Course knowledge graph construction method based on multi-knowledge-point intelligent identification

Abstract

The invention discloses a curriculum knowledge graph construction method based on multi-knowledge-point intelligent recognition, which comprises the following steps: preprocessing test question data, extracting test question features, further performing knowledge point recognition, and finding out a hyperplane which can correctly divide a training set into different types of data and has the maximum interval through an SVM hyperplane solving step. The test questions corresponding to different knowledge points are distinguished through the hyperplane, then course knowledge graph construction is carried out, and finally knowledge point query is carried out. According to the method, SVM classification is carried out on various knowledge points, multi-label classification of test question knowledge points is realized, a test question word frequency matrix is constructed, then test question features capable of completely covering all test questions of different knowledge points are extracted from the test questions of the different knowledge points by adopting a TF-TDF method, the word frequency matrix is optimized by the test question features of all the knowledge points, and the test question features of all the knowledge points are extracted. The SVM model is trained through the optimized word frequency matrix, classification of single knowledge points of the test questions is achieved, and finally recognition of the knowledge points of the test questions is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of educational data mining, and in particular to a method for constructing a course knowledge graph based on intelligent identification of multiple knowledge points. Background Art

[0002] In existing teaching resources, knowledge points in the knowledge system are distributed in the form of chapter levels. However, in the face of fiercely competitive academic assessments, the difficulty of various test questions is increasing, and the focus is on the integration of knowledge points. Therefore, the knowledge system distributed in the form of chapter levels can no longer meet the needs of students' learning and teachers' teaching. The present invention believes that the knowledge system with a network structure can show the relationship between knowledge points and better meet the needs of students and teachers.

[0003] In the teaching process, the class teaching system is the most important and basic teaching organization form in my country. The class teaching system is conducive to the economical and efficient training of talents on a large scale, and is conducive to teaching management and teaching inspection. However, the relationship between teachers and students who teach each subject in the class is one-to-many, and teachers cannot accurately grasp the learning status of each student. Teachers assign the same homework to each student during the teaching process. In the face of fiercely competitive academic assessments, students have to rely on the "sea of ​​questions tactics" to improve their ability to apply knowledge points. However, due to the different mastery of each student, the sea of ​​questions tactics will waste a lot of time learning the knowledge that has been mastered, and some students are not good at summarizing, but just do questions blindly without understanding the situation of the knowledge points they have not mastered.

[0004] A method for constructing a course knowledge graph based on GMM is disclosed in a Chinese patent document with announcement number CN114595337B. The method for constructing a course knowledge graph based on GMM includes: grouping test questions by chapters, preprocessing and Chinese word segmentation of test questions in each chapter to obtain structured test question data, and then performing Jieba word segmentation on valid test question strings to generate a word frequency matrix for the test questions; based on the processed test question word frequency matrix, using the GMM model to cluster test question knowledge points and extract features to generate a test question knowledge point model; at the same time, based on the structured test question data, feature recognition and extraction of chapter knowledge points are performed to generate a chapter knowledge point model; based on the generated test question knowledge point model and chapter knowledge point model, the knowledge graph technology is used to integrate the two types of knowledge points, test questions and chapters, into a course knowledge graph. The present invention takes a large number of course test questions as the research object, uses the Gaussian mixture clustering method to identify the test question knowledge points and their associations, and combines the existing chapter knowledge point system and uses the knowledge graph technology to realize the reconstruction of the course knowledge system. However, the method of constructing a course knowledge graph based on GMM adopts the GMM assumption that the data is generated by a mixture of multiple Gaussian distributions, and describes the distribution of the data by estimating the parameters and mixing ratio of each Gaussian distribution. It cannot directly process linearly separable and linearly inseparable data.

[0005] A method, device, terminal and storage medium for constructing a course knowledge graph disclosed in a Chinese patent document with announcement number CN111797222B include: extracting domain keywords from the full text data and arbitrarily combining them into keyword pairs, calculating the co-occurrence features of the keyword pairs in the same text data; clustering the keyword pairs according to the co-occurrence features, and screening out multiple positive keyword pairs; determining the first keyword as the first-level knowledge point based on multiple positive keyword pairs and determining multiple second-level knowledge points corresponding to the first keyword; extracting the problem points corresponding to each knowledge point and extracting the related courses corresponding to the knowledge points; constructing a course knowledge graph based on the knowledge points, problem points and related courses, and recommending courses. The present invention can be applied to smart education, and the quality and efficiency of course recommendations can be improved by establishing a course knowledge graph. In addition, the present invention also relates to a blockchain, and the course knowledge graph is stored in a blockchain node. However, the method, device, terminal and storage medium for constructing the course knowledge graph mainly focus on the step-by-step construction of knowledge points, and have poor ability to review knowledge points and summarize weak knowledge points.

[0006] The existing teaching knowledge framework is mostly distributed in the form of chapter levels, which cannot meet the needs of students' learning and teachers' teaching.

[0007] In order to solve the deficiencies in the above-mentioned prior art, it is a problem worth studying to provide a method for constructing a course knowledge graph based on intelligent identification of multiple knowledge points. Summary of the invention

[0008] The purpose of the present invention is to overcome the shortcomings of existing teaching knowledge architecture that is mostly distributed in the form of chapter levels and cannot meet the needs of student learning and teacher teaching, and to provide a method for constructing a course knowledge graph based on intelligent identification of multiple knowledge points, thereby achieving the technical effect of forming a network structure knowledge system according to the student's learning situation.

[0009] The purpose of the present invention is achieved through the following technical solutions:

[0010] A method for constructing a course knowledge graph based on intelligent identification of multiple knowledge points includes the following steps:

[0011] Step A: Test question data preprocessing: first introduce the key attributes of "chapter", "question type" and "knowledge point" to create a test question data model, then build a test question word frequency matrix and perform word segmentation on the test questions;

[0012] Step B: Extract question features. By calculating the term frequency (TF) and inverse document frequency (IDF) of each phrase in the term frequency matrix, the TF-IDF value is obtained to extract question features. For each question, the TF-IDF values of all phrases form a vector, and the vector serves as the feature vector of this question, which is subsequently used for question classification and knowledge point association operations;

[0013] Step C: Knowledge point identification. Through the SVM hyperplane solution step, find the hyperplane that can correctly divide the training set into different categories of data and has the largest margin, and distinguish the questions corresponding to different knowledge points through this hyperplane;

[0014] Step D: Course knowledge graph construction. By designing a course knowledge graph construction algorithm, add all the nodes in the node file to the knowledge graph, and then traverse the triple file, and add the triples that meet the conditions (both the head node and the tail node are in the node set) to the knowledge graph. Finally, a complete knowledge graph is obtained, and this knowledge graph is used to represent the relationships between course knowledge;

[0015] Step E: Knowledge point query. By designing a knowledge point query algorithm to query relevant or irrelevant knowledge points in the constructed knowledge graph, a knowledge graph related to the query knowledge point is obtained.

[0016] The said Step A includes:

[0017] Step A1: Picture cleaning, process the text information in the test paper;

[0018] Step A2: Question data modeling. When modeling the question data, introduce the attributes of "chapter", "question type", and "knowledge point" of the question, and create a question data model, that is, the question data is defined as where q i = W, P, c, l, k represents the i-th question, W represents the question data after cleaning the picture, P represents the word segmentation result, c represents the chapter to which the question belongs, l represents the type of the question, k represents the question knowledge point, and n represents the number of questions;

[0019] Step A3: Construct a question term frequency matrix. Use the Chinese word segmentation component jieba to segment the questions to obtain P s , after completing the word segmentation, clean the word segmentation result, and filter out irrelevant phrases such as prepositions and adjectives P a , so that q i .P = P s -P a , the complete set of phrases after word segmentation and screening of all question data is represented as p s = {p j |0 < j ≤ m}, m represents the number of phrases, p jrepresents the jth phrase in the phrase set, and then constructs the word frequency matrix M based on the phrase set. <f ij >, where f ij Indicates whether the jth phrase appears in the pth j For each test question, we construct a test question word frequency matrix and a test question data model to realize intelligent recognition of the test question and quickly identify the chapter to which the test question belongs, the question type and the knowledge points involved.

[0020] The step B comprises:

[0021] Step B1: Calculation of the word frequency of each phrase in the word frequency matrix, i.e., TF, represents the frequency of the phrase appearing in the test questions corresponding to the knowledge point. The word frequency matrix M is obtained from step A3. The word frequency calculation formula is as follows:

[0022]

[0023] Among them, t c It indicates the number of test questions corresponding to the knowledge point in which the word or phrase appears. 总 Refers to the total number of test questions corresponding to this knowledge point;

[0024] Step B2: Calculate the inverse document frequency of each phrase in the word frequency matrix, i.e., IDF. This method uses IDF to obtain the general importance measure of each phrase. The IDF calculation formula is as follows:

[0025]

[0026] Among them, T 总 It represents the total number of test questions in the test question bank. In order to avoid the situation where the denominator is 0, the denominator in the formula needs to be added with 1;

[0027] Step B3: TF-IDF calculation is used to evaluate the importance of a phrase to a phrase in a test question set. The calculation formula is as follows:

[0028] TF-IDF=TF*IDF

[0029] The test questions are characterized by extracting test questions data, optimizing the word frequency matrix based on these test questions characteristics, and using the optimized word frequency matrix to train the classification model to identify questions with different knowledge points. The TF-IDF value can accurately reflect the key features of the test questions content and can more accurately determine the category of the test questions and the associated knowledge points.

[0030] The step C comprises:

[0031] Step C1: Define the training set D = {(x1,y1),(x2,y2),…,(x M ,y M)}, where x is an n-dimensional feature vector with a value of 1 or -1, select the kernel function H(x,z) and the penalty coefficient C, calculate the hyperplane when the interval is the largest, construct the constrained optimization problem as shown in the following formula, and express the solution of the vector as follows,

[0032]

[0033] 0≤α i ≤C

[0034] Among them, α is the Lagrange multiplier vector, and the C value represents the degree of penalty for misclassified points. The larger the C value, the greater the penalty for misclassified points.

[0035] Step C2: SVM classification decision function, using the α calculated in step C1 * , select α * A positive component of Calculate b * , the formula is as follows:

[0036]

[0037] The SVM classification decision function is shown in the following formula:

[0038]

[0039] The step D comprises:

[0040] Step D1: Definition of knowledge graph, definition Among them, Z stands for "chapter", S stands for "section" and "K" stands for knowledge point, represents the head entity, represents the relationship, represents the tail entity, represents the relationship between chapters and sections in the chapter level, represents the relationship between sections and knowledge points in the chapter level, and represents the relationship between knowledge points;

[0041] Step D2: Definition of node file E and HRT, extracting entities "chapter", "section" and "knowledge point" from the textbook through manual mode, and defining the relationship between entities "chapter" and "section", "section" and "knowledge point" in the form of chapter hierarchy of knowledge points in the textbook, E includes chapter entity, section entity and knowledge point entity, and contains the color and size of the entity to indicate the entity type and the frequency of the entity in the test paper; this method recognizes the knowledge point information from the test questions according to the test question knowledge point recognition algorithm proposed in step B and step C;

[0042] Step D3: Construction of the knowledge graph. Based on the node file E obtained in step D1 and step D2, and the triple file HRT, first add all the nodes in the node file to the knowledge graph, then traverse the triple file, and according to the information stored in the triple, add the triples that meet the conditions (the head node and the tail node are both in the node set) to the knowledge graph, and finally obtain a complete knowledge graph. The course knowledge graph provides a guide for resource construction for educational managers, which helps them to reasonably allocate and optimize teaching resources according to the structure and relationship of the knowledge graph.

[0043] The step E queries the knowledge graph related to the given knowledge point. First, the knowledge point e1 to be queried is input, and a link is made to the neo4j database Z storing related information. The knowledge point to be queried is checked to see if it is in the database. If it is, all triples with e1 as the head entity are queried from the database Z, that is, h=e1, and the results are stored in h1, r1, and t1 respectively. The same operation is performed to query with e1 as the tail entity, and new triples h2, r2, and t2 are constructed. The query results are merged, and the results of the query with e1 as the head entity and the tail entity are merged to obtain h3, r3, and t3 to construct the knowledge graph G(e1) and the relationship list L(e1) related to the knowledge point e1 and output them. If the query e1 is not in Z, the output shows that the entity is not added.

[0044] Positive and beneficial effects: 1. The course knowledge graph construction method based on intelligent identification of multiple knowledge points realizes multi-label classification of test knowledge points by performing SVM classification on various knowledge points respectively. First, the test data is preprocessed, and the Chinese word segmentation component jieba is selected to segment the test data. The irrelevant phrases of adjectives and prepositions are filtered to construct a test word frequency matrix. Then, the TF-TDF method is used to extract test features that can completely cover all test questions of the knowledge point for different knowledge points. The word frequency matrix is ​​optimized by the test features of all knowledge points. The SVM model is trained through the optimized word frequency matrix to realize the classification of single knowledge points of the test questions, and finally the recognition of test knowledge points is realized.

[0045] 2. This course knowledge graph construction method is based on intelligent identification of multiple knowledge points. The course knowledge graph provides a guide for resource construction for educational managers, helping them to reasonably allocate and optimize teaching resources based on the structure and relationship of the knowledge graph.

[0046] 3. The course knowledge graph construction method based on intelligent identification of multiple knowledge points displays the course knowledge in a structured way through the relationship between nodes and edges, making the hierarchy and relationship between knowledge clearer. With the help of graphic visualization tools, the course knowledge graph presents abstract and complex concepts and relationships in an intuitive way, which is easy for students to understand and master. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a model diagram of the construction method of the present invention;

[0048] Figure 2 This is the trend chart of F1 value changing with the proportion of training set;

[0049] Figure 3 This is the trend chart of F1 value changing with gamma proportion;

[0050] Figure 4 This is the trend chart of F1 value changing with C proportion;

[0051] Figure 5 This is a graph showing the F1 value changing with different test sets. DETAILED DESCRIPTION

[0052] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0053] Example 1

[0054] like Figure 1 As shown, a method for constructing a course knowledge graph to implement a course knowledge graph system includes the following steps:

[0055] Step A: test data preprocessing; In order to improve the accuracy of test question recognition, this method first introduces the key attributes of "chapter", "question type" and "knowledge point" to create a test question data model, providing a structured data basis for subsequent processing; then construct a test question word frequency matrix and perform word segmentation on the test questions, so as to more quantitatively reflect the occurrence of each phrase in different test questions.

[0056] Step B: Extract test question features. This method calculates the term frequency (TF) and inverse document frequency (IDF) of each phrase in the term frequency matrix to obtain the TF-IDF value to extract test question features. For each test question, the TF-IDF values ​​of all phrases are combined into a vector. This vector is used as the feature vector of the test question and is subsequently used for test question classification and knowledge point association operations.

[0057] Step C: Knowledge point identification: This method uses the SVM hyperplane solution step to find a hyperplane that can correctly divide the training set into different categories of data and has the largest interval, and uses this hyperplane to distinguish the test questions corresponding to different knowledge points.

[0058] Step D: Construction of Course Knowledge Graph; In this method, by designing an algorithm for constructing a course knowledge graph, all nodes in the node file are added to the knowledge graph, and then the triple file is traversed, and triples that meet the condition (both the head node and the tail node are in the node set) are added to the knowledge graph. Finally, a complete knowledge graph is obtained, which is used to represent the relationships between course knowledge, such as the dependency relationships and the order of precedence between different course contents.

[0059] Step E: Query of Knowledge Points; In this method, relevant or irrelevant knowledge points of the constructed knowledge graph are mainly queried by designing an algorithm for querying knowledge points, and a knowledge graph related to the queried knowledge points is obtained.

[0060] Furthermore, Step A includes:

[0061] Step A1: Image Cleaning; The method proposed in the present invention only processes the text information in the test paper and does not involve the knowledge features in the images. Therefore, the images of the test paper are first cleaned to obtain the test question data in pure text.

[0062] Step A2: Modeling of Test Question Data; When modeling the test question data, the attributes of "chapter", "question type", and "knowledge point" of the test questions are introduced to create a test question data model, that is, the test question data is defined as where q i = W, P, c, l, k represents the i-th test question, W represents the test question data after cleaning the image, P represents the word segmentation result, c represents the chapter to which the test question belongs, l represents the type of the test question, k represents the knowledge point of the test question, and n represents the number of test questions.

[0063] Step A3: Construction of Test Question Word Frequency Matrix; In this method, the Chinese word segmentation component jieba is used to segment the test questions to obtain P s , and after the word segmentation is completed, the word segmentation result is cleaned, and irrelevant phrases such as prepositions and adjectives P1 are screened out, so that q i .P = P s - P a , and the complete set of phrases after word segmentation and screening of all test question data is represented as p s = {p j |0 < j ≤ m}, m represents the number of phrases, p j represents the j-th phrase in the complete set of phrases, and then a word frequency matrix Μ = <f ij > is constructed according to the complete set of phrases, where f ij represents whether the j-th phrase appears in the p j -th test question.

[0064] By introducing the key attributes of "chapter", "question type" and "knowledge point", we can more accurately describe and classify test questions, provide a more accurate data basis for the subsequent construction of the knowledge graph, and perform word segmentation on the test questions, splitting them into independent words or phrases to facilitate subsequent analysis and modeling. By constructing a test question word frequency matrix, we can further reveal the frequency and importance of each word in the test questions, help identify key knowledge points and concepts, enhance the practicality and relevance of the knowledge graph, and optimize the efficiency and effectiveness of knowledge graph construction. Furthermore, by constructing a test question word frequency matrix and a test question data model, we can achieve intelligent recognition of test questions, quickly identify the chapters, question types and knowledge points involved in the test questions, and based on the relevance and hierarchical structure of the knowledge graph, achieve personalized recommendations for learners, such as recommending relevant test questions and learning resources based on the learners' learning progress and interests.

[0065] Further, step B comprises:

[0066] Step B1: Calculation of the word frequency of each phrase in the word frequency matrix, i.e., TF, represents the frequency of the phrase appearing in the test questions corresponding to the knowledge point. The word frequency matrix M is obtained from step A3. The word frequency calculation formula is as follows:

[0067]

[0068] Among them, t c It indicates the number of test questions corresponding to the knowledge point in which the word or phrase appears. 总 Refers to the total number of test questions corresponding to this knowledge point.

[0069] Step B2: Calculate the inverse document frequency of each phrase in the word frequency matrix, i.e., IDF. This method uses IDF to obtain the general importance measure of each phrase. The IDF calculation formula is as follows:

[0070]

[0071] Among them, T 总 It represents the total number of test questions in the test question bank. In order to avoid the situation where the denominator is 0, the denominator in the formula needs to be increased by 1.

[0072] Step B3: TF-IDF calculation is used to evaluate the importance of a phrase to a phrase in a test set. The calculation formula is as follows:

[0073] TF-IDF=TF*IDF

[0074] By extracting test features from the test data, optimizing the word frequency matrix based on these test features, and using the optimized word frequency matrix to train the classification model, questions with different knowledge points can be identified.

[0075] The TF-IDF value combines the term frequency (TF) and the inverse document frequency (IDF). The former measures the number of occurrences of a word in a document (here, the test questions), and the latter measures the uniqueness of a word in the entire document collection. Words with high TF-IDF values are usually the key information in the document content and can effectively reflect the core features of the test questions. By calculating the TF-IDF value, some common but meaningless words (such as "de" (的) and "shi" (是)) are filtered out. These words have a high term frequency in the document but a low IDF value, so their contribution to the TF-IDF value is small. The TF-IDF value can also reduce the noise brought by the high-dimensional sparse matrix and improve the effectiveness of the feature vector. Since the TF-IDF value can accurately reflect the key features of the test question content, when performing classification and association operations based on these feature vectors, it can more accurately judge the category of the test question and the associated knowledge points.

[0076] Furthermore, step C includes:

[0077] Step C1: Define the training set D = {(x1, y1), (x2, y2), …, (x M , y M )}, where x is an n-dimensional feature vector with values of 1 or -1. Select the kernel function H(x, z) and the penalty coefficient C, calculate the hyperplane when the margin is maximized, and construct the constrained optimization problem as shown in the following formula. The solution of the vector is represented by.

[0078]

[0079] 0 ≤ α i ≤ C

[0080] where α is the Lagrange multiplier vector, and the value of C represents the degree of penalty for misclassified points. The larger the value of C, the greater the penalty for misclassified points.

[0081] Step C2: The SVM classification decision function; through the α * calculated in step C1, select a positive component of α * to calculate b , and the formula is as follows: * The SVM classification decision function obtained is as shown in the following formula:

[0082]

[0083]

[0084]

[0085] ​The core of the SVM algorithm is to find an optimal hyperplane that can maximize the interval between two different categories (here, the test questions corresponding to different knowledge points). This strategy of maximizing the interval makes SVM more accurate and robust in classification. Through the learning of the training set, SVM can find those support vectors that play a key role in classification. These support vectors are located on the classification boundary and are crucial to determining the position of the hyperplane. SVM can efficiently identify the test questions corresponding to different knowledge points in classification. In the test question classification, the boundaries between different knowledge points may not be linear. Through the kernel function of SVM, this application can map the features of the test questions to a high-dimensional space and find a nonlinear hyperplane to accurately divide the test questions corresponding to different knowledge points. The decision boundary of VM is defined by the support vector, which makes the model relatively easy to explain. In the test question classification, this application can understand which features play a key role in classification by analyzing the support vector, which helps this application to better understand the mechanism and principle of test question classification, thereby further optimizing and improving the classification algorithm.

[0086] Further, step D comprises:

[0087] Step D1: Definition of knowledge graph; in this method, define Among them, Z stands for "chapter", S stands for "section" and "K" stands for knowledge point, represents the head entity, represents the relationship, represents the tail entity, represents the relationship between chapters and sections in the chapter level, represents the relationship between sections and knowledge points in the chapter level, and represents the relationship between knowledge points.

[0088] Step D2: Definition of node file E and HRT; This method extracts entities "chapter", "section" and "knowledge point" from the textbook through manual mode, and defines the relationship between entities "chapter" and "section", "section" and "knowledge point" in the form of chapter hierarchy of knowledge points in the textbook. E includes chapter entity, section entity and knowledge point entity, and contains the color and size of the entity to indicate the entity type and the frequency of the entity in the test paper; This method identifies knowledge point information from the test questions according to the test question knowledge point recognition algorithm proposed in steps B and C: the co-occurrence of knowledge points in the test questions and the examination frequency of knowledge points, and then judges the dependency between knowledge points and the examination intensity of knowledge points in the test paper. Finally, the chapter, section, knowledge point and the relationship between them are stored in the neo4j graph database in the form of triple HRT, and the relationship strength is defined according to the dependency between knowledge points, and the node size is defined by the examination intensity of knowledge points in the test paper.

[0089] Step D3: Construction of knowledge graph: In this method, the knowledge graph is constructed according to the following algorithm.

[0090] Based on the node file E obtained in step D1 and step D2, and the triple file HRT, first add all the nodes in the node file to the knowledge graph, then traverse the triple file, and according to the information stored in the triple, add the triples that meet the conditions (the head node and the tail node are both in the node set) to the knowledge graph, and finally obtain a complete knowledge graph.

[0091] The course knowledge graph displays course knowledge in a structured way through the relationship between nodes and edges, making the hierarchy and relationship between knowledge clearer. With the help of graphic visualization tools, the course knowledge graph can present abstract and complex concepts and relationships in an intuitive way, which is easy for students to understand and master. The course knowledge graph also provides a guide for resource construction for educational managers, helping them to reasonably allocate and optimize teaching resources based on the structure and relationship of the knowledge graph.

[0092] Step E includes: querying the knowledge graph related to a given knowledge point; first input the knowledge point e1 to be queried, link to the neo4j database Z that stores relevant information, check whether the knowledge point to be queried is in the database, if it exists, query all triples with e1 as the head entity from the database Z, that is, h=e1, and store the results in h1, r1, and t1 respectively, perform the same operation to query with e1 as the tail entity, and construct new triples h2, r2, and t2, merge the query results, merge the results of the query with e1 as the head entity and the tail entity to obtain h3, r3, and t3, construct the knowledge graph G(e1) and relationship list L(e1) related to the knowledge point e1 and output them, if the query e1 is not in Z, output that the entity is not added,

[0093] By performing SVM classification on each type of knowledge point respectively, multi-label classification of test knowledge points is achieved. First, the test data is preprocessed. The Chinese word segmentation component jieba is selected to segment the test data. The adjective and preposition-irrelevant phrases are filtered to build a test word frequency matrix. Then, the TF-TDF method is used to extract test features that can completely cover all test questions of different knowledge points. The word frequency matrix is ​​optimized by the test features of all knowledge points. The SVM model is trained through the optimized word frequency matrix to achieve classification of single knowledge points in the test, and finally the recognition of test knowledge points is achieved.

[0094] Example 2

[0095] The SVM model is optimized by adjusting the proportion, value and value of the training set to evaluate the model performance of the test set F1 value. The method of the present invention is used to construct a knowledge point label map for the mechanical motion part of the first chapter of the physics course of the eighth grade of junior high school (upper) on the subject network, simulating 13740 test questions.

[0096] like Figures 2 to 4As shown in the figure, it means adjusting the training set ratio, gamma value and C value to adjust the model performance, such as Figure 5 As shown, the figure shows the F1 value of the model in different test sets, and the average F1 value is 89.66%, which proves that the support vector machine algorithm used in the present invention can effectively identify the knowledge points in the test set.

[0097] The model uses the TF-IDF algorithm to extract features from the test data and the SVM algorithm to identify the test knowledge points, build a course knowledge graph, design and implement a course knowledge graph system, and use a network structured knowledge system to represent the relationship between knowledge points. The method first acquires data, including manually extracting chapter, section, and knowledge point information from the textbook, and manually marking the knowledge points of the test questions based on the extracted knowledge point information. Then, the test data is preprocessed, and the test knowledge point recognition algorithm is implemented for the test data with marked knowledge points. The test data is modeled by introducing attributes such as chapters, question types, and knowledge points. The test word frequency matrix is ​​constructed using the Chinese word segmentation component, and the TF-IDF method is used to identify the test questions. The data is used for feature extraction to optimize the word frequency matrix, and the SVM classification algorithm is used to identify the knowledge points in the test questions, and then the knowledge point information in the test questions is obtained. Next, the chapter, section, and knowledge point information extracted from the textbooks and the knowledge point information identified from the test questions (the frequency of knowledge points in the test papers and the degree of association between knowledge points in the test questions) are used to store the chapters, sections, knowledge points and the relationship between the three in the form of triples in the neo4j graph database, and the relationship strength is defined according to the dependency between knowledge points (here the node size represents the strength). According to the node files and triple files, the course knowledge graph is constructed according to the designed algorithm, and finally the course knowledge graph is used to query the knowledge points.

[0098] The experimental results show that the average F1 value of the SVM algorithm used in this application in different test segments is 89.66%, which can effectively identify all knowledge points in the test set. The method proposed in this application to construct a course knowledge graph based on textbooks and test papers can clearly show the distribution of knowledge points in chapters, the frequency of examination in test papers, and the dependency relationship between knowledge points in test questions.

Claims

1. A method for constructing a course knowledge graph based on intelligent identification of multiple knowledge points, characterized in that: The following steps are involved: Step A: Preprocessing of test data: first introduce the key attributes of "chapter", "question type" and "knowledge point" to create a test data model, then construct a test word frequency matrix and perform word segmentation on the test questions; Step B: Extract test question features. By calculating the term frequency (TF) and inverse document frequency (IDF) of each phrase in the term frequency matrix, we get the TF-IDF value to extract test question features. For each test question, we combine the TF-IDF values ​​of all phrases into a vector. The vector is used as the feature vector of the test question and is subsequently used for test question classification and knowledge point association operations. Step C: Knowledge point identification, through the SVM hyperplane solution step, find the hyperplane that can correctly divide the training set into different categories of data and has the largest interval, and use this hyperplane to distinguish the test questions corresponding to different knowledge points; Step D: Course knowledge graph construction. By designing a course knowledge graph construction algorithm, all nodes in the node file are added to the knowledge graph. Then, the triple file is traversed and the triples that meet the conditions (the head node and the tail node are both in the node set) are added to the knowledge graph. Finally, a complete knowledge graph is obtained, which is used to represent the relationship between course knowledge. Step E: Knowledge point query, by designing a knowledge point query algorithm to query the constructed knowledge graph for relevant or irrelevant knowledge points, and obtain a knowledge graph related to the query knowledge point.

2. According to claim 1, a method for constructing a course knowledge graph based on intelligent identification of multiple knowledge points is characterized in that: The step A comprises: Step A1: Image cleaning, processing the text information in the test paper; Step A2: Test data modeling. When modeling test data, the test data attributes of "chapter", "question type" and "knowledge point" are introduced to create a test data model. That is, the test data is defined as where q i =W,P,c,l,k represents the i-th test question, W represents the test question data after cleaning the image, P represents the word segmentation result, c represents the chapter to which the test question belongs, l represents the type of the test question, k represents the test question knowledge point, and n represents the number of test questions; Step A3: Construct the test question word frequency matrix. Use the Chinese word segmentation component jieba to segment the test questions to obtain P s , after the word segmentation is completed, clean the word segmentation results, and screen out irrelevant phrases such as prepositions and adjectives P a , so that q i .P = P s -P a , the complete set of phrases after word segmentation and screening of all test question data is represented as p s = {p j |0 < j ≤ m}, where m represents the number of phrases, and p j represents the j-th phrase in the complete set of phrases. Then, construct the word frequency matrix Μ = <f ij >, where f ij indicates whether the j-th phrase appears in the p j -th test question.

3. According to claim 2, a method for constructing a course knowledge graph based on intelligent identification of multiple knowledge points is characterized in that: The step B comprises: Step B1: Calculation of the word frequency of each phrase in the word frequency matrix, i.e., TF, represents the frequency of the phrase appearing in the test questions corresponding to the knowledge point. The word frequency matrix M is obtained from step A3. The word frequency calculation formula is as follows: Among them, t c It indicates the number of test questions corresponding to the knowledge point in which the word or phrase appears. 总 Refers to the total number of test questions corresponding to this knowledge point; Step B2: Calculate the inverse document frequency of each phrase in the word frequency matrix, i.e., IDF. This method uses IDF to obtain the general importance measure of each phrase. The IDF calculation formula is as follows: Among them, T 总 It represents the total number of test questions in the test question bank. In order to avoid the situation where the denominator is 0, the denominator in the formula needs to be added with 1; Step B3: TF-IDF calculation is used to evaluate the importance of a phrase to a phrase in a test question set. The calculation formula is as follows: TF-IDF=TF*IDF By extracting test features from the test data, optimizing the word frequency matrix based on these test features, and using the optimized word frequency matrix to train the classification model, questions with different knowledge points can be identified.

4. According to claim 3, a method for constructing a course knowledge graph based on intelligent identification of multiple knowledge points is characterized in that: The step C comprises: Step C1: Define the training set D = {(x1,y1),(x2,y2),…,(x M ,y M )}, where x is an n-dimensional feature vector with a value of 1 or -1, select the kernel function H(x,z) and the penalty coefficient C, calculate the hyperplane when the interval is the largest, construct the constrained optimization problem as shown in the following formula, and express the solution of the vector as follows, 0≤α i ≤C Among them, α is the Lagrange multiplier vector, and the C value represents the degree of penalty for misclassified points. The larger the C value, the greater the penalty for misclassified points. Step C2: SVM classification decision function, using the α calculated in step C1 * , select α * A positive component of Calculate b * , the formula is as follows: The SVM classification decision function is shown in the following formula:

5. According to claim 4, a method for constructing a course knowledge graph based on intelligent identification of multiple knowledge points is characterized in that: The step D comprises: Step D1: Knowledge Graph Definition, Definition Among them, z represents "chapter", S represents "section" and "K" represents knowledge point, represents head entity, represents relationship, represents tail entity, represents the relationship between chapter and section in chapter level, represents the relationship between section and knowledge point in chapter level, represents the relationship between knowledge point and knowledge point; Step D2: Definition of node file E and HRT, extracting entities "chapter", "section" and "knowledge point" from the textbook through manual mode, and defining the relationship between entities "chapter" and "section", "section" and "knowledge point" in the form of chapter hierarchy of knowledge points in the textbook, E includes chapter entity, section entity and knowledge point entity, and includes the color and size of the entity to indicate the entity type and the frequency of the entity in the test paper; this method recognizes the knowledge point information from the test questions according to the test question knowledge point recognition algorithm proposed in step B and step C; Step D3: Construction of the knowledge graph, based on the node file E obtained in step D1 and step D2, and the triple file HRT, first add all the nodes in the node file to the knowledge graph, then traverse the triple file, and according to the information stored in the triple, add the triples that meet the conditions (the head node and the tail node are both in the node set) to the knowledge graph, and finally obtain a complete knowledge graph.

6. A method for constructing a course knowledge graph based on intelligent identification of multiple knowledge points according to claim 5, characterized in that: The step E includes: querying the knowledge graph related to a given knowledge point, first inputting the knowledge point e1 to be queried, linking to the neo4j database Z that stores relevant information, checking whether the knowledge point to be queried is in the database, if it exists, querying all triples with e1 as the head entity from the database Z, that is, h=e1, and storing the results in h1, r1, and t1 respectively, performing the same operation to query with e1 as the tail entity, and constructing new triples h2, r2, and t2, merging the query results, merging the results queried with e1 as the head entity and the tail entity to obtain h3, r3, and t3, constructing the knowledge graph G(e1) and relationship list L(e1) related to the knowledge point e1 and outputting them, if the query e1 is not in Z, output that the entity is not added.

Images