Knowledge graph construction method and device

Abstract

The embodiment of the invention provides a knowledge graph construction method and device, and the method comprises the steps: carrying out the knowledge point analysis of a target document through a knowledge point analysis model, obtaining a plurality of initial document knowledge points in the target document, carrying out the relation recognition of the plurality of initial document knowledge points, and carrying out the relation recognition of the plurality of initial document knowledge points; obtaining an initial association relationship between the initial document knowledge points; performing knowledge point merging on the plurality of initial document knowledge points to obtain a plurality of target document knowledge points, and performing relationship analysis on the plurality of target document knowledge points and the initial association relationship by using a relationship analysis model to obtain a target association relationship among the target document knowledge points; based on the multiple target document knowledge points and the target association relationship among the target document knowledge points, constructing a target knowledge graph; the problems of low efficiency and poor accuracy in the knowledge graph construction process are avoided, and efficient and accurate construction of the knowledge graph is realized.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of knowledge graph processing technology, and in particular to a knowledge graph construction method. One or more embodiments of this specification also relate to a knowledge graph construction device, a computing device, a computer-readable storage medium, and a computer program product. Background Art

[0002] With the development of information technology and computers, knowledge graphs, as an emerging form of knowledge organization, have been widely used in various fields, such as search engines, recommendation systems, intelligent question-answering systems, etc.

[0003] The current construction of knowledge graphs mainly relies on inefficient manual processing methods, or uses low-accuracy automatic extraction technology to extract complex knowledge points and associations. Since the current methods have problems of low efficiency and poor accuracy, how to efficiently and accurately construct knowledge graphs has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] In view of this, the embodiments of this specification provide a method for constructing a knowledge graph. One or more embodiments of this specification also involve a knowledge graph construction device, a computing device, a computer-readable storage medium, and a computer program product to solve the technical defects existing in the prior art.

[0005] According to a first aspect of an embodiment of this specification, a knowledge graph construction method is provided, including:

[0006] Performing knowledge point analysis on the target document using the knowledge point analysis model to obtain a plurality of initial document knowledge points in the target document, and performing relationship identification on the plurality of initial document knowledge points to obtain an initial association relationship between the initial document knowledge points;

[0007] Merging the multiple initial document knowledge points to obtain multiple target document knowledge points, and using a relationship analysis model to perform relationship analysis on the multiple target document knowledge points and the initial association relationships to obtain target association relationships between the target document knowledge points;

[0008] A target knowledge graph is constructed based on the multiple target document knowledge points and the target association relationships between the target document knowledge points.

[0009] According to a second aspect of an embodiment of this specification, a knowledge graph construction device is provided, including:

[0010] A first knowledge point processing module is configured to perform knowledge point analysis on a target document using a knowledge point analysis model to obtain a plurality of initial document knowledge points in the target document, and perform relationship recognition on the plurality of initial document knowledge points to obtain an initial association relationship between the initial document knowledge points;

[0011] A second knowledge point processing module is configured to merge the multiple initial document knowledge points to obtain multiple target document knowledge points, and use a relationship analysis model to perform relationship analysis on the multiple target document knowledge points and the initial association relationships to obtain target association relationships between the target document knowledge points;

[0012] The graph construction module is configured to construct a target knowledge graph based on the multiple target document knowledge points and the target association relationships between the target document knowledge points.

[0013] According to a third aspect of an embodiment of this specification, a computing device is provided, including:

[0014] Memory and processor;

[0015] The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, the steps of the above-mentioned knowledge graph construction method are implemented.

[0016] According to the fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores a computer program / instruction, and when the computer program / instruction is executed by a processor, the steps of the above-mentioned knowledge graph construction method are implemented.

[0017] According to the fifth aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instruction, which implements the steps of the above-mentioned knowledge graph construction method when executed by a processor.

[0018] The knowledge graph construction method provided in one or more embodiments of this specification introduces artificial intelligence technology in the process of constructing the target knowledge graph. Through the knowledge point analysis model and the relationship analysis model, the knowledge points and association relationships can be automatically and efficiently acquired, thereby improving the efficiency of constructing the target knowledge graph. In addition, this method uses a two-stage acquisition method to accurately acquire knowledge points and association relationships. In the first stage, the initial document knowledge points and initial association relationships are obtained; in the second stage, more accurate target document knowledge points and target association relationships are obtained based on the initial document knowledge points and initial association relationships, thereby ensuring the accuracy of the target knowledge graph. Therefore, the problems of low efficiency and poor accuracy in the knowledge graph construction process are avoided, and efficient and accurate construction of the knowledge graph is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is an application schematic diagram of a knowledge graph construction method provided by an embodiment of this specification;

[0020] Figure 2 It is a flowchart of a knowledge graph construction method provided by an embodiment of this specification;

[0021] Figure 3 It is a process flow chart of a knowledge graph construction method provided by an embodiment of this specification;

[0022] Figure 4 It is a schematic diagram of the structure of a knowledge graph construction device provided by an embodiment of this specification;

[0023] Figure 5 It is a structural block diagram of a computing device provided by an embodiment of this specification. DETAILED DESCRIPTION

[0024] Many specific details are described in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of this specification, so this specification is not limited to the specific implementation disclosed below.

[0025] The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a", "said" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0026] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0027] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0028] In one or more embodiments of this specification, a large model refers to a deep learning model with large-scale model parameters, which usually contains hundreds of millions, tens of billions, hundreds of billions, trillions, or even more than ten trillion model parameters. A large model can also be called a foundation model / foundation model (Foundation Model 1). The large model is pre-trained with large-scale unlabeled corpus to produce a pre-trained model with more than 100 million parameters. This model can adapt to a wide range of downstream tasks, and the model has good generalization ability, such as a large-scale language model (Large Language Model 1, LLM), a multi-modal pre-training model (multi-modal pre-training model 1), etc.

[0029] When the big model is used in practice, only a small number of samples are needed to fine-tune the pre-trained model and it can be applied to different tasks. The big model can be widely used in natural language processing (NLP), computer vision and other fields. Specifically, it can be applied to computer vision tasks such as visual question answering (VQA), image description (IC), image generation, as well as natural language processing tasks such as text-based sentiment classification, text summary generation, and machine translation. The main application scenarios of the big model include digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, etc.

[0030] First, the terms involved in one or more embodiments of this specification are explained.

[0031] NLP (Natural Language Processing) model: refers to the natural language processing model.

[0032] GPT (Generative Pre-trained Transformer): is a large NLP model that uses a multi-layer Transformer structure to predict the probability distribution of the next word.

[0033] BERT (Bidirectional Encoder Representations from Transformers): is a representation learning model based on a bidirectional encoder of Transformer.

[0034] NER (Named Entity Recognition): refers to named entity recognition, NER refers to identifying entities with specific meanings in text, mainly including names of people, places, institutions, proper nouns, etc. The purpose of NER is to extract meaningful nouns or phrases from text.

[0035] Knowledge graph: a technology that displays knowledge points and their relationships in the form of nodes (entities) and edges (relationships), which can help to systematically organize and display complex knowledge structures. In the field of education, knowledge graphs built based on textbook content can provide teaching with auxiliary decision-making, personalized learning path recommendations, and intelligent retrieval of knowledge points, greatly improving learning efficiency and the utilization of teaching resources.

[0036] With the development of information technology, knowledge graphs, as an emerging form of knowledge organization, have been widely used in many fields, such as search engines, recommendation systems, and intelligent question-answering systems. At present, knowledge acquisition mainly relies on manual input and organization, but with the explosive growth of educational resources and information resources, the demand for automated knowledge acquisition and structured representation is becoming increasingly urgent. Especially in the field of education, a large number of information resources such as textbooks and courseware are difficult to organize and associate effectively, which limits the in-depth utilization of educational resources and the efficiency of knowledge transfer. The challenge of converting textbook content into knowledge points is that textbooks, as the main carrier of knowledge transfer, contain a large number of professional terms, concepts, principles, and inferences. However, this information usually exists in the form of text, lacks a systematic structure, and is difficult to automatically process and understand efficiently. In addition, the hierarchical and complex content of textbooks makes the extraction and association construction of knowledge points based solely on keyword matching inefficient and cannot fully express the deep relationship between knowledge points. Therefore, how to automatically extract knowledge points and their associations from textbooks and construct efficient knowledge graphs has become an urgent problem to be solved.

[0037] Regarding the above-mentioned technical solution of manual input and organization, specifically, knowledge points and relationship networks are constructed through manual annotation. However, this method has the problems of high cost and low efficiency, and it is difficult to cope with the constantly updated textbook content.

[0038] Moreover, current automatic extraction technology mostly relies on natural language processing (NLP) technology, but for complex subject knowledge points and multi-level associations, the accuracy and automation level of this technology still need to be improved. Therefore, the technology of automatically extracting knowledge points and generating knowledge graphs based on textbook content can solve the efficiency problem of traditional manual annotation and make up for the limitations of existing automatic extraction technology, which has important application value.

[0039] In response to the above problems, this manual provides a knowledge point extraction solution based on a deep learning model; with the development of technologies such as artificial intelligence, natural language processing, and deep learning, knowledge point extraction and graph construction technologies based on big data and deep learning are rapidly emerging. In particular, the improvement of multimodal data processing capabilities enables knowledge graphs to extract information from multiple forms such as text, images, and videos. This technology can be well applied in the field of intelligent education and provide strong technical support for educational informatization.

[0040] Therefore, the solution uses artificial intelligence to process textbook content to extract knowledge points and their relationships to generate a knowledge graph. This technology can be a method based on natural language processing (NLP) and data science, which aims to extract knowledge points from textbook content and present these knowledge points and their relationships in the form of a knowledge graph. The main components of this technology are:

[0041] 1. Text analysis and knowledge point extraction: First, the technology uses natural language processing technology to analyze the textbook text, identify and extract key concepts and knowledge points, involving vocabulary analysis, grammar analysis, entity recognition and other technologies.

[0042] 2. Knowledge graph construction: Next, the knowledge points extracted from the textbook text are constructed into a knowledge graph through data science methods. In this process, knowledge points are represented as nodes in the graph, and the relationships between them are represented as edges in the graph. This involves technologies such as machine learning algorithms and graph theory algorithms.

[0043] 3. Educational Applications: Finally, this technology can be applied in the field of education to provide personalized learning experiences and support teaching activities. Education experts can use this knowledge graph to design courses, develop teaching plans, recommend learning resources, etc.

[0044] However, this technical solution has major defects, specifically:

[0045] (1) Insufficient accuracy of knowledge point extraction:

[0046] The knowledge point extraction technologies of the above technical solutions, especially those based on natural language processing (NLP) and simple keyword matching, are often difficult to accurately capture the key knowledge points in textbooks. The complex subject structure and rich semantic expressions in textbooks make these methods have a high misjudgment rate when extracting deep knowledge concepts and professional terms.

[0047] (2) The construction of knowledge relationships lacks systematicity:

[0048] The knowledge graph construction method of the above technical solutions is usually based on static rules or predefined templates, resulting in the lack of flexibility and systematicity in the relationship expression of the generated knowledge graph. Especially in textbook content, the relationship between knowledge points may be expressed as a complex hierarchical structure or a many-to-many nonlinear relationship. Existing technologies are difficult to effectively mine these deep-level associations.

[0049] (3) Poor adaptability and difficulty in handling diverse teaching materials:

[0050] The above-mentioned technical solutions are often highly dependent on textbook contents in specific formats or fields and have poor adaptability. Especially when dealing with textbooks of different subjects and structures, the extraction effect is greatly reduced.

[0051] (4) Low degree of automation and reliance on a large amount of manual intervention:

[0052] The construction of the knowledge graph of the above technical solutions still requires a lot of manual annotation and intervention, which leads to low efficiency and difficulty in ensuring the quality and consistency of annotation. This method is not only time-consuming and labor-intensive, but also unable to cope with the frequent updates of textbook content.

[0053] (5) Unable to dynamically update and expand:

[0054] The above technical solutions usually generate static knowledge graphs. Once built, it is difficult to dynamically adjust and expand them according to new knowledge or updated teaching materials. However, modern educational resources and knowledge systems are developing rapidly, and static graphs cannot meet the growing needs of knowledge management.

[0055] (6) Insufficient user interaction and applicability:

[0056] In actual applications, the knowledge graph system in the above technical solution has weak user interactivity and is difficult to provide intuitive and effective assistance in teaching or learning scenarios.

[0057] Based on this, in this specification, a knowledge graph construction method is provided. This specification also involves a knowledge graph construction device, a computing device, and a computer-readable storage medium, which are described in detail one by one in the following embodiments.

[0058] See also Figure 1, Figure 1 The schematic diagram of the application of a knowledge graph construction method provided according to an embodiment of the present specification is shown; considering the huge amount of model parameters of the model and the limited computing resources of the mobile terminal, the knowledge graph construction method provided in the embodiment of the present application can be applied to Figure 1 The application scenarios shown are not limited to these. Figure 1 In the application scenario shown, the model is deployed in the server 10, and the server 10 can be connected to one or more client devices 20 through a local area network connection, a wide area network connection, an Internet connection, or other types of data networks. The client device 20 here may include but is not limited to: a smart phone, a tablet computer, a laptop computer, a PDA, a personal computer, a smart home device, a vehicle-mounted device, etc. The client device 20 can interact with the user through a graphical user interface to call the model, thereby implementing the method provided in the embodiment of this specification.

[0059] In an embodiment of the present specification, the system composed of the client device 20 and the server 10 can execute the following steps: the client device 20 executes to provide the target teaching material to the server 10; the server 10 performs three steps of deep learning-based knowledge point extraction, dynamic knowledge relationship generation, and knowledge graph construction for the target teaching material; wherein, deep learning-based knowledge point extraction refers to: using a deep learning model to have a deeper understanding and analysis of the contextual semantics in the teaching material, so as to identify the knowledge points (i.e., initial document knowledge points) and the association relationships between the knowledge points (i.e., initial association relationships); wherein, dynamic knowledge relationship generation refers to: merging the knowledge points in the teaching material content to obtain merged knowledge points (i.e., target document knowledge points), and constructing complex multi-level relationships between knowledge points (i.e., target association relationships) through natural language understanding and reasoning technology; wherein, constructing a knowledge graph refers to: converting the extracted knowledge points and their relationships into graph structure data to obtain a target knowledge graph.

[0060] It should be noted that, when the operating resources of the client device can meet the deployment and operating conditions of the model, the embodiments of the present application can be carried out in the client device.

[0061] See also Figure 2 , Figure 2 A flowchart of a knowledge graph construction method provided according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0062] Step 202: Utilize the knowledge point analysis model to perform knowledge point analysis on the target document to obtain a plurality of initial document knowledge points in the target document, and perform relationship identification on the plurality of initial document knowledge points to obtain initial association relationships between the initial document knowledge points.

[0063] Among them, the target document can be understood as a document that needs to construct a knowledge graph. For example, the target document can be a target teaching material, a technical guidance document, an equipment manual, etc. There is no specific restriction on the target document here.

[0064] The knowledge point analysis model can be understood as a model that can obtain document knowledge points from the target document; the knowledge point analysis model can be a language model, a deep learning model, a large model, etc., and no specific restrictions are made here; for example, the knowledge point analysis model can be a knowledge point recognition algorithm, a NER model; the knowledge point analysis model is a pre-trained model.

[0065] Document knowledge points can be understood as the knowledge content or knowledge information recorded in the target document; the document knowledge points can be technical concepts, terms, definitions, etc. in the target document. For example, the target tutorial of the target document, the document knowledge points can be terms with specific subject meanings such as "calculus" and "periodic table".

[0066] The initial document knowledge points can be understood as document knowledge points obtained by performing knowledge point analysis on the target document using a knowledge point analysis model.

[0067] The association relationship can be understood as the logical relationship between document knowledge points; the association relationship can be a causal relationship, hierarchical relationship, time relationship, etc. between concepts. For example, "addition is the basis of multiplication", "oxygen and hydrogen can generate water", etc.

[0068] The initial association relationship can be understood as the association relationship between the initial document knowledge points.

[0069] Taking the application of the knowledge graph construction method provided in this specification in an educational scenario as an example, the knowledge graph construction method is described. The target document can be a target textbook. Based on this, the method obtains the knowledge points and associations from the textbook as follows:

[0070] 1. Automatically identify important concepts, terms, definitions and other knowledge points (i.e., initial document knowledge points) in the textbooks through the knowledge point recognition algorithm (i.e., knowledge point analysis model). For example, through the trained NER model (i.e., knowledge point analysis model), mark out terms with specific subject meanings such as "calculus" and "periodic table".

[0071] 2. After identifying the knowledge points, further use the relation extraction technology to identify the logical relationship between the knowledge points (i.e. the initial association relationship). These relationships can be causal relationships, hierarchical relationships, time relationships, etc. between concepts. For example, "addition is the basis of multiplication", "oxygen and hydrogen can generate water", etc.

[0072] In one or more embodiments provided in this specification, before performing knowledge point analysis on the target document using the knowledge point analysis model to obtain a plurality of initial document knowledge points in the target document, the method further includes:

[0073] The document to be processed is preprocessed to obtain the target document.

[0074] The text preprocessing may be understood as segmenting, sentence segmenting, removing irrelevant symbols, etc. the document to be processed.

[0075] Using the above example, this method can extract knowledge points based on deep learning in the process of constructing the knowledge graph corresponding to the textbook; considering that the existing technology relies on keyword matching, rule engines or shallow natural language processing (NLP) technology to identify and extract knowledge points from textbooks. However, the existing technology is prone to errors when processing complex language structures and ambiguous knowledge points, resulting in inaccurate recognition.

[0076] Based on this, this method can provide a deeper understanding and analysis of the contextual semantics in teaching materials by introducing deep learning models, such as pre-trained language models (such as BERT, GPT, etc.).

[0077] Before model processing, the textbook content needs to be preprocessed: the specific method is as follows:

[0078] First, determine the content of the textbook; the textbook content can be used as text data as the input of the pre-trained language model. The input data usually contains paragraphs, chapters, titles, and other structured text information. The text data (target document) can be an original book scan (processed by optical character recognition), an e-book, or a digital textbook resource.

[0079] Secondly, the textbook content is preprocessed; the input content of the pre-trained language model usually needs to be preprocessed, such as segmentation, sentence segmentation, and removal of irrelevant symbols (such as punctuation marks, charts, etc.) to ensure that the input content meets the requirements of the language model. Therefore, by performing the above preprocessing operations on the original textbook content (i.e., the document to be processed), the textbook content (i.e., the target document) that meets the requirements of the language model is obtained.

[0080] Moreover, the input content of the model, in addition to the original textbook text (i.e., original textbook content), may also include some content-related metadata, such as chapter number, textbook subject, grade information and other textbook information; therefore, this method can use these textbook information and textbook content that meets the requirements of the language model as the input content of the model (i.e., the target document), thereby helping the language model to understand the context of the text or the hierarchy of knowledge.

[0081] In one or more embodiments provided in this specification, the method of performing knowledge point analysis on a target document using a knowledge point analysis model to obtain a plurality of initial document knowledge points in the target document, and performing relationship identification on the plurality of initial document knowledge points to obtain an initial association relationship between the initial document knowledge points includes:

[0082] Inputting the target document into a language model for vectorization processing to obtain a plurality of word vectors corresponding to the target document, wherein the word vectors contain semantic information of document words in the target document;

[0083] Determine a plurality of document chapters contained in the target document, and determine a chapter word vector corresponding to the target document chapter from the plurality of word vectors, wherein the target document chapter is any one of the plurality of document chapters;

[0084] Inputting the chapter word vector corresponding to the target document chapter into the knowledge point analysis model to perform knowledge point recognition, and obtaining a plurality of initial document knowledge points corresponding to the target document chapter;

[0085] Relationship identification is performed on the multiple initial document knowledge points corresponding to the target document chapters to obtain initial association relationships between the initial document knowledge points corresponding to the target document chapters.

[0086] The word vector may refer to a vector corresponding to a document word in a target document, and the word vector may be embedded.

[0087] The chapter word vector may refer to a word vector corresponding to a document chapter; the chapter word vector corresponding to a document chapter may represent the semantic information of all document words in a document chapter.

[0088] Using the above example, this method introduces a deep learning model to gain a deeper understanding and analysis of the contextual semantics in the textbook, thereby identifying the knowledge points in the textbook and the logical relationships between the knowledge points. The specific implementation method is as follows:

[0089] 1. Use the pre-trained language model to vectorize the text of the teaching materials.

[0090] Specifically, this method converts the textbook content (target document) into context-aware word vectors (embedded ing) through a pre-trained language model. These word vectors contain the semantic information of the words and details related to the context of the textbook content.

[0091] It should be noted that before using the pre-trained language model to perform text vectorization processing on the textbook, it is necessary to initialize the pre-trained language model; the initialization of the pre-trained language model means: using the pre-trained language model (such as BERT, GPT, etc.) as the basis for building the knowledge graph. The language model has been pre-trained through a large-scale corpus and has strong text understanding and generation capabilities; the pre-training of the pre-trained language model can be fine-tuning the language model to adapt to specific textbook content extraction tasks.

[0092] 2. Perform knowledge point recognition.

[0093] After determining the word vector, we can apply the knowledge point recognition algorithm to the word vector corresponding to each textbook chapter (i.e. document chapter) in the textbook to automatically identify the important concepts, terms, definitions and other knowledge points of each textbook chapter in the textbook. For example, through the trained NER model, we can mark the terms (initial document knowledge points) with specific subject meanings such as "calculus" and "periodic table".

[0094] 3. Perform relationship extraction.

[0095] After identifying the knowledge points, the relation extraction technology can be further used to identify the logical relationship (i.e., initial association relationship) between the knowledge points in the document chapters. These logical relationships can be causal relationships, hierarchical relationships, time relationships, etc. between concepts. For example, "addition is the basis of multiplication", "oxygen and hydrogen can generate water", etc. In this example, the system will infer that there is an inheritance relationship between "Newton's Second Law" and "Newton's First Law", and there is a conditional derivation relationship between "Law of Conservation of Momentum" and "Newton's Second Law".

[0096] Based on the above embodiments, this method can use pre-trained language models or other deep learning models to automatically extract knowledge points from the textbook content; thereby improving the accuracy and efficiency of automated knowledge point extraction; specifically, this method uses deep learning models (such as BERT or GPT) for semantic analysis and knowledge point extraction, which has higher accuracy than existing methods based on keyword matching or rule engines. Existing technologies can often only process surface keywords or simple fixed patterns, and it is difficult to effectively identify complex syntactic structures and knowledge points with implicit meanings, which can easily lead to missed detections or misjudgments.

[0097] Considering that deep learning models have stronger context understanding capabilities, they can learn and extract semantic features from a large amount of training data and identify key knowledge points under complex semantics. In this way, this method significantly reduces the errors caused by keyword matching in the prior art, especially when processing long texts or containing complex knowledge concepts, ensuring that important knowledge points will not be missed. This automated extraction method not only improves the accuracy of knowledge point recognition, but also greatly reduces the need for manual intervention, thereby improving efficiency and is suitable for the rapid processing of large-scale textbook content.

[0098] Step 204: Merge the multiple initial document knowledge points to obtain multiple target document knowledge points, and use the relationship analysis model to perform relationship analysis on the multiple target document knowledge points and the initial association relationships to obtain target association relationships between the target document knowledge points.

[0099] The target document knowledge point may be understood as a document knowledge point obtained by merging a plurality of initial document knowledge points; and the target association relationship may be understood as an association relationship determined by using a relationship analysis model.

[0100] The relationship analysis model can be understood as a model that can identify the association relationship between document knowledge points. For example, the relationship analysis model can be an inference algorithm, a large model for semantic reasoning, etc.

[0101] Continuing with the above example, after analyzing the sentence structure and vocabulary meaning in the textbook chapters through natural language processing technology to identify key physical concepts (initial document knowledge points), this method can merge multiple initial document knowledge points based on the textbook context information and reconstruct the logical relationship for the merged knowledge points; among which, the physical probability can be "force", "acceleration", "mass" and "momentum" etc.

[0102] That is to say, in the process of executing the first stage (i.e., step 202), the knowledge points corresponding to chapters 1, 2, 3, and 4 of the textbook can be extracted, and they will be extracted 4 times respectively. Then in the process of executing the second stage (i.e., step 204), the same knowledge points will be merged to obtain the merged knowledge points (i.e., the target document knowledge points). The knowledge point merging operation is equivalent to further abstracting and combing the knowledge system presented in the textbook, so as to obtain the knowledge system corresponding to the textbook.

[0103] For instance, taking limit as an example, the knowledge point of limit may belong to the first section in the first chapter of the textbook (a second-level knowledge point). At the same time, in the second chapter of the textbook, the knowledge point of limit may be a knowledge point under the second section of the second chapter (a third-level knowledge point). Therefore, in the process of merging knowledge points, the knowledge points under different chapters can be merged according to the levels, so that the knowledge points under different chapters can be merged into the same level.

[0104] After the knowledge points are merged, a multi-level association relationship (ie, target association relationship) between the knowledge points can be generated through an inference algorithm based on the context information and the logical order of the knowledge points in the textbook (teaching material) (ie, hierarchical relationship, knowledge system).

[0105] In one or more embodiments provided in this specification, the multiple initial document knowledge points are multiple initial document knowledge points corresponding to a target document chapter, and the target document chapter is any one of the multiple document chapters of the target document;

[0106] The step of merging the plurality of initial document knowledge points to obtain a plurality of target document knowledge points, and performing relationship analysis on the plurality of target document knowledge points and the initial association relationships using a relationship analysis model to obtain target association relationships between the target document knowledge points includes:

[0107] Determining at least two identical initial document knowledge points and other initial document knowledge points other than the at least two identical initial document knowledge points from the multiple initial document knowledge points corresponding to the multiple document chapters;

[0108] Merging the at least two identical initial document knowledge points to obtain a merged knowledge point, and determining the other initial document knowledge points and the merged knowledge point as the multiple target document knowledge points;

[0109] Inputting the plurality of target document knowledge points and the initial association relationships into the relationship analysis model;

[0110] In the relationship analysis model, based on the document context information of the target document and the initial association relationship, the target document knowledge points are subjected to relationship analysis to obtain the target association relationship between the target document knowledge points.

[0111] The merged knowledge point may be understood as a document knowledge point obtained by merging or fusing at least two identical initial document knowledge points.

[0112] In the embodiments of the present specification, the initial document knowledge points may be deduplicated to obtain deduplicated initial document knowledge points; subsequently, knowledge point merging processing may be performed based on the deduplicated initial document knowledge points.

[0113] Following the above example, this method also introduces a dynamic knowledge relationship generation mechanism in the process of knowledge graph construction; considering that traditional knowledge graph construction usually relies on manually defined relationships between knowledge points or based on predefined rule templates, it is difficult to flexibly deal with complex knowledge point associations in different textbooks. This method introduces a dynamic generation mechanism based on relational reasoning, which can automatically analyze the knowledge points in the textbook content and construct complex multi-level relationships between knowledge points through natural language understanding and reasoning technology.

[0114] Take the high school physics textbook (i.e. the target document) as an example, which involves knowledge points of mechanics, such as "Newton's First Law", "Newton's Second Law" and "Law of Conservation of Momentum". Through natural language understanding and reasoning technology, the system can automatically identify these core knowledge points from the textbook content and analyze the relationship between them.

[0115] After obtaining the core knowledge points (ie, the initial document knowledge points) and the association relationships between the knowledge points (ie, the initial association relationships), multiple initial document knowledge points can be merged based on the teaching material context information, and the logical relationships for the merged knowledge points can be reconstructed.

[0116] The method for merging the multiple initial document knowledge points is as follows: determining multiple similar core knowledge points (ie, at least two identical initial document knowledge points) from the multiple core knowledge points; and then merging them to obtain merged knowledge points.

[0117] Among them, reconstructing the logical relationship for the merged knowledge points means that the knowledge graph construction system used in this method can generate the association relationship (i.e., the initial association relationship) between the knowledge points through the inference algorithm based on the context information and the logical order of the knowledge points in the textbook, and then perform multi-level complex relationship generation, specifically in the following ways:

[0118] First, the context information of the teaching material (ie, document context information) and the association relationship between knowledge points are input into a relational analysis model (eg, a NER model).

[0119] Secondly, by analyzing the contents of multiple chapters, the system can use the relationship analysis module to dynamically generate more complex multi-level relationships. For example, when discussing collision problems, the system can automatically infer that the "law of conservation of momentum" and "Newton's second law" act together on the collision system, and generate a higher-level relationship diagram that reflects the comprehensive application of these laws in solving complex physical problems.

[0120] It should be noted that, based on the above-mentioned dynamic knowledge relationship generation mechanism, this method can realize dynamic adaptation of the knowledge graph.

[0121] This dynamic adaptation means that the knowledge graph generation method provided by this method is not a pre-set static relationship, but a dynamic adjustment based on the different contents and knowledge points of different chapters of the textbook. For example, for a textbook focusing on dynamics, the system will focus more on generating deep relationships related to "force, acceleration, and mass", while in another textbook focusing on energy, the system will dynamically adjust the relevance of knowledge points according to the context and generate a graph related to "energy conservation".

[0122] Through this dynamic generation method, this method can automatically adapt and generate the relationship between knowledge points according to the content of different textbooks, no longer relying on manual annotation or preset templates. This flexibility improves the efficiency and accuracy of graph generation, especially when facing textbooks in different disciplines or fields, and can ensure the accuracy and completeness of the relationship expression in the knowledge graph.

[0123] Based on the above embodiments, it can be seen that the method can be based on a knowledge point extraction algorithm of semantic understanding, can handle complex syntactic structures and ambiguity, and dynamically generate associations between knowledge points to automatically construct a knowledge graph; thereby achieving the flexibility and adaptability of the dynamic knowledge graph construction mechanism.

[0124] The flexibility and adaptability of the dynamic knowledge graph construction mechanism means that this method can automatically adapt to different types of teaching materials and automatically generate complex relationships between knowledge points by introducing dynamic knowledge graph generation technology based on relational reasoning. Compared with the existing technology that relies on manual annotation or preset templates, the dynamic generation mechanism of this method greatly improves the flexibility of graph construction.

[0125] Existing technologies require artificially setting relationship rules between knowledge points. This approach often fails to take into account diversity and complexity when dealing with interdisciplinary and cross-field teaching materials, and is prone to produce one-sided or distorted knowledge relationships. The dynamic generation mechanism of this method can automatically mine the hierarchical structure and logical relationships between knowledge points from the context of the teaching materials through natural language understanding and reasoning. This automated process not only improves the efficiency of construction, but also can flexibly adjust the structure of the knowledge graph according to the knowledge characteristics of different disciplines to make it more in line with actual needs. This technology improves the adaptability of the graph, especially in the field of education, and can be widely used in various teaching materials.

[0126] Step 206: Construct a target knowledge graph based on the multiple target document knowledge points and the target association relationships between the target document knowledge points.

[0127] In one or more embodiments provided in this specification, the target document is a target teaching material, the initial document knowledge points are initial teaching material knowledge points, and the target document knowledge points are target teaching material knowledge points;

[0128] The constructing a target knowledge graph based on the multiple target document knowledge points and the target association relationships between the target document knowledge points includes:

[0129] Based on multiple target textbook knowledge points and the target association relationships between the target textbook knowledge points, the target knowledge graph corresponding to the target textbook is constructed.

[0130] Using the above example, the knowledge graph generation rule of this method can be to convert the extracted knowledge points (i.e., target textbook knowledge points) and their relationships (i.e., target association relationships) into graph structure data (i.e., target knowledge graph). According to the rule setting, the knowledge points are used as nodes in the graph, and the relationships are used as edges to generate a complete knowledge graph. A graph database (such as Neo4j) can be used to store and manage the knowledge graph.

[0131] After processing by one or more of the above embodiments, the method will output a target knowledge graph (KnowledgeGraph); that is, the final output of the method is a structured knowledge graph. The knowledge graph (i.e., the target knowledge graph) displays various knowledge points and their relationships in the form of nodes and edges, which can be used for display, query, and reasoning. Moreover, the knowledge graph can be further output in different formats, such as JSON, XML, RDF, etc., for subsequent integration, display, or analysis.

[0132] It should be noted that after obtaining the target knowledge graph, this method can perform feedback and iterative operations on the target knowledge graph; the feedback and iteration mean that the output knowledge graph can be optimized through manual correction and feedback mechanism, and the verified and adjusted knowledge graph can be used to further improve the effect of the pre-training model, forming a virtuous iteration.

[0133] Based on this embodiment, it can be seen that the knowledge graph construction method in this specification provides an automated knowledge graph generation method based on relational reasoning, which is adapted to the relational structure of different types of teaching materials. Through the knowledge point extraction technology based on semantic understanding, the recognition accuracy of key knowledge points in complex teaching materials is significantly improved. In addition, the deep learning model can capture the connection between a wider range of contexts and implicit knowledge points, avoiding the common omission and false detection problems of keyword matching technology, thereby ensuring higher quality of the knowledge graph.

[0134] In one or more embodiments provided in this specification, after constructing the target knowledge graph based on the multiple target document knowledge points and the target association relationships between the target document knowledge points, the method further includes:

[0135] Determining an update document corresponding to the target document, and performing knowledge point analysis on the update document using the knowledge point analysis model to obtain a plurality of newly added knowledge points in the update document;

[0136] Using the multiple newly added knowledge points, updating the multiple target document knowledge points in the target knowledge graph to obtain multiple updated knowledge points;

[0137] Using a relationship analysis model, performing relationship analysis on the multiple updated knowledge points and the target association relationship to obtain an updated association relationship between the updated knowledge points;

[0138] Based on the multiple updated knowledge points and the updated association relationship between the updated knowledge points, the target knowledge graph is updated to obtain an updated target knowledge graph.

[0139] Among them, the updated document can be understood as a document used to supplement the content of the target document, or a document used to correct the content of the target document; the newly added knowledge points can be understood as the document knowledge points contained in the updated document; the updated knowledge points can be understood as the document knowledge points after the knowledge points are updated; the updated association relationship can be understood as the association relationship between the updated knowledge points.

[0140] Following the above example, after building the target knowledge graph, this method also introduces an automated knowledge graph update mechanism; considering that existing knowledge graph technology is usually static after construction, it is difficult to update according to new teaching materials or new knowledge points. This method designs an automated update mechanism so that the system can dynamically update the knowledge graph as new teaching materials (i.e. updated documents) or new knowledge points (updated knowledge points) are added; this includes:

[0141] The language model is used to analyze the knowledge points of the updated teaching materials (i.e., updated documents) to obtain multiple new knowledge points in the updated documents, thereby determining the new knowledge points introduced (i.e., new knowledge points); the new knowledge points are added to the historical knowledge points to obtain the supplemented knowledge points (i.e., updated knowledge points).

[0142] Alternatively, the method can also utilize new knowledge points to correct errors in historical knowledge points, thereby correcting known erroneous relationships or erroneous knowledge points.

[0143] Finally, the NER model is used to adjust the relationship between knowledge points to obtain the adjusted logical relationship (i.e., updated association relationship), and a new knowledge graph (i.e., updated target knowledge graph) is constructed based on the knowledge points and logical relationships.

[0144] This automatic update mechanism allows the knowledge graph to keep pace with the times and adapt to the rapidly developing knowledge system and the constantly updated textbook content. Compared with the traditional static graph, the dynamic update function of this method significantly improves the continuous use value of the system and avoids frequent manual maintenance and updates.

[0145] Based on this embodiment, it can be seen that this method introduces the efficiency of dynamic updating and maintenance; specifically, this method designs an automated updating mechanism so that the knowledge graph can be automatically updated according to the new textbook content and the latest knowledge points, while most existing technologies use static knowledge graphs, and the update requires manual intervention, which is time-consuming and labor-intensive.

[0146] And considering the rapid changes in educational resources and knowledge systems, the knowledge graph must be updated in a timely manner to reflect the latest knowledge and textbook content. Once the static graph in the prior art is built, maintenance and updates often require a large number of repeated manual operations, which is time-consuming and prone to consistency problems. However, this method uses an automated update mechanism to automatically adjust the graph structure and content through incremental updates when new textbooks are introduced or knowledge points change, ensuring that it always keeps pace with the latest knowledge; in this way, the system not only reduces maintenance costs, but also significantly improves the timeliness and accuracy of the knowledge graph.

[0147] In one or more embodiments provided in this specification, after constructing the target knowledge graph based on the multiple target document knowledge points and the target association relationships between the target document knowledge points, steps one to three are also included:

[0148] Step 1: Determine the multimodal data corresponding to the target document, and perform knowledge point recognition on the multimodal data to obtain multimodal features corresponding to the multimodal data.

[0149] Among them, multimodal data can be understood as multimodal data related to the target document or document knowledge points in the target document. For example, the multimodal data can be data in various forms such as text, images, tables, formulas, etc. corresponding to the target teaching materials.

[0150] Following the above example, after completing the construction of the knowledge graph, this method introduces multimodal data processing capabilities in order to improve the data richness of the knowledge graph; since most existing technologies only process pure text textbook content and lack the ability to process multimodal information such as images, tables, formulas, etc., the scope of knowledge point extraction and relationship construction is limited. Therefore, this method introduces multimodal data processing capabilities, which can process various forms in textbooks, such as text, charts, images, formulas, etc., and extract and integrate knowledge points (i.e., multimodal knowledge points) and their related information.

[0151] In one or more embodiments provided in this specification, the step of performing knowledge point recognition on the multimodal data to obtain multimodal features corresponding to the multimodal data includes:

[0152] The multimodal data is input into a multimodal data processing model, and at least two multimodal feature extraction units in the multimodal data processing model are used to perform feature extraction on the multimodal data to obtain the multimodal features.

[0153] Following the above example, this method introduces a multimodal data processing module (i.e., a multimodal data processing model) in the process of executing multimodal data processing, which can process various forms in teaching materials, such as text, charts, images, formulas, etc.; for the processing and integration of multimodal data, feature extraction needs to be performed first, and the specific feature extraction method is as follows:

[0154] Each modality of data is converted into a specific feature representation (i.e., multimodal features). For text, natural language processing techniques (such as word embedding models or pre-trained language models) are usually used. For images, computer vision techniques (such as convolutional neural networks, CNN) may be used to extract visual features. Audio can be transcribed into text or sound features can be extracted through automatic speech recognition (ASR).

[0155] Step 2: Based on the multimodal features, determine the multimodal knowledge points corresponding to the multimodal data, and determine similar document knowledge points corresponding to the multimodal knowledge points from the multiple target document knowledge points, wherein the similar document knowledge points are target document knowledge points among the multiple target document knowledge points that are similar to the multimodal knowledge points.

[0156] Among them, multimodal knowledge points can be understood as knowledge points extracted from multimodal data; for example, knowledge points are extracted from data in various forms such as text, images, tables, formulas, etc.

[0157] In one or more embodiments provided in this specification, determining, from the plurality of target document knowledge points, similar document knowledge points corresponding to the multimodal knowledge point includes:

[0158] Using the knowledge point recognition unit in the multimodal data processing model, the knowledge point recognition is performed according to the multimodal features to obtain the multimodal knowledge points corresponding to the multimodal data;

[0159] Utilizing the knowledge point feature extraction unit in the multimodal data processing model, feature extraction is performed on the plurality of target document knowledge points to obtain a plurality of knowledge point features;

[0160] Using the feature alignment unit in the multimodal data processing model, the plurality of knowledge point features and the multimodal features are feature aligned to obtain a plurality of aligned knowledge point features and aligned multimodal features;

[0161] Utilizing the feature fusion unit in the multimodal data processing model, the similarity between each aligned knowledge point feature and the aligned multimodal feature is calculated;

[0162] Based on the similarity, a similar knowledge point feature similar to the aligned multimodal feature is determined from the multiple aligned knowledge point features, and the target document knowledge point corresponding to the similar knowledge point feature is determined as the similar document knowledge point corresponding to the multimodal knowledge point.

[0163] The knowledge point recognition unit can be understood as a sub-model or network layer in a multi-modal data processing model that can perform knowledge point recognition on multi-modal data. For example, the knowledge point recognition unit can be a model such as BERT, GPT, ResNet, EfficientNet, etc. There can be multiple knowledge point recognition units, and each modal data can have a corresponding knowledge point recognition unit dedicated to processing the corresponding modal data.

[0164] The knowledge point feature extraction unit can be understood as a sub-model or network layer in the multimodal data processing model that can extract features of the target document knowledge points.

[0165] The feature alignment unit can be understood as a sub-model or network layer that implements feature alignment in a multimodal data processing model.

[0166] The feature fusion unit can be understood as a sub-model or network layer that fuses features in a multimodal data processing model.

[0167] Using the above example, in the process of performing multimodal data processing, this method needs to perform the following operations after performing feature extraction in order to process and integrate multimodal data:

[0168] 1. Modal alignment: The temporal and semantic features of data in different modalities may be different, so they need to be mapped to the same semantic space through modal alignment technology. For example, a cross-modal neural network model (i.e., feature alignment unit) can be used to ensure that data in different modalities are consistent in the same knowledge dimension. The cross-modal neural network model can be a cross-modal transformer or contrastive learning.

[0169] It should be noted that the multimodal data processing model in this method includes a knowledge point feature extraction network layer (i.e., a knowledge point feature extraction unit) for the target document knowledge points, which is used to extract features from multiple target document knowledge points and obtain multiple knowledge point features to facilitate subsequent feature alignment and feature fusion.

[0170] 2. Modal fusion: After alignment, the system uses a fusion mechanism to integrate information from different modalities. This fusion can be performed at the feature level (by concatenating feature vectors of different modalities) or at the decision level (by weighted averaging the prediction results of each modality); common methods include multi-head attention mechanism, multimodal contrastive learning, etc.

[0171] Based on this, the system used in this method (i.e., the knowledge graph construction system) can use the following specific technologies to implement multimodal data processing:

[0172] 1. Introducing pre-trained models (i.e., multimodal data processing models): Multimodal pre-trained models, such as CLIP (Contrastive Language-Image Pre-training), can be used to process the combination of text and images. This model can perform efficient information alignment and retrieval in a multimodal semantic space by jointly training text and image data.

[0173] 2. The multimodal pre-training model includes a natural language processing module: knowledge points are extracted from the text through language models such as BERT and GPT (i.e., knowledge point recognition units), and the contextual understanding capabilities of these models are used to generate potential relationships between different knowledge points.

[0174] 3. The multimodal pre-training model includes a computer vision module (i.e., knowledge point recognition unit): through deep learning networks (such as ResNet, EfficientNet), relevant concepts are extracted from images in the textbooks, especially schematic diagrams, flow charts, etc., which often have implicit correlations between concepts.

[0175] 4. The multimodal pre-training model includes a modal fusion layer (i.e., feature fusion unit): After feature extraction and feature alignment of data of different modalities, a multi-head attention mechanism or Transformer architecture (i.e., feature fusion unit) is used to perform feature fusion processing, so that the system can establish associations between multimodal data, thereby better constructing a knowledge graph.

[0176] It should be noted that the feature fusion processing in this method refers to calculating the similarity between multimodal features and knowledge point features, thereby determining the correspondence between multimodal features and knowledge point features (i.e., similar knowledge point features) through the calculated similarity, and then determining the correspondence between multimodal knowledge points and document knowledge points (i.e., similar document knowledge points), so that multimodal data can be associated with document knowledge points, thereby achieving the purpose of fusing multimodal data with corresponding document knowledge points.

[0177] Step three: Determine the multimodal data as reference data for the similar document knowledge points.

[0178] The reference data may be understood as data that supplements or references the content of similar document knowledge points.

[0179] Continuing with the above example, through the cooperation of the above modules, the multimodal data processing model can effectively integrate various types of information in the textbooks, help the system understand knowledge points from different dimensions and generate a richer knowledge graph.

[0180] This model provides the overall system with the ability to fuse multi-source information, ensuring that knowledge extraction not only relies on text information, but can also be combined with other data forms such as images, making it more comprehensive and accurate.

[0181] The ability to process multimodal data means that this method is not limited to text analysis, but can also extract knowledge points and their relationships from various non-text information in textbooks. This can significantly improve the breadth and depth of the knowledge graph, especially when it comes to science and engineering textbooks, the ability to process formulas and images becomes a major advantage.

[0182] Multimodal processing is effective because each modality can provide complementary semantic information. For example, a text may describe a formula, while the corresponding image shows the application scenario of the formula. By integrating information from different modalities, the system can obtain a more comprehensive and accurate knowledge representation.

[0183] Based on the above embodiments, it can be seen that the present method supports multimodal data input, expands the scope of knowledge graph construction, and thus realizes the comprehensiveness of multimodal data processing capabilities; specifically, the present method overcomes the limitation that the prior art can only process plain text by supporting multimodal data processing, including knowledge point extraction and graph construction of information such as images, tables, and formulas, and expands the application scope and depth of the knowledge graph.

[0184] Considering that many textbooks, especially science and engineering textbooks, contain a lot of non-text information, such as formulas, charts, images, etc., these contents often contain core knowledge points and logical reasoning chains. When processing such content, existing technologies often lose a lot of important information due to the lack of support for non-text information. By introducing multimodal data processing technology, this method can extract corresponding knowledge points and their relationships from these non-text contents and integrate them into the knowledge graph. This comprehensive data processing capability ensures the integrity of the knowledge graph, allowing the graph to cover more types of knowledge information, not just text descriptions, thereby more accurately reflecting the full picture of subject knowledge.

[0185] In one or more embodiments provided in this specification, after constructing the target knowledge graph based on the multiple target document knowledge points and the target association relationships between the target document knowledge points, the method further includes:

[0186] Determine the knowledge learning record data corresponding to the target knowledge graph, wherein the knowledge learning record data is the learning record data generated by the user during the knowledge learning process based on the target knowledge graph;

[0187] Analyze the knowledge learning record data using a learning status analysis module to obtain knowledge point learning status information;

[0188] Performing a learning status evaluation based on the knowledge point learning status information and the target knowledge graph to obtain a learning status evaluation result;

[0189] Based on the learning status evaluation result, the knowledge points to be learned by the user are determined from the target knowledge graph.

[0190] Among them, the knowledge learning record data can be understood as the user's learning behavior data for the target document knowledge points in the target knowledge graph. For example, the knowledge learning record data can be answer records, homework submissions, course progress, and interactive feedback.

[0191] The knowledge point learning status information can be understood as the user's mastery of the knowledge points in the target document.

[0192] The learning status assessment result can be understood as the assessment result of the user's mastery of the knowledge points in the target document.

[0193] The knowledge points to be learned can be understood as target document knowledge points among multiple target document knowledge points that the user still needs to learn; the knowledge points to be learned can be multiple associated target document knowledge points; the multiple associated target document knowledge points can serve as the user's learning path, thereby recommending the learning path to the user.

[0194] Using the above example, after constructing the target knowledge graph, this method can recommend personalized learning paths based on the target knowledge graph. The specific implementation method is as follows:

[0195] Since existing knowledge graphs are mostly used to display knowledge points and their relationships, they lack the ability to adapt to learners' personalized needs. Therefore, this method combines learners' learning progress and needs and provides a personalized learning path recommendation function based on knowledge graphs. This function intelligently generates the optimal learning path by analyzing the knowledge points that learners have mastered and the content that needs to be learned.

[0196] One of the core functions of the system (knowledge graph construction system) is to analyze the knowledge points that learners have mastered. The system modules and execution steps to achieve this function include:

[0197] 1. Data collection module: collects learners’ learning behavior data (i.e. knowledge learning record data) through various channels. This includes but is not limited to answer records, homework submissions, course progress, and interactive feedback. These data exist in multimodal forms, including text, voice, video, and user interaction logs.

[0198] 2. Data processing and analysis module: Process the collected learning data, use natural language processing (NLP) technology to analyze text data, combine deep learning models for semantic understanding, and extract the learner's knowledge points (i.e. knowledge point learning status information). For example, by analyzing the answer records, the system can determine whether the learner has a complete grasp of certain knowledge points, and determine weak points based on error rates or repeated errors.

[0199] 3. Knowledge point mastery assessment: The system calculates the learner's mastery of each knowledge point (i.e., learning status assessment results) by comparing it with the relevant knowledge points in the knowledge graph. The assessment algorithm can adopt a weighted scoring mechanism to consider the influence of different types of data, such as the correctness of oral expression and the accuracy of answering questions.

[0200] 4. After determining the learner's current knowledge mastery, the system will generate the knowledge content to be learned next (i.e. the knowledge points to be learned). The implementation process includes:

[0201] 1) Knowledge point gap analysis: The system compares the learner's knowledge mastery map with the complete textbook knowledge map to identify the knowledge points that the learner has not yet mastered or does not understand deeply. By combining prior knowledge (such as the learning order in the textbook), the system can predict the learner's learning needs.

[0202] 2) Learning path dependency modeling: In the structure of the knowledge graph, there are dependencies between knowledge points. By analyzing these relationships, the system identifies the logical sequence between knowledge points that have not yet been mastered, ensuring that the learning path is coherent and that necessary prerequisite knowledge is not omitted.

[0203] The generation of the optimal learning path is an intelligent reasoning process based on the knowledge graph, the learner's mastery and the learning goals.

[0204] Specifically as shown in step 3):

[0205] 3) Path optimization algorithm: The system uses a path optimization algorithm (such as A* algorithm or reinforcement learning algorithm) to search for the optimal path (i.e., the knowledge points to be learned) in the knowledge graph. The algorithm comprehensively considers the learner's current knowledge status, learning priorities (for example, certain knowledge points are more important in practical applications), and the learner's personal preferences (such as learning speed and learning method).

[0206] It should be noted that after determining the knowledge content that the user wants to learn, this method can recommend the knowledge content to the user; this process can be implemented using a personalized recommendation system. Specifically, based on a recommendation system model (such as collaborative filtering or content-based recommendation), the system can optimize the path selection of the current learner based on the learning paths of similar learners. By comparing the learning results of different learners, the system can dynamically adjust the recommended content.

[0207] In addition, this method can perform dynamic path adjustment; since the learning process is nonlinear, the system has the function of dynamically adjusting the learning path. After the learner completes the learning of a certain knowledge point, the system updates its knowledge map in real time according to its mastery level and adjusts the subsequent learning plan. For knowledge points that have not been fully mastered, the system will guide the learner to review or provide more relevant resources for reinforcement.

[0208] This personalized learning path recommendation not only improves the pertinence and efficiency of learning, but also transforms the knowledge graph from a passive knowledge display tool to an active teaching assistance system. Learners can efficiently master the core knowledge points in the textbooks according to the recommended learning path and improve their learning experience.

[0209] Based on the content of this embodiment, it can be seen that this method provides a personalized learning path generation algorithm based on the knowledge graph, thereby achieving the effectiveness and practicality of personalized learning recommendations; specifically, unlike the prior art, this method provides intelligent and personalized learning path recommendations by combining the user's learning progress and knowledge graph. This not only improves the display function of the knowledge graph, but also makes it a tool for learners to acquire knowledge efficiently.

[0210] Existing knowledge graph technologies are mostly used for knowledge point display, lacking the ability to adapt to individual learners' needs and making it difficult to provide targeted learning guidance. This method combines learners' personalized needs, analyzes the knowledge points they have mastered and the content they have not mastered, and automatically generates a learning path suitable for learners through knowledge graphs. This personalized recommendation function based on knowledge point associations not only improves the targeted nature of learning, but also optimizes learning efficiency, allowing learners to master core knowledge in the shortest time. Compared with the single knowledge display function of existing technologies, this method significantly improves the practicality of the system by providing active personalized learning guidance.

[0211] In one or more embodiments provided in this specification, after constructing the target knowledge graph based on the multiple target document knowledge points and the target association relationships between the target document knowledge points, the method further includes:

[0212] Sending the target knowledge graph to a client, so that the client displays the target knowledge graph to a user through an interactive interface;

[0213] receiving a graph relationship adjustment instruction sent by the client, wherein the graph relationship adjustment instruction is sent by the client when the user performs an association relationship adjustment operation based on the interactive interface;

[0214] Determining the association relationship identifier and the association relationship adjustment information from the graph relationship adjustment instruction;

[0215] Based on the association relationship identifier, determining the association relationship to be adjusted from the target association relationship in the target knowledge graph;

[0216] Based on the association relationship adjustment information, the association relationship to be adjusted in the target knowledge graph is adjusted to obtain an adjusted target knowledge graph.

[0217] Among them, the graph relationship adjustment instruction can be understood as an instruction for adjusting the target knowledge graph; the association relationship identifier can be understood as the identifier of the target association relationship; the association relationship adjustment information can be understood as information for adjusting the target association relationship, for example, the association relationship adjustment information can be a new association relationship, a new target document knowledge point, etc.

[0218] The interactive interface may be a human-computer interactive interface in the client, such as a network, an application program interface, etc. The association adjustment operation may be understood as an operation performed by the user based on the interactive interface to adjust the target association, such as click, drag, and the like.

[0219] Following the above example, after constructing the target knowledge graph, this method can realize automatic visualization and user interaction; considering that in the existing technology, the visualization and user interaction functions of the knowledge graph are relatively limited, usually just simple graphic display, unable to provide an in-depth interactive experience. This method introduces more advanced graph visualization technology and interactive design; the target knowledge graph is displayed to the user using the human-computer interaction interface, and the user can explore the nodes and their relationships in the knowledge graph through interactive operations with the interactive interface (such as clicking, dragging, etc.), to obtain a more intuitive and in-depth learning experience.

[0220] The user can reorganize the structure of knowledge points through interactive operations (drag and drop operations) with the client's interactive interface; after the user performs the interactive operation, the client can send a structure reconstruction instruction (i.e., a graph relationship adjustment instruction) to the server; wherein the server is the server used by the knowledge graph construction method in this specification.

[0221] After receiving the structure reconstruction instruction, the server determines the association relationship to be adjusted from the knowledge graph according to the association relationship identifier carried in the structure reconstruction instruction.

[0222] Furthermore, the new association relationship carried in the structural reconstruction instruction is used to replace the association relationship to be adjusted, thereby achieving the reorganization of the structure of the knowledge points; or, the new target document knowledge point carried in the structural reconstruction instruction is used to change the historical knowledge points connected by the association relationship to be adjusted to the new target document knowledge point, thereby achieving the reorganization of the structure of the knowledge points.

[0223] After completing the reconstruction of the knowledge points, the reconstructed knowledge graph (i.e., the adjusted target knowledge graph) can be obtained, and the reconstructed knowledge graph can be displayed to the user through an interactive interface.

[0224] The enhanced visualization and interaction functions in this method greatly improve the user-friendliness and usage effect of the knowledge graph. Users can not only browse the knowledge points in the graph, but also explore, retrieve and personalize knowledge through interaction with the graph, which greatly improves the practicality of the system.

[0225] Based on the content of this embodiment, it can be seen that this method introduces interactive knowledge graph visualization technology, which supports users to freely browse and operate knowledge points and their relationships, thereby achieving improved interactivity and visualization; specifically, this method has made significant improvements in the visualization and user interaction of the knowledge graph, allowing users to explore, learn and retrieve the knowledge graph through intuitive operations (such as clicking, dragging, etc.), while most of the existing technologies are just simple static graph displays.

[0226] Most existing knowledge graph visualization systems lack in-depth interactive functions. Users can only browse the already constructed graphs, and cannot flexibly operate, search, or customize the view. This method provides rich interactive functions through advanced visualization technology design, allowing users to freely explore and adjust in the graph according to their learning needs. For example, users can view detailed information or associations by clicking on knowledge points, or reorganize the structure of knowledge points by dragging and dropping. This interactivity enhances the user's sense of participation and learning experience, making the knowledge graph not only a passive display tool, but also an active knowledge exploration platform.

[0227] The knowledge graph construction method provided in one or more embodiments of this specification introduces artificial intelligence technology in the process of constructing the target knowledge graph. Through the knowledge point analysis model and the relationship analysis model, the knowledge points and association relationships can be automatically and efficiently acquired, thereby improving the efficiency of constructing the target knowledge graph. In addition, this method uses a two-stage acquisition method to accurately acquire knowledge points and association relationships. In the first stage, the initial document knowledge points and initial association relationships are obtained; in the second stage, more accurate target document knowledge points and target association relationships are obtained based on the initial document knowledge points and initial association relationships, thereby ensuring the accuracy of the target knowledge graph. Therefore, the problems of low efficiency and poor accuracy in the knowledge graph construction process are avoided, and efficient and accurate construction of the knowledge graph is achieved.

[0228] The following combination Figure 3 , taking the application of the knowledge graph construction method provided in this specification in an educational scenario as an example, the knowledge graph construction method is further explained. Figure 3 A process flow chart of a knowledge graph construction method provided by an embodiment of this specification is shown, which specifically includes the following steps.

[0229] Step 302: Construct a knowledge graph corresponding to the teaching material.

[0230] The process of constructing the knowledge graph corresponding to the textbook in this method is divided into three steps: knowledge point extraction based on deep learning, dynamic knowledge relationship generation mechanism, and knowledge graph construction:

[0231] Among them, knowledge point extraction based on deep learning refers to using deep learning models to more deeply understand and analyze the contextual semantics in the textbooks, thereby identifying knowledge points. The specific implementation method is as follows:

[0232] 1. Preprocess the teaching materials content.

[0233] First, determine the content of the textbook; the textbook content can be used as text data as input to the pre-trained language model. The input data usually contains paragraphs, chapters, titles, and other structured text information. The text data can be original book scans (processed by optical character recognition), e-books, or digital textbook resources.

[0234] Secondly, the textbook content is preprocessed; the input content of the pre-trained language model usually needs to be preprocessed, such as segmentation, sentence segmentation, and removal of irrelevant symbols (such as punctuation marks, charts, etc.) to ensure that the input content meets the requirements of the language model. Therefore, by performing the above preprocessing operations on the original textbook content, the textbook content that meets the requirements of the language model is obtained.

[0235] 2. Use the pre-trained language model to vectorize the text of the teaching materials.

[0236] Specifically, this method converts the textbook content into context-aware word vectors (embedded ing) through a pre-trained language model. These word vectors contain the semantic information of the words and details related to the context of the textbook content.

[0237] 3. Perform knowledge point recognition.

[0238] After determining the word vector, the knowledge point recognition algorithm can be applied to the word vector corresponding to each textbook chapter in the textbook to automatically identify important concepts, terms, definitions and other knowledge points in each textbook chapter.

[0239] For example, through the trained NER model, terms with specific subject meanings such as "calculus" and "periodic table" can be marked.

[0240] 4. Perform relationship extraction.

[0241] After identifying the knowledge points, the relation extraction technology can be further used to identify the logical relationships between the knowledge points in the document chapters. These logical relationships can be causal relationships, hierarchical relationships, time relationships, etc. between concepts. For example, "addition is the basis of multiplication", "oxygen and hydrogen can generate water", etc. In this example, the system will infer that there is an inheritance relationship between "Newton's Second Law" and "Newton's First Law", and there is a conditional derivation relationship between "Law of Conservation of Momentum" and "Newton's Second Law".

[0242] Among them, the dynamic knowledge relationship generation mechanism refers to automatically analyzing the knowledge points in the textbook content and constructing complex multi-level relationships between knowledge points through natural language understanding and reasoning technology.

[0243] It should be noted that after analyzing the sentence structure and vocabulary meaning in the textbook chapters through natural language processing technology to identify key physical concepts, this method can merge multiple initial document knowledge points based on the textbook context information and reconstruct the logical relationship for the merged knowledge points.

[0244] That is to say, in the process of executing the first stage (i.e., knowledge point extraction based on deep learning), the knowledge points corresponding to chapters 1, 2, 3, and 4 of the textbook can be extracted, and they will be extracted 4 times respectively. Then, in the process of executing the second stage (i.e., the dynamic knowledge relationship generation mechanism), the same knowledge points will be merged to obtain the merged knowledge points. This knowledge point merging operation is equivalent to further abstracting and combing the knowledge system presented in the textbook, so as to obtain the knowledge system corresponding to the textbook.

[0245] For instance, taking limit as an example, the knowledge point of limit may belong to the first section in the first chapter of the textbook (a second-level knowledge point). At the same time, in the second chapter of the textbook, the knowledge point of limit may be a knowledge point under the second section of the second chapter (a third-level knowledge point). Therefore, in the process of merging knowledge points, the knowledge points under different chapters can be merged according to the levels, so that the knowledge points under different chapters can be merged into the same level.

[0246] After the knowledge points are merged, a multi-level association relationship between the knowledge points can be generated through an inference algorithm based on the context information and the logical order of the knowledge points in the textbook (i.e., the hierarchical relationship, the knowledge system).

[0247] The specific implementation of the dynamic knowledge relationship generation mechanism is as follows:

[0248] 1. Merge multiple initial document knowledge points:

[0249] A plurality of similar core knowledge points are determined from the plurality of core knowledge points; and then they are merged to obtain merged knowledge points.

[0250] 2. Reconstruct the logical relationship for the merged knowledge points:

[0251] The knowledge graph construction system used in this method can generate the association relationship between knowledge points based on context information and the logical order of knowledge points in the textbook through reasoning algorithms, and then perform multi-level complex relationship generation. The specific method is as follows:

[0252] First, the context information of the teaching material and the association relationship between the knowledge points are input into a relational analysis model (eg, a NER model).

[0253] Secondly, by analyzing the contents of multiple chapters, the system can use the relationship analysis module to dynamically generate more complex multi-level relationships. For example, when discussing collision problems, the system can automatically infer that the "law of conservation of momentum" and "Newton's second law" act together on the collision system, and generate a higher-level relationship diagram that reflects the comprehensive application of these laws in solving complex physical problems.

[0254] It should be noted that, based on the above-mentioned dynamic knowledge relationship generation mechanism, this method can realize dynamic adaptation of the knowledge graph.

[0255] This dynamic adaptation means that the knowledge graph generation method provided by this method is not a pre-set static relationship, but a dynamic adjustment based on the different contents and knowledge points of different chapters of the textbook. For example, for a textbook focusing on dynamics, the system will focus more on generating deep relationships related to "force, acceleration, and mass", while in another textbook focusing on energy, the system will dynamically adjust the relevance of knowledge points according to the context and generate a graph related to "energy conservation".

[0256] Through this dynamic generation method, this method can automatically adapt and generate the relationship between knowledge points according to the content of different textbooks, no longer relying on manual annotation or preset templates. This flexibility improves the efficiency and accuracy of graph generation, especially when facing textbooks in different disciplines or fields, and can ensure the accuracy and completeness of the relationship expression in the knowledge graph.

[0257] Among them, constructing a knowledge graph means: converting the extracted knowledge points and their relationships into graph structure data. According to the rule setting, knowledge points are used as nodes in the graph, and relationships are used as edges to generate a complete knowledge graph; the knowledge graph displays each knowledge point and their mutual relationships in the form of nodes and edges, which can be used for display, query and reasoning. The knowledge graph can also be further output in different formats, such as JSON, XML, RDF, etc., for subsequent integration, display or analysis.

[0258] Step 304: Process the constructed knowledge graph.

[0259] Specifically, after constructing the knowledge graph, this method can perform operations on the knowledge graph such as: multimodal data association, personalized learning path recommendation, automated visualization and user interaction, and / or automated knowledge graph update.

[0260] Among them, multimodal data association means: associating corresponding multimodal data for knowledge points in the knowledge graph.

[0261] In the process of executing multimodal data processing, this method introduces a multimodal data processing module, which can process various forms in teaching materials, such as text, charts, images, formulas, etc.; the specific implementation method for the processing and integration of multimodal data is as follows:

[0262] 1. Perform feature extraction.

[0263] Each modality of data is converted into a specific feature representation. For text, natural language processing techniques (such as word embedding models or pre-trained language models) are usually used. For images, computer vision techniques (such as convolutional neural networks, CNN) may be used to extract visual features. Audio can be transcribed into text or sound features can be extracted through automatic speech recognition (ASR).

[0264] 2. Modal alignment: The temporal and semantic features of data in different modalities may be different, so they need to be mapped to the same semantic space through modal alignment technology. For example, a cross-modal neural network model can be used to ensure that data in different modalities are consistent in the same knowledge dimension. The cross-modal neural network model can be a cross-modal transformer or contrastive learning.

[0265] It should be noted that the multimodal data processing model in this method includes a knowledge point feature extraction network layer for the target document knowledge points, which is used to extract features of multiple target document knowledge points and obtain multiple knowledge point features to facilitate subsequent feature alignment and feature fusion.

[0266] 3. Modal fusion: After alignment, the system uses a fusion mechanism to integrate information from different modalities. This fusion can be performed at the feature level (by concatenating feature vectors of different modalities) or at the decision level (by weighted averaging of prediction results of each modality); common methods include multi-head attention mechanism, multimodal contrastive learning, etc.

[0267] Based on this, the system used in this method (i.e., the knowledge graph construction system) can use the following specific technologies to implement multimodal data processing:

[0268] 1. Pre-training models are introduced: Multimodal pre-training models, such as CLIP (Contrast ive Language-Image Pre-training), can be used to process the combination of text and images. This model can perform efficient information alignment and retrieval in a multimodal semantic space by jointly training text and image data.

[0269] 2. The multimodal pre-training model includes a natural language processing module: knowledge points are extracted from the text through language models such as BERT and GPT, and the contextual understanding capabilities of these models are used to generate potential relationships between different knowledge points.

[0270] 3. The multimodal pre-trained model includes a computer vision module: through deep learning networks (such as ResNet, EfficientNet), relevant concepts are extracted from images in the textbooks, especially schematic diagrams, flow charts, etc., which often have implicit correlations between concepts.

[0271] 4. The multimodal pre-training model includes a modal fusion layer: After feature extraction and feature alignment of data of different modalities, feature fusion processing is performed using a multi-head attention mechanism or Transformer architecture, so that the system can establish associations between multimodal data, thereby better constructing a knowledge graph.

[0272] It should be noted that the feature fusion processing in this method refers to calculating the similarity between multimodal features and knowledge point features, thereby determining the correspondence between multimodal features and knowledge point features through the calculated similarity, and then determining the correspondence between multimodal knowledge points and document knowledge points, so that multimodal data can be associated with document knowledge points, thereby achieving the purpose of fusing multimodal data with corresponding document knowledge points.

[0273] Through the cooperation of the above modules, the multimodal data processing model of this method can effectively integrate various types of information in the textbooks, help the system understand knowledge points from different dimensions and generate a richer knowledge graph.

[0274] This model provides the overall system with the ability to fuse multi-source information, ensuring that knowledge extraction not only relies on text information, but can also be combined with other data forms such as images, making it more comprehensive and accurate.

[0275] The ability to process multimodal data means that this method is not limited to text analysis, but can also extract knowledge points and their relationships from various non-text information in textbooks. This can significantly improve the breadth and depth of the knowledge graph, especially when it comes to science and engineering textbooks, the ability to process formulas and images becomes a major advantage.

[0276] Among them, personalized learning path recommendation refers to: personalized learning path recommendation based on the target knowledge graph.

[0277] After constructing the target knowledge graph, this method can recommend personalized learning paths based on the target knowledge graph. The specific implementation method is as follows:

[0278] Since existing knowledge graphs are mostly used to display knowledge points and their relationships, they lack the ability to adapt to learners' personalized needs. Therefore, this method combines learners' learning progress and needs and provides a personalized learning path recommendation function based on knowledge graphs. This function intelligently generates the optimal learning path by analyzing the knowledge points that learners have mastered and the content that needs to be learned.

[0279] One of the core functions of the system (knowledge graph construction system) is to analyze the knowledge points that learners have mastered. The system modules and execution steps to achieve this function include:

[0280] 1. Data collection module: collects learners' learning behavior data through various channels. This includes but is not limited to answer records, homework submissions, course progress, and interactive feedback. These data exist in multimodal forms, including text, voice, video, and user interaction logs.

[0281] 2. Data processing and analysis module: Process the collected learning data, analyze text data using natural language processing (NLP) technology, combine deep learning models for semantic understanding, and extract the learner's knowledge mastery. For example, by analyzing the answer records, the system can determine whether the learner has mastered certain knowledge points completely, and determine the weak points based on the error rate or repeated errors.

[0282] 3. Knowledge point mastery assessment: The system calculates the learner's mastery of each knowledge point by comparing it with the relevant knowledge points in the knowledge graph. The assessment algorithm can adopt a weighted scoring mechanism to consider the influence of different types of data, such as the correctness of oral expression and the accuracy of answering questions.

[0283] 4. After determining the learner's current knowledge mastery, the system will generate the knowledge content that needs to be learned next; the implementation process includes:

[0284] 1) Knowledge point gap analysis: The system compares the learner's knowledge mastery map with the complete textbook knowledge map to identify the knowledge points that the learner has not yet mastered or does not understand deeply. By combining prior knowledge (such as the learning order in the textbook), the system can predict the learner's learning needs.

[0285] 2) Learning path dependency modeling: In the structure of the knowledge graph, there are dependencies between knowledge points. By analyzing these relationships, the system identifies the logical sequence between knowledge points that have not yet been mastered, ensuring that the learning path is coherent and that necessary prerequisite knowledge is not omitted.

[0286] The generation of the optimal learning path is an intelligent reasoning process based on the knowledge graph, the learner's mastery and the learning goals.

[0287] Specifically as shown in step 3):

[0288] 3) Path optimization algorithm: The system uses a path optimization algorithm (such as A* algorithm or reinforcement learning algorithm) to search for the optimal path in the knowledge graph. The algorithm comprehensively considers the learner's current knowledge status, learning priority (for example, certain knowledge points are more important in practical applications), and the learner's personal preferences (such as learning speed, learning method).

[0289] It should be noted that after determining the knowledge content that the user wants to learn, this method can recommend the knowledge content to the user; this process can be implemented using a personalized recommendation system. Specifically, based on a recommendation system model (such as collaborative filtering or content-based recommendation), the system can optimize the path selection of the current learner based on the learning paths of similar learners. By comparing the learning results of different learners, the system can dynamically adjust the recommended content.

[0290] Among them, automated visualization and user interaction refers to: displaying the knowledge graph to the user and realizing the interaction between the knowledge graph and the user.

[0291] After constructing the target knowledge graph, this method can realize automatic visualization and user interaction. Considering that in the prior art, the visualization and user interaction functions of the knowledge graph are relatively limited, usually just simple graphic display, unable to provide an in-depth interactive experience. This method introduces more advanced graph visualization technology and interactive design; the target knowledge graph is displayed to the user using a human-computer interaction interface, and the user can explore the nodes and their relationships in the knowledge graph through interactive operations (such as clicking, dragging, etc.) with the interactive interface, to obtain a more intuitive and in-depth learning experience.

[0292] The user can reorganize the structure of knowledge points through interactive operations (drag and drop operations) with the client's interactive interface; after the user performs the interactive operation, the client can send a structure reconstruction instruction to the server; wherein, the server is the server used by the knowledge graph construction method in this specification.

[0293] After receiving the structure reconstruction instruction, the server determines the association relationship to be adjusted from the knowledge graph according to the association relationship identifier carried in the structure reconstruction instruction.

[0294] Furthermore, the new association relationship carried in the structural reconstruction instruction is used to replace the association relationship to be adjusted, thereby achieving the reorganization of the structure of the knowledge points; or, the new target document knowledge point carried in the structural reconstruction instruction is used to change the historical knowledge points connected by the association relationship to be adjusted to the new target document knowledge point, thereby achieving the reorganization of the structure of the knowledge points.

[0295] After completing the reconstruction of the knowledge points, a reconstructed knowledge graph can be obtained, and the reconstructed knowledge graph can be displayed to the user through an interactive interface.

[0296] Among them, automated knowledge graph updating refers to: automated updating of the knowledge graph.

[0297] After building the target knowledge graph, this method also introduces an automated knowledge graph update mechanism; considering that existing knowledge graph technology is usually static after construction, it is difficult to update according to new teaching materials or new knowledge points. This method designs an automated update mechanism so that the system can dynamically update the knowledge graph as new teaching materials or new knowledge points are added. This includes:

[0298] The language model is used to analyze the knowledge points of the updated textbooks to obtain multiple new knowledge points in the updated documents, thereby determining the new knowledge points to be introduced; the new knowledge points are added to the historical knowledge points to obtain the supplemented knowledge points.

[0299] Alternatively, the method can also utilize new knowledge points to correct errors in historical knowledge points, thereby correcting known erroneous relationships or erroneous knowledge points.

[0300] Finally, the NER model is used to adjust the relationship between knowledge points, obtain the adjusted logical relationship, and construct a new knowledge graph based on the knowledge points and logical relationships.

[0301] Based on the above steps, it can be seen that the knowledge graph construction method in this specification provides a technology for extracting knowledge points and their relationships based on textbook content to generate a knowledge graph. The technical effects achieved by the knowledge graph construction method include:

[0302] 1. Personalized learning experience: This technology can provide students with personalized learning resources and guidance based on their learning needs and interests. By analyzing the content of the textbooks and building a knowledge graph, the system can better understand each student's knowledge level and learning preferences, thereby tailoring a learning path for them and improving learning outcomes.

[0303] 2. Provide teaching aids: For teachers, this technology can be used as a teaching aid to help them better organize course content, develop teaching plans, and provide personalized teaching suggestions. Teachers can use knowledge graphs to discover potential connections and key concepts in courses, so as to impart knowledge more effectively and stimulate students' interest in learning.

[0304] 3. Optimize educational resources: By analyzing the content of textbooks and building knowledge graphs, educational institutions can optimize the use of educational resources. They can better understand the key points and difficulties in the courses and adjust teaching strategies and resource allocation accordingly to improve the quality and efficiency of education.

[0305] 4. Provide support for textbook editing and development: For textbook editors and developers, this technology can provide valuable reference and support. They can use knowledge graphs to better organize and present textbook content, making the textbooks more structured and easier to understand.

[0306] Moreover, compared with the above-mentioned knowledge point extraction scheme based on deep learning model, this method takes into account the problem of insufficient knowledge point extraction accuracy in this scheme. Therefore, this method can effectively improve the extraction accuracy and ensure the complete extraction of key knowledge points through a more intelligent knowledge point recognition mechanism.

[0307] Considering that the scheme lacks systematicity in constructing knowledge relationships, this method provides an automated relationship mining technology based on deep learning or semantic analysis, which can better reflect the multi-dimensional associations between knowledge points.

[0308] Taking into account the poor adaptability of this solution and its difficulty in handling the diverse content of teaching materials, this method, by establishing a general knowledge point extraction framework, can adapt to a variety of teaching materials. Both structured teaching materials and unstructured teaching materials can be efficiently processed, thus expanding the scope of application of the technology.

[0309] Considering that this solution has the problems of low automation and reliance on a lot of manual intervention, the technology proposed in this method can greatly reduce manual intervention and improve construction efficiency through automated knowledge point extraction and relationship generation, ensuring that the knowledge graph can be quickly generated and updated in a timely manner.

[0310] Taking into account the problem that this solution cannot be updated and expanded dynamically, the technology of this method can support the dynamic expansion and updating of the knowledge graph, so that the system can keep pace with the times and maintain the timeliness and cutting-edge nature of the graph content.

[0311] Taking into account the problems of insufficient user interaction and applicability of this solution, this method provides a more friendly user interface and interactive functions, supports intelligent retrieval based on knowledge graphs, personalized learning recommendations and other functions, and improves the practicality and operability of knowledge graphs.

[0312] Corresponding to the above method embodiment, this specification also provides a knowledge graph construction device embodiment, Figure 4 FIG. 1 shows a schematic diagram of a knowledge graph construction device provided by an embodiment of the present specification. Figure 4As shown, the device comprises:

[0313] The first knowledge point processing module 402 is configured to perform knowledge point analysis on the target document using the knowledge point analysis model to obtain a plurality of initial document knowledge points in the target document, and perform relationship recognition on the plurality of initial document knowledge points to obtain initial association relationships between the initial document knowledge points;

[0314] The second knowledge point processing module 404 is configured to merge the multiple initial document knowledge points to obtain multiple target document knowledge points, and use the relationship analysis model to perform relationship analysis on the multiple target document knowledge points and the initial association relationships to obtain target association relationships between the target document knowledge points.

[0315] The graph construction module 406 is configured to construct a target knowledge graph based on the multiple target document knowledge points and the target association relationships between the target document knowledge points.

[0316] Optionally, the first knowledge point processing module 402 is further configured to:

[0317] Inputting the target document into a language model for vectorization processing to obtain a plurality of word vectors corresponding to the target document, wherein the word vectors contain semantic information of document words in the target document;

[0318] Determine a plurality of document chapters contained in the target document, and determine a chapter word vector corresponding to the target document chapter from the plurality of word vectors, wherein the target document chapter is any one of the plurality of document chapters;

[0319] Inputting the chapter word vector corresponding to the target document chapter into the knowledge point analysis model to perform knowledge point recognition, and obtaining a plurality of initial document knowledge points corresponding to the target document chapter;

[0320] Relationship identification is performed on the multiple initial document knowledge points corresponding to the target document chapters to obtain initial association relationships between the initial document knowledge points corresponding to the target document chapters.

[0321] Optionally, the multiple initial document knowledge points are multiple initial document knowledge points corresponding to a target document chapter, and the target document chapter is any one of multiple document chapters of the target document;

[0322] The second knowledge point processing module 404 is further configured to:

[0323] Determining at least two identical initial document knowledge points and other initial document knowledge points other than the at least two identical initial document knowledge points from the multiple initial document knowledge points corresponding to the multiple document chapters;

[0324] Merging the at least two identical initial document knowledge points to obtain a merged knowledge point, and determining the other initial document knowledge points and the merged knowledge point as the multiple target document knowledge points;

[0325] Inputting the plurality of target document knowledge points and the initial association relationships into the relationship analysis model;

[0326] In the relationship analysis model, based on the document context information of the target document and the initial association relationship, the target document knowledge points are subjected to relationship analysis to obtain the target association relationship between the target document knowledge points.

[0327] Optionally, the knowledge graph construction device further includes a graph updating module configured to:

[0328] Determining an update document corresponding to the target document, and performing knowledge point analysis on the update document using the knowledge point analysis model to obtain a plurality of newly added knowledge points in the update document;

[0329] Using the multiple newly added knowledge points, updating the multiple target document knowledge points in the target knowledge graph to obtain multiple updated knowledge points;

[0330] Using a relationship analysis model, performing relationship analysis on the multiple updated knowledge points and the target association relationship to obtain an updated association relationship between the updated knowledge points;

[0331] Based on the multiple updated knowledge points and the updated association relationship between the updated knowledge points, the target knowledge graph is updated to obtain an updated target knowledge graph.

[0332] Optionally, the knowledge graph construction device further includes a multimodal data processing module configured to:

[0333] Determining multimodal data corresponding to the target document, and performing knowledge point recognition on the multimodal data to obtain multimodal features corresponding to the multimodal data;

[0334] Based on the multimodal features, determine the multimodal knowledge points corresponding to the multimodal data, and determine similar document knowledge points corresponding to the multimodal knowledge points from the multiple target document knowledge points, wherein the similar document knowledge points are target document knowledge points similar to the multimodal knowledge points among the multiple target document knowledge points;

[0335] The multimodal data is determined as reference data for the similar document knowledge points.

[0336] Optionally, the multimodal data processing module is further configured to:

[0337] The multimodal data is input into a multimodal data processing model, and at least two multimodal feature extraction units in the multimodal data processing model are used to perform feature extraction on the multimodal data to obtain the multimodal features.

[0338] Optionally, the multimodal data processing module is further configured to:

[0339] Using the knowledge point recognition unit in the multimodal data processing model, the knowledge point recognition is performed according to the multimodal features to obtain the multimodal knowledge points corresponding to the multimodal data;

[0340] Utilizing the knowledge point feature extraction unit in the multimodal data processing model, feature extraction is performed on the plurality of target document knowledge points to obtain a plurality of knowledge point features;

[0341] Using the feature alignment unit in the multimodal data processing model, the plurality of knowledge point features and the multimodal features are feature aligned to obtain a plurality of aligned knowledge point features and aligned multimodal features;

[0342] Utilizing the feature fusion unit in the multimodal data processing model, the similarity between each aligned knowledge point feature and the aligned multimodal feature is calculated;

[0343] Based on the similarity, a similar knowledge point feature similar to the aligned multimodal feature is determined from the multiple aligned knowledge point features, and the target document knowledge point corresponding to the similar knowledge point feature is determined as the similar document knowledge point corresponding to the multimodal knowledge point.

[0344] Optionally, the knowledge graph construction device further includes a module for determining knowledge points to be learned, which is configured to:

[0345] Determine the knowledge learning record data corresponding to the target knowledge graph, wherein the knowledge learning record data is the learning record data generated by the user during the knowledge learning process based on the target knowledge graph;

[0346] Analyze the knowledge learning record data using a learning status analysis module to obtain knowledge point learning status information;

[0347] Performing a learning status evaluation based on the knowledge point learning status information and the target knowledge graph to obtain a learning status evaluation result;

[0348] Based on the learning status evaluation result, the knowledge points to be learned by the user are determined from the target knowledge graph.

[0349] Optionally, the knowledge graph construction device further includes a graph adjustment module configured to:

[0350] Sending the target knowledge graph to a client, so that the client displays the target knowledge graph to a user through an interactive interface;

[0351] receiving a graph relationship adjustment instruction sent by the client, wherein the graph relationship adjustment instruction is sent by the client when the user performs an association relationship adjustment operation based on the interactive interface;

[0352] Determining the association relationship identifier and the association relationship adjustment information from the graph relationship adjustment instruction;

[0353] Based on the association relationship identifier, determining the association relationship to be adjusted from the target association relationship in the target knowledge graph;

[0354] Based on the association relationship adjustment information, the association relationship to be adjusted in the target knowledge graph is adjusted to obtain an adjusted target knowledge graph.

[0355] Optionally, the target document is a target textbook, the initial document knowledge points are initial textbook knowledge points, and the target document knowledge points are target textbook knowledge points;

[0356] The graph construction module 406 is further configured to:

[0357] Based on multiple target textbook knowledge points and the target association relationships between the target textbook knowledge points, the target knowledge graph corresponding to the target textbook is constructed.

[0358] The knowledge graph construction device provided in one or more embodiments of this specification introduces artificial intelligence technology in the process of constructing the target knowledge graph. Through the knowledge point analysis model and the relationship analysis model, the knowledge points and association relationships can be automatically and efficiently acquired, thereby improving the efficiency of constructing the target knowledge graph. In addition, the method adopts a two-stage acquisition method to accurately acquire knowledge points and association relationships. In the first stage, the initial document knowledge points and initial association relationships are obtained; in the second stage, more accurate target document knowledge points and target association relationships are obtained based on the initial document knowledge points and initial association relationships, thereby ensuring the accuracy of the target knowledge graph. Therefore, the problems of low efficiency and poor accuracy in the knowledge graph construction process are avoided, and efficient and accurate construction of the knowledge graph is achieved.

[0359] The above is a schematic scheme of a knowledge graph construction device of this embodiment. It should be noted that the technical scheme of the knowledge graph construction device and the technical scheme of the above-mentioned knowledge graph construction method belong to the same concept. For details not described in detail in the technical scheme of the knowledge graph construction device, please refer to the description of the technical scheme of the above-mentioned knowledge graph construction method.

[0360] Figure 5 The structure block diagram of a computing device 500 provided according to an embodiment of the present specification is shown. The components of the computing device 500 include but are not limited to a memory 510 and a processor 520. The processor 520 is connected to the memory 510 via a bus 530, and the database 550 is used to store data.

[0361] The computing device 500 also includes an access device 540 that enables the computing device 500 to communicate via one or more networks 560. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 540 may include one or more of any type of network interface, wired or wireless (e.g., a network interface card (NIC)), such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a World Wide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, and a Near Field Communication (NFC).

[0362] In one embodiment of the present specification, the above components of the computing device 500 and Figure 5 Other components not shown in the figure may also be connected to each other, for example, via a bus. It should be understood that Figure 5 The computing device structure block diagram shown is only for the purpose of illustration, and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0363] The computing device 500 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 500 may also be a mobile or stationary server.

[0364] Among them, the processor 520 is used to execute the following computer program / instructions, which, when executed by the processor, implement the steps of the above-mentioned knowledge graph construction method.

[0365] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the computing device embodiment, since it is basically similar to the knowledge graph construction method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the knowledge graph construction method embodiment.

[0366] An embodiment of the present specification also provides a computer-readable storage medium storing a computer program / instruction, which implements the steps of the above-mentioned knowledge graph construction method when executed by a processor.

[0367] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the computer-readable storage medium embodiment, since it is basically similar to the knowledge graph construction method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the knowledge graph construction method embodiment.

[0368] An embodiment of the present specification also provides a computer program product, including a computer program / instruction, which implements the steps of the above-mentioned knowledge graph construction method when executed by a processor.

[0369] The above is a schematic scheme of a computer program product of this embodiment. It should be noted that the technical scheme of the computer program product and the technical scheme of the above-mentioned knowledge graph construction method belong to the same concept, and the details not described in detail in the technical scheme of the computer program product can be found in the description of the technical scheme of the above-mentioned knowledge graph construction method.

[0370] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0371] The computer instructions include computer program codes, which may be in source code form, object code form, executable files or some intermediate forms, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0372] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0373] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0374] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to only the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that technicians in the relevant technical field can well understand and use this specification. This specification is only limited by the claims and their full scope and equivalents.

Claims

1. A knowledge graph construction method, characterized in that: include: Performing knowledge point analysis on the target document using the knowledge point analysis model to obtain a plurality of initial document knowledge points in the target document, and performing relationship identification on the plurality of initial document knowledge points to obtain an initial association relationship between the initial document knowledge points; Merging the multiple initial document knowledge points to obtain multiple target document knowledge points, and using a relationship analysis model to perform relationship analysis on the multiple target document knowledge points and the initial association relationships to obtain target association relationships between the target document knowledge points; A target knowledge graph is constructed based on the multiple target document knowledge points and the target association relationships between the target document knowledge points.

2. The knowledge graph construction method according to claim 1, characterized in that: The method of using the knowledge point analysis model to perform knowledge point analysis on the target document to obtain a plurality of initial document knowledge points in the target document, and performing relationship identification on the plurality of initial document knowledge points to obtain initial association relationships between the initial document knowledge points includes: Inputting the target document into a language model for vectorization processing to obtain a plurality of word vectors corresponding to the target document, wherein the word vectors contain semantic information of document words in the target document; Determine a plurality of document chapters contained in the target document, and determine a chapter word vector corresponding to the target document chapter from the plurality of word vectors, wherein the target document chapter is any one of the plurality of document chapters; Inputting the chapter word vector corresponding to the target document chapter into the knowledge point analysis model to perform knowledge point recognition, and obtaining a plurality of initial document knowledge points corresponding to the target document chapter; Relationship identification is performed on the multiple initial document knowledge points corresponding to the target document chapters to obtain initial association relationships between the initial document knowledge points corresponding to the target document chapters.

3. The knowledge graph construction method according to claim 1, characterized in that: The multiple initial document knowledge points are multiple initial document knowledge points corresponding to a target document chapter, and the target document chapter is any one of the multiple document chapters of the target document; The step of merging the plurality of initial document knowledge points to obtain a plurality of target document knowledge points, and performing relationship analysis on the plurality of target document knowledge points and the initial association relationships using a relationship analysis model to obtain target association relationships between the target document knowledge points includes: Determining at least two identical initial document knowledge points and other initial document knowledge points other than the at least two identical initial document knowledge points from the multiple initial document knowledge points corresponding to the multiple document chapters; Merging the at least two identical initial document knowledge points to obtain a merged knowledge point, and determining the other initial document knowledge points and the merged knowledge point as the multiple target document knowledge points; Inputting the plurality of target document knowledge points and the initial association relationships into the relationship analysis model; In the relationship analysis model, based on the document context information of the target document and the initial association relationship, the target document knowledge points are subjected to relationship analysis to obtain the target association relationship between the target document knowledge points.

4. The method for constructing a knowledge graph according to any one of claims 1 to 3, characterized in that: After constructing the target knowledge graph based on the multiple target document knowledge points and the target association relationships between the target document knowledge points, the method further includes: Determining an update document corresponding to the target document, and performing knowledge point analysis on the update document using the knowledge point analysis model to obtain a plurality of newly added knowledge points in the update document; Using the multiple newly added knowledge points, updating the multiple target document knowledge points in the target knowledge graph to obtain multiple updated knowledge points; Using a relationship analysis model, performing relationship analysis on the multiple updated knowledge points and the target association relationship to obtain an updated association relationship between the updated knowledge points; Based on the multiple updated knowledge points and the updated association relationship between the updated knowledge points, the target knowledge graph is updated to obtain an updated target knowledge graph.

5. The method for constructing a knowledge graph according to any one of claims 1 to 3, characterized in that: After constructing the target knowledge graph based on the multiple target document knowledge points and the target association relationships between the target document knowledge points, the method further includes: Determining multimodal data corresponding to the target document, and performing knowledge point recognition on the multimodal data to obtain multimodal features corresponding to the multimodal data; Based on the multimodal features, determine the multimodal knowledge points corresponding to the multimodal data, and determine similar document knowledge points corresponding to the multimodal knowledge points from the multiple target document knowledge points, wherein the similar document knowledge points are target document knowledge points similar to the multimodal knowledge points among the multiple target document knowledge points; The multimodal data is determined as reference data for the similar document knowledge points.

6. The knowledge graph construction method according to claim 5, characterized in that: The performing knowledge point recognition on the multimodal data to obtain multimodal features corresponding to the multimodal data includes: The multimodal data is input into a multimodal data processing model, and at least two multimodal feature extraction units in the multimodal data processing model are used to perform feature extraction on the multimodal data to obtain the multimodal features.

7. The knowledge graph construction method according to claim 6, characterized in that: The step of determining, based on the multimodal features, a multimodal knowledge point corresponding to the multimodal data, and determining, from the plurality of target document knowledge points, a similar document knowledge point corresponding to the multimodal knowledge point, comprises: Using the knowledge point recognition unit in the multimodal data processing model, the knowledge point recognition is performed according to the multimodal features to obtain the multimodal knowledge points corresponding to the multimodal data; Utilizing the knowledge point feature extraction unit in the multimodal data processing model, feature extraction is performed on the plurality of target document knowledge points to obtain a plurality of knowledge point features; Using the feature alignment unit in the multimodal data processing model, the plurality of knowledge point features and the multimodal features are feature aligned to obtain a plurality of aligned knowledge point features and aligned multimodal features; Utilizing the feature fusion unit in the multimodal data processing model, the similarity between each aligned knowledge point feature and the aligned multimodal feature is calculated; Based on the similarity, a similar knowledge point feature similar to the aligned multimodal feature is determined from the multiple aligned knowledge point features, and the target document knowledge point corresponding to the similar knowledge point feature is determined as the similar document knowledge point corresponding to the multimodal knowledge point.

8. The method for constructing a knowledge graph according to any one of claims 1 to 3, characterized in that: After constructing the target knowledge graph based on the multiple target document knowledge points and the target association relationships between the target document knowledge points, the method further includes: Determine the knowledge learning record data corresponding to the target knowledge graph, wherein the knowledge learning record data is the learning record data generated by the user during the knowledge learning process based on the target knowledge graph; Analyze the knowledge learning record data using a learning status analysis module to obtain knowledge point learning status information; Performing a learning status evaluation based on the knowledge point learning status information and the target knowledge graph to obtain a learning status evaluation result; Based on the learning status evaluation result, the knowledge points to be learned by the user are determined from the target knowledge graph.

9. The knowledge graph construction method according to any one of claims 1 to 3, characterized in that: After constructing the target knowledge graph based on the multiple target document knowledge points and the target association relationships between the target document knowledge points, the method further includes: Sending the target knowledge graph to a client, so that the client displays the target knowledge graph to a user through an interactive interface; receiving a graph relationship adjustment instruction sent by the client, wherein the graph relationship adjustment instruction is sent by the client when the user performs an association relationship adjustment operation based on the interactive interface; Determining the association relationship identifier and the association relationship adjustment information from the graph relationship adjustment instruction; Based on the association relationship identifier, determining the association relationship to be adjusted from the target association relationship in the target knowledge graph; Based on the association relationship adjustment information, the association relationship to be adjusted in the target knowledge graph is adjusted to obtain an adjusted target knowledge graph.

10. The knowledge graph construction method according to any one of claims 1 to 3, characterized in that: The target document is a target textbook, the initial document knowledge points are initial textbook knowledge points, and the target document knowledge points are target textbook knowledge points; The constructing a target knowledge graph based on the multiple target document knowledge points and the target association relationships between the target document knowledge points includes: Based on multiple target textbook knowledge points and the target association relationships between the target textbook knowledge points, the target knowledge graph corresponding to the target textbook is constructed.

11. A knowledge graph construction device, characterized in that: include: A first knowledge point processing module is configured to perform knowledge point analysis on a target document using a knowledge point analysis model to obtain a plurality of initial document knowledge points in the target document, and perform relationship recognition on the plurality of initial document knowledge points to obtain an initial association relationship between the initial document knowledge points; A second knowledge point processing module is configured to merge the multiple initial document knowledge points to obtain multiple target document knowledge points, and use a relationship analysis model to perform relationship analysis on the multiple target document knowledge points and the initial association relationships to obtain target association relationships between the target document knowledge points; The graph construction module is configured to construct a target knowledge graph based on the multiple target document knowledge points and the target association relationships between the target document knowledge points.

12. A computing device, characterized in that: include: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, the steps of the method described in any one of claims 1 to 10 are implemented.

13. A computer-readable storage medium, characterized in that: It stores a computer program / instruction, which implements the steps of the method described in any one of claims 1 to 10 when executed by a processor.

14. A computer program product, characterized in that The invention comprises a computer program / instruction, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 10.

Images