If knowledge point A is a prerequisite for knowledge point B, the test questions corresponding to knowledge point A may have a relatively high assessment accuracy rate, but the assessment accuracy rate corresponding to knowledge point B is relatively low, then the mastery level corresponding to knowledge point A should be reduced; the output layer is used to perform a fully connected operation on the strengthened knowledge point mastery feature to obtain the deep mastery feature data of knowledge points; Feature fusion is performed on the feature data of knowledge point mastery based on the educational content knowledge graph through a graph convolutional layer to construct a fused knowledge point mastery feature, which specifically includes the following: obtaining a corresponding dependency adjacency matrix based on the educational content knowledge graph. The size of the dependency adjacency matrix is N×N, where N is the total number of knowledge point entities. The data stored in the i-th row and j-th column of the dependency adjacency matrix is the dependency relationship vector between the i-th knowledge point entity and the j-th knowledge point entity, and i, j = 1, 2, 3, …, N. The dependency relationship vector includes the forward dependency strength, correlation strength, and auxiliary enhancement strength between the i-th knowledge point entity and the j-th knowledge point entity. It should be noted that the educational content knowledge graph is a directed graph, and the forward dependency strength and auxiliary enhancement strength will only store the corresponding values when there is a directed connection. In the case of no directed connection, the values of the corresponding items of the forward dependency strength and auxiliary enhancement strength are zero. The graph convolution operation is performed through the following formula: H(t + 1) = σ(H(t)AW(t)), where H(t) is the fused feature map output at the t-th iteration, A is the normalized dependency adjacency matrix, W(t) is the graph convolution weight matrix corresponding to the t-th iteration, and σ(·) is the sigmoid function; until the number of iterations reaches the iteration upper limit E, the feature fusion is completed. The fused feature map output at the last iteration is denoted as the fused knowledge point mastery feature, and E is generally 3; Feature enhancement is performed on the fused knowledge point mastery feature based on the educational content knowledge graph through a dependency relationship enhancement layer to construct an enhanced knowledge point mastery feature, which specifically includes the following: constructing a corresponding knowledge point feature value vector V and knowledge point feature key vector K based on the fused knowledge point mastery feature, constructing a corresponding knowledge point feature query vector Q based on the dependency adjacency matrix, and performing a self-attention mechanism operation through the following formula: G = softmax(QK T / D 0.5 )V, where G is the enhanced knowledge point mastery feature, T is the matrix transpose operation, and D is the dimension size of the knowledge point feature key vector K to achieve feature enhancement; It should be noted that the self-attention mechanism operation refers to the Transformer model. In the self-attention mechanism operation of the Transformer model, the input features are respectively multiplied by the corresponding value weight matrix, key weight matrix, and query weight matrix to perform matrix multiplication operations to construct the corresponding value vector, key vector, and query vector, and then the self-attention mechanism operation is achieved through the formula G = softmax(QK T / D 0.5 )V. In this application, the basis of the query vector is no longer the input feature (fused knowledge point mastery feature) itself, but comes from the dependency adjacency matrix of the educational content knowledge graph, which can strengthen the mastery degree corresponding to each knowledge point based on the dependency relationship between knowledge points; Determine the set of missing knowledge points based on the in-depth mastery feature data of knowledge points and the educational content knowledge graph, which specifically includes the following steps: Traverse the knowledge points in the in-depth mastery feature data of knowledge points, and perform the following operations for the selected knowledge points: Judge whether the in-depth mastery degree value corresponding to the knowledge point is lower than the first mastery threshold, which is set manually and is generally 0.6. If the in-depth mastery degree value corresponding to the knowledge point is lower than the first mastery threshold, mark the selected knowledge point as a missing knowledge point. If the in-depth mastery degree value corresponding to the knowledge point is not lower than the first mastery threshold, it means that the user's mastery degree of this knowledge point meets the expectation and no operation is performed; until all the knowledge points in the in-depth mastery feature data of knowledge points are traversed, output all the missing knowledge points; Extract from the educational content knowledge graph the pre-dependency educational content knowledge directed graph and the auxiliary enhancement educational content knowledge directed graph that only have pre-dependency relationships and auxiliary enhancement relationships. Traverse all the missing knowledge points, and perform the following operations for each missing knowledge point: In the pre-dependency educational content knowledge directed graph, traverse backwards along the pre-dependency relationship based on the missing knowledge point, record the knowledge point entities, and obtain the pre-dependency path. In the auxiliary enhancement educational content knowledge directed graph, traverse backwards along the auxiliary enhancement relationship based on the missing knowledge point, record the knowledge point entities, and obtain the auxiliary enhancement path. Both the pre-dependency path and the auxiliary enhancement path represent the influence of different knowledge points on the missing knowledge point; until all the missing knowledge points are traversed, output all the pre-dependency paths and auxiliary enhancement paths. For all the pre-dependency paths, form the first set of candidate knowledge points by combining all the knowledge point entities that overlap in all the pre-dependency paths. Traverse all the knowledge point entities in the first set of candidate knowledge points, and mark the knowledge points corresponding to the knowledge point entities with an in-depth mastery degree value lower than the second mastery threshold as missing knowledge points. Here, the second mastery threshold is also set manually and is higher than the first mastery threshold. When supplementing the knowledge points with a lower mastery degree, since the knowledge points are interdependent, it is necessary to judge the mastery degree of the previously learned knowledge points again and relax the judgment requirements. For all the auxiliary enhancement paths, form the second set of candidate knowledge points by combining all the knowledge point entities that overlap in all the auxiliary enhancement paths. Traverse all the knowledge point entities in the second set of candidate knowledge points, and mark the knowledge points corresponding to the knowledge point entities with an in-depth mastery degree value lower than the second mastery threshold as missing knowledge points; Combine all the missing knowledge points into a set of missing knowledge points.

[0016] Determine the educational content learning path based on the set of missing knowledge points and the educational content knowledge graph, which specifically includes the following steps: Map all the missing knowledge points concentrated by the missing knowledge point concentration to the educational content knowledge graph, segment the educational content learning path that retains the pre-dependence relationship and the auxiliary enhancement relationship, and associate each knowledge point entity in the educational content learning path with the corresponding educational content.

[0017] Train the mastery analysis model, which specifically includes the following steps: Obtain several mastery analysis training samples. The mastery analysis training samples include knowledge point mastery feature data and the educational content knowledge graph. Here, the knowledge point mastery feature data is obtained artificially based on the evaluation data collected from the actual online education platform. Form all the mastery analysis training samples into a mastery analysis training set, and perform unsupervised training on the mastery analysis model through the mastery analysis training set. The training objective is that for all the mastery analysis training samples of the same user at the same time period, the deep knowledge point mastery feature data output by the mastery analysis model has consistency. The consistency can be calculated in the following way: perform clustering analysis on all the deep knowledge point mastery feature data output by the mastery analysis model, take the cluster center as the average value, and use the calculated mean square error value as the consistency.

[0018] Example 2, An online education platform data-driven teaching content recommendation system, see Figure 1 , including: The knowledge point mastery feature data construction module is used to obtain the evaluation data for the user. During the learning process of the user on the online education platform, there will be evaluations in the form of after-class exercises and comprehensive test papers, etc., to reflect the user's mastery of the learned content. Here, the evaluation data refers to the evaluation accuracy rate corresponding to each knowledge point under various historical evaluation methods. It should be noted that each exercise in the evaluation corresponds to a knowledge point, and a one-to-one mapping between the exercise and the knowledge point is established. Here, the knowledge point is obtained from the educational content knowledge graph constructed based on the corresponding educational content. For example, the knowledge points corresponding to "High school mathematics functions and derivatives" include "Function monotonicity", "Function value range", and "Geometric meaning of derivatives", etc.; construct the knowledge point mastery feature data based on the user's evaluation data. Here, the knowledge point mastery feature data includes the feature vector corresponding to each knowledge point and the user's evaluation accuracy rate under the corresponding knowledge point. Specifically, during the historical evaluation process of the user, there are several questions corresponding to each knowledge point. Count the accuracy rate of all the questions under each knowledge point, which is the evaluation accuracy rate corresponding to the knowledge point. The evaluation accuracy rate can reflect the user's mastery of the knowledge point. The feature vector corresponding to the knowledge point is specifically the data after the word embedding operation; The mastery degree analysis module is used to process the knowledge point mastery feature data and the educational content knowledge graph in the mastery degree analysis model, and output the deep mastery feature data of knowledge points. The deep mastery feature data of knowledge points includes the feature vector corresponding to each knowledge point and the deep mastery degree value under the corresponding knowledge point. It should be noted that during the user's assessment process, there may be errors in the test questions caused by the user's carelessness or other reasons, which affect the judgment of the mastery degree of knowledge points. On the other hand, an important aspect of the mastery degree of knowledge points is their comprehensive application. For example, if knowledge point A is a prerequisite for knowledge point B, the test questions corresponding to knowledge point A may have a high assessment accuracy rate, but the assessment accuracy rate corresponding to knowledge point B is low. This indicates that the comprehensive application mastery degree of knowledge point A may be low, which will affect the judgment of the mastery degree of knowledge point A. Therefore, by analyzing the educational content knowledge graph in the mastery degree analysis model, studying the dependency relationship between knowledge points, and then adjusting the judgment of the mastery degree of different knowledge points; The gap knowledge point determination module is used to determine the set of gap knowledge points based on the deep mastery feature data of knowledge points and the educational content knowledge graph. The set of gap knowledge points includes several gap knowledge points. Here, the gap knowledge point refers to the knowledge point with low user mastery degree, that is, the educational content that the user needs to supplement; The educational content recommendation module is used to determine the educational content learning path based on the set of gap knowledge points and the educational content knowledge graph. The educational content learning path includes educational content arranged in the learning order. Here, the educational content includes textbooks, explanatory videos, etc. The recommendation of educational content is realized through the educational content learning path.

[0019] It should be understood that those of ordinary skill in the art can make improvements or modifications according to the above description, and all such improvements and modifications shall fall within the protection scope of the appended claims of the present invention. The parts not described in detail in this specification belong to the prior art well-known to those of ordinary skill in the art.

Claims

1. A data-driven teaching content recommendation method for an online education platform, characterized in that: include: Acquire evaluation data for the user, and construct knowledge point mastery feature data based on the user's evaluation data. The knowledge point mastery feature data includes a feature vector corresponding to each knowledge point and the user's evaluation accuracy rate under the corresponding knowledge point; The knowledge point mastery feature data and the educational content knowledge graph are sent to the mastery degree analysis model for processing, and the deep mastery feature data of the knowledge point is output; Determine a gap knowledge point set based on the deep-level feature data of knowledge points and the knowledge graph of educational content, where the gap knowledge point set includes several gap knowledge points; Determine the educational content learning path based on the gap knowledge point set and the educational content knowledge graph. The educational content learning path includes educational content arranged in a learning order. The educational content is recommended through the educational content learning path. The mastery degree analysis model includes a graph convolution layer, a dependency reinforcement layer, and an output layer. The graph convolution layer is used to fuse the knowledge point mastery feature data based on the educational content knowledge graph to construct the fused knowledge point mastery feature. The dependency enhancement layer is used to enhance the mastery features of fused knowledge points based on the educational content knowledge graph and construct enhanced mastery features of knowledge points; The output layer is used to fully connect the features of the enhanced knowledge points to obtain the deep mastery feature data of the knowledge points.

2. According to claim 1, a data-driven teaching content recommendation method for an online education platform is characterized in that: The educational content knowledge graph is used to describe the dependencies between all knowledge points in the educational content and is constructed in the following way: Construct several groups of knowledge point entity-dependency-knowledge point entity triplets, and organize all knowledge point entity-dependency-knowledge point entity triplets into an educational content knowledge graph. The dependency relationships include pre-dependency relationships, correlation relationships, and auxiliary enhancement relationships. The weight values ​​corresponding to the pre-dependency relationships, correlation relationships, and auxiliary enhancement relationships are pre-dependency strength, correlation strength, and auxiliary enhancement degree, respectively.

3. The method for recommending teaching content driven by online education platform according to claim 2, characterized in that: The knowledge point mastery feature data and the educational content knowledge graph are sent to the mastery degree analysis model for processing, and the deep mastery feature data of the knowledge point is output, which specifically includes the following steps: Through the graph convolution layer, the feature data of knowledge point mastery is fused based on the educational content knowledge graph to construct the fused knowledge point mastery feature, which specifically includes the following contents: based on the educational content knowledge graph, the corresponding dependency adjacency matrix is ​​obtained. The size of the dependency adjacency matrix is ​​N×N, where N is the total number of knowledge point entities. The data stored in the i-th row and j-th column of the dependency adjacency matrix is ​​the dependency vector between the i-th knowledge point entity and the j-th knowledge point entity, and satisfies i, j=1,2,3,…,N. The dependency vector includes the i-th knowledge point entity and The strength of the pre-dependency relationship, the strength of the correlation, and the degree of auxiliary enhancement between the entities of the j-th knowledge point are calculated by performing the graph convolution operation through the following formula: H(t+1)=σ(H(t)AW(t)), where H(t) is the fused feature map output by the t-th iteration, A is the normalized dependency adjacency matrix, W(t) is the graph convolution weight matrix corresponding to the t-th iteration, and σ(·) is the sigmoid function; until the number of iterations reaches the upper limit of the number of iterations E, the feature fusion is completed, and the fused feature map output by the last iteration is recorded as the fused knowledge point mastery feature; The dependency enhancement layer is used to enhance the fused knowledge point mastery features based on the educational content knowledge graph, and to construct enhanced knowledge point mastery features, including the following: constructing the corresponding knowledge point feature value vector V and knowledge point feature key vector K based on the fused knowledge point mastery features, constructing the corresponding knowledge point feature query vector Q based on the dependency adjacency matrix, and performing the self-attention mechanism operation through the following formula: G=softmax(QK T / D 0.5 ) V, where G is the feature of strengthening the knowledge point, T is the matrix transposition operation, and D is the dimension size of the knowledge point feature key vector K to achieve feature strengthening; The features of strengthening the mastery of knowledge points are fully connected to obtain the feature data of deep mastery of knowledge points.

4. The method for recommending teaching content driven by online education platform according to claim 3, characterized in that: Based on the deep-level feature data of knowledge points and the knowledge graph of educational content, the set of gap knowledge points is determined, which specifically includes the following steps: Traverse the knowledge points in the knowledge point deep mastery feature data, and perform the following operations for the selected knowledge points: determine whether the deep mastery degree value corresponding to the knowledge point is lower than the first mastery threshold, if the deep mastery degree value corresponding to the knowledge point is lower than the first mastery threshold, record the selected knowledge point as a gap knowledge point, if the deep mastery degree value corresponding to the knowledge point is not lower than the first mastery threshold, no operation; Until all knowledge points in the feature data of the knowledge points are traversed and all missing knowledge points are output; A pre-dependency educational content knowledge directed graph and an auxiliary enhancement educational content knowledge directed graph with only pre-dependency and auxiliary enhancement relationships are separated from the educational content knowledge graph, all missing knowledge points are traversed, and the following operations are performed for each missing knowledge point: in the pre-dependency educational content knowledge directed graph, traverse the pre-dependency relationship based on the missing knowledge points, record the knowledge point entities, and obtain the pre-dependency path; in the auxiliary enhancement educational content knowledge directed graph, traverse the auxiliary enhancement relationship based on the missing knowledge points, record the knowledge point entities, and obtain the auxiliary enhancement path; until all missing knowledge points are traversed, all pre-dependency paths and Auxiliary enhancement path, for all preceding dependent paths, all knowledge point entities overlapping with all preceding dependent paths are formed into a first candidate knowledge point set, all knowledge point entities in the first candidate knowledge point set are traversed, and the knowledge points corresponding to the knowledge point entities with a deep mastery value lower than the second mastery threshold are recorded as gap knowledge points, and the second mastery threshold is higher than the first mastery threshold. For all auxiliary enhancement paths, all knowledge point entities overlapping with all auxiliary enhancement paths are formed into a second candidate knowledge point set, all knowledge point entities in the second candidate knowledge point set are traversed, and the knowledge points corresponding to the knowledge point entities with a deep mastery value lower than the second mastery threshold are recorded as gap knowledge points; All the gap knowledge points are grouped into a gap knowledge point set.

5. The method for recommending teaching content driven by online education platform data according to claim 4, characterized in that: Determining the educational content learning path based on the gap knowledge point set and the educational content knowledge graph specifically includes the following steps: mapping all the gap knowledge points in the gap knowledge point set to the educational content knowledge graph, segmenting the educational content learning path that retains the pre-dependency relationship and auxiliary enhancement relationship, and associating each knowledge point entity in the educational content learning path to the corresponding educational content.

6. The method for recommending teaching content driven by online education platform data according to claim 5, characterized in that: Training the mastery analysis model includes the following steps: A number of mastery analysis training samples are obtained, which include knowledge point mastery feature data and educational content knowledge graphs. All mastery analysis training samples are grouped into a mastery analysis training set. The mastery analysis model is trained unsupervisedly through the mastery analysis training set. The training goal is to ensure that the deep mastery feature data of knowledge points output by the mastery analysis model are consistent for all mastery analysis training samples of the same user in the same period.

7. A data-driven teaching content recommendation system for an online education platform, characterized in that: The system applies the online education platform data-driven teaching content recommendation method according to any one of claims 1 to 6, including: A knowledge point mastery feature data construction module is used to obtain evaluation data for users and construct knowledge point mastery feature data based on the user's evaluation data. The knowledge point mastery feature data includes a feature vector corresponding to each knowledge point and the user's evaluation accuracy under the corresponding knowledge point; The mastery degree analysis module is used to send the knowledge point mastery feature data and the educational content knowledge graph into the mastery degree analysis model for processing, and output the knowledge point deep mastery feature data; A gap knowledge point determination module is used to determine a gap knowledge point set based on the knowledge point deep-level mastery feature data and the educational content knowledge graph, where the gap knowledge point set includes several gap knowledge points; The educational content recommendation module is used to determine the educational content learning path based on the gap knowledge point set and the educational content knowledge graph. The educational content learning path includes educational content arranged in the order of learning, and the recommendation of educational content is realized through the educational content learning path.

Images