Differentiated educational knowledge graph construction method, system, medium and device

Abstract

The invention discloses an educational knowledge graph construction method supporting differentiation, which comprises the following steps: S1, collecting learner information and knowledge point information, and constructing a knowledge point knowledge graph; s2, predicting and complementing knowledge points in the knowledge point mapping knowledge domain and complex relationships among the knowledge points; s3, determining the centrality of each knowledge point in the knowledge point mapping knowledge domain; s4, calculating the knowledge point learning probability of the learner according to the historical answer data of the learner; and S5, comprehensively calculating the importance of each knowledge point through the centrality of the knowledge points and the knowledge point learning probability values, optimizing a user learning path according to the importance, obtaining a learning path with the minimum weight, and constructing an education knowledge map of the user according to the shortest path. According to the method, the mastering condition of the learner for the new knowledge points is judged, the optimal recommendation method for learning the new knowledge points is customized, and the learning efficiency of the learner is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent education technology, and in particular to a method, system, medium and device for constructing an educational knowledge graph that supports differentiation. Background Art

[0002] As my country's digitalization process continues to accelerate, online education technology has been more and more widely used. At present, academic social networks and educational platforms such as MOOC have obtained a large number of academic and educational resources. After the strict time and location restrictions are lifted, learners can obtain various learning resources they need at any time through various platforms. Therefore, in order to better improve the learning effect of learners, it is very necessary to construct a knowledge graph for each attribute of each learner and use it in the education recommendation system.

[0003] Most of the current related studies judge the degree of adaptation between learners and learning resources based on learners' cognitive level, learning style, learning tendency and other characteristics, and sort learning resources and recommend paths according to corresponding rules. However, they do not utilize information such as learners' historical data and the relationship between knowledge points. The generated learning paths may violate the internal logic between knowledge points and fail to match learners' changing knowledge levels.

[0004] The knowledge graph is a huge network knowledge base linked by entities, entity attributes and entity relationships. The knowledge points that learners have mastered, the target knowledge points and the historical interactive learning sequence are collected from various educational resources, and then the various knowledge points and their attributes involved are integrated to establish a knowledge graph, so as to better recommend learning content for different learners. The existing knowledge graph recommendation model randomly samples and selects neighborhoods, which easily leads to inappropriate recommendation results and learners cannot make good use of online learning resources. Therefore, it is very necessary to propose an educational knowledge graph that supports learner differentiation to better recommend learning resources. Summary of the invention

[0005] The purpose of the present invention is to overcome the above-mentioned defects in the prior art and provide a differentiated educational knowledge graph construction method. The present invention first judges the learner's mastery of new knowledge points based on the learner's own characteristics and other attributes, and then based on the learner's previous test-taking situation and knowledge point search situation, and personalizes the optimal recommendation method for learning new knowledge points to improve the learner's learning efficiency.

[0006] To achieve the above objectives, the present invention provides a differentiated educational knowledge graph construction method, comprising:

[0007] S1: Collect learner information and knowledge point information, and build a knowledge graph of knowledge points;

[0008] S2: predict and complete the knowledge points in the knowledge graph and the complex relationships between the knowledge points;

[0009] S3: Determine the centrality of each knowledge point in the knowledge graph;

[0010] S4: Calculate the probability of the learner's knowledge point acquisition based on the learner's historical answer data;

[0011] S5: The importance of each knowledge point is calculated by comprehensively calculating the centrality of the knowledge point and the probability of knowledge point acquisition. The user's learning path is optimized according to the importance to obtain the learning path with the minimum weight, and the user's educational knowledge graph is constructed with the shortest path.

[0012] Furthermore, the knowledge point knowledge graph in S1 uses the TransR model to predict and complete the complex relationships between knowledge point entities and knowledge point entities;

[0013] The above-mentioned TtansR model is:

[0014] First, the knowledge points and the relationships between them are encoded and mapped to integer indices to construct a relationship graph between knowledge points, namely, a knowledge graph triple (h, r, t), where (h, r, t) represents the head knowledge point, the relationship, and the tail knowledge point, respectively. The mapping relationship is h, t∈Z |E| and r∈Z |R| , where |E| represents the number of knowledge points, |R| represents the number of relations, and Z is an integer index set;

[0015] Then define and randomly initialize the knowledge point vector Relationship Vector and the projection matrix where d e and d r Represent the dimensions of entity vector and relationship vector respectively;

[0016] For each relationship, use the projection matrix M r , project the head knowledge point vector, the tail knowledge point vector and the relationship vector into the same relationship space, and optimize the parameters so that the projected head knowledge point vector plus the projected relationship vector are closer to the tail knowledge point vector in the same relationship space.

[0017] Furthermore, the head knowledge point vector, the tail knowledge point vector and the relationship vector are projected into the same relationship space, which are represented as h r and t r , where h r =hM r and t r =tM rThe projected head knowledge point vector plus the projected relationship vector in the same relationship space should be closer to the tail knowledge point vector. The quality is calculated by the following formula:

[0018] f r (h,r,t)=|h r +rt r |

[0019] In order to make the real relationship score as high as possible and the non-existent relationship score as low as possible, and improve the accuracy of the model, the loss function of the model is set to:

[0020]

[0021] in is the set of positive sample triplets, is a set of negative sample triplets, the positive sample triplets are real knowledge point triplets from the knowledge point knowledge graph; the negative sample triplets are false knowledge point triplets generated based on the real positive sample triplets; h' r , t' r represents the projection of the head knowledge point and the tail knowledge point in the relational projection space; h r , r, t r They represent the vector representations of the head knowledge point, relationship, and tail knowledge point in the original space respectively; γ is the margin hyperparameter, which is used to measure the interval between positive samples and negative samples to ensure that the model makes the positive sample score higher than the negative sample score as much as possible;

[0022] The Adam optimizer is then used to minimize the optimization objective and update the model parameters. In each iteration, a batch of positive and negative sample pairs are randomly selected, their scores are calculated, and the model parameters are updated according to the optimization objective.

[0023] Furthermore, the centrality of each knowledge point is determined in S3, that is, by defining and calculating the adaptability index Score(i,j), where i represents the central entity and j represents the neighbor entity. The adaptability index Score(i,j) includes the similarity metric Sim(i,j) and the structural hole metric Hole(i). The similarity metric reflects the similarity between the central knowledge point i and the neighbor knowledge point j, while the structural hole metric reflects the closeness of the structural relationship between them. The sum of the two multiplications is the value of the adaptability index. The calculation formula of Sim(i,j) is as follows:

[0024] Sim(i,j)=(|N(i)∩N(j)|+1) / |N(i)∪N(j)|

[0025] Among them, N(i) and N(j) represent the number of neighbor nodes of knowledge point node i and knowledge point node j respectively, |N(i)∩N(j)| represents the number of common neighbors of two entity nodes, |N(i)∪N(j)| represents the sum of neighbor nodes of node i and node j, and the addition of one in (|N(i)∩N(j)|+1) is to avoid the situation where the adaptability is zero when there are no common neighbors between two entity nodes.

[0026] Furthermore, the probability of acquiring the knowledge points in S4 is to use the temporal convolutional knowledge tracking ATCKT model with an integrated attention mechanism to judge the learner's knowledge mastery and predict the probability of the user's correct answers to all exercises corresponding to the knowledge points that have not been mastered.

[0027] Furthermore, the calculation formula of the importance in S5 is:

[0028] important=a*Score(i,j)+(1-a)*probability, where probability is the learning probability value in S4, and the parameter a in the formula is manually assigned and adjusted appropriately according to the learner's preference.

[0029] Furthermore, in S5, the Dijkstra algorithm is used to calculate the weighted shortest path from each mastered knowledge point to the target knowledge point in the personalized knowledge point directed graph G, and the path with the smallest weight among all weighted shortest paths is selected to construct the user's educational knowledge graph.

[0030] A differentiated educational knowledge graph construction system, including:

[0031] Data storage module, used to store learner information, knowledge point set, educational resource information and their attribute relationship information;

[0032] Backend service layer: The backend service layer is responsible for processing the data and requests transmitted from the frontend, and performing corresponding calculations and processing. It should include: data acquisition function, which is mainly responsible for obtaining historical data such as the user's historical answer status; knowledge graph construction function, that is, building a complete knowledge graph and using TransR to complete the knowledge graph. The relationships between them include knowledge points, concepts, and associations of various disciplines to support the system's deep understanding and recommendation of knowledge; personalized recommendation function, integrating personalized recommendation algorithms, and using the ATCKT model to recommend suitable knowledge content to learners based on their personalized characteristics and behavior data;

[0033] The front-end display layer includes the user login page, data input interface, knowledge graph display interface, knowledge point viewing and question answering page.

[0034] A computer storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the above-mentioned method for constructing an educational knowledge graph that supports differentiation is completed.

[0035] An electronic device includes a memory, a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, the method for constructing an educational knowledge graph that supports differentiation as described above is completed.

[0036] Compared with the prior art, the present invention has the following advantages:

[0037] The design of the recommendation model in the present invention aims at the problem of relationship sparsity in the knowledge graph composed of knowledge points and the relationships between knowledge points. The use of the TransR model to complete the knowledge graph can more accurately capture the semantic information of the relationship, predict the complex relationship between knowledge points, improve the accuracy of relationship reasoning and the generalization ability of the model, thereby improving the quality of the knowledge graph of knowledge points and the effect of completion. In particular, in view of the differences between learners, the use of the ATCKT model can better judge the learners' knowledge mastery, capture the learning characteristics and behavior patterns of each learner, and achieve higher prediction accuracy and personalized learning guidance.

[0038] Through testing and verification on multi-type and multi-source data sets, the accuracy, precision and AUC value of the present invention have been significantly improved compared with traditional models.

[0039] The present invention helps to provide learners with a personalized education recommendation system, assists learners to better learn target knowledge points, provides personalized learning guidance, and improves learners' learning efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 Flowchart of the method for building a knowledge graph to support differentiation in education;

[0042] Figure 2 This is a sample diagram of the knowledge graph of advanced mathematics course knowledge points;

[0043] Figure 3 This is the design architecture diagram of the ATCKT model;

[0044] Figure 4It is the architecture diagram of the residual block;

[0045] Figure 5 Construct a system architecture diagram for supporting learners’ differentiated educational knowledge graph. DETAILED DESCRIPTION

[0046] The technical scheme in this embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in this embodiment of the present invention. Obviously, the described embodiment is one embodiment of the present invention, not all embodiments of the present invention. Based on this embodiment of the present invention, all other embodiments of the present invention obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as described in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0048] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features.

[0049] The embodiments of the present invention provide a method, system, medium and device for constructing a differentiated educational knowledge graph, as follows:

[0050] A method for constructing an educational knowledge graph that supports learner differentiation, including:

[0051] S1: Collect learner information and knowledge point information, and build a knowledge graph of knowledge points;

[0052] S2: predict and complete the knowledge points in the knowledge graph and the complex relationships between the knowledge points;

[0053] S3: Determine the centrality of each knowledge point in the knowledge graph;

[0054] S4: Calculate the probability of the learner's knowledge point acquisition based on the learner's historical answer data;

[0055] S5: The importance of each knowledge point is calculated by comprehensively calculating the centrality of the knowledge point and the probability of knowledge point acquisition. The user's learning path is optimized according to the importance to obtain the learning path with the minimum weight, and the user's educational knowledge graph is constructed with the shortest path.

[0056] The steps are as follows Figure 1The following is a detailed description of each step:

[0057] Step S1: Construct knowledge graph of knowledge points

[0058] The knowledge point knowledge graph is a representation that organizes learning resources into a graphical network. The knowledge point knowledge graph uses the knowledge points that the user has learned and is yet to master as nodes and the relationships between the knowledge points as edges, reflecting the logical relationships and connections between different knowledge points. First, various knowledge points involved are manually collected from resources such as textbooks, academic papers, course outlines, and online education platforms, and attribute information (definition of knowledge points, related concepts, related theorems or formulas, examples, etc.) is determined for each knowledge point. Then, the relationships between the collected knowledge points are analyzed to determine the relationships between the knowledge points, including prerequisite relationships, dependency relationships, similarity relationships, etc.

[0059] By building a knowledge graph of knowledge points, students can more clearly understand the relationship between knowledge points, discover the difficulty of knowledge and the learning path, and provide support for personalized learning and intelligent recommendations. Figure 2 For example, the relationship between different knowledge points in mathematics is shown. The connections between nodes in the knowledge graph represent the relationship between these knowledge points, such as prerequisite relationships, dependency relationships, and application relationships. For example, linear equations and quadratic equations are the basis of equations and inequalities, systems of equations depend on linear equations and quadratic equations, and polynomial equations are derived from systems of equations, etc. Such a knowledge graph can help students better understand the structure and relationship of knowledge, and help them learn and master the knowledge of algebra in a more systematic way.

[0060] Step S2: Predict and complete the knowledge points in the knowledge point knowledge graph and the complex relationships between the knowledge points:

[0061] In this embodiment, the TransR model is used to complete the knowledge graph, aiming to solve some problems in the traditional knowledge graph representation learning method, such as the diversity and complexity between entities and relationships. The solution is to improve the completeness and accuracy of the knowledge graph by filling or predicting the missing knowledge points and the relationships between the knowledge points. The main idea is to project the relationships between knowledge points into different spaces to better capture the semantic relationships between knowledge points.

[0062] First, the knowledge points and the relationships between them are encoded and mapped to integer indices to construct a relationship graph between knowledge points, namely, a knowledge graph triple (h, r, t), where (h, r, t) represents the head knowledge point, the relationship, and the tail knowledge point, respectively. The mapping relationship is h, t∈Z |E| and r∈Z |R|, where |E| represents the number of knowledge points, |R| represents the number of relations, and Z is an integer index set. Then randomly initialize the knowledge point vector Relationship Vector and the projection matrix where d e and d r Represent the dimensions of entity vector and relationship vector respectively.

[0063] For each relationship, use the projection matrix M r , project the head knowledge point vector, tail knowledge point vector and relationship vector into the same relationship space, denoted as h r and t r , where h r =hM r and t r =tM r The projected head knowledge point vector plus the projected relationship vector in the same relationship space should be closer to the tail knowledge point vector, and its quality can be calculated by the following formula:

[0064] f r (h,r,t)=|j r +rt r |

[0065] In order to make the real relationship score as high as possible and the non-existent relationship score as low as possible, and improve the accuracy of the model, the loss function of the model is set to:

[0066]

[0067] in is the set of positive sample triplets, is a set of negative sample triplets, and the positive sample triplets are real knowledge point triplets from the knowledge graph. Negative sample triplets are false knowledge point triplets generated based on real positive sample triplets. h' r , t' r represents the projection of the head knowledge point and the tail knowledge point in the relational projection space, h r , r, t r They represent the vector representation of the head knowledge point, relationship, and tail knowledge point in the original space respectively. γ is a margin hyperparameter, which is used to measure the interval between positive samples and negative samples. It can ensure that the model makes the positive sample score higher than the negative sample score as much as possible, thereby improving the accuracy and performance of the model.

[0068] The Adam optimizer is then used to minimize the optimization objective and update the model parameters. In each iteration, a batch of positive and negative sample pairs are randomly selected, their scores are calculated, and the model parameters are updated according to the optimization objective.

[0069] Use the existing knowledge graph data and TransR model to train through the above steps, and then use the trained model to predict and complete the relationship between knowledge points. Through the above steps, a TransR model is obtained to complete the knowledge graph of knowledge points. In this way, a more complete and accurate knowledge graph of knowledge points can be obtained, which helps to deeply understand the relationship and structure between knowledge.

[0070] Step S3: Calculate and determine the centrality of each knowledge point in the knowledge graph

[0071] By analyzing the connection relationship between knowledge point nodes in the knowledge graph, especially the connection between the central knowledge point node and its neighboring knowledge point nodes, the centrality of each knowledge point is determined to improve the efficiency and quality of the recommendation results.

[0072] In order to determine the centrality of each knowledge point, an adaptability index Score(i,j) is proposed, where i represents the central knowledge point node and j represents its first-order neighbor knowledge point. This adaptability index is used to measure the importance of the neighbor knowledge point node to the central knowledge point node, that is, the degree of its influence on the central knowledge point. If a neighbor knowledge point node has a greater influence on the central knowledge point node, the higher the value of its adaptability index, otherwise it will be lower.

[0073] The calculation of the adaptability index Score(i,j) adopts a formula, which is as follows:

[0074]

[0075] In this formula, N(i) is the neighbor of node i, which includes two factors, namely the similarity metric Sim(i,j) and the structural hole metric Hole(i). The similarity metric is used to measure the similarity between the central knowledge point and the neighbor knowledge point, while the structural hole metric is used to measure the closeness of the structural relationship between them. The product of these two factors is the value of the adaptability index. The calculation formula of Sim(i,j) is as follows:

[0076] Sim(i,j)=(|N(i)∩N(j)|+1) / |N(i)∪N(j)|

[0077] In this formula, N(i) and N(j) represent the number of neighbor nodes of knowledge point node i and knowledge point node j respectively, |N(i)∩N(j)| represents the number of common neighbors of two entity nodes, |N(i)∪N(j)| represents the sum of neighbor nodes of node i and node j, and the "+1" in (|N(i)∩N(j)|+1) is to avoid the situation where the adaptability is zero when there are no common neighbors between two entity nodes. Hole(i) represents the weight of the entity node, and the structural hole index is used to calculate the node weight. The formula is as follows:

[0078] Hole(i)=(Σ(j=1) N (d i -d j )) / (N-1)

[0079] Among them, d i Indicates the degree of knowledge point node i, d j Indicates the degree of other knowledge point nodes j in the connected component. The larger the structural hole index, the more special or important the role of the entity knowledge point node in the graph structure. A knowledge point node with a low structural hole index may mean that the role of the knowledge point node in the graph structure is relatively small or relatively ordinary, it is connected to fewer other knowledge point nodes in the graph structure, is in a marginal position, or the link method is relatively localized, and only plays a role in a specific field or topic.

[0080] Step S4: Calculate the learner's knowledge point acquisition probability based on the learner's historical answer data

[0081] In this invention, the temporal convolutional knowledge tracking (ATCKT) with fusion attention mechanism is used to judge the knowledge mastery of learners. ATCKT uses a neural network model to more accurately predict the knowledge mastery of students in the learning process. Compared with traditional knowledge tracking methods, ATCKT can better capture students' learning characteristics and behavior patterns, and achieve higher prediction accuracy and personalized learning guidance. The following is a detailed description of the ATCKT model in this system:

[0082] The specific process of the ATCKT model is as follows Figure 3 As shown in the figure, in this system, the input of the ATCKT model is the user's historical answer sequence, represented as a one-dimensional array X, where x i for (q i ,0 / 1), q is the question, q i It represents the question selected by the user in step i, and the second item is correct / wrong. The output of the model is the predicted result of the next answering action for each user.

[0083] First, input the historical answer sequence X and randomly initialize the question embedding matrix E K×Q), and each row vector e i Represents a question q i The embedding vectors of the questions are K, Q and Q, respectively. During the training of the model, these embedding vectors e will automatically encode the information of the questions through learning.

[0084] Then the unanswered questions will be t and questions that the user has answered before i Perform a linear transformation:

[0085] Q=e t W query

[0086] A=e i W answer ,i∈(1,T)

[0087] Where W query ∈R( K×K ),W answer ∈R( K×K ), these two transformation matrices are used to transform e t ,e i Then, do a dot product operation on Q and A, and use the softmax activation function to normalize to get the attention weight α. At the same time, concatenate the historical answer data H and convert the behavior h of step i into i Expressed as (q,h), q i is the code of the question answered in step i, h i Is the answer to the question in step i correct? If correct, it is (1,...,1), if wrong, it is (0,...,0), its length is K, and then the attention weight α is used i For historical interaction data h i The weighted representation of the user's historical comprehensive performance P i .

[0088] Then through the user's historical comprehensive performance P i Enter the temporal convolutional network (TCN) to extract the user's knowledge state at each moment. Each convolutional layer of TCN has a fixed-size sliding window for performing convolution operations on the input sequence. The residual block structure in the model is as follows: Figure 4 As shown, it includes: causal dilated convolution, layer normalization to normalize all neurons in the layer, ReLU activation function and Dropout layer to prevent overfitting.

[0089] TCN extracts the user's knowledge state matrix S∈R at each moment N×K , N represents the number of questions the user has answered. t ∈R K, which indicates the student's mastery of knowledge at that moment, and the student's knowledge state s t The question to be answered at the next moment t+1 Perform a dot product operation to get y t+1 , then for y t+1 The sigmoid activation function is used to get the probability of answering the next question correctly. The model outputs the probability of each possible answer option, so the loss function L is defined as the cross entropy loss function, a t is the true value, p t For the predicted value, the formula is as follows:

[0090]

[0091] The model parameters are updated through the back-propagation algorithm and the Adam optimizer to minimize the loss function.

[0092] The above model is used to comprehensively measure students' personalized prior knowledge by training users' historical data of practice answers, that is, the model is used to predict the probability value of users' correct answers to all exercises corresponding to knowledge points that have not been mastered. Since each question only tests one knowledge point, the probability value of the correct answer to the exercise calculated by the model can be equivalently regarded as the probability value (probabilty value) of the knowledge point related to the exercise that the student answers correctly. The larger the probability value of a certain knowledge point, the more sufficient the student's prior knowledge about the knowledge point is, and the more likely it is that the student will spend a shorter time learning and mastering the knowledge point in the near future.

[0093] Step S5: The importance of each knowledge point is calculated by comprehensively calculating the centrality of the knowledge point and the probability value of knowledge point acquisition, and the user's learning path is optimized according to the importance to obtain the learning path with the minimum weight, and the user's educational knowledge graph is constructed with the shortest path.

[0094] The calculation method of the importance (important value) of each knowledge point in the knowledge graph is shown in the following formula:

[0095] important=a*Score(i,j)+(1-a)*probability

[0096] The Score(i,j) value is obtained by calculating the knowledge point centrality using the algorithm in step 3, and the probability value is obtained by calculating the probability of knowledge point acquisition using the ATCKT model. The parameter a in the formula is manually assigned and can be adjusted appropriately based on the learner's preferences. For example, some students have limited energy and prefer to achieve the target knowledge point in a shorter period of time, so the system can appropriately reduce the student's a value; or some students want to learn more knowledge points and have a more complete knowledge system, so their a value can be increased. Experimental results show that the best effect is when a is 0.5, so a is set to 0.5 in this article.

[0097] Calculate the knowledge point v j After the important value is obtained, it is used as a directed edge in the directed graph of personalized knowledge points. <v i ,v j >, and then use the Dijkstra algorithm to calculate the weighted shortest path from each mastered knowledge point to the target knowledge point in the personalized knowledge point directed graph G, and select the path with the smallest weight among all weighted shortest paths to build the user's knowledge graph. In the learning process, the paths recommended to students are important and easy for learners to master, so that learners can learn quickly, get feedback, and complete the learning of target knowledge along the shortest path.

[0098] An embodiment of the present invention also provides a differentiated education knowledge graph construction system, including: realizing user input data, calculation processing and result display through front-end and back-end interaction and data storage, providing users with efficient and accurate education knowledge recommendation services. Its main architecture is as follows Figure 5 As shown, including:

[0099] Data storage module: The data storage layer can store various data required for system operation, such as learner information, knowledge point collection, educational resource information, and its attribute relationship information;

[0100] The back-end service layer is responsible for processing the data and requests transmitted from the front-end, and performing corresponding calculations and processing; it should include: data acquisition function, which is mainly responsible for obtaining historical data such as the user's historical answers; knowledge graph construction function, that is, building a complete knowledge graph and using TransR to complete the knowledge graph. The relationships include knowledge points, concepts, and associations of various disciplines to support the system's deep understanding and recommendation of knowledge; personalized recommendation function, integrating personalized recommendation algorithms, and using the ATCKT model to recommend suitable knowledge content to learners based on their personalized characteristics and behavioral data.

[0101] Front-end display layer: The front-end display layer is the interface for users to interact with the system. Its main functions include user login page, data input interface, knowledge graph display interface, knowledge point viewing, question answering page, etc. The front-end interface is concise and clear, easy to use, able to attract users and provide a good user experience; it should also support various forms of knowledge presentation, including text, pictures, videos, audio, etc., to meet the learning habits and needs of different learners.

[0102] An embodiment of the present invention also provides a computer storage medium for storing computer instructions, which, when executed by a processor, complete the above-mentioned method for constructing an educational knowledge graph that supports differentiation.

[0103] An embodiment of the present invention also provides an electronic device, including a memory, a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, the method for constructing an educational knowledge graph that supports differentiation as described above is completed.

[0104] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for constructing a differentiated educational knowledge graph, characterized in that: include: S1: Collect learner information and knowledge point information, and build a knowledge graph of knowledge points; S2: predict and complete the knowledge points in the knowledge graph and the complex relationships between the knowledge points; S3: Determine the centrality of each knowledge point in the knowledge graph; S4: Calculate the probability of the learner's knowledge point acquisition based on the learner's historical answer data; S5: The importance of each knowledge point is calculated by comprehensively calculating the centrality of the knowledge point and the probability of knowledge point acquisition. The user's learning path is optimized according to the importance to obtain the learning path with the minimum weight, and the user's educational knowledge graph is constructed with the shortest path.

2. The method for constructing a differentiated educational knowledge graph according to claim 1, characterized in that: The knowledge point knowledge graph in S1 uses the TransR model to predict and complete the complex relationships between knowledge point entities and knowledge point entities; The above-mentioned TtansR model is: First, the knowledge points and the relationships between them are encoded and mapped to integer indices to construct a relationship graph between knowledge points, namely, a knowledge graph triple (h, r, t), where (h, r, t) represents the head knowledge point, the relationship, and the tail knowledge point, respectively. The mapping relationship is h, t∈Z |E| and r∈Z |R| , where |E| represents the number of knowledge points, |R| represents the number of relations, and Z is an integer index set; Then define and randomly initialize the knowledge point vector Relationship Vector and the projection matrix where d e and d r Represent the dimensions of entity vector and relationship vector respectively; For each relationship, use the projection matrix M r , project the head knowledge point vector, the tail knowledge point vector and the relationship vector into the same relationship space, and optimize the parameters so that the projected head knowledge point vector plus the projected relationship vector are closer to the tail knowledge point vector in the same relationship space.

3. The method for constructing a differentiated educational knowledge graph according to claim 2, characterized in that: The head knowledge point vector, the tail knowledge point vector and the relationship vector are projected into the same relationship space and are represented as h r and t r , where h r =hM r and t r =tM r The projected head knowledge point vector plus the projected relationship vector in the same relationship space should be closer to the tail knowledge point vector. The quality is calculated by the following formula: f r (h,r,t)=|h r +r-t r | In order to make the real relationship score as high as possible and the non-existent relationship score as low as possible, and improve the accuracy of the model, the loss function of the model is set to: in is the set of positive sample triplets, is a set of negative sample triplets, the positive sample triplets are real knowledge point triplets from the knowledge point knowledge graph; the negative sample triplets are false knowledge point triplets generated based on the real positive sample triplets; h' r , t' r represents the projection of the head knowledge point and the tail knowledge point in the relational projection space; h r , r, t r They represent the vector representations of the head knowledge point, relationship, and tail knowledge point in the original space respectively; γ is the margin hyperparameter, which is used to measure the interval between positive samples and negative samples to ensure that the model makes the positive sample score higher than the negative sample score as much as possible; The Adam optimizer is then used to minimize the optimization objective and update the model parameters. In each iteration, a batch of positive and negative sample pairs are randomly selected, their scores are calculated, and the model parameters are updated according to the optimization objective.

4. The method for constructing a differentiated educational knowledge graph according to claim 1, characterized in that: The centrality of each knowledge point is determined in S3, that is, by defining and calculating the adaptability index Score(i,j), where i represents the central entity and j represents the neighbor entity. The adaptability index Score(i,j) includes the similarity metric Sim(i,j) and the structural hole metric Hole(i). The similarity metric reflects the similarity between the central knowledge point i and the neighbor knowledge point j, while the structural hole metric reflects the closeness of the structural relationship between them. The sum of the two multiplies is the value of the adaptability index. The calculation formula of Sim(i,j) is as follows: Sim(i,j)=(|N(i)∩N(j)|+1) / |N(i)∪N(j)| N(i) and N(j) represent the number of neighbor nodes of knowledge point node i and knowledge point node j respectively, |N(i)∩N(j)| represents the number of common neighbors of two entity nodes, |N(i)∪N(j)| represents the sum of neighbor nodes of node i and node j, and the addition of one in (|N(i)∩N(j)|+1) is to avoid the situation where the adaptability is zero when there are no common neighbors between two entity nodes.

5. The method for constructing a differentiated educational knowledge graph according to claim 4, characterized in that: The probability of acquiring knowledge points in S4 is to use the temporal convolutional knowledge tracking ATCKT model with an integrated attention mechanism to judge the learner's knowledge mastery and predict the probability of the user correctly answering the exercises corresponding to all the knowledge points that have not been mastered.

6. The method for constructing a differentiated educational knowledge graph according to claim 5, characterized in that: The calculation formula of the importance in S5 is: important=a*Score(i,j)+(1-a)*probability, where probability is the learning probability value in S4, and the parameter a in the formula is manually assigned and adjusted appropriately according to the learner's preference.

7. The method for constructing a differentiated educational knowledge graph according to claim 6, characterized in that: In S5, the Dijkstra algorithm is used to calculate the weighted shortest path from each mastered knowledge point to the target knowledge point in the personalized knowledge point directed graph G, and the path with the smallest weight among all weighted shortest paths is selected to construct the user's educational knowledge graph.

8. A differentiated educational knowledge graph construction system, characterized in that: include: Data storage module, used to store learner information, knowledge point set, educational resource information and their attribute relationship information; Backend service layer: The backend service layer is responsible for processing the data and requests transmitted from the frontend, and performing corresponding calculations and processing; it should include: data acquisition function, which is mainly responsible for obtaining historical data such as the user's historical answer status; knowledge graph construction function, that is, building a complete knowledge graph and using TransR to complete the knowledge graph. The relationships between them include knowledge points, concepts, and associations of various disciplines to support the system's deep understanding and recommendation of knowledge; personalized recommendation function, integrating personalized recommendation algorithms, and using the ATCKT model to recommend suitable knowledge content to learners based on their personalized characteristics and behavior data; The front-end display layer includes the user login page, data input interface, knowledge graph display interface, knowledge point viewing and question answering page.

9. A computer storage medium, characterized in that Used to store computer instructions, which, when executed by a processor, complete any one of the methods for constructing a differentiated educational knowledge graph in claims 1-7.

10. An electronic equipment device, characterized in that: It includes a memory, a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, the method for constructing an educational knowledge graph that supports differentiation as described in any one of claims 1 to 7 is completed.

Images