Intelligent teaching method based on knowledge graph

Abstract

The invention relates to the technical field of information, in particular to an intelligent teaching method based on a knowledge graph, and the method comprises the steps: obtaining knowledge points, carrying out the feature extraction of the knowledge points, and constructing the knowledge graph, the knowledge graph is constructed through a knowledge point topological structure, nodes in the knowledge point topological structure represent a knowledge point text, and nodes in the knowledge point topological structure represent a knowledge point text; the edges represent a hyponymy relation or a cross association relation between the knowledge point texts; acquiring student user information, performing feature extraction on the student user information, and constructing a learning path diagram for each student user; generating a plurality of questions based on the knowledge graph and the learning path graph, and screening the questions to obtain a question set for each student user; and guiding a corresponding student user to learn based on the question set, and completing intelligent teaching based on the knowledge graph. According to the invention, personalized and high-quality learning experience can be provided for students, and the teaching effect is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an intelligent teaching method based on knowledge graph. Background Art

[0002] In the process of constructing the knowledge graph, it is necessary to topologically sort the knowledge points according to the upper and lower relationships to determine the order of learning. This process seems simple, but there are some technical difficulties in actual operation. First, the relationship between knowledge points is intricate, and it is difficult to completely distinguish which are superordinate concepts and which are subordinate concepts. Some knowledge points are cross-related and cannot be simply divided into levels. Secondly, the number of knowledge points is huge, and it is a huge workload to manually sort out the upper and lower relationships, and it is easy to make mistakes. If you want to achieve automated topological sorting, you also need to deeply analyze the text description and attributes of the knowledge points, extract key information, and design appropriate algorithms, which puts high demands on both natural language processing and graph algorithms. There are also certain technical challenges in inserting questions that guide students to think actively on the basis of topological sorting. Questions should be able to help students establish logical connections between knowledge points and play a role in connecting the upper and lower levels. They should not be too simple, but not too difficult, and should be in line with the students' cognitive level. Generating such high-quality questions requires a deep understanding of the knowledge points, and the use of natural language generation technology to dynamically generate question texts based on the context. In addition, different students have different learning levels, and the difficulty of the questions should be differentiated. This requires the ability to profile students and generate personalized questions based on their characteristics. To achieve the above functions, it is necessary to comprehensively use multiple technologies such as knowledge graphs, natural language processing, and personalized recommendations. Therefore, a smart teaching method based on knowledge graphs is proposed. Summary of the invention

[0003] The purpose of this invention is to provide an intelligent teaching method based on knowledge graph, to provide students with personalized, high-quality learning experience, and to effectively improve teaching effectiveness.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A smart teaching method based on knowledge graph, including:

[0006] Acquire knowledge points, perform feature extraction on the knowledge points and construct a knowledge graph, wherein the knowledge graph is constructed by a knowledge point topological structure, in which nodes represent knowledge point texts, and edges represent hierarchical relationships or cross-correlation relationships between knowledge point texts;

[0007] Acquire student user information, perform feature extraction on the student user information and construct a learning path map for each student user;

[0008] Generate a number of questions based on the knowledge graph and the learning path graph, and filter the questions to obtain a question set for each student user;

[0009] Based on the question set, the corresponding student users are guided to study, and the intelligent teaching based on the knowledge graph is completed.

[0010] Optionally, acquiring the knowledge point includes:

[0011] Crawler technology is used to automatically capture knowledge points in online education platforms and academic papers, and the knowledge points are constructed into a knowledge point text collection.

[0012] Optionally, extracting features from the knowledge points and constructing a knowledge graph includes:

[0013] Using natural language processing technology to perform word segmentation and part-of-speech tagging on each knowledge point text in the knowledge point text set, obtaining a structured representation of the knowledge point text, and extracting keywords and subject words from the structured representation of the knowledge point text through semantic analysis to obtain a semantic feature vector of each knowledge point text;

[0014] Acquire the hierarchical relationship and cross-correlation relationship between the knowledge point texts in the knowledge point text set, and construct a semantic correlation matrix of the knowledge point text set;

[0015] The semantic feature vector and the semantic association matrix are fused and dimensionally reduced to obtain an embedded representation of each knowledge point text;

[0016] The knowledge point text set is clustered based on the embedded representation of each knowledge point text, and the topological structure of the knowledge point text set is constructed according to the clustering result to complete the construction of the knowledge graph.

[0017] Optionally, obtaining the student user information includes:

[0018] The learning behavior records of student users are collected, and the learning ability, knowledge mastery and learning interests of the student users are extracted based on the learning behavior records to complete the acquisition of the student user information.

[0019] Optionally, extracting features from the student user information and constructing a learning path map for each student user includes:

[0020] Construct a correlation map of all knowledge points within a subject;

[0021] Find similar users of the student user according to their learning interests, extract knowledge points learned by similar users, and construct a preliminary learning path map in combination with the association map;

[0022] The difficulty of the preliminary learning path map is optimized and adjusted based on the learning ability and knowledge mastery of the student user to obtain a final learning path map, wherein the final learning path map is a hierarchical learning path map divided according to difficulty.

[0023] Optionally, generating several questions based on the knowledge graph and the learning path graph includes:

[0024] Determine the related knowledge points according to the positions of the knowledge points learned by the student users in the knowledge graph;

[0025] Natural language generation technology is used to generate a number of questions based on the associated knowledge points and the learning path map of the student user.

[0026] Optionally, filtering the questions to obtain a question set for each student user includes:

[0027] Evaluate the difficulty of the generated questions and obtain the difficulty coefficient of each question;

[0028] Screening questions based on a preset question difficulty coefficient threshold to obtain a primary question set, wherein the question difficulty coefficient threshold is set according to the learning ability and knowledge mastery of the student user;

[0029] The questions in the primary question set are ranked by quality score, and the top N questions with the highest ranking scores are taken as the final question set.

[0030] Optionally, the method further includes updating the knowledge graph by regularly capturing new knowledge points from online education platforms and academic papers.

[0031] The beneficial effects of the present invention are:

[0032] The present invention uses natural language processing technology to perform semantic analysis on knowledge points and construct a knowledge graph. It uses collaborative filtering and content-based recommendation algorithms to generate personalized learning paths based on student user portraits. At the same time, the present invention can also automatically generate guiding questions and dynamically adjust the difficulty according to student characteristics. It also uses an incremental method to continuously update the knowledge graph through knowledge point updates, which can provide students with personalized, high-quality learning experiences and effectively improve teaching results. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0034] Figure 1 This is a flow chart of a knowledge graph-based intelligent teaching method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0037] This embodiment provides a smart teaching method based on knowledge graph, such as Figure 1 As shown, including:

[0038] Step S101: Acquire knowledge points, extract features from the knowledge points and construct a knowledge graph;

[0039] Through natural language processing technology, semantic analysis is performed on the knowledge point text, keywords and subject terms are extracted, and combined with the pre-constructed domain knowledge graph, the hierarchical and cross-correlation relationships between knowledge points are determined to construct a knowledge graph. The knowledge graph includes the topological structure of knowledge points. The nodes in the topological structure represent the knowledge point text, and the edges represent the hierarchical and cross-correlation relationships between the knowledge point texts.

[0040] Specifically, crawler technology is used to automatically capture knowledge point data from online education platforms and academic papers to build a knowledge point text set; for each knowledge point text in the knowledge point text set, natural language processing technology is used to perform word segmentation and part-of-speech tagging to obtain a structured representation of the knowledge point text; for the structured representation of the knowledge point text, keywords and subject terms are extracted through a semantic analysis algorithm to obtain the semantic feature vector corresponding to the knowledge point text; a pre-constructed domain knowledge graph is obtained, and the hierarchical and cross-correlation relationships between the knowledge point texts in the knowledge point text set are obtained according to the domain knowledge graph, and a semantic association matrix of the knowledge point text set is constructed; the semantic features of the knowledge point text set are combined into a matrix. The eigenvector is fused with the semantic association matrix, and the fused knowledge points are mapped to the low-dimensional semantic space using a graph embedding algorithm to obtain the embedded representation of each knowledge point text in the knowledge point text set; based on the embedded representation of each knowledge point text in the knowledge point text set, the knowledge point text set is clustered using a clustering algorithm to obtain the clustering result of the knowledge point text set; based on the clustering result, the topological structure of the knowledge point text set is constructed, the nodes in the topological structure represent the knowledge point texts, and the edges represent the hierarchical relationship or cross-correlation relationship between the knowledge point texts; the topological structure of the knowledge point text set is visualized to complete the construction of the knowledge graph, which can provide interactive exploration and navigation functions.

[0041] Specifically, this embodiment first obtains a knowledge point text set from various sources such as literature, textbooks, and papers, including 500 articles in the fields of machine learning, deep learning, and natural language processing. Then, the jieba word segmentation tool is used to segment and tag each article, and the part-of-speech sequence is obtained as a structured representation of the knowledge point text. Then, the TextRank algorithm is used to extract 20 keywords and 5 subject words for each article to generate a 100-dimensional semantic feature vector. At the same time, a knowledge graph in the field of artificial intelligence is constructed based on the CN-DBpedia knowledge base, and the upper and lower relationships between knowledge point texts such as "machine learning-supervised learning-classification" and cross-correlation relationships such as "neural network-deep learning" are obtained through knowledge reasoning to construct a 500*500 semantic association matrix. After concat the semantic feature vector and the semantic association matrix, they are input into the GraphSAGE model to learn the 128-dimensional knowledge point embedding representation. Clustering is performed based on the embedded representation using the K-Means algorithm, and the optimal number of clusters K=20 is obtained by evaluating the silhouette coefficient. Finally, the Gephi tool is used to visualize the clustering results and generate a topological structure diagram of knowledge points. It also provides interactive functions such as node filtering, local zooming, and information floating box to facilitate users to explore the connections and hierarchical structures between knowledge points and tap into potential knowledge context.

[0042] According to the domain knowledge graph, the upper and lower relations and cross-correlation relations between the knowledge point texts in the knowledge point text set are obtained. If there is a upper and lower relation or cross-correlation relation between the knowledge point texts, the corresponding position in the semantic correlation matrix is ​​marked as 1, otherwise it is marked as 0. The semantic correlation matrix is ​​trained by the deep learning algorithm to obtain the semantic similarity matrix between the knowledge point texts. According to the preset similarity threshold, the matrix position corresponding to the knowledge point text with a semantic similarity greater than the threshold is marked as 1, indicating that the two knowledge point texts are semantically related. According to the semantic similarity matrix, the clustering algorithm is used to cluster the knowledge point text set, and the semantically related knowledge point texts are aggregated into a knowledge point cluster. The knowledge point texts in the knowledge point cluster are obtained, and the keywords of the knowledge point texts in the knowledge point cluster are extracted by natural language processing technology. According to the co-occurrence relationship between the keywords, the semantic network of the knowledge point cluster is constructed. For each knowledge point text in the knowledge point text set, the importance of the knowledge point text is calculated by the random walk algorithm according to its association relationship with other knowledge point texts in the semantic correlation matrix. According to the importance ranking results of knowledge point texts, the importance of knowledge point text nodes in the semantic network is annotated, and the knowledge point text nodes with high importance are highlighted. The semantic network is visualized to form an intuitive knowledge point text semantic association map, which helps users quickly grasp the semantic context and association structure of the knowledge point text set.

[0043] When obtaining the relationship between each knowledge point text in the knowledge point text set according to the domain knowledge graph, a rule-based method can be used. For example, a series of rules and patterns are defined, such as "A is a kind of B" indicates that there is a hierarchical relationship between A and B, "A is related to B" indicates that there is a cross-correlation relationship between A and B, etc., and then natural language processing technologies such as syntactic analysis and dependency analysis are used to identify the relationship patterns in the knowledge point text and mark the corresponding positions in the semantic association matrix. When training the semantic association matrix to obtain the semantic similarity matrix, a word embedding model such as Word2Vec can be used, with the knowledge point text as input, and the low-dimensional dense vector representation of the knowledge point text is learned through a neural network, and then the cosine similarity between the vectors is calculated to obtain the semantic similarity matrix. Set the similarity threshold to 8, and mark the matrix position corresponding to the knowledge point text with a similarity greater than 8 as 1. The clustering algorithm can select a hierarchical clustering algorithm, such as the AGNES algorithm, which takes the semantic similarity matrix as input and gradually aggregates the semantically related knowledge point texts through a bottom-up hierarchical clustering process to form a knowledge point cluster. When constructing the semantic network of knowledge point clusters, the TextRank algorithm can be used to extract keywords from the knowledge point text within the knowledge point cluster. By counting the co-occurrence frequency of keywords and calculating the correlation between keywords, the edges of the semantic network are generated. The PageRank algorithm is used to calculate the importance of the semantic network. The importance score is calculated based on the in-degree and out-degree of the knowledge point text node. The higher the score, the more important the node is in the network. Visual attributes such as node size and color can be used to highlight important nodes. Finally, the semantic network is visualized using graph databases and visualization tools such as Gephi and Neo4j. Based on the force-directed layout algorithm, a semantic association map of knowledge point text, namely the knowledge map, is generated. Interactive operation functions are provided to facilitate users to explore the semantic connections between knowledge points.

[0044] Step S102: Obtain student user information, extract features of student users, and construct a learning path map for each student user;

[0045] Based on student user portrait data, collaborative filtering algorithm and content-based recommendation algorithm are used to personalize and recommend knowledge points, and generate learning path maps suitable for different students' cognitive levels.

[0046] Specifically, the personal portrait data of student users is obtained, including learning ability, knowledge mastery, learning interests, etc., to construct a user feature vector. Semantic analysis and association mining are performed on subject knowledge points to construct an association map between knowledge points and determine the prerequisite and subsequent relationship of knowledge points. A collaborative filtering algorithm is used to find similar users based on the learning behavior and preferences of student users, and knowledge points are personalized and sorted to generate a preliminary learning path map. Through a content-based recommendation algorithm, the knowledge points that students have mastered and those that have not yet mastered are analyzed, the importance and difficulty of the knowledge points are evaluated, and the preliminary learning path map is optimized and adjusted. According to the students' cognitive level and learning ability, the difficulty of the optimized learning path map is divided to generate a hierarchical learning path map that meets the students' personal situation. In the process of generating the learning path map, the students' learning progress and mastery are dynamically tracked, and the learning path map is adjusted and updated in real time based on feedback data. Finally, a personalized learning path map for each student is output, and a visual learning path navigation and knowledge point recommendation are provided to guide students to conduct independent learning and exploration.

[0047] Specifically, in order to obtain the personal portrait data of student users. Collect students' learning behavior records, such as homework completion status, test scores, etc., and evaluate students' learning ability, knowledge mastery and learning interest through data analysis and machine learning algorithms. For example, according to the students' answer accuracy and completion time on different knowledge points, calculate the students' mastery score of each knowledge point, and then divide the students into different ability levels through clustering algorithms. At the same time, perform natural language processing and semantic analysis on subject knowledge points, calculate the similarity and correlation between knowledge points, and construct a knowledge point correlation map. Use graph algorithms, such as the shortest path algorithm, to determine the optimal learning order of knowledge points. When generating personalized learning paths, use collaborative filtering algorithms to calculate the similarity between students based on students' learning behaviors and preferences, such as learning time, access frequency, etc., find the Top-N users most similar to the target students, and use the knowledge points learned by these users as candidate sets. Combined with the students' ability level and the difficulty of the knowledge points, sort the knowledge points in the candidate set to generate a preliminary learning path map. Next, through the content-based recommendation algorithm, the knowledge points that students have mastered and have not mastered are analyzed, the importance score and difficulty score of each knowledge point are calculated, and the knowledge points in the preliminary learning path map are optimized and adjusted, such as adjusting the order of knowledge points, adding or deleting certain knowledge points, etc. The difficulty of the optimized learning path map is divided according to the students' cognitive level and learning ability. For example, the knowledge points are divided into different difficulty levels such as basic, advanced, and improved to generate a hierarchical learning path map. During the students' learning process, the learning behavior and performance of the students, such as the correct answer rate and completion time, are tracked in real time, the students' mastery is calculated, and the learning path is adjusted dynamically. For example, when students have difficulty mastering a certain knowledge point, the prerequisite knowledge points of the knowledge point are automatically added to strengthen the learning of basic knowledge. Finally, a personalized learning path map is output, and a visual learning path navigation interface is provided to clearly display the logical relationship and recommended learning order of each knowledge point, guiding students to conduct independent learning and exploration.

[0048] A collaborative filtering algorithm is used to find similar users based on the learning behavior and preferences of student users, personalize the sorting of knowledge points, and generate a preliminary learning path map.

[0049] Obtain the learning behavior data and learning preference data of student users, analyze the learning behavior data, and extract learning behavior features; based on the learning behavior features, use a collaborative filtering algorithm to calculate the similarity between different users and find the user group that is most similar to the target student user; obtain the learning preference data of similar user groups, analyze the learning preference data, and obtain the degree of preference of similar user groups for different knowledge points; based on the degree of preference of similar user groups for knowledge points, use a sorting algorithm to perform personalized sorting of knowledge points to obtain the personalized knowledge point sorting results for the target student users; based on the personalized knowledge point sorting results, use a path generation algorithm to generate a preliminary version of the personalized learning path map for the target student users; obtain the dependency and prerequisite relationships between knowledge points pre-defined by subject experts, optimize the preliminary version of the personalized learning path map, adjust the order of knowledge points, and ensure that the learning path conforms to the dependency and prerequisite relationships of knowledge points; visualize the optimized personalized learning path map to generate the final personalized learning path map, and provide it to the target student users for personalized learning.

[0050] Specifically, when obtaining the learning behavior data of student users, data such as the learning time, learning frequency, and number of clicks on learning resources on the student online learning platform can be collected, and data mining techniques such as association rule mining algorithms can be used to discover students' learning behavior patterns and extract learning behavior feature vectors. Then, the cosine similarity algorithm is used to calculate the similarity between different users, and the top 20% of users with the highest similarity to the target student are selected as similar user groups. The learning preference data of similar user groups are analyzed, and the TF-IDF algorithm is used to calculate the preference weight of each knowledge point to obtain the preference degree vector of similar user groups for different knowledge points. According to the vector, the knowledge points are sorted in descending order to obtain the personalized knowledge point sorting results of the target students. Then, the shortest path algorithm is used to generate a preliminary personalized learning path map with the sorted knowledge points as nodes and the dependency relationships between knowledge points as edges. Finally, the preliminary learning path map is optimized using the knowledge point prerequisite relationship matrix pre-defined by subject experts, and the order of knowledge points is adjusted by the topological sorting algorithm to ensure that the learning path conforms to the dependency and prerequisite relationships of knowledge points. The optimized personalized learning path map can be displayed using a directed graph visualization algorithm to provide clear and intuitive learning path guidance for target students.

[0051] Step S103: Generate a number of questions based on the knowledge graph and the learning path graph, and screen the questions to obtain a question set for each student user;

[0052] Based on the topological sorting in the knowledge graph, natural language generation technology is used in combination with the students' learning path map to generate questions that guide students' thinking through question generation strategies. Questions are screened according to difficulty adjustment strategies and question quality assessment strategies to obtain a set of questions for each student user.

[0053] According to the position of the knowledge points currently learned by students in the knowledge graph, the knowledge points associated with them are determined, and the difficulty coefficients of the associated knowledge points are obtained; natural language generation technology is used to automatically generate questions to guide students' thinking based on the students' learning path map and the associated knowledge point information; the difficulty of the generated questions is evaluated to obtain the difficulty coefficient of the questions, and the difficulty evaluation is combined with the difficulty of the knowledge points involved in the questions and the complexity of the question statements; according to the students' knowledge mastery and learning ability, the threshold of the difficulty coefficient of the questions is dynamically adjusted, and the questions are screened according to the threshold to obtain a primary question set; further comprehensively consider the relevance, importance, difficulty matching and other indicators of the questions in the first question set, score and sort the generated questions, and screen out high-quality questions as the final question set for actual teaching applications.

[0054] Specifically, based on the knowledge graph and the student's learning path map, graph algorithms such as the shortest path algorithm are used to obtain several knowledge points that are closest to the current knowledge point and have the highest correlation, and their difficulty and quality are evaluated. The difficulty evaluation can comprehensively consider factors such as the complexity of the knowledge point and the student's mastery level, and the quality evaluation comprehensively considers indicators such as the relevance, importance, and difficulty matching of the question. When generating guiding questions, natural language processing technology is used to perform semantic analysis and key information extraction on the text description of the knowledge point, and then combined with the student's learning path map, the question text is automatically constructed through predefined question templates and generation rules.

[0055] The difficulty of questions can be assessed by establishing a difficulty assessment model, taking into account the difficulty coefficient of the knowledge points involved in the question, the semantic complexity of the question description, the cognitive difficulty of the question type, etc., and calculating the difficulty coefficient of the question. According to the students' knowledge mastery and learning ability, the threshold of the difficulty coefficient of the question can be dynamically adjusted, for example, setting an interval range of 6 to 8 to generate a question sequence that meets the students' personal abilities.

[0056] Question quality assessment can construct a question quality assessment model, comprehensively consider indicators such as the relevance, importance, and difficulty matching of the questions, score and rank the generated questions, and screen out high-quality questions for actual teaching applications.

[0057] For each question in the primary question set, extract the text feature vector of the question; determine the relevance of each question to the teaching topic to obtain the relevance score of each question; determine the importance of each question to the teaching objective to obtain the importance score of each question; obtain the target students' knowledge mastery data, determine the matching degree of each question with the target students' knowledge level, and obtain the difficulty matching score of each question; perform weighted summation of the relevance score, importance score and difficulty matching score of each question to obtain the comprehensive quality score of each question; sort the primary question set in descending order according to the comprehensive quality score, and use the top N questions with the highest scores after sorting as the final question set, which are output for actual teaching applications, where N is a preset positive integer.

[0058] Specifically, for each question in the primary question set, the pre-trained BERT model is used to extract its text feature vector to obtain a 768-dimensional embedded representation. Then, a CNN network model pre-trained on a large-scale question data set is used, and the question feature vector is used as input to determine its relevance to the teaching topic, and a relevance score between 0 and 1 is output. At the same time, another pre-trained LSTM network model is used, and the Word2Vec representation of the question features and teaching objectives is used as input to evaluate the importance of the question to the teaching objective and output the importance score. In addition, according to the knowledge mastery of the target students in previous tests, the Bayesian knowledge tracking model is used to estimate their current knowledge level, and the relationship between the difficulty of the problem and the student's ability is fitted using a logistic regression model, thereby determining the degree of match between each question and the student's level and generating a difficulty matching score. Finally, the three-dimensional scores of each question are weighted and summed, and the weights can be adjusted according to actual teaching needs. In this embodiment, it is assumed that the weights of relevance, importance, and difficulty matching are 3, 5, and 2, respectively, thereby obtaining a comprehensive quality score for each question. The question set is sorted from high to low according to the comprehensive quality score, and the top 20 questions are taken as the final question set, which is applied to actual teaching. Through this question quality assessment process, we can more comprehensively examine the various attributes of generated questions and automatically screen out high-quality, highly relevant questions that match students' cognitive levels, thereby improving the practicality and effectiveness of the intelligent teaching system.

[0059] Step S104: guiding the corresponding student users to learn based on the question set, and obtaining the learning effect of the student users, and optimizing the question generation strategy and difficulty adjustment strategy through the feedback of the learning effect.

[0060] The final set of questions generated is presented to students to obtain their answers and feedback information; based on the answers and feedback information, the question generation strategy and difficulty adjustment strategy are continuously optimized.

[0061] Specifically, the generated question sequence is presented to students in real time, and students' answers and feedback information are recorded, such as the correct answer rate, answering time, subjective evaluation, etc. By collecting and analyzing this data and using machine learning algorithms, the question generation strategy and difficulty adjustment strategy are continuously optimized to achieve intelligent and personalized question push. For example, when a student answers incorrectly three times in a row, the difficulty coefficient of the next question can be automatically reduced, and when a student answers correctly five times in a row, the difficulty is appropriately increased to maintain the challenge and fun of learning. Through this dynamic adjustment mechanism, real-time optimization and personalized adaptation of the learning process are achieved.

[0062] To further optimize the technical solution, the present embodiment provides a knowledge graph-based intelligent teaching method, which also includes: regularly optimizing the knowledge graph.

[0063] In order to solve the problem of large number of knowledge points and complex relationships, an incremental knowledge graph construction method is adopted. By regularly acquiring new knowledge points from online education platforms, academic papers and other channels, and using natural language processing technology to parse and integrate them, the existing knowledge graph is continuously expanded and optimized.

[0064] Specifically, knowledge point data from online education platforms and academic papers are obtained, and semantic analysis of the knowledge point data is performed through natural language processing technology to obtain key concepts and relationship information; the key concepts and relationship information are merged with the pre-built knowledge graph to obtain a fused knowledge graph. If the similarity between the key concepts and relationship information and the existing concepts and relationships in the knowledge graph exceeds a preset threshold, the key concepts and relationship information are merged into the corresponding existing nodes and edges. If the key concepts and relationship information cannot be matched with the knowledge graph, new nodes and connection edges are created in the knowledge graph based on their semantic attribute information; a representation learning algorithm based on a graph neural network is used to generate a fused knowledge graph based on the fused knowledge graph. The topological structure and node attribute information of the fusion knowledge graph are obtained to obtain a low-dimensional vector representation of each concept node; the semantic similarity between different concept nodes in the fused knowledge graph is calculated through the knowledge graph embedding model, and potential association relationships are mined to obtain an enriched knowledge graph; for the key concept nodes in the enriched knowledge graph, a knowledge representation model based on the attention mechanism is constructed, and the weight of the node in the enriched knowledge graph is dynamically adjusted according to its importance; the quality of the enriched knowledge graph is regularly evaluated, and the noise data and erroneous relationships in the enriched knowledge graph are identified through manual annotation and verification, and the enriched knowledge graph is corrected to obtain an optimized knowledge graph.

[0065] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A smart teaching method based on knowledge graph, characterized in that: include: Acquire knowledge points, perform feature extraction on the knowledge points and construct a knowledge graph, wherein the knowledge graph is constructed by a knowledge point topological structure, in which nodes represent knowledge point texts, and edges represent hierarchical relationships or cross-correlation relationships between knowledge point texts; Acquire student user information, perform feature extraction on the student user information and construct a learning path map for each student user; Generate a number of questions based on the knowledge graph and the learning path graph, and filter the questions to obtain a question set for each student user; Based on the question set, the corresponding student users are guided to study, and the intelligent teaching based on the knowledge graph is completed.

2. The intelligent teaching method based on knowledge graph according to claim 1 is characterized in that: Acquiring the knowledge points includes: Crawler technology is used to automatically capture knowledge points in online education platforms and academic papers, and the knowledge points are constructed into a knowledge point text collection.

3. The intelligent teaching method based on knowledge graph according to claim 2 is characterized in that: Extracting features from the knowledge points and constructing a knowledge graph includes: Using natural language processing technology to perform word segmentation and part-of-speech tagging on each knowledge point text in the knowledge point text set, obtaining a structured representation of the knowledge point text, and extracting keywords and subject words from the structured representation of the knowledge point text through semantic analysis to obtain a semantic feature vector of each knowledge point text; Acquire the hierarchical relationship and cross-correlation relationship between the knowledge point texts in the knowledge point text set, and construct a semantic correlation matrix of the knowledge point text set; The semantic feature vector and the semantic association matrix are fused and dimensionally reduced to obtain an embedded representation of each knowledge point text; The knowledge point text set is clustered based on the embedded representation of each knowledge point text, and the topological structure of the knowledge point text set is constructed according to the clustering result to complete the construction of the knowledge graph.

4. The intelligent teaching method based on knowledge graph according to claim 1 is characterized in that: Acquiring the student user information includes: The learning behavior records of student users are collected, and the learning ability, knowledge mastery and learning interests of the student users are extracted based on the learning behavior records to complete the acquisition of the student user information.

5. The intelligent teaching method based on knowledge graph according to claim 4 is characterized in that: Extracting features from the student user information and constructing a learning path map for each student user includes: Construct a correlation map of all knowledge points within a subject; Find similar users of the student user according to their learning interests, extract knowledge points learned by similar users, and construct a preliminary learning path map in combination with the association map; The difficulty of the preliminary learning path map is optimized and adjusted based on the learning ability and knowledge mastery of the student user to obtain a final learning path map, wherein the final learning path map is a hierarchical learning path map divided according to difficulty.

6. The intelligent teaching method based on knowledge graph according to claim 1 is characterized in that: Based on the knowledge graph and learning path graph, several questions are generated, including: Determine the related knowledge points according to the positions of the knowledge points learned by the student users in the knowledge graph; Natural language generation technology is used to generate a number of questions based on the associated knowledge points and the learning path map of the student user.

7. The intelligent teaching method based on knowledge graph according to claim 6 is characterized in that: The questions are screened to obtain a set of questions for each student user, including: Evaluate the difficulty of the generated questions and obtain the difficulty coefficient of each question; Screening questions based on a preset question difficulty coefficient threshold to obtain a primary question set, wherein the question difficulty coefficient threshold is set according to the learning ability and knowledge mastery of the student user; The questions in the primary question set are ranked by quality score, and the top N questions with the highest ranking scores are taken as the final question set.

8. The intelligent teaching method based on knowledge graph according to any one of claims 1 to 7, characterized in that: The method also includes updating the knowledge graph by regularly capturing new knowledge points from online education platforms and academic papers.

Images