Knowledge graph-based teaching display system and method, medium

Abstract

The application discloses a knowledge graph-based teaching display system and method and a medium, relates to the field of artificial intelligence and education technology, and comprises the following steps: collecting multi-modal original teaching data periodically, and constructing a teaching knowledge graph; dynamically updating the relationship weight of the teaching knowledge graph; acquiring the cognitive dependence coefficient between each knowledge point corresponding to each learning terminal according to the potential ability coefficient of each learning terminal, the discrimination coefficient and the difficulty coefficient of a plurality of questions corresponding to each knowledge point, and the relationship weight between each knowledge point; generating the individualized learning path of each learning terminal according to the cognitive dependence coefficient and the learning target; adding the interactive attribute of the node in the teaching knowledge graph according to the individualized learning path, generating a VR virtual scene and visual data; and performing real-time learning monitoring on the individualized learning path of each learning terminal, and performing real-time learning feedback operation according to the real-time learning monitoring result, so that the overall teaching quality is significantly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and education technology, in particular to a teaching display system and method based on a knowledge graph, and a medium. BACKGROUND

[0002] A Chinese patent with publication number CN114896417A discloses a method for constructing a computer education knowledge graph based on a knowledge graph, including the following steps: step one: constructing an education knowledge graph, S1: data acquisition, S2: knowledge extraction, S3: knowledge identification, S4: knowledge storage, S5: knowledge fusion, S6: quality control, step two: platform construction of the education knowledge graph, A: building a webpage, which includes an education knowledge graph display module, an intelligent question answering module, and a knowledge point query module.

[0003] A Chinese patent with publication number CN119271339A discloses a textbook content visualization system and method based on data analysis, which includes: through dynamic updating of a knowledge graph and optimization of a learning path, combining reinforcement learning and a knowledge graph, the optimal learning path most suitable for the current learning situation of students can be found out to help students reinforce weak knowledge points.

[0004] In the current education and teaching process, there are problems such as scattered teaching resources, lack of effective correlation display between knowledge points, and difficulty in individualized teaching according to individual differences of students, and the knowledge graph is static, the traditional system constructs the graph based on the preset ontology, it is difficult to incorporate new knowledge (such as the frontier of the subject, real-time feedback of students) in time, and the traditional teaching display method is often limited to the presentation of single course content, teachers are difficult to grasp the whole subject knowledge system from a macro perspective and carry out systematic teaching, and students are also difficult to clearly understand the internal logical relationship between knowledge points, resulting in poor learning effect. At the same time, due to the inability to accurately analyze the learning situation of students, it is difficult to provide targeted learning suggestions and resource recommendations, and it is difficult to meet the individualized learning needs of different students, the interactive form is single: mainly two-dimensional graphics, lacking immersive learning experience. SUMMARY

[0005] In order to solve the above technical problems, the purpose of the present application is to provide a teaching display system and method based on a knowledge graph, and a medium.

[0006] The first aspect of the present application provides a teaching display system based on a knowledge graph, comprising a cloud, the cloud is in communication connection with a learning terminal, a knowledge graph construction module, a graph dynamic updating module, a cognitive dependence analysis module, an individualized path planning module, a teaching visualization module, and a real-time monitoring feedback module.

[0007] The knowledge graph construction module is configured to periodically collect multi-modal original teaching data, perform entity naming and interest extraction on the multi-modal original teaching data, and construct a teaching knowledge graph.

[0008] The graph dynamic updating module is configured to construct a graph dynamic updating model and dynamically update relationship weights of the teaching knowledge graph.

[0009] The cognitive dependency analysis module is configured to obtain potential ability coefficients of each learning terminal, discrimination coefficients of each question, and difficulty coefficients according to answer records of each learning terminal, construct a knowledge path directed graph, and obtain cognitive dependency coefficients between each knowledge point corresponding to each learning terminal according to the potential ability coefficients of each learning terminal, the discrimination coefficients and the difficulty coefficients of a plurality of questions corresponding to each knowledge point, and relationship weights between the knowledge points.

[0010] The personalized path planning module is configured to generate a personalized learning path of each learning terminal according to the cognitive dependency coefficients between each knowledge point corresponding to each learning terminal and a learning goal.

[0011] The teaching visualization module is configured to add interactive attributes to nodes in the teaching knowledge graph according to the personalized learning path, generate VR virtual scenes and visualized data, and feed back to the learning terminal.

[0012] The real-time monitoring and feedback module is configured to perform real-time learning monitoring on the personalized learning path of each learning terminal, and perform real-time learning feedback operation according to the real-time learning monitoring result.

[0013] Further, the knowledge graph construction module periodically collects multi-modal original teaching data, performs entity naming and interest extraction on the multi-modal original teaching data, and constructs a teaching knowledge graph, and the process includes:

[0014] The multi-modal original teaching data is periodically collected through a plurality of channels and marked with a collection period, the collection period of the present embodiment is one month, and the multi-modal original teaching data includes teaching materials and teaching plans, academic papers, online course resources, and student homework and examination data. The multi-modal original teaching data is subjected to format conversion (including converting a PDF format teaching material document into a text format), word segmentation and sentence segmentation, and part-of-speech tagging preprocessing operations. For text data, word segmentation is performed and the part-of-speech of each word is tagged.

[0015] The preprocessed multi-modal original teaching data is input into a preset entity naming determination model to determine entities, and the multi-modal original teaching data includes a plurality of entities. The entity naming determination model is trained by a large amount of annotated text data, and can automatically identify entities such as course names, knowledge points, and teachers. A relationship extraction model is constructed based on machine learning, the multi-modal original teaching data and the plurality of entities are input into the relationship extraction model to extract relationships, and the domain semantic connection relationships between the plurality of entities are obtained.

[0016] The plurality of entities are used as nodes of a knowledge graph, and the domain semantic connection relationships between the plurality of entities are used as connection relationships between the nodes, to construct a teaching knowledge graph. The relationship weights of the edges between the nodes in the teaching knowledge graph are set.

[0017] Further, the process of setting the relationship weights of the edges between the nodes in the teaching knowledge graph includes:

[0018] The preprocessed multi-modal original teaching data is scanned, and the number of times each pair of entities appears simultaneously is counted. For example, entity A and entity B appear together n times in num data records, and the relationship weight ω AB is obtained by the formula For example, in the teaching text data, the two entities "function" and "derivative" appear together 50 times, and the total number of data records is 200, so the relationship weight between them is 0.25. The more times the two entities appear together in the multi-modal original teaching data, the stronger the association between the two entities, and the higher the corresponding relationship weight can be set.

[0019] Further, the graph dynamic updating module constructs a graph dynamic updating model, and the process of dynamically updating the relationship weights of the teaching knowledge graph includes:

[0020] The teaching knowledge graph is divided into different collection periods according to the collection period, and the teaching knowledge graph G t = (V, E t ), t = 1, 2,..., N, N represents the total number of collection periods, wherein V represents the node set of the teaching knowledge graph, E t represents the relationship weight set of the edge corresponding to the collection period t, and ω ij t represents the relationship weight of node i and j in the tth collection period. According to the teaching knowledge graph of different collection periods, a time series graph {G1, G2,..., G Nnode feature analysis is performed on the time series graph, and a node feature sequence is extracted, wherein the node features include static features (including subject of knowledge points, prior difficulty) and dynamic features (including number of paper mentions of the node in the last three collection periods, and collection period of the node embedding);

[0021] The DySAT architecture is adopted, graph convolution and time convolution are combined to construct a graph dynamic update model, the time series graph and the node feature sequence are taken as the training set and the verification set, the graph dynamic update model is iteratively trained, the model parameters of the graph dynamic update model are optimized, and meanwhile, when the knowledge graph construction module collects the multi-modal original teaching data of the latest collection period, the graph dynamic update model is incrementally learned;

[0022] The predicted relationship weight is obtained according to the output layer of the graph dynamic update model, and the teaching knowledge graph is dynamically updated according to the predicted relationship weight, for example, according to IEEE Xplore, arXiv and other academic platforms, the monthly newly added papers are obtained through the API interface, the keyword is set as quantum computing, and the relationship weight is dynamically updated (such as the correlation strength of quantum computing-linear algebra) according to the graph dynamic update model.

[0023] Further, the specific process of constructing the graph dynamic update model includes:

[0024] The DySAT architecture is adopted, the mean square error loss function is selected, G1~G N-1 is taken as the training set, and G N is taken as the verification set, the graph dynamic update model is iteratively trained, and the model parameters W (l) of the graph dynamic update model are optimized.

[0025] The model input layer: time series graph {G1, G2,..., G N} and node feature sequence;

[0026] The time embedding layer: E t = PositionalEncoding (t), t ∈ {1,..., N}; wherein PositionalEncoding () represents an encoding function, which is used to encode the time information (month) into a continuous vector;

[0027] The graph convolution layer: wherein H t (l+1) represents the updated node representation at the current time step, σ represents a nonlinear activation function, represents a degree matrix, represents a normalized adjacency matrix, W (l) represents a weight matrix;

[0028] Temporal convolution layer: one-dimensional convolution is performed on the time series features of each node to capture long-term dependencies;

[0029] According to the output layer, the prediction relationship weight ω of the next collection cycle is predicted ij N+1 ; According to the prediction relationship weight ω ij N+1 Update the atlas, dynamically update the rule: ω ij new = α· ω ij N+1 +(1-α)· ω ij old , wherein a represents a forgetting factor, the above formulas are to remove the dimension and take the numerical value, the formula is obtained by collecting a large amount of data to simulate software to obtain a formula closest to the real situation, the preset parameters and the preset threshold in the formula are set by the person skilled in the art according to the actual situation or obtained by a large amount of data simulation;

[0030] When new multi-modal original teaching data of each new collection cycle is added, G N is added to the training set and retrained.

[0031] Further, the process of acquiring the potential ability coefficient of each learning terminal, the discrimination coefficient of each question and the difficulty coefficient by the cognitive dependence analysis module according to the answer record of each learning terminal includes:

[0032] According to the answer record of each learning terminal, a joint answer matrix is constructed, the row of the joint answer matrix represents a learning terminal (k learning terminals), the column represents a question (m questions), and the element is an answer result (correct = 1, wrong = 0). The potential ability coefficient of each learning terminal, the discrimination coefficient of each question and the difficulty coefficient are randomly initialized in the answer matrix. The correct answer probability of each learning terminal for each question is acquired according to the potential ability coefficient, the discrimination coefficient and the difficulty coefficient. According to the joint answer matrix and the correct answer probability of each learning terminal for each question, a log-likelihood function is constructed. According to the log-likelihood function, the correct answer probability of all learning terminals for each question under the condition of given student ability and question parameters is determined.

[0033] Wherein, the specific formula for acquiring the correct answer probability of each learning terminal for each question and the log-likelihood function is:

[0034]

[0035] Wherein, P(θ p ,a j ,b j ) represents the correct answer probability of learning terminal p for question j, θp denotes the latent ability coefficient of the learning terminal p;

[0036]

[0037] wherein lnL denotes the log-likelihood function, X pj denotes the answer result (0 or 1) of the learning terminal p to the question j;

[0038] Since the latent ability coefficient of the learning terminal and the discrimination coefficient and the difficulty coefficient of the question are mutually dependent, the latent ability coefficient of the learning terminal and the discrimination coefficient and the difficulty coefficient of the question are solved by separate iterations, the discrimination coefficient and the difficulty coefficient of each question are fixed, the latent ability coefficient is iteratively updated for the log-likelihood function, then the latent ability coefficient of each learning terminal is fixed, the discrimination coefficient and the difficulty coefficient are iteratively updated for the log-likelihood function, the above process is repeated until the latent ability coefficient of each learning terminal, the discrimination coefficient and the difficulty coefficient of each question converge, and the latent ability coefficient of each learning terminal, the discrimination coefficient and the difficulty coefficient of each question are output.

[0039] Further, the process of fixing the discrimination coefficient and the difficulty coefficient of each question and iteratively updating the latent ability coefficient for the log-likelihood function includes:

[0040]

[0041] and the process of fixing the latent ability coefficient of each learning terminal and iteratively updating the discrimination coefficient and the difficulty coefficient for the log-likelihood function includes:

[0042]

[0043] wherein argmax denotes the parameter that makes the function maximum.

[0044] Further, the cognitive dependency analysis module constructs a knowledge path directed graph, and according to the latent ability coefficient of each learning terminal, the discrimination coefficient and the difficulty coefficient of a plurality of questions corresponding to each knowledge point, and the relationship weight between each knowledge point, the cognitive dependency coefficient between each knowledge point corresponding to each learning terminal is obtained.

[0045] The knowledge point set S = {S1, S2,... S n1and the domain semantic connection relationship between each knowledge point in the knowledge point set is extracted, each knowledge point is taken as a node, and the domain semantic connection relationship between the nodes is taken as a directed edge between the nodes (for example, S1→S2 indicates that S1 is a prerequisite knowledge of S2), a knowledge path directed graph is constructed, the relationship weight of the directed edge in the knowledge path directed graph is marked according to the teaching knowledge graph, and the potential ability coefficient of each learning terminal, the discrimination coefficient of each question and the difficulty coefficient are obtained according to the answer record of each learning terminal, and the corresponding question set of each knowledge point in the knowledge path directed graph is obtained;

[0046] When the learning target of the learning terminal is received, the conditional mastery probability between each node in the knowledge path directed graph is obtained according to the potential ability coefficient of the learning terminal, the discrimination coefficient and the difficulty coefficient of each question corresponding to each knowledge point in the knowledge path directed graph and the relationship weight between each knowledge point, the conditional mastery probability between each node is standardized, and the conditional mastery probability between each node is converted into a cognitive dependence coefficient.

[0047] The calculation formula of the conditional mastery probability between each node in the knowledge path directed graph is:

[0048] Suppose that the nodes of the knowledge path directed graph are S={S1, S2, S i ,..., S j ,..., S n1}, and the calculation formula of the conditional mastery probability between two nodes (i, j) in the knowledge path directed graph knowledge point is:

[0049]

[0050] wherein P(j|i, θ) represents the conditional probability of the learning terminal mastering the knowledge point j under the condition of mastering the knowledge point i, w ij represents the relationship weight between the knowledge point i and the knowledge point j, θ represents the potential ability coefficient of the learning terminal, represents the discrimination coefficient of the knowledge point j, a j represents the discrimination coefficient of the question j (a j > 0, the greater the value, the stronger the discrimination ability of the question to high and low ability students), b j represents the difficulty coefficient of the question j (the greater the value, the more difficult the question), represents the discrimination coefficient of the knowledge point i, b i represents the difficulty coefficient of the question i, represents the discrimination coefficient of the knowledge point j, M i represents the question set corresponding to the knowledge point i, |NM i | represents the total number of questions corresponding to the knowledge point i, M jNM j | represents the total number of questions corresponding to knowledge point j, Δ ij represents the cognitive promotion amount of knowledge point i to j, the higher the difficulty of the prerequisite knowledge point i, the greater the promotion to the subsequent knowledge point j, for example, if i is a directly connected prerequisite knowledge point of j, and Δ ij > 0, indicating that after mastering i, the difficulty of j is relatively reduced, the probability P(j|i, θ) is increased, and α represents an adjustment coefficient for converting the conditional mastery probability between nodes into a cognitive dependence coefficient. The specific formula is:

[0051]

[0052] where dp ij represents the cognitive dependence coefficient between knowledge point i and knowledge point j, pre(j) represents the set of all prerequisite knowledge points of knowledge point j, and pathlength(i, j) represents the shortest path length from i to j in the knowledge path directed graph (reflecting the cognitive span), where ∑ d∈pre(j) P(d|i, θ) ensures that the sum of all outgoing edge weights of the same node is 1, avoiding preference for long paths, and in order to punish long paths, direct dependence relationships are preferred. In the formula, divide by pathlength(i, j), the above formulas are dimensionless, and the numerical value is calculated, the formula is obtained by software simulation of a large amount of data to obtain a formula closest to the real situation, and the preset parameters and preset thresholds in the formula are set by a person skilled in the art according to the actual situation or obtained by a large amount of data simulation.

[0053] Further, the personalized path planning module generates a personalized learning path for each learning terminal according to the cognitive dependence coefficients between each knowledge point corresponding to each learning terminal and the learning goal, and the process includes:

[0054] According to the learning goal of the learning terminal, the starting node and the ending node in the knowledge path directed graph are obtained, all paths between the starting node and the ending node in the knowledge path directed graph are extracted, the cognitive dependence of all paths is aggregated, the cumulative cognitive dependence coefficient of each path is obtained, and the path with the largest cumulative cognitive dependence coefficient is selected to generate a personalized learning path. The specific process includes finding the path with the largest cumulative cognitive dependence coefficient (indicating the strongest cognitive dependence) from the starting node to the ending node, defining the state dpx[i] as the maximum cumulative cognitive dependence coefficient to reach knowledge point i, then: dpx[j] = max i∈pre(j) (dpx[i] + w ij ), and the personalized learning path can be obtained by path backtracking.

[0055] Further, the teaching visualization module adds interactive attributes to the nodes in the teaching knowledge graph according to the personalized learning path, generates VR virtual scenes and visualization data, and feeds back to the learning terminal, including:

[0056] The correct answer probability, prerequisite relationship, and subject attribute of each knowledge point in the personalized learning path are obtained, and interactive attributes are added to the nodes to which each knowledge point belongs in the teaching knowledge graph according to the correct answer probability, prerequisite relationship, and subject attribute of each knowledge point. The interactive attributes include the color of the knowledge point: subject classification (red - mathematics, blue - physics); size: correct answer probability (the lower the correct answer probability, the larger the node); transparency: prerequisite relationship (pre-knowledge points are displayed semi-transparently); node pop-up layer display: knowledge point definition, related examples, learning videos, historical error rate, gesture operation: double-finger sliding rotation graph, fist click to expand sub-knowledge points, etc.; voice operation: query knowledge points, request help or navigation operations through voice commands; head tracking and gaze interaction operation: use the head tracking function of the VR headset to realize gaze interaction, when the user gazes at a certain knowledge point node, the system can display related detailed information or pop up an interactive menu;

[0057] The calculation formula of the correct answer probability of each knowledge point is:

[0058]

[0059] A VR virtual scene is created according to the nodes with added interactive attributes in the teaching knowledge graph, including using 3D modeling software (such as Unity, UnrealEngine, etc.) to create a VR teaching scene, designing appropriate scene layout and environment according to the subject attribute of the knowledge point, for example, in geography teaching, a virtual earth model and related geographical landscape can be created, constructing a knowledge graph visualization model, outputting visualization data of nodes with added interactive attributes (such as using spheres to represent nodes and lines to represent relationships) according to the knowledge graph visualization model, the knowledge graph visualization model presents the nodes and edges in the teaching knowledge graph in a visual form to the virtual scene, different shapes, colors, and sizes are used to present nodes with different interactive attributes, and the VR virtual scene and visualization data are fed back to the learning terminal, and the user interacts with the knowledge graph using the VR device of the learning terminal according to the VR virtual scene and visualization data.

[0060] Further, the real-time monitoring feedback module monitors the personalized learning path of each learning terminal in real time, and the process of performing real-time learning feedback operation according to the real-time learning monitoring result includes:

[0061] According to the progress segmentation of each knowledge point in the personalized learning path, one knowledge point is taken as one stage to form a progress stage, then the learning time of the knowledge point is estimated according to the cognitive dependence coefficient between the knowledge point and the previous knowledge point in the personalized learning path to form the estimated time of the progress stage, the end time stamp of each progress stage is taken as a detection time point, and the correct answer probability of each knowledge point in the personalized learning path is set as the judgment standard of each progress stage;

[0062] At the detection time point, the actual correct answer probability of the learning terminal in the current progress stage is obtained, the actual correct answer probability is compared with the judgment standard of the current progress stage, if the actual correct answer probability is less than the judgment standard, the answer record of the learning terminal in the current progress stage is input into the cognitive dependence analysis module, then the personalized learning path is regenerated according to the personalized path planning module, and if the actual correct answer probability is greater than or equal to the judgment standard, the next progress stage is automatically entered into the judgment standard comparison process (the actual correct answer probability of the next progress stage is compared with the judgment standard of the next progress stage).

[0063] The second aspect of the present application also provides a teaching display method based on a knowledge graph, comprising the following steps:

[0064] Step s1: periodically collecting multi-modal original teaching data, performing entity naming and interest extraction on the multi-modal original teaching data, constructing a teaching knowledge graph, constructing a graph dynamic updating model, and dynamically updating the relationship weight of the teaching knowledge graph;

[0065] Step s2: obtaining the potential ability coefficient of each learning terminal, the discrimination coefficient of each question and the difficulty coefficient according to the answer record of each learning terminal, constructing a knowledge path directed graph, and obtaining the cognitive dependence coefficient between each knowledge point corresponding to each learning terminal according to the potential ability coefficient of each learning terminal, the discrimination coefficient and the difficulty coefficient of a plurality of questions corresponding to each knowledge point, and the relationship weight between each knowledge point;

[0066] Step s3: generating the personalized learning path of each learning terminal according to the cognitive dependence coefficient between each knowledge point corresponding to each learning terminal and the learning target;

[0067] Step s4: adding the interactive attribute of the node in the teaching knowledge graph according to the personalized learning path to generate a VR virtual scene and visual data and feed back to the learning terminal;

[0068] Step s5: real-time learning monitoring of the personalized learning path of each learning terminal, and real-time learning feedback operation according to the real-time learning monitoring result.

[0069] The third aspect of the present application also provides a computer readable storage medium, the computer readable storage medium stores computer program instructions, when the computer program instructions are executed, the steps of the knowledge graph-based teaching display system as any one of the first aspect are realized.

[0070] Compared with the prior art, the present application has the following advantages:

[0071] 1. Multimodal data utilization: The knowledge graph construction module regularly collects multimodal raw teaching data, covering various forms such as text, images, and audio, which can comprehensively and deeply reflect the teaching content. Compared with the traditional method of relying only on single text data, it can provide more abundant and three-dimensional knowledge sources, making the knowledge graph more complete and accurate in presenting the knowledge system of the teaching field. For example, in a science course, not only can the theoretical knowledge described in words be obtained, but also videos of experimental operations, experimental data charts, etc. can be included, making the knowledge graph more vivid and comprehensive.

[0072] 2. Dynamic update: The dynamic update model of the graph constructed by the graph dynamic update module can divide the teaching knowledge graph by time and update it according to the collection period. It can timely integrate new teaching content, academic achievements, etc. into the knowledge graph, maintaining the timeliness of knowledge. For example, in the field of computer science, with the rapid development of technology, new algorithms, programming language features, etc. can be updated to the knowledge graph in a timely manner to ensure that students learn the latest knowledge. At the same time, through the analysis and learning of time series graphs and node feature sequences, the relationship weights between nodes can be accurately adjusted to reflect the dynamic changes of the relationship between knowledge and optimize the structure of the knowledge graph.

[0073] 3. Precise cognitive dependence analysis: The cognitive dependence analysis module can accurately obtain the student's potential ability coefficient, question discrimination degree, and difficulty coefficient through deep mining of the learning terminal answer record. On this basis, combined with the knowledge point relationship weight in the knowledge graph, a knowledge path directed graph is constructed and the cognitive dependence coefficient is calculated. This enables the system to deeply understand the mastery of different knowledge points by each student and the learning dependence relationship between knowledge points. For example, in mathematics learning, the cognitive association between students and different knowledge blocks such as functions and geometry can be clearly defined, providing precise basis for personalized learning path planning.

[0074] 4. Personalized path planning: The personalized path planning module generates a personalized learning path based on the cognitive dependence coefficient and the student's learning goal. Each student can obtain a learning route that meets their own knowledge mastery and learning needs. For students with weak foundation, the system plans a path that gradually deepens from basic knowledge; while for students with surplus, it provides more challenging and expansive learning paths. This greatly improves learning efficiency, avoids wasting time on learning content that is not suitable for the student, and improves learning enthusiasm and effectiveness.

[0075] 5. Immersive learning experience: The teaching visualization module adds interactive attributes to the teaching knowledge graph nodes to generate VR virtual scenes and visualized data. Students can learn in an immersive VR environment, enhancing the interest and participation in learning. For example, in history teaching, students can be placed in a historical scene and interact with knowledge nodes in the virtual environment, such as clicking on historical figures to learn about their lives, touching historical artifacts to view related introductions, etc., making the learning process more intuitive and vivid, and deepening understanding and memory of knowledge.

[0076] 6. Real-time monitoring feedback: The real-time monitoring feedback module monitors the student's individualized learning path in real time, segments by knowledge points and estimates the learning time, sets the judgment standard. At the detection time point, compare the actual correct answer probability with the judgment standard to find out the problems in the student's learning in time. If the student does not meet the standard, the system automatically re-plans the learning path. This realizes the dynamic tracking and adjustment of the student's learning process, ensures that the student always learns at his own pace and difficulty, and continuously optimizes the learning effect. BRIEF DESCRIPTION OF DRAWINGS

[0077] Figure 1 The principle diagram of the teaching display system based on the knowledge graph of the embodiment of the present application.

[0078] Figure 2 The principle diagram of the teaching display method based on the knowledge graph of the embodiment of the present application. DETAILED DESCRIPTION

[0079] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0080] As shown in Figure 1 The first aspect of the present application provides a teaching display system based on a knowledge graph, comprising a cloud, the cloud being communicatively connected with a learning terminal, a knowledge graph construction module, a graph dynamic updating module, a cognitive dependence analysis module, an individualized path planning module, a teaching visualization module and a real-time monitoring feedback module;

[0081] The knowledge graph construction module is used for periodically collecting multi-modal original teaching data, performing entity naming and interest extraction on the multi-modal original teaching data, and constructing a teaching knowledge graph;

[0082] The graph dynamic updating module is used for constructing a graph dynamic updating model and dynamically updating the relationship weight of the teaching knowledge graph;

[0083] The cognitive dependency analysis module is configured to obtain the latent ability coefficients of the learning terminals, the discrimination coefficients of the questions, and the difficulty coefficients according to the answer records of the learning terminals, construct a knowledge path directed graph, and obtain the cognitive dependency coefficients between the knowledge points corresponding to each learning terminal according to the latent ability coefficients of the learning terminals, the discrimination coefficients and the difficulty coefficients of the questions corresponding to each knowledge point, and the relationship weights between the knowledge points.

[0084] The personalized path planning module is configured to generate the personalized learning paths of the learning terminals according to the cognitive dependency coefficients between the knowledge points corresponding to each learning terminal and the learning goals.

[0085] The teaching visualization module is configured to add interactive attributes to the nodes in the teaching knowledge graph according to the personalized learning paths, generate VR virtual scenes and visual data, and feed back to the learning terminals.

[0086] The real-time monitoring and feedback module is configured to perform real-time learning monitoring on the personalized learning paths of the learning terminals, and perform real-time learning feedback operations according to the real-time learning monitoring results.

[0087] It should be further explained that, in the specific implementation process, the knowledge graph construction module periodically collects multi-modal original teaching data, performs entity naming and relation extraction on the multi-modal original teaching data, and the process of constructing the teaching knowledge graph includes:

[0088] The multi-modal original teaching data is collected through several channels and marked with a collection period. In this embodiment, the collection period is one month. The multi-modal original teaching data includes teaching materials and teaching plans, academic papers, online course resources, and student homework and examination data. The multi-modal original teaching data is preprocessed by format conversion (including converting PDF format teaching material documents into text format), word segmentation, and part-of-speech tagging. For text data, word segmentation is performed and the part-of-speech of each word is tagged.

[0089] The multi-modal original teaching data after the preprocessing operation is input into a preset entity naming determination model to determine the entities included in the multi-modal original teaching data. The entity naming determination model is trained by a large amount of annotated text data and can automatically identify course names, knowledge points, teachers, and other entities. A relation extraction model is constructed based on machine learning, the multi-modal original teaching data and the entities are input into the relation extraction model to extract the relationship between the entities, and the domain semantic connection relationship between the entities is obtained.

[0090] The process of constructing the relation extraction model based on deep learning includes:

[0091] The field semantic connection relationship in the teaching field is determined, the field semantic connection relationship includes a knowledge level relationship, such as a containing relationship, a contained relationship, for example, “mathematics” contains sub-disciplines such as “algebra” and “geometry”; “algebra” further contains knowledge points such as “function” and “equation”; a sequence relationship, such as a prerequisite relationship, “advanced mathematics” is a prerequisite for “probability theory and mathematical statistics”; a cause-and-effect relationship, in the physics discipline, “force action” causes “change in the motion state of an object”; a correlation relationship, a certain connection between different knowledge points, for example, “literary works” are associated with “historical background”; a large number of annotated sample text data are then collected, each sample includes two entities and a field semantic connection relationship therebetween, features related to the field semantic connection relationship are extracted from the text, such as context information, part of speech, and syntax structure of the entities, and a relationship extraction model is obtained by training the sample text data and the features using a neural network.

[0092] A number of entities are taken as nodes of a knowledge graph, and a field semantic connection relationship between a number of entities is taken as a connection relationship between the nodes, a teaching knowledge graph is constructed, and a relationship weight of an edge between each node in the teaching knowledge graph is set.

[0093] It should be further explained that, in the specific implementation process, the process of setting the relationship weight of the edge between each node in the teaching knowledge graph includes:

[0094] The preprocessed multi-modal original teaching data are scanned, and the number of times that each pair of entities appears simultaneously is counted, for example, entity A and entity B appear together n times in num data records, and the relationship weight ω AB is obtained by the formula , for example, in the teaching text data, the two entities “function” and “derivative” appear together 50 times, and the total data records are 200, so the relationship weight between them is 0.25, the more times the two entities appear together in the multi-modal original teaching data, the stronger the correlation between the two entities, and the higher the corresponding relationship weight can be set.

[0095] It should be further explained that, in the specific implementation process, the process of constructing the graph dynamic updating model by the graph dynamic updating module and dynamically updating the relationship weight of the teaching knowledge graph includes:

[0096] The teaching knowledge graph is divided into different collection periods according to a collection period, and teaching knowledge graphs G t = (V, E t ), t = 1, 2,... N, N represents the total number of collection periods, wherein V represents a node set of the teaching knowledge graph, E t represents a relationship weight set of the edge corresponding to the collection period t, and ωij t denotes the relationship weight between nodes i and j in the t-th collection cycle, and a time series graph {G1, G2,...G N} is constructed according to the teaching knowledge graphs in different collection cycles, node feature analysis is performed on the time series graph, and a node feature sequence is extracted, wherein the node features include static features (including the subject of the knowledge point, prior difficulty) and dynamic features (including the number of paper mentions of the node in the last three collection cycles, and the collection cycle in which the node is embedded);

[0097] The DySAT architecture is adopted, the graph convolution and the time convolution are combined to construct the graph dynamic update model, the time series graph and the node feature sequence are taken as the training set and the verification set, the graph dynamic update model is iteratively trained, the model parameters of the graph dynamic update model are optimized, and meanwhile, when the knowledge graph construction module collects the multi-modal original teaching data in the latest collection cycle, the graph dynamic update model is incrementally learned;

[0098] The predicted relationship weight is obtained according to the output layer of the graph dynamic update model, and the teaching knowledge graph is dynamically updated according to the predicted relationship weight, for example, according to the IEEE Xplore, arXiv and the like academic platforms, the monthly newly added papers are obtained through the API interface, the keyword is set as quantum computing, and the relationship weight is dynamically updated (such as the correlation strength of quantum computing-linear algebra) according to the graph dynamic update model.

[0099] It needs to be further explained that the specific process of constructing the graph dynamic update model includes:

[0100] The DySAT architecture is adopted, the mean square error loss function is selected, G1~G N-1 is taken as the training set, and G N is taken as the verification set, the graph dynamic update model is iteratively trained, and the model parameters W (l) of the graph dynamic update model are optimized, wherein the graph dynamic update model comprises:

[0101] The model input layer: the time series graph {G1, G2,...G N} and the node feature sequence;

[0102] The time embedding layer: E t = PositionalEncoding(t), t∈{1,...,N}; wherein PositionalEncoding() represents an encoding function, which is used to encode the time information (month) into a continuous vector;

[0103] The graph convolution layer: Wherein H t (l+1)denotes the updated node representation at the current time step, and denotes a nonlinear activation function, denotes the degree matrix, denotes the normalized adjacency matrix, and W (l) denotes the weight matrix.

[0104] Time convolution layer: one-dimensional convolution is performed on the time series features of each node to capture long-term dependencies.

[0105] According to the output layer, the prediction relationship weight ω of the next collection period is predicted. ij N+1 According to the prediction relationship weight ω ij N+1 The graph is updated, and the dynamic update rule is ω ij new = α·ω ij N+1 + (1-α)·ω ij old wherein, α represents a forgetting factor, the above formulas are all dimensionless values, the formula is obtained by software simulation of a large amount of data to obtain a formula closest to the actual situation, and the preset parameters and the preset threshold in the formula are set by the person skilled in the art according to the actual situation or obtained by a large amount of data simulation.

[0106] When new multi-modal original teaching data of a new collection period are added each time, G N is added to the training set and retrained.

[0107] It needs to be further explained that, in the specific implementation process, the process in which the cognitive dependence analysis module obtains the potential ability coefficient of each learning terminal, the discrimination coefficient of each question and the difficulty coefficient according to the answer record of each learning terminal includes:

[0108] A joint answer matrix is constructed according to the answer record of each learning terminal, the row of the joint answer matrix represents a learning terminal (k learning terminals), the column represents a question (m questions), and the element is an answer result (correct = 1, wrong = 0). The potential ability coefficient of each learning terminal, the discrimination coefficient of each question and the difficulty coefficient are randomly initialized in the answer matrix. The correct answer probability of each learning terminal for each question is obtained according to the potential ability coefficient, the discrimination coefficient and the difficulty coefficient. According to the joint answer matrix and the correct answer probability of each learning terminal for each question, a log-likelihood function is constructed. According to the log-likelihood function, the correct answer probability of all learning terminals for each question is determined under the condition that the given student ability and question parameters are given.

[0109] Wherein, the specific formula for obtaining the correct answer probability of each learning terminal for each question and the log-likelihood function is:

[0110]

[0111] wherein P (θ p ,a j ,b j ) represents the correct answer probability of the learning terminal p to the question j, θ p represents the latent ability coefficient of the learning terminal p;

[0112]

[0113] wherein lnL represents the log-likelihood function, X pj represents the answer result (0 or 1) of the learning terminal p to the question j;

[0114] Since the latent ability coefficient of the learning terminal and the discrimination coefficient and the difficulty coefficient of the question are mutually dependent, the latent ability coefficient of the learning terminal and the discrimination coefficient and the difficulty coefficient of the question are solved by separate iterations, the discrimination coefficient and the difficulty coefficient of each question are fixed, the latent ability coefficient of the log-likelihood function is iteratively updated, then the latent ability coefficient of each learning terminal is fixed, the discrimination coefficient and the difficulty coefficient of the log-likelihood function are iteratively updated, the above process is repeated until the latent ability coefficient of each learning terminal, the discrimination coefficient and the difficulty coefficient of each question converge, and the latent ability coefficient of each learning terminal, the discrimination coefficient and the difficulty coefficient of each question are output.

[0115] It needs to be further explained that, in the specific implementation process, the process of fixing the discrimination coefficient and the difficulty coefficient of each question and iteratively updating the latent ability coefficient of the log-likelihood function includes:

[0116]

[0117] and the process of fixing the latent ability coefficient of each learning terminal and iteratively updating the discrimination coefficient and the difficulty coefficient of the log-likelihood function includes:

[0118]

[0119] wherein argmax represents the parameter that makes the function take the maximum value.

[0120] It needs to be further explained that, in the specific implementation process, the process of constructing the knowledge path directed graph by the cognitive dependence analysis module, and obtaining the cognitive dependence coefficient between each knowledge point corresponding to each learning terminal includes:

[0121] A knowledge point set S = {S1, S2,... S n1} is extracted from a plurality of entities of a teaching knowledge graph, and a domain semantic connection relationship between each knowledge point in the knowledge point set is extracted, each knowledge point is taken as a node, and a domain semantic connection relationship between nodes is taken as a directed edge between nodes (for example, S1→S2 indicates that S1 is a prerequisite knowledge of S2), a knowledge path directed graph is constructed, a relationship weight of a directed edge in the knowledge path directed graph is marked according to the teaching knowledge graph, meanwhile, a potential ability coefficient of each learning terminal, a discrimination coefficient of each question and a difficulty coefficient are obtained according to an answer record of each learning terminal, and a corresponding question set of each knowledge point in the knowledge path directed graph is obtained.

[0122] When the learning target of the learning terminal is received, a conditional mastery probability between each node in the knowledge path directed graph is obtained according to the potential ability coefficient of the learning terminal, the discrimination coefficient and the difficulty coefficient of each question corresponding to each knowledge point in the knowledge path directed graph and the relationship weight between each knowledge point, the conditional mastery probability between each node is standardized, and the conditional mastery probability between each node is converted into a cognitive dependence coefficient.

[0123] The calculation formula of the conditional mastery probability between each node in the knowledge path directed graph is as follows:

[0124] Suppose that nodes S = {S1, S2, S i ,..., S j ,... S n1} of the knowledge path directed graph, and the calculation formula of the conditional mastery probability between two nodes (i, j) in the knowledge point of the knowledge path directed graph is as follows:

[0125]

[0126] Wherein, P(j|i, θ) represents a conditional probability that the learning terminal masters the knowledge point j under the condition of mastering the knowledge point i, w ij represents a relationship weight between the knowledge point i and the knowledge point j, θ represents a potential ability coefficient of the learning terminal, represents a discrimination coefficient of the knowledge point j, a j represents a discrimination coefficient of the question j (a j > 0, the greater the value, the stronger the discrimination ability of the question to high and low ability students), b j represents a difficulty coefficient of the question j (the greater the value, the more difficult the question), represents a discrimination coefficient of the knowledge point i, b i represents a difficulty coefficient of the question i, represents a discrimination coefficient of the knowledge point j, M i represents a question set corresponding to the knowledge point i, and |NMi | represents the total number of questions corresponding to knowledge point i, M j | represents the total number of questions corresponding to knowledge point i, M j | represents the total number of questions corresponding to knowledge point i, M ij | represents the total number of questions corresponding to knowledge point i, M | represents the total number of questions corresponding to knowledge point i, M ij | represents the total number of questions corresponding to knowledge point i, M

[0127]

[0128] | represents the total number of questions corresponding to knowledge point i, M ij | represents the total number of questions corresponding to knowledge point i, M d∈pre(j) | represents the total number of questions corresponding to knowledge point i, M i∈pre(j) | represents the total number of questions corresponding to knowledge point i, M ij | represents the total number of questions corresponding to knowledge point i, M

[0129] It should be further pointed out that in the specific implementation process, the personalized path planning module generates personalized learning paths for each learning terminal according to the cognitive dependency coefficients between each knowledge point corresponding to each learning terminal and the learning goal, which includes:

[0130] According to the learning goal of the learning terminal, the starting node and the ending node in the knowledge path directed graph are obtained, all paths between the starting node and the ending node in the knowledge path directed graph are extracted, the cognitive dependency of all paths is aggregated, the cumulative cognitive dependency coefficient of each path is obtained, and the path with the largest cumulative cognitive dependency coefficient is selected to generate the personalized learning path, and the specific process includes finding the path with the largest cumulative cognitive dependency coefficient (representing the strongest cognitive dependency) from the starting node to the ending node, defining the state dpx[i] as the maximum cumulative cognitive dependency coefficient to reach knowledge point i, then: dpx[j] = max i∈pre(j) (dpx[i] + w ij ), and the personalized learning path can be obtained by path backtracking.

[0131] It needs to be further explained that, in the specific implementation process, the teaching visualization module adds interactive attributes to the nodes in the teaching knowledge graph according to the personalized learning path, generates VR virtual scenes and visualization data, and feeds back to the learning terminal, which includes:

[0132] Obtain the correct answer probability, prerequisite relationship and subject attribute of each knowledge point in the personalized learning path, add interactive attributes to the nodes to which each knowledge point belongs in the teaching knowledge graph according to the correct answer probability, prerequisite relationship and subject attribute of each knowledge point, the interactive attributes include the color of the knowledge point: subject classification (red - mathematics, blue - physics); size: correct answer probability (the lower the correct answer probability, the larger the node); transparency: prerequisite relationship (the prerequisite knowledge point is displayed semi-transparently); node pop-up layer display: knowledge point definition, related examples, learning video, historical error rate, gesture operation: double-finger sliding rotation graph, fist click to expand sub-knowledge points, etc.; Voice operation: query knowledge points, request help or navigation operations through voice commands; head tracking and gaze interaction operation: use the head tracking function of the VR headset to realize gaze interaction, when the user gazes at a certain knowledge point node, the system can display related detailed information or pop-up interaction menu;

[0133] Wherein, the calculation formula of the correct answer probability of each knowledge point is:

[0134]

[0135] According to the nodes in the teaching knowledge graph that complete the interactive attribute addition, create a VR virtual scene, the specific process includes using 3D modeling software (such as Unity, UnrealEngine, etc.) to create a VR teaching scene, according to the subject attribute of the knowledge point, design appropriate scene layout and environment, for example, in geography teaching, a virtual earth model and related geographical landscape can be created, build a knowledge graph visualization model, output the visualization data of the nodes that complete the interactive attribute addition (for example, use a sphere to represent a node, and use a line to represent a relationship) according to the knowledge graph visualization model, the knowledge graph visualization model presents the nodes and edges in the teaching knowledge graph in a visual form to the virtual scene, different shapes, colors and sizes are used to present nodes with different interactive attributes, and the VR virtual scene and the visualization data are fed back to the learning terminal, and the user and the knowledge graph are interacted with the VR device of the learning terminal according to the VR virtual scene and the visualization data.

[0136] It needs to be further explained that, in the specific implementation process, the real-time monitoring feedback module monitors the personalized learning path of each learning terminal in real time, and the process of performing real-time learning feedback operation according to the real-time learning monitoring result includes:

[0137] According to the progress segmentation of each knowledge point in the personalized learning path, a knowledge point is taken as a stage to form a progress stage, then the learning time of the knowledge point is estimated according to the cognitive dependence coefficient between the knowledge point and the previous knowledge point in the personalized learning path to form the estimated time of the progress stage, the end time stamp of each progress stage is taken as a detection time point, and the correct answer probability of each knowledge point in the personalized learning path is set as the judgment standard of each progress stage;

[0138] At the detection time point, the actual correct answer probability of the learning terminal in the current progress stage is obtained, the actual correct answer probability is compared with the judgment standard of the current progress stage, if the actual correct answer probability is less than the judgment standard, the answer record of the learning terminal in the current progress stage is extracted and input into the cognitive dependence analysis module, then the personalized learning path is regenerated according to the personalized path planning module, and if the actual correct answer probability is greater than or equal to the judgment standard, the next progress stage is automatically entered for the judgment standard comparison process (the actual correct answer probability of the next progress stage is compared with the judgment standard of the next progress stage).

[0139] As shown in Figure 2 The second aspect of the present application also provides a teaching display method based on a knowledge graph, comprising the following steps:

[0140] Step s1: periodically collecting multi-modal original teaching data, performing entity naming and interest extraction on the multi-modal original teaching data, constructing a teaching knowledge graph, constructing a graph dynamic updating model, and dynamically updating the relationship weight of the teaching knowledge graph;

[0141] Step s2: obtaining the potential ability coefficient of each learning terminal, the discrimination coefficient of each question and the difficulty coefficient according to the answer record of each learning terminal, constructing a knowledge path directed graph, and obtaining the cognitive dependence coefficient between each knowledge point corresponding to each learning terminal according to the potential ability coefficient of each learning terminal, the discrimination coefficient and the difficulty coefficient of a plurality of questions corresponding to each knowledge point, and the relationship weight between each knowledge point;

[0142] Step s3: generating the personalized learning path of each learning terminal according to the cognitive dependence coefficient between each knowledge point corresponding to each learning terminal and the learning goal;

[0143] Step s4: adding the interactive attribute of the node in the teaching knowledge graph according to the personalized learning path, generating a VR virtual scene and visual data, and feeding back to the learning terminal;

[0144] Step s5: real-time learning monitoring of the personalized learning path of each learning terminal, and real-time learning feedback operation according to the real-time learning monitoring result.

[0145] The third aspect of the present application also provides a computer readable storage medium, which stores computer program instructions, and the computer program instructions, when executed, implement the steps of the knowledge graph-based teaching display system according to any one of the first aspect.

[0146] The above embodiments are only used to illustrate the technical method of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present application can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present application.

Claims

1. A knowledge graph-based teaching demonstration system, characterized in that, This includes a cloud platform, which has communication connections to a learning terminal, a knowledge graph construction module, a knowledge graph dynamic update module, a cognitive dependency analysis module, a personalized path planning module, a teaching visualization module, and a real-time monitoring and feedback module. The knowledge graph construction module is used to periodically collect multimodal raw teaching data, perform entity naming and relation extraction on the multimodal raw teaching data, and construct a teaching knowledge graph; The graph dynamic update module is used to divide the teaching knowledge graph into time periods according to the collection period, generate teaching knowledge graphs for different collection periods, construct time series graphs based on teaching knowledge graphs for different collection periods, perform node feature analysis on the time series graphs, and extract node feature sequences. A dynamic knowledge graph update model is constructed, using time series graphs and node feature sequences as training and validation sets. The dynamic knowledge graph update model is iteratively trained and its parameters are optimized. When the knowledge graph construction module collects multimodal raw teaching data from the latest collection period, the dynamic knowledge graph update model is incrementally trained. The predicted relation weights are obtained from the output layer of the dynamic update model based on the graph, and the relation weights of the edges between nodes in the teaching knowledge graph are updated based on the predicted relation weights. The cognitive dependency analysis module is used to obtain the potential ability coefficient, discrimination coefficient, and difficulty coefficient of each learning terminal based on the answer records of each learning terminal. It extracts the set of knowledge points and the domain semantic connection relationship between each knowledge point from several entities in the teaching knowledge graph. It constructs a directed knowledge path graph by treating each knowledge point as a node and the domain semantic connection relationship between nodes as directed edges between nodes. It then labels the directed edges in the directed knowledge path graph with relation weights based on the teaching knowledge graph to obtain the set of questions corresponding to each knowledge point in the directed knowledge path graph. Based on the potential ability coefficient of the learning terminal, the discrimination coefficient and difficulty coefficient of each question corresponding to each knowledge point in the directed knowledge path graph, and the relationship weight between each knowledge point, the condition mastery probability between each node in the directed knowledge path graph is obtained. The condition mastery probability between each node is standardized and converted into cognitive dependence coefficient. The specific formula for converting the probability of conditional mastery between various nodes into a cognitive dependence coefficient is as follows: ； in, This represents the conditional probability that a learning terminal can master knowledge point j given that it has already mastered knowledge point i. This represents the cognitive dependency coefficient between knowledge point i and knowledge point j. Let j represent the set of all prerequisite knowledge points for knowledge point j. This represents the length of the shortest path from i to j in the directed graph of knowledge paths. Ensure that the sum of the weights of all outgoing edges from the same node is 1; The personalized path planning module is used to generate personalized learning paths for each learning terminal based on the cognitive dependency coefficient between each knowledge point and the learning objectives. The teaching visualization module is used to add interactive attributes to nodes in the teaching knowledge graph according to the personalized learning path, generate VR virtual scenes and visualization data, and feed them back to the learning terminal. The real-time monitoring and feedback module is used to monitor the personalized learning paths of each learning terminal in real time and to provide real-time learning feedback based on the monitoring results.

2. The knowledge graph-based teaching demonstration system according to claim 1, characterized in that, The knowledge graph construction module periodically collects multimodal raw teaching data, performs entity naming and relation extraction on the multimodal raw teaching data, and the process of constructing the teaching knowledge graph includes: Regularly collect multimodal raw teaching data and set collection cycles, and perform format conversion, word segmentation, sentence segmentation, and part-of-speech tagging preprocessing operations on the multimodal raw teaching data; The preprocessed multimodal raw teaching data is input into a preset entity naming judgment model for entity judgment, and several entities included in the multimodal raw teaching data are obtained. A relation extraction model is constructed based on machine learning, and the multimodal raw teaching data and several entities are input into the relation extraction model for relation extraction to obtain the domain semantic connection relationship between several entities. By using several entities as nodes in a knowledge graph and the domain semantic connections between these entities as connections between nodes, a teaching knowledge graph is constructed, and the relational weights of the edges between nodes in the teaching knowledge graph are set.

3. The knowledge graph-based teaching demonstration system according to claim 2, characterized in that, The cognitive dependency analysis module obtains the potential ability coefficients of each learning terminal, the discrimination coefficients of each question, and the difficulty coefficients through the following process: Construct a joint answer matrix based on the answer records of each learning terminal, and randomly initialize the potential ability coefficient of each learning terminal, the discrimination coefficient of each question, and the difficulty coefficient in the answer matrix. The probability of each learning terminal answering each question correctly is obtained based on the potential ability coefficient, discrimination coefficient, and difficulty coefficient. Based on the joint answer matrix and the probability of each learning terminal answering each question correctly, a log-likelihood function is constructed. The discrimination coefficient and difficulty coefficient of each question are fixed, and the potential ability coefficient of the log-likelihood function is iteratively updated. Then, the potential ability coefficient of each learning terminal is fixed, and the discrimination coefficient and difficulty coefficient of the log-likelihood function are iteratively updated. The above process is repeated until the potential ability coefficient of each learning terminal, the discrimination coefficient and difficulty coefficient of each question converge. The potential ability coefficient of each learning terminal, the discrimination coefficient and difficulty coefficient of each question are output.

4. The knowledge graph-based teaching demonstration system according to claim 3, characterized in that, The personalized learning path planning module generates personalized learning paths for each learning terminal based on the cognitive dependency coefficients between knowledge points and learning objectives. This process includes: Based on the learning objectives of the learning terminal, obtain the starting and ending nodes of the directed knowledge path graph, extract all paths between the starting and ending nodes of the directed knowledge path graph, perform cognitive dependency aggregation on all paths, obtain the cumulative cognitive dependency coefficient of each path, and select the path with the largest cumulative cognitive dependency coefficient to generate a personalized learning path.

5. The knowledge graph-based teaching demonstration system according to claim 4, characterized in that, The instructional visualization module adds interactive attributes to nodes in the instructional knowledge graph based on personalized learning paths, generates VR virtual scenes and visualization data, and feeds it back to the learning terminal. The process includes: Obtain the correct answer probability, prerequisite relationships, and subject attributes for each knowledge point in the personalized learning path, and add interactive attributes to the nodes to which each knowledge point belongs in the teaching knowledge graph based on the correct answer probability, prerequisite relationships, and subject attributes for each knowledge point. A VR virtual scene is created based on the nodes in the teaching knowledge graph where interactive attributes have been added. A knowledge graph visualization model is constructed, and the visualization data of the nodes where interactive attributes have been added is output based on the knowledge graph visualization model. The VR virtual scene and visualization data are then fed back to the learning terminal.

6. The knowledge graph-based teaching demonstration system according to claim 5, characterized in that, The real-time monitoring and feedback module monitors the personalized learning paths of each learning terminal in real time, and the process of providing real-time learning feedback based on the monitoring results includes: The progress is segmented according to each knowledge point in the personalized learning path, with each knowledge point as a stage, forming a progress stage. Then, the learning time for each knowledge point is estimated based on the cognitive dependency coefficient between the knowledge point and the previous knowledge point in the personalized learning path, forming the estimated time for the progress stage. The end timestamp of each progress stage is used as the detection time point, and the judgment criteria for each progress stage are set according to the correct answer probability of each knowledge point in the personalized learning path. At the detection time point, the actual correct answer probability of the learning terminal at the current progress stage is obtained. The actual correct answer probability is compared with the judgment standard of the current progress stage. If the actual correct answer probability is less than the judgment standard, the answer record of the learning terminal at the current progress stage is extracted and input into the cognitive dependency analysis module. Then, a personalized learning path is regenerated according to the personalized path planning module. If the actual correct answer probability is greater than or equal to the judgment standard, the process of comparing the judgment standard of the next progress stage is automatically started.

7. A knowledge graph-based teaching demonstration method, specifically applied to the knowledge graph-based teaching demonstration system described in any one of claims 1 to 6, characterized in that, Includes the following steps: Step s1: Regularly collect multimodal raw teaching data, perform entity naming and relation extraction on the multimodal raw teaching data, construct a teaching knowledge graph, construct a dynamic update model for the graph, and dynamically update the relation weights of the teaching knowledge graph. Step s2: Based on the answer records of each learning terminal, obtain the potential ability coefficient, the discrimination coefficient and the difficulty coefficient of each question for each learning terminal, construct a directed graph of knowledge path, and obtain the cognitive dependency coefficient between each knowledge point corresponding to each learning terminal based on the potential ability coefficient of each learning terminal, the discrimination coefficient and difficulty coefficient of several questions corresponding to each knowledge point, and the relationship weight between each knowledge point. Step s3: Generate personalized learning paths for each learning terminal based on the cognitive dependency coefficients between the knowledge points corresponding to each learning terminal and the learning objectives. Step s4: Add interactive attributes to the nodes in the teaching knowledge graph according to the personalized learning path, generate VR virtual scenes and visualization data, and feed them back to the learning terminal; Step s5: Perform real-time learning monitoring on the personalized learning paths of each learning terminal, and perform real-time learning feedback operations based on the real-time learning monitoring results.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, which, when executed, implement the steps of the knowledge graph-based teaching and demonstration system as described in any one of claims 1 to 6.

Images