Digital teaching material intelligent generation system based on digital education

Abstract

The invention relates to the technical field of digital education, in particular to a digital teaching material intelligent generation system based on digital education, which comprises an education resource acquisition module, a knowledge graph construction module, a teaching material content generation module, a knowledge tracking module and a teaching material presentation module, according to the method, a graph embedding method based on a hyperbolic space graph convolutional network and a hierarchical clustering algorithm with improved robustness are introduced, low-dimensional embedding and multilayer structure optimization are performed on an initial knowledge graph, and an educational knowledge graph with a semantic hierarchical structure is constructed; a graph structured learning behavior modeling method is introduced, a student-knowledge point heterogeneous graph structure is constructed, a heterogeneous graph convolutional network is adopted to model interaction behaviors of students and knowledge points, a knowledge state graph of each student is obtained, and the accuracy of individual knowledge mastery degree modeling is improved.

Description

Technical Field

[0001] The present invention relates to the field of digital education technology, and specifically refers to an intelligent digital textbook generation system based on digital education. Background Art

[0002] With the rapid development of information technology, digital education has gradually become an important development direction in the education industry. The static content and fixed structure of traditional textbooks can no longer meet the needs of personalized teaching and intelligent education. However, in the existing digital textbook generation process, there are still problems such as limited educational resource processing capabilities, single knowledge graph structure, weak semantic hierarchy, and inaccurate tracking of students' knowledge states. Most existing systems rely on manual extraction and structural templates, unable to efficiently integrate multi-source heterogeneous teaching resources, lacking the ability of in-depth semantic understanding and structured organization of resource content. Currently, knowledge graphs are mostly constructed using shallow point-to-point association methods, failing to build a multi-layer knowledge graph that reflects the semantic level and abstract structure of educational content, restricting the systematic organization and reasoning application of knowledge. Current learning tracking methods are mostly based on answering data statistics, lacking in-depth analysis based on graph-structured behavior modeling, and it is difficult to comprehensively and accurately depict students' mastery of each knowledge point, resulting in inaccurate textbook recommendations and personalized presentations. Summary of the Invention

[0003] In view of the above situation, to overcome the defects of the prior art, the present invention provides an intelligent digital textbook generation system based on digital education. To address the problem of limited educational resource processing capabilities, the present invention uses an educational resource collection module to perform semantic parsing and entity extraction on teaching content data through natural language processing (NLP) and image recognition technologies, constructing a semantic metadata set to achieve efficient integration and semantic unit representation of multi-source educational resources. To address the problems of single knowledge graph structure and weak semantic hierarchy, the present invention introduces a graph embedding method based on hyperbolic space graph convolutional network and a robust improved hierarchical clustering algorithm to perform low-dimensional embedding and multi-layer structure optimization on the initial knowledge graph, constructing an educational knowledge graph with a semantic hierarchical structure to enhance the systematic expression and hierarchical understanding capabilities between knowledge. To address the problem of inaccurate tracking of students' knowledge states, the present invention introduces a graph-structured learning behavior modeling method, constructs a student-knowledge point heterogeneous graph structure, and uses a heterogeneous graph convolutional network to model the interaction behavior between students and knowledge points, obtaining a knowledge state graph for each student to improve the accuracy of individual knowledge mastery modeling.

[0004] The technical solution adopted by the present invention is as follows: The intelligent digital textbook generation system based on digital education provided by the present invention includes an educational resource collection module, a knowledge graph construction module, a textbook content generation module, a knowledge tracking module, and a textbook presentation module, specifically including the following:

[0005] The education resource collection module collects teaching content data, uses NLP and image recognition technologies to perform semantic parsing and entity extraction on the teaching content data, obtains education semantic units, and constructs a semantic metadata set;

[0006] The knowledge graph construction module performs structured processing on the semantic metadata set to construct an initial knowledge graph, which includes nodes and edges. An embedding method based on a hyperbolic space graph convolutional network is introduced to perform low-dimensional representation learning on the initial knowledge graph to obtain an education knowledge graph with a semantic hierarchical structure;

[0007] The teaching material content generation module is based on the education knowledge graph with a semantic hierarchical structure, and uses a graph structure expansion and semantic reasoning method to cluster relevant knowledge points and expand paths for the target teaching topic, and generates corresponding teaching material content units;

[0008] The knowledge tracking module introduces a graph-structured learning behavior modeling method, and obtains the knowledge state graph of each student by constructing a student-knowledge point heterogeneous graph structure;

[0009] The teaching material presentation module provides a visual teaching material content presentation method based on the teaching material content unit and the education knowledge graph with a semantic hierarchical structure, in combination with the knowledge state graph of each student.

[0010] Furthermore, in the education resource collection module, the teaching content data includes textbook originals, teaching plans, teacher's explanations, question bank materials, online course content, and multimedia resources, and the education semantic units include concept entities, knowledge points, teaching objectives, and topic labels;

[0011] Furthermore, in the teaching material content generation module, the teaching material content units include concept explanations, example explanations, charts, and exercise questions.

[0012] Furthermore, the knowledge graph construction module specifically includes the following steps:

[0013] Step S1: Structured processing, clarify and standardize the semantic metadata, map the education semantic units to nodes, and establish edges according to the relationships between the various education semantic units. Each edge represents the semantic connection between different nodes;

[0014] Step S2: Construct the initial knowledge graph, and construct an initial knowledge graph of knowledge points - teaching objectives - concept entities according to the nodes and edges;

[0015] Step S3: Embedding based on the hyperbolic space graph convolutional network, introduce the hyperbolic space graph convolutional network to perform low-dimensional representation learning on the initial knowledge graph, embed the nodes into the hyperbolic space, and obtain the embedding vector representation of each node;

[0016] Step S4: Calculation Structure. Based on the embedded vector representations of each node, use a hierarchical clustering algorithm improved based on robustness to cluster educational semantic units and form a multi-level semantic grouping structure;

[0017] Step S5: Structure Optimization. Jointly evaluate the similarity between nodes and the original connection relationship, and use a hierarchical adjustment algorithm to adjust the connection method of edges to obtain an educational knowledge graph with a semantic hierarchical structure.

[0018] Furthermore, in Step S4, when using a hierarchical clustering algorithm improved based on robustness to cluster educational semantic units, it specifically includes the following steps:

[0019] Step S41: Denoising Processing. Conduct density analysis on the embedded vector representations of each educational information unit, use a density-based noise discrimination mechanism to identify and remove abnormal samples, and obtain a set of denoised educational information units;

[0020] Step S42: Construct a k-nearest neighbor directed graph. Based on the set of denoised educational information units, construct a k-nearest neighbor directed graph, map each educational information unit to a node in the graph, and determine its k nearest neighbor nodes according to the Euclidean distance between the embedded vectors, connect to form directed edges, and set the weights of the directed edges according to the relative distance between nodes;

[0021] Step S43: Graph Merging and Clustering. Group nodes with close embedded vector distances into one category to form clusters, and multiple clusters are constructed to form a semantic hierarchical structure. Iteratively traverse the directed edges in the k-nearest neighbor directed graph in the order of the weights of the directed edges, and perform graph merging operations until all directed edges are traversed;

[0022] Step S44: Determination of the Number of Clusters. During the graph merging process in Step S43, record the changes in the number of clusters after each graph merging operation, construct a mapping sequence of the iteration rounds and the number of clusters, count the merging iteration times corresponding to each number of clusters, and select the number of clusters corresponding to the most merging iteration times as the final clustering result;

[0023] Step S45: Construction of Semantic Abstract Representation. Perform mean pooling processing on the embedded vectors of educational information units within each cluster to obtain the semantic center vector of the cluster, which serves as the abstract semantic representation of the current hierarchical cluster and identifies the common features of this type of educational semantic unit;

[0024] Step S46: Generation of Multi-level Structure. Recursively execute Step S41 - Step S45 using a top-down hierarchical strategy to construct a multi-level semantic hierarchical structure.

[0025] Furthermore, the knowledge tracking module specifically includes the following steps:

[0026] Step Q1: Learning behavior data collection. Collect the learning behavior data of students, where the learning behavior data includes answering records, homework submission situations, learning resource access frequencies, and video learning durations.

[0027] Step Q2: Construction of student-knowledge point heterogeneous graph. According to the collected learning behavior data, construct a heterogeneous graph structure composed of student nodes and knowledge point nodes. Each student is represented as a student node, and each knowledge point is represented as a knowledge point node. Directed edges in the heterogeneous graph are generated based on the interaction behaviors between students and knowledge points.

[0028] Step Q3: Graph embedding representation learning. Use a heterogeneous graph convolutional network to perform node representation learning on the student-knowledge point heterogeneous graph. Combining graph structure information and node attribute information, map each node to a low-dimensional embedding vector.

[0029] Step Q4: Calculate the knowledge mastery scores of each student on different knowledge points based on the low-dimensional embedding vectors and the embedding features of their adjacent knowledge point nodes, and form a knowledge state graph of the students.

[0030] The beneficial effects achieved by the present invention using the above solution are as follows:

[0031] (1) Aiming at the problem of limited educational resource processing capabilities, the present invention uses an educational resource collection module to perform semantic parsing and entity extraction on teaching content data using natural language processing (NLP) and image recognition technologies, construct a semantic metadata set, and achieve efficient integration and semantic unit representation of multi-source educational resources.

[0032] (2) Aiming at the problems of single knowledge graph structure and weak semantic hierarchy, the present invention introduces a graph embedding method based on a hyperbolic space graph convolutional network and a robustly improved hierarchical clustering algorithm to perform low-dimensional embedding and multi-layer structure optimization on the initial knowledge graph, construct an educational knowledge graph with a semantic hierarchical structure, and enhance the systematic expression and hierarchical understanding ability between knowledge.

[0033] (3) Aiming at the problem of inaccurate tracking of students' knowledge states, the present invention introduces a graph-structured learning behavior modeling method, constructs a student-knowledge point heterogeneous graph structure, uses a heterogeneous graph convolutional network to model the interaction behaviors between students and knowledge points, obtains a knowledge state graph of each student, and improves the accuracy of individual knowledge mastery modeling. Description of the Drawings

[0034] Figure 1 It is a block diagram of a digital textbook intelligent generation system based on digital education proposed by the present invention.

[0035] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. Detailed Embodiments

[0036] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0037] Embodiment 1. Refer to Figure 1 , the digital textbook intelligent generation system based on digital education provided by the present invention includes an educational resource collection module, a knowledge graph construction module, a textbook content generation module, a knowledge tracking module, and a textbook presentation module, and specifically includes the following:

[0038] The educational resource collection module collects teaching content data, uses NLP and image recognition technologies to perform semantic parsing and entity extraction on the teaching content data, obtains educational semantic units, and constructs a semantic metadata set;

[0039] The knowledge graph construction module performs a structured process on the semantic metadata set, constructs an initial knowledge graph, including nodes and edges, and introduces an embedding method based on a hyperbolic space graph convolutional network to perform low-dimensional representation learning on the initial knowledge graph, obtaining an educational knowledge graph with a semantic hierarchical structure;

[0040] The textbook content generation module is based on the educational knowledge graph with a semantic hierarchical structure, and uses a graph structure expansion and semantic reasoning method to perform relevant knowledge point clustering and path expansion on the target teaching topic, generating corresponding textbook content units;

[0041] The knowledge tracking module introduces a graph-structured learning behavior modeling method, and obtains the knowledge state graph of each student by constructing a student-knowledge point heterogeneous graph structure;

[0042] The textbook presentation module is based on the textbook content units and the educational knowledge graph with a semantic hierarchical structure, and combines the knowledge state graph of each student to provide a visual textbook content presentation method;

[0043] In the educational resource collection module, the teaching content data includes textbook originals, teaching lesson plans, teacher explanations, question bank materials, online course content, and multimedia resources, and the educational semantic units include concept entities, knowledge points, teaching objectives, and topic labels;

[0044] In the teaching material content generation module, the teaching material content unit includes concept explanations, example explanations, charts, and practice questions.

[0045] Embodiment 2, based on the above embodiment, the knowledge graph construction module specifically includes the following steps:

[0046] Step S1: Structured processing, clarify and standardize the semantic metadata, map the educational semantic units to nodes, and establish edges according to the relationships between the educational semantic units. Each edge represents the semantic connection between different nodes;

[0047] Step S2: Construct an initial knowledge graph, and construct an initial knowledge graph of knowledge points - teaching objectives - concept entities according to the nodes and edges;

[0048] Step S3: Embedding based on the hyperbolic space graph convolutional network, introduce the hyperbolic space graph convolutional network to perform low-dimensional representation learning on the initial knowledge graph, embed the nodes into the hyperbolic space, and obtain the embedded vector representation of each node;

[0049] Step S4: Calculate the structure, based on the embedded vector representation of each node, use a hierarchical clustering algorithm improved based on robustness to cluster the educational semantic units to form a multi-level semantic grouping structure;

[0050] Step S5: Structure optimization, jointly evaluate the similarity between nodes and the original connection relationship, and use a hierarchical adjustment algorithm to adjust the connection method of the edges to obtain an educational knowledge graph with a semantic hierarchical structure.

[0051] In this embodiment, in the teaching topic of "junior high school mathematics - linear function", the application process of the knowledge graph construction module is demonstrated, specifically including the following steps:

[0052] Structured processing: The educational semantic units include: linear function, function graph, function expression, slope, intercept, monotonicity of the function, image and practical problem modeling, teaching objective A (master the form of the linear function expression), teaching objective B (understand the connection between the image and algebra);

[0053] Convert the above semantic units into nodes in the knowledge graph, and identify the semantic connection relationships between them. There is a "has image" relationship between "linear function" and "function graph", an "inclusion" relationship between "function expression" and "slope", and a "target coverage" relationship between "teaching objective A" and "function expression";

[0054] Construct an initial knowledge graph according to the above nodes and edges;

[0055] Form the following multi-level structure:

[0056] First-level clustering:

[0057] [Linear function]

[0058] Secondary clustering:

[0059] [Function expression, slope, intercept]

[0060] [Function graph, monotonicity of the function]

[0061] Tertiary clustering:

[0062] [Teaching objective A, Teaching objective B]

[0063] A semantic stratification from "core concepts" to "constituent elements" to "teaching objectives" is formed.

[0064] Example 3, this example is based on the above example. In step S4, a hierarchical clustering algorithm based on robustness improvement is used to cluster educational semantic units, which specifically includes the following steps:

[0065] Step S41: Denoising processing. Perform density analysis on the embedded vector representation of each educational information unit, use a density-based noise discrimination mechanism to identify and remove abnormal samples, and obtain a set of denoised educational information units;

[0066] Step S42: Construct a k-nearest neighbor directed graph. Construct a k-nearest neighbor directed graph based on the set of denoised educational information units. Map each educational information unit to a node in the graph, and determine its k nearest neighbor nodes according to the Euclidean distance between the embedded vectors. Connect to form directed edges, and set the weights of the directed edges according to the relative distance between the nodes;

[0067] Step S43: Graph merging and clustering. Group nodes with close embedded vector distances into one class to form clusters. Multiple clusters are constructed to form a semantic hierarchy. Iteratively traverse the directed edges in the k-nearest neighbor directed graph in the order of the weights of the directed edges, and perform graph merging operations until all directed edges are traversed;

[0068] Step S44: Determination of the number of clusters. During the graph merging process in step S43, record the change in the number of clusters after each graph merging operation, construct a mapping sequence of the iteration rounds and the number of clusters, count the merging iteration times corresponding to each number of clusters, and select the number of clusters corresponding to the most merging iteration times as the final clustering result;

[0069] Step S45: Construction of semantic abstract representation. Perform mean pooling processing on the embedded vectors of educational information units within each cluster to obtain the semantic center vector of the cluster, which is used as the abstract semantic representation of the current hierarchical cluster to identify the common features of this type of educational semantic unit;

[0070] Step S46: Generation of a multi-layer structure. Recursively execute Steps S41 - S45 using a top-down hierarchical strategy to construct a multi-level semantic hierarchical structure.

[0071] In this embodiment, the educational information units extracted by the system include the semantic contents corresponding to the following embedding vectors: linear function, function graph, function expression, slope, intercept, increasing and decreasing properties of the function, x-axis intersection point, non-linear function, teaching objective A, teaching objective B;

[0072] Among them, "non-linear function" appears in some semantic contexts, but its relevance to the "linear function" theme is not strong. It is detected as an outlier through a density-based noise discrimination mechanism and is removed;

[0073] Calculate the Euclidean distance for the embedding vectors of any two semantic units:

[0074] "Function expression" and "slope": 0.21;

[0075] "Function expression" and "intercept": 0.25;

[0076] "Function expression" and "linear function": 0.28;

[0077] "Function expression" and "teaching objective A": 0.49;

[0078] Calculate the rest in sequence to obtain a complete distance matrix;

[0079] (2) Determine the 3 nearest neighbors of each node

[0080] Taking "function expression" as an example, the 3 semantic units with the closest embedding vectors to it are:

[0081] Slope (0.21);

[0082] Intercept (0.25);

[0083] Linear function (0.28);

[0084] Add three directed edges pointing to its neighbors to the "function expression" node:

[0085] Function expression → Slope (weight 0.21);

[0086] Function expression → Intercept (weight 0.25);

[0087] Function expression → Linear function (weight 0.28);

[0088] Linear function:

[0089] Function expression (0.28);

[0090] Function graph (0.32);

[0091] Increasing and decreasing nature of the function (0.36);

[0092] Slope:

[0093] Intercept (0.19);

[0094] Function expression (0.21);

[0095] x - axis intersection point (0.33);

[0096] Function graph:

[0097] Linear function (0.32);

[0098] Increasing and decreasing nature of the function (0.29);

[0099] Teaching objective B (0.44);

[0100] Increasing and decreasing nature of the function:

[0101] Function graph (0.29);

[0102] Linear function (0.36);

[0103] x - axis intersection point (0.38);

[0104] x - axis intersection point:

[0105] Intercept (0.27);

[0106] Function expression (0.30);

[0107] Slope (0.33);

[0108] Teaching objective A:

[0109] Teaching objective B (0.20);

[0110] Function graph (0.43);

[0111] Linear function (0.45);

[0112] Teaching objective B:

[0113] Teaching objective A (0.20);

[0114] Increasing and decreasing nature of the function (0.42);

[0115] Intercept (0.48);

[0116] Construct graph structure:

[0117] The system finally constructs a directed graph with 9 nodes, each node having 3 outgoing edges. Each edge represents the nearest neighbor relationship between semantic units in the form of starting point → ending point, and at the same time, the weight of the edge is saved as a distance metric;

[0118] 9 educational semantic units (nodes):

[0119] V = {v1: linear function, v2: function graph, v3: function expression, v4: slope, v5: intercept, v6: monotonicity of function, v7: x-axis intersection point, v8: teaching objective A, v9: teaching objective B};

[0120] Initially, each node is an independent cluster;

[0121] The first round of merging:

[0122] Edge: teaching objective A → teaching objective B (0.20);

[0123] Operation: Merge v8 (teaching objective A) and v9 (teaching objective B) into a cluster C1 = {v8, v9};

[0124] Update the number of clusters: 9 → 8;

[0125] The second round of merging:

[0126] Edge: slope → intercept (0.19);

[0127] Operation: Merge v4 and v5 into a cluster C2 = {v4, v5};

[0128] Update the number of clusters: 8 → 7;

[0129] The third round of merging:

[0130] Edge: function expression → slope (0.21);

[0131] Note: v4 and v5 are already in C2, and the function expression (v3) is added to C2 → C2 = {v3, v4, v5};

[0132] Update the number of clusters: 7 → 6;

[0133] The fourth round of merging:

[0134] Edge: function expression → linear function (0.28);

[0135] v3 is in C2 and v1 has not been merged → merge v1 into C2 → C2 = {v1, v3, v4, v5};

[0136] Update the number of clusters: 6 → 5;

[0137] The 5th round of merging:

[0138] Edge: x-axis intersection point → intercept (0.27);

[0139] v5 is already in C2, v7 is not merged → merge v7 → C2 = {v1, v3, v4, v5, v7};

[0140] Update the number of clusters: 5 → 4;

[0141] The 6th round of merging:

[0142] Edge: Increasing and decreasing nature of the function → function graph (0.29);

[0143] v2, v6 → merged into a new cluster C3 = {v2, v6};

[0144] Update the number of clusters: 4 → 3;

[0145] The 7th round of merging:

[0146] Edge: x-axis intersection point → function expression (0.30);

[0147] v7 and v3 are already in C2, skipped (already in the same cluster);

[0148] The 8th round of merging:

[0149] Edge: Linear function → function graph (0.32);

[0150] v1 ∈ C2, v2 ∈ C3 → merge C2 ∪ C3 → C2 + C3 = {v1, v2, v3, v4, v5, v6, v7};

[0151] Update the number of clusters: 3 → 2;

[0152] Current cluster status:

[0153] C2' = {v1, v2, v3, v4, v5, v6, v7};

[0154] C1 = {v8, v9};

[0155] The 9th round of merging:

[0156] Edge: Teaching objective B → intercept (0.48);

[0157] v9 ∈ C1, v5 ∈ C2' → merge C1 and C2';

[0158] Final clustering: all semantic units are merged into one large cluster;

[0159] The situation where the number of clusters is 3 appears for two consecutive rounds. This is the hierarchical clustering number with the most stable structure and the best ability to express semantic levels. The clustering result is determined to be 3 categories;

[0160] Perform mean pooling on the nodes in each clustering cluster, extract the central embedding vector of the cluster, and use it as the semantic abstraction of the educational semantic unit of this class;

[0161] Cluster A (function concept class): {v1: linear function, v3: function expression, v4: slope, v5: intercept, v7: x-axis intersection point};

[0162] Cluster B (image and change class): {v2: function image, v6: monotonicity of function};

[0163] Cluster C (teaching objective class): {v8: teaching objective A, v9: teaching objective B}.

[0164] Example 4, this example is based on the above example. The knowledge tracking module specifically includes the following steps:

[0165] Step Q1: Learning behavior data collection. Collect students' learning behavior data, where the learning behavior data includes answer records, homework submission situations, learning resource access frequencies, and video learning durations;

[0166] Step Q2: Construction of student-knowledge point heterogeneous graph. According to the collected learning behavior data, construct a heterogeneous graph structure composed of student nodes and knowledge point nodes. Each student is represented as a student node, and each knowledge point is represented as a knowledge point node. Directed edges in the heterogeneous graph are generated based on the interaction behaviors between students and knowledge points;

[0167] Step Q3: Graph embedding representation learning. Use a heterogeneous graph convolutional network to perform node representation learning on the student-knowledge point heterogeneous graph, and combine the graph structure information and node attribute information to map each node to a low-dimensional embedding vector;

[0168] Step Q4: Calculate the knowledge mastery score of each student on different knowledge points according to the low-dimensional embedding vector and the embedding features of its adjacent knowledge point nodes, and form a knowledge state graph of the students.

[0169] In this example, the collected student behavior data:

[0170] Student ID Knowledge point Answer correctly Homework submission Video viewing duration Access frequency A1 K1 1 1 320 5 A1 K2 0 1 180 3 A2 K1 1 0 260 4 A2 K3 0 0 90 2

[0171] Construct the following heterogeneous graph structure:

[0172] Node set = {S1, S2, K1, K2, K3}

[0173] S1 --> K1 (because answering questions + assignments + videos + access ≥ threshold)

[0174] S1 --> K2

[0175] S2 --> K1

[0176] S2 --> K3

[0177] The code for graph embedding representation learning is as follows:

[0178] import torch

[0179] import torch.nn.functional as F

[0180] from torch_geometric.nn import RGCNConv

[0181] class HeteroGNN(torch.nn.Module):

[0182] def __init__(self, in_channels, hidden_channels, out_channels, num_relations):

[0183] super().__init__()

[0184] self.conv1 = RGCNConv(in_channels, hidden_channels, num_relations)

[0185] self.conv2 = RGCNConv(hidden_channels, out_channels, num_relations)

[0186] def forward(self, x, edge_index, edge_type):

[0187] x = F.relu(self.conv1(x, edge_index, edge_type))

[0188] x = self.conv2(x, edge_index, edge_type)

[0189] return x

[0190] z_S1 = [0.8, 0.3, 0.6]

[0191] z_K1 = [0.9, 0.2, 0.7]

[0192] z_K2 = [0.2, 0.6, 0.5]

[0193] def mastery_score(student_embed, kp_embed):

[0194] return torch.sigmoid(torch.dot(student_embed, kp_embed))

[0195] score_S1_K1 = mastery_score(z_S1, z_K1) # The output may be 0.91, indicating a relatively high degree of mastery;

[0196] score_S1_K2 = mastery_score(z_S1, z_K2) # The output may be 0.62, indicating a moderate degree of mastery.

[0197] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device.

[0198] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

[0199] The above describes the present invention and its implementation manners. Such description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. In short, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. A digital teaching material intelligent generation system based on digital education, characterized by: It includes educational resource acquisition module, knowledge graph construction module, textbook content generation module, knowledge tracking module and textbook presentation module, specifically including the following contents: The educational resource acquisition module collects teaching content data, uses NLP and image recognition technology to perform semantic analysis and entity extraction on the teaching content data, obtains educational semantic units and constructs a semantic metadata set; The knowledge graph construction module performs structured processing on the semantic metadata set to construct an initial knowledge graph, including nodes and edges, and introduces an embedding method based on a hyperbolic space graph convolutional network to perform low-dimensional representation learning on the initial knowledge graph to obtain an educational knowledge graph with a semantic hierarchical structure; The teaching material content generation module is based on an educational knowledge graph with a semantic hierarchy, and uses graph structure expansion and semantic reasoning methods to cluster relevant knowledge points and expand paths for target teaching topics to generate corresponding teaching material content units; The knowledge tracking module introduces a graph-structured learning behavior modeling method to obtain the knowledge state map of each student by constructing a student-knowledge point heterogeneous graph structure; The teaching material presentation module provides a visual teaching material content presentation method based on teaching material content units and an educational knowledge graph with a semantic hierarchy, combined with a knowledge status graph of each student.

2. The digital teaching material intelligent generation system based on digital education according to claim 1 is characterized by: In the educational resource acquisition module, the teaching content data includes textbook original text, teaching lesson plans, teacher explanation content, question bank materials, online course content and multimedia resources, and the educational semantic units include concept entities, knowledge points, teaching objectives and subject tags; In the teaching material content generation module, the teaching material content units include concept explanations, example explanations, diagrams and exercises.

3. The digital teaching material intelligent generation system based on digital education according to claim 1 is characterized in that: The knowledge graph construction module specifically includes the following steps: Step S1: Structural processing, clarifying and standardizing the semantic metadata, mapping the educational semantic units into nodes, and establishing edges based on the relationship between each educational semantic unit. Each edge represents the semantic connection between different nodes; Step S2: construct an initial knowledge graph, and construct an initial knowledge graph of knowledge points-teaching objectives-concept entities based on nodes and edges; Step S3: Based on the embedding of the hyperbolic space graph convolutional network, a hyperbolic space graph convolutional network is introduced to perform low-dimensional representation learning on the initial knowledge graph, embed the nodes into the hyperbolic space, and obtain the embedded vector representation of each node; Step S4: Calculate the structure, based on the embedding vector representation of each node, cluster the educational semantic units using a hierarchical clustering algorithm based on robustness improvement to form a multi-level semantic grouping structure; Step S5: Structural optimization, jointly evaluate the similarity between nodes and the original connection relationship, and use the hierarchical adjustment algorithm to adjust the connection method of the edges to obtain an educational knowledge graph with a semantic hierarchical structure.

4. The digital teaching material intelligent generation system based on digital education according to claim 3 is characterized by: Step S4, clustering the educational semantic units using a hierarchical clustering algorithm based on robustness improvement, specifically including the following steps: Step S41: De-noising, performing density analysis on the embedded vector representation of each educational information unit, using a density-based noise discrimination mechanism to identify and remove abnormal samples, and obtaining a denoised educational information unit set; Step S42: construct a k-nearest-neighbor directed graph, construct a k-nearest-neighbor directed graph according to the denoised education information unit set, map each education information unit to a node in the graph, and determine its k nearest neighbor nodes according to the Euclidean distance between the embedded vectors, connect them to form directed edges, and set the weights of the directed edges according to the relative distances between the nodes; Step S43: Graph merging and clustering: a group of nodes with close embedding vector distances are classified into one category to form a cluster. Multiple clusters are constructed to form a semantic hierarchy. In the k-nearest neighbor directed graph, the directed edges are iteratively traversed in the order of their weights, and the graph merging operation is performed until all directed edges are traversed. Step S44: Determine the number of clusters. During the graph merging process of step S43, record the change in the number of clusters after each graph merging operation, construct a mapping sequence between iteration rounds and the number of clusters, count the number of merging iterations corresponding to each number of clusters, and select the number of clusters corresponding to the maximum number of merging iterations as the final clustering result. Step S45: constructing a semantic abstract representation, performing mean pooling processing on the embedding vectors of the educational information units in each cluster, obtaining the semantic center vector of the cluster as the abstract semantic representation of the current level cluster, and identifying the common features of this type of educational semantic units; Step S46: Multi-layer structure generation, using a top-down hierarchical strategy to recursively execute steps S41 to S45 to construct a multi-layer semantic hierarchy.

Images