A knowledge system construction method based on multi-source teaching materials

Abstract

The application discloses a kind of knowledge system construction methods based on multi-source teaching material, belong to knowledge system construction field.The application includes: extracting the course list of target major, and through expert cooperation, the course list is enhanced and checked, and forms authoritative course label library;For different source teaching materials of target major, after preprocessing, automatically locate and structure analysis the directory content of target teaching material, reconstruct chapter nesting relationship, and map target teaching material to the only course label in the authoritative course label library;For different teaching materials under the same course label, bottom-up semantic clustering and structure backtracking fusion are carried out, and the structured knowledge system is output based on the teaching order of the course label.The application realizes cross-teaching material semantic alignment and structure fusion, and finally generates unified, authoritative, hierarchical clear and semantically consistent knowledge system.

Description

Technical Field

[0001] This invention relates to the field of knowledge system construction, and specifically to a method for constructing a knowledge system based on textbooks from multiple sources. Background Technology

[0002] With the accelerated advancement of digital transformation in education, building a structured, standardized, and computable curriculum knowledge system has become a core infrastructure for applications such as intelligent education, personalized learning, and AI-assisted teaching systems. In recent years, artificial intelligence technologies such as knowledge graphs and Large Language Models (LLM) have been widely explored for knowledge organization. However, current mainstream methods mostly focus on knowledge extraction from single textbooks or general semantic modeling, lacking deep alignment with educational standards and struggling to address the heterogeneity of multiple textbook sources in terms of expression, structure, and version. Especially in higher education, the complex curriculum, frequent textbook updates, and significant differences in teaching logic have led to the long-standing dilemma of "no standards, difficulty in alignment, and low reusability" in the construction of automated knowledge systems. Therefore, there is an urgent need for a technical solution that can integrate authoritative educational standards, expert knowledge, and content from multiple textbook sources to achieve the automatic generation of a high-fidelity, highly consistent knowledge system.

[0003] Existing knowledge system construction or course knowledge graph generation technologies mainly suffer from the following three shortcomings:

[0004] (1) The lack of a unified and authoritative semantic benchmark for courses leads to a chaotic and unusable tagging system. Currently, most educational knowledge graphs rely on manual definition or direct extraction of course or knowledge point names from textbooks, and are not linked to legal teaching norms. The naming, boundaries, and connotations of courses vary across different projects, platforms, and even different versions of the same platform, resulting in serious semantic drift. In addition, course tags are mostly labeled by non-domain personnel, lacking comprehensive verification and cutting-edge supplementation, making them difficult to serve as reliable anchors for subsequent automated processing.

[0005] (2) Textbook catalog parsing relies heavily on manual processing or rule-based templates, making it difficult to adapt to the format noise and structural heterogeneity of real publications. Existing technologies mostly use regular expressions, fixed templates, or simple OCR post-processing to extract textbook catalogs. When faced with common problems such as PDF scans, misformatted layouts, mixed multilingual text, embedded page numbers, and missing chapter numbers, the accuracy drops sharply. More importantly, traditional methods cannot recover the chapter nesting relationships lost due to format conversion, leading to the break in the subsequent knowledge unit hierarchy and making it impossible to support the construction of a structured knowledge tree.

[0006] (3) The multi-textbook knowledge fusion method breaks the semantic consistency and the integrity of the teaching logic, resulting in low-quality fusion results. Current knowledge fusion methods mostly adopt pure semantic clustering, which can merge knowledge points with similar expressions, but completely ignores the parent-child hierarchy in the original textbooks, resulting in "logical jumps" or "structural collapse" in the fused knowledge tree. Conversely, if the structure of a single textbook is forcibly retained, it is impossible to eliminate cross-textbook redundancy and ambiguity, resulting in poor generalization ability. Existing technologies have not yet effectively balanced the core contradiction between "semantic deduplication" and "structural fidelity". Summary of the Invention

[0007] To address the shortcomings of existing technologies, the present invention aims to provide a method for constructing a knowledge system based on multiple sources of teaching materials, thereby solving the problem that existing methods cannot fully cover the three-level collaboration of "teaching standards - teaching materials - knowledge".

[0008] The objective of this invention is achieved through the following technical solution: a method for constructing a knowledge system based on multi-source textbooks, comprising the following steps:

[0009] Extract the course list for the target major, and enhance and verify the course list through expert collaboration to form an authoritative course tag library;

[0010] For textbooks from different sources for the target major, the table of contents of the target textbooks are automatically located and structured after preprocessing, the chapter nesting relationship is reconstructed, and the target textbooks are mapped to unique course tags in the authoritative course tag library;

[0011] For different teaching materials under the same course tag, bottom-up semantic clustering and structural backtracking are integrated to output a structured knowledge system based on the teaching sequence of the course tag.

[0012] Furthermore, the course list for the target major includes required courses, core courses, and recommended courses for the target major.

[0013] Furthermore, the enhancement and verification include:

[0014] Completeness verification: Based on the actual teaching situation of the target major, the completeness of the courses in the course list is checked. If there are any missing courses, the missing courses are added to the course list.

[0015] Cutting-edge updates: Based on the development trends of the target major, determine whether to add emerging courses;

[0016] The teaching logic is sequenced by organizing and optimizing the courses of the target major in a hierarchical manner based on the knowledge progression logic, according to the expert's understanding of the target major.

[0017] Furthermore, the preprocessing includes: for textbooks of different formats, a multimodal document processing workflow is adopted to uniformly convert the content of textbooks from different sources into plain text representation, and an intelligent document processing framework is used to finely divide the full text into blocks according to the title level, retaining the title and content information of the blocks;

[0018] If the teaching material is in image format or a scanned PDF, an OCR engine is used to recognize the text and obtain the text content; if the teaching material is a PDF with optional Chinese text, a structured parsing tool is used to extract the original text and its layout information to obtain the text content.

[0019] Furthermore, the automatic location and structured parsing of the table of contents of textbooks from different sources, and the reconstruction of chapter nesting relationships, include:

[0020] By using multilingual keyword matching technology, potential directory start positions are located in the preprocessed block sequence;

[0021] Based on the starting position, the information of subsequent blocks is input into the large language model one by one. Based on the multi-objective collaborative judgment method, it is determined whether it belongs to the directory content, and the directory is extracted to obtain the original directory. If a non-directory judgment result occurs for a preset number of consecutive times, the directory extraction is stopped.

[0022] The original directory is reconstructed hierarchically to build a complete directory structure;

[0023] Remove non-knowledge-related noise from the reconstructed hierarchical directory, retain only the knowledge unit titles, and translate these titles into Chinese to obtain a structured directory tree.

[0024] Furthermore, the multi-target collaborative judgment method is as follows: the large language model judges the block information by judging targets including text length, format regularity, content semantics and numbering continuity. If the number of correct judgment targets is greater than a preset value, the block is judged to belong to the directory content.

[0025] Furthermore, the unique course tag that maps the target teaching material to the authoritative course tag library includes:

[0026] Based on a large language model, the target textbook is subjected to deep semantic understanding. The target textbook is mapped to the single course tag with the highest matching degree in the authoritative course tag library, and a matching confidence score is attached. If the matching confidence score is lower than the preset value, it is confirmed by experts for a second time.

[0027] Furthermore, the bottom-up semantic clustering and structural backtracking fusion for different teaching materials under the same course tag includes:

[0028] Before integrating semantic clustering and structural backtracking, a multi-dimensional evaluation of different textbooks under the same course tag is conducted based on a large language model. The dimensions include knowledge point coverage, content depth, and authority.

[0029] Based on the multi-dimensional evaluation, a comprehensive quality score is generated for each textbook. Textbooks with the highest comprehensive quality scores are grouped into a set of high-quality textbooks, which serve as input for subsequent semantic clustering and structural backtracking fusion.

[0030] Furthermore, the bottom-up semantic clustering and structural backtracking fusion includes:

[0031] Extract the leaf node titles without sub-chapter titles from the textbook catalog structure of the aforementioned collection of high-quality teaching materials to form an initial set of knowledge units;

[0032] The leaf node titles in the initial knowledge unit set are converted into semantic vectors, clustered based on semantic similarity to form an initial leaf title cluster set, and a unified knowledge point name is generated for each cluster to form a knowledge point set.

[0033] The parent node title of each title in the original textbook table of contents is traced back to the cluster, merged with the single-level knowledge points to form a new title set, and semantic clustering and standardization operations are performed on it;

[0034] When it is impossible to trace back to a higher-level parent node, stop iterating and output the complete knowledge system.

[0035] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0036] The beneficial effects of this invention are as follows:

[0037] 1. This invention establishes an end-to-end, scalable, and reusable knowledge system construction pipeline, significantly improving the automation level and scalability of educational knowledge engineering. Existing technologies are mostly isolated modules, requiring significant manual intervention and making them difficult to extend to different majors or courses. This invention, through a three-level closed-loop architecture of "standard-driven—intelligent parsing—structural fusion," achieves full-process automation from educational standards to multi-source textbooks to a unified knowledge tree, possessing good versatility and industrial application value.

[0038] 2. Constructing an authoritative course tag library based on educational standards and expert collaboration. Existing methods often directly extract course names from textbooks as tags, leading to naming chaos, blurred boundaries, inconsistent versions, and severe semantic drift. This invention is the first to systematically transform legally mandated teaching standards into a structured course tag library and introduces an expert collaboration mechanism for verification and cutting-edge supplementation, fundamentally solving the problem of unreliable tag sources and providing a highly reliable and unified semantic anchor for subsequent automated processing.

[0039] 3. Breaking away from the strong dependence of traditional rule templates on textbook formats, this invention achieves highly robust intelligent parsing of tables of contents in unstructured publications. Existing technologies mostly use regular expressions or fixed templates to extract tables of contents, which suffer from a sharp drop in accuracy when faced with real-world scenarios such as scanned PDFs, misformatted layouts, mixed multilingual text, and embedded page numbers, and cannot recover chapter nesting relationships. This invention proposes a three-stage mechanism of "keyword triggering + semantic judgment of a large language model + dynamic termination conditions," which can not only accurately locate the table of contents area, but also use a large language model to reconstruct lost hierarchical structures, clean up redundant information, and unify multilingual expressions, significantly improving the fidelity of conversion from original textbooks to structured outlines.

[0040] 4. An innovative dual-constraint fusion mechanism of "bottom-up semantic clustering + original structure backtracking" effectively bridges the gap between semantic differences and structural heterogeneity among multiple textbooks. Existing knowledge fusion methods either perform only pure semantic clustering, resulting in the neglect of original structural information from the textbooks; or they are limited to building a knowledge system within a single textbook, preserving the integrity of local structures but failing to achieve cross-textbook knowledge alignment and generalization. This invention achieves cross-textbook terminology unification through semantic embedding clustering at the leaf knowledge point level. During the upper-level fusion process, it actively backtracks and introduces parent-child dependencies from the original catalogs of each textbook as structural priors, forcibly constraining the clustering boundaries. Thus, while fusion from multiple sources, it fully preserves the reasonable teaching logic structure, truly achieving the construction of a high-quality knowledge system that is both deduplicated and faithful to the original. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a schematic diagram of the process of the present invention;

[0043] Figure 2 This is a flowchart of the intelligent parsing and course annotation process for the textbook catalog of this invention;

[0044] Figure 3 This invention provides a flowchart for constructing a knowledge system by integrating multiple teaching materials and multi-level knowledge. Detailed Implementation

[0045] To enable those skilled in the art to better understand the technical solutions in the embodiments of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art should fall within the protection scope of the embodiments of this application.

[0046] like Figure 1 As shown, this invention provides a method for constructing a knowledge system based on multi-source textbooks. It extracts hierarchical knowledge units with high precision from unstructured, heterogeneous textbook publications, integrates educational standards and expert knowledge, achieves cross-textbook semantic alignment and structural fusion, and ultimately generates a unified, authoritative, clearly hierarchical, and semantically consistent knowledge system. The specific steps include:

[0047] S1. Construction of Professional Curriculum System: Extract the course list of the target major, and enhance and verify the course list through expert collaboration to form an authoritative course tag library; wherein, the course list of the target major includes required courses, core courses and recommended courses of the target major.

[0048] S2. Intelligent parsing and course labeling of textbook catalogs: For textbooks from different sources for the target major, the catalog content of the target textbooks is automatically located and structured after preprocessing, the chapter nesting relationship is reconstructed, and the target textbooks are mapped to unique course tags in the authoritative course tag library.

[0049] S3. Integration and Construction of Knowledge System from Multiple Teaching Materials: For different teaching materials under the same course tag, perform bottom-up semantic clustering and structural backtracking integration, and output a structured knowledge system based on the teaching sequence of the course tag.

[0050] As a preferred embodiment, the professional curriculum system construction mentioned in step S1 aims to establish a legally authoritative and cutting-edge adaptable curriculum semantic benchmark, serving as the sole anchor point for subsequent textbook annotation and knowledge integration. Specifically, it includes the following sub-steps:

[0051] S1.1. Extraction of Standard Courses

[0052] Collect normative documents related to teaching quality standards, and extract a list of required courses, core courses and recommended courses explicitly stipulated for the target major. This results in a course list, and a structured "major → course set" mapping table is constructed, where each course is stored in a standardized manner using its official name.

[0053] S1.2. Expert knowledge supplementation and verification

[0054] Based on the initial course list, a review panel composed of front-line university teachers in the target professional field, course leaders, members of the teaching guidance committee, or senior experts from relevant industries was invited to systematically enhance and authoritatively verify the curriculum system. Optimization was carried out in three specific aspects:

[0055] Completeness verification identifies and supplements course content that is not explicitly listed in educational standards but is essential for actual teaching or industry practice.

[0056] The curriculum is updated to reflect cutting-edge development trends and the dynamics of disciplinary evolution, and supplemented with emerging courses that reflect the latest research and industry directions, ensuring that the curriculum system is adaptable to the times.

[0057] Based on cognitive patterns and teaching arrangements, the course is organized hierarchically and optimized according to the progressive knowledge logic of "basics → core → application → cutting-edge".

[0058] After completing systematic enhancement and authoritative verification, an authoritative course tag library that combines legal authority, pedagogical rationality, and technological foresight is output, serving as the unique, stable, and highly reliable semantic benchmark for all subsequent automatic textbook annotation and knowledge integration processes.

[0059] For example, this invention provides a case study of the steps for constructing a professional curriculum system. For the target major: Physics, courses are extracted from educational standards such as "Mechanics," "Thermodynamics," "Electromagnetism," "Optics," "Atomic Physics," and "Theoretical Mechanics," with expert-supplemented courses including "Introduction to Modern Physics," "Semiconductor Physics and Devices," and "Electromagnetic Fields and Waves." The final output is a course tag library including: "Mechanics," "Thermodynamics," "Electromagnetism," "Optics," "Atomic Physics," "Theoretical Mechanics," "Introduction to Modern Physics," "Semiconductor Physics and Devices," and "Electromagnetic Fields and Waves."

[0060] As a preferred embodiment, the intelligent parsing and course annotation of the textbook catalog mentioned in step S2 is used to achieve high-fidelity structured conversion and course annotation, such as... Figure 2 As shown, the specific steps include the following:

[0061] S2.1. Preprocessing of teaching materials from multiple sources

[0062] For the target major, we collected various teaching materials widely used in actual teaching, including mainstream textbooks, university-compiled lecture notes, experimental guides, and original foreign language textbooks, covering multiple file formats such as editable PDFs, scanned image PDFs, and EPUB e-books.

[0063] To address the issues of format heterogeneity and content noise, a multi-modal document processing flow is adopted: First, for image-based or scanned PDFs, a high-precision OCR engine (such as PaddleOCR, Tesseract) is called for text recognition; for selectable text-based PDFs, structured parsing tools (such as PyMuPDF, pdfplumber) are combined to extract the original text and its layout information. Subsequently, the textbook content from all sources is uniformly converted into a plain text representation, and using intelligent document processing frameworks such as LangChain, the full text is refined and segmented into blocks according to the title hierarchy, retaining the title and content information of the blocks, providing a high-quality input basis for subsequent directory location and structure reconstruction.

[0064] S2.2. Directory Location and Structured Extraction

[0065] A three-stage approach is adopted to accurately identify and extract the directory content:

[0066] Trigger stage: Quickly locate potential directory start positions in the preprocessed block sequence through multi-language keyword matching (such as Chinese "目录", English "Contents", "Table of Contents", etc.).

[0067] Expansion stage: Starting from the start position identified in the trigger stage, the subsequent block information is input into the large language model (LLM) one by one. The large language model comprehensively judges whether it belongs to the directory content for directory extraction to obtain the original directory. Among them, the judgment basis includes context clues such as text length (usually shorter), format regularity (such as containing numbers, indents or alignment features), content semantics (such as chapter title style), and number continuity.

[0068] Termination stage: Once three or more long text blocks determined by the large language model as "non-directory" are detected, which usually indicates the start of the main text, the extraction is automatically stopped, effectively preventing main text content from mistakenly entering the directory area and significantly improving the robustness of boundary recognition.

[0069] Preferably, for the extracted original directory text, the LLM is called to perform the following post-processing:

[0070] Hierarchical structure reconstruction: By analyzing the chapter numbering system (such as "1.2.3"), visual indents, font levels, or implicit logical relationships, the parent-child nested structure lost due to PDF conversion, scanning distortion, or typesetting disorder is automatically restored to construct a complete directory structure.

[0071] Redundancy cleaning: Identify and remove non-knowledgeable noises such as number prefixes, page numbers, header and footer residues, appendix identifiers, reference identifiers, etc., and retain the pure knowledge unit titles. <000,0152>

[0072] Multilingual normalization: Translate and standardize English and mixed language titles into Chinese terms to obtain a structured directory tree.

[0073] S2.3. Automatic Course Marking

[0074] Using textbook metadata (including title, ISBN, author, publisher, etc.) and the structured directory tree output in step S2.2 as input, a large language model is used to perform deep semantic understanding of the overall knowledge content of the textbook. Specifically, the large language model comprehensively analyzes high-frequency professional terms, knowledge coverage, and teaching depth in the directory to determine the course tag with the highest fit for the textbook. The textbook is then precisely mapped to the single course tag that best matches the authoritative course tag library, along with a matching confidence score generated by the large language model. For annotation results with a confidence score below a preset threshold (e.g., 0.7), the system automatically adds them to a manual review queue for secondary confirmation by domain experts to ensure the accuracy and reliability of the annotations. This approach effectively balances automation efficiency with the professionalism and rigor of course attribution determination in educational scenarios.

[0075] Finally, a structured, hierarchical, and linguistically consistent directory tree and a single course tag are output for each textbook, serving as the basic input for subsequent integration of knowledge from multiple textbooks.

[0076] For example, this invention provides a case study of intelligent parsing and course annotation of textbook catalogs. For textbooks A and B in the physics major, the catalog is extracted after preprocessing. The catalog of textbook A includes:

[0077] 1. Force and Motion

[0078] 1.1 Particle Kinematics

[0079] 1.1.1 Description of Particle Motion

[0080] 1.1.2 Position vector and equation of motion of a particle

[0081] 1.1.3 Displacement and Velocity

[0082] 1.1.4 Acceleration

[0083] 1.1.5 Two Types of Kinematic Problems

[0084] 1.2 Particle Dynamics

[0085] 1.2.1 Newton's Laws of Motion

[0086] 1.2.2 Interaction Forces

[0087] 1.2.3 Application of Newton's Laws of Motion

[0088] 1.2.4 Non-inertial frames of reference / inertial forces

[0089] The table of contents for textbook B includes:

[0090] 1 Introduction to Mechanics

[0091] 1.1 The object of study in mechanics

[0092] 1.2 Mathematical Methods in Mechanics

[0093] 1.3 Fundamental Physical Quantities in Mechanics

[0094] 1.3.1 Basic Quantities and Derived Quantities / Units / Dimensions

[0095] 1.3.2 Scope of Application of Newtonian Mechanics

[0096] 2 Kinematics

[0097] 2.1 Vector description of particle motion.

[0098] 2.1.1 Displacement and Elementary Displacement

[0099] 2.1.2 Speed

[0100] 2.1.3 Acceleration

[0101] 2.2 Coordinate Description of Particle Motion (I)

[0102] 2.2.1 Rectangular coordinate system

[0103] 2.2.2 Rectangular coordinate description of particle motion

[0104] 2.3 Coordinate Description of Particle Motion (II)

[0105] 2.3.1 Planar Polar Coordinate System

[0106] 2.3.2 Circular Motion, Angular Velocity and Angular Acceleration

[0107] 2.3.3 Natural Coordinate System

[0108] 2.4 Relativity in the Description of Motion

[0109] 2.4.1 Absolute Spacetime View

[0110] 2.4.2 Relativity of Motion Description

[0111] 2.4.3 Such as the Galilean transformation

[0112] Based on the case study of the steps in constructing a professional curriculum system, course tags are selected for textbook A and textbook B from the output course tag library. The course tag for both is "Mechanics".

[0113] As a preferred embodiment, the multi-textbook knowledge system fusion construction mentioned in step S3 specifically generates a high-quality, logically consistent fusion knowledge tree through a dual constraint mechanism of "semantic clustering + structural backtracking," such as... Figure 3 As shown, it includes the following sub-steps:

[0114] S3.1 Selection of High-Quality Teaching Materials

[0115] Based on the completion of textbook course annotation, for a specific target course, all textbooks automatically or manually annotated as such are first aggregated to form an initial candidate set. To ensure the quality and representativeness of subsequent knowledge integration, this step introduces a large language model to conduct a multi-dimensional, fine-grained evaluation of the instructional suitability of each candidate textbook. The evaluation mainly unfolds from the following three core dimensions:

[0116] Knowledge coverage: Assess whether the knowledge units reflected in the textbook catalog fully cover the core content stipulated in the educational standards or authoritative curriculum syllabus.

[0117] The depth of content is examined to see if the textbook provides a systematic explanation of principles and mechanisms, rather than just focusing on the use of tools or application examples, in order to distinguish introductory reading materials from advanced textbooks.

[0118] Authority is determined by comprehensively considering external factors such as the author's academic background, the publisher's reputation, the textbook's widespread use, and the frequency of academic citations.

[0119] Based on the aforementioned dimensions, different weights were assigned to each dimension. A comprehensive quality score for each textbook was generated using a large language model, and the scores were sorted in descending order. Finally, the top-N (e.g., N=20) textbooks were selected as high-quality knowledge sources for constructing the course's knowledge system. This evaluation and screening effectively filtered out textbooks with incomplete content, insufficient depth, or questionable authority, ensuring that the integration results are based on high-quality, representative teaching resources, significantly improving the professionalism, completeness, and pedagogical applicability of the final knowledge system.

[0120] S3.2. Multi-level knowledge integration

[0121] This step adopts a bottom-up, iterative fusion strategy, combining semantic similarity and the original teaching structure as dual constraints, to achieve high-quality alignment and systematic integration of knowledge from multiple sources of teaching materials.

[0122] For a specific target course, firstly, extract all leaf nodes—that is, the smallest knowledge units without child chapters—from the structured table of contents of all high-quality textbooks to form an initial set of leaf titles. .

[0123] Subsequently, advanced semantic embedding models (such as Sentence-BERT) are used to transform each leaf title into a high-dimensional vector representation, and clustering is performed based on the semantic similarity between vectors (such as HDBSCAN) to form an initial set of leaf title clusters. .

[0124] For each cluster, the large language model is invoked to synthesize all its internal original expressions, generating a standardized, concise, unambiguous, and pedagogically compliant unified knowledge point name, thus forming a knowledge point set. This eliminates differences in terminology across different textbooks.

[0125] Next, traverse For each cluster, trace back the direct parent node title of each heading within that cluster in the original textbook's table of contents to form a heading set. At the same time, if there exists All headings within its cluster are single-level (without explicit parent nodes) in the original textbook directories. Join middle.

[0126] The new title set obtained by merging the parent node title with the single-level knowledge points Semantic embedding and clustering are performed again to generate a higher-level set of title / knowledge point clusters. The large language model generates standardized names for each new cluster, forming a set of knowledge points. .

[0127] This iterative process continues until all current clusters can no longer trace back to their parent nodes at higher levels, at which point the upward tracing stops, completing the entire knowledge fusion process. Through this mechanism, the resulting knowledge system achieves both semantic deduplication and terminology unification across different textbooks, while also fully preserving the teaching logic structure upon which educational scenarios rely, effectively resolving the core contradictions of "semantic fragmentation" and "structural collapse" in existing methods.

[0128] S3.3. Integration of Knowledge Tree Assembly

[0129] After completing the multi-level knowledge integration, based on the clustering relationships between knowledge clusters at each level and the parent-child dependency structure retained from the original textbooks, a logically rigorous and hierarchically clear multi-textbook integrated course knowledge tree is constructed from the bottom up. Each node in the tree contains meta-information such as: standardized name, list of source textbooks, original expression variants, and teaching level. Finally, the knowledge trees of each course are systematically integrated according to the teaching order defined in the authoritative course tag library constructed in Module 1, outputting a three-level structured knowledge system of "major → course → knowledge system tree".

[0130] For example, this invention provides a case study of the integrated construction of knowledge systems from multiple textbooks. Based on the case study in the steps of constructing a professional curriculum system and intelligent parsing and marking textbook catalogs, the catalogs of textbook A and textbook B are integrated to obtain a knowledge tree as follows:

[0131] 1 Mechanics

[0132] 1.1 Fundamentals and Introduction to Mechanics

[0133] 1.1.1 The basic research objects and mathematical and physical foundations of mechanics

[0134] 1.2 Particle Kinematics

[0135] 1.2.1 Displacement, Velocity and Acceleration

[0136] 1.2.2 Multiple Coordinate Description Methods for Particle Motion

[0137] 1.2.3 Two fundamental problems in kinematics

[0138] 1.2.4 Galilean Principle of Relativity and Galilean Transformation

[0139] 1.3 Particle Dynamics

[0140] 1.3.1 Newton's Laws of Motion and Their Scope of Application

[0141] 1.3.2 Application of Newton's Laws in Interaction Forces

[0142] 1.3.3 Non-inertial reference frames and inertial forces

[0143] This invention constructs a three-tiered collaborative framework of "teaching standards-textbooks-knowledge," using educational standards as the top-level design basis, an expert-enhanced course tag library as the mid-level semantic anchor, and intelligent parsing and knowledge fusion of multi-source textbook catalogs as the bottom-level data driver. The three work together organically to form a scalable, reusable, and highly standardized knowledge system construction pipeline, achieving system-level integrated innovation and significantly improving the standardization and efficiency of knowledge system construction.

[0144] This invention transforms the core curriculum requirements in education standards into a structured curriculum tagging system, and introduces domain experts to verify and supplement the courses, forming a dynamic tagging library that combines legal authority with cutting-edge adaptability.

[0145] This invention uses a three-stage mechanism of "keyword triggering + semantic judgment of large language model + dynamic termination condition" to accurately locate the table of contents area, and uses a large language model to restore the chapter nesting relationship lost due to PDF conversion or scanning. At the same time, it cleans up redundant information and unifies multilingual expressions, realizing high-fidelity automated conversion from original publications to machine-readable knowledge outlines.

[0146] This invention designs a dual-constraint mechanism of "bottom-up semantic clustering + original structure backtracking" for the fusion of knowledge from multiple teaching materials. At the leaf knowledge point level, clustering and deduplication are performed based on semantic embedding and standardized names are generated by a large language model. In the upper-level fusion process, the parent-child hierarchical relationship in the original catalog of each teaching material is forcibly introduced as a structural prior, ensuring that the fused knowledge tree eliminates ambiguity in expression while retaining reasonable teaching logic.

[0147] This invention also provides a computer-readable storage medium storing a program thereon, which, when executed by a processor, implements the knowledge system construction method based on multi-source textbooks described in the above embodiments.

[0148] The computer-readable storage medium can be an internal storage unit of any data processing device described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be any data processing device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0149] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process may be rearranged without departing from the scope of this disclosure. The appended method claims provide elements of various steps in an exemplary order and are not intended to limit the scope to the specific order or hierarchy described.

Claims

1. A method for constructing a knowledge system based on textbooks from multiple sources, characterized in that, Includes the following steps: Extract the course list for the target major, and enhance and verify the course list through expert collaboration to form an authoritative course tag library; For textbooks from different sources for the target major, the table of contents of the target textbooks are automatically located and structured after preprocessing, the chapter nesting relationship is reconstructed, and the target textbooks are mapped to unique course tags in the authoritative course tag library; For different teaching materials under the same course tag, bottom-up semantic clustering and structural backtracking are fused to output a structured knowledge system based on the teaching sequence of the course tag, including: Before integrating semantic clustering and structural backtracking, a multi-dimensional evaluation of different textbooks under the same course tag is conducted based on a large language model. The dimensions include knowledge point coverage, content depth, and authority. Based on the multi-dimensional evaluation, a comprehensive quality score is generated for each textbook. Textbooks with the highest comprehensive quality scores are grouped into a set of high-quality textbooks, which serve as input for subsequent semantic clustering and structural backtracking fusion. Extract the leaf node titles without sub-chapter titles from the textbook catalog structure of the aforementioned collection of high-quality teaching materials to form an initial set of knowledge units; The leaf node titles in the initial knowledge unit set are converted into semantic vectors, clustered based on semantic similarity to form an initial leaf title cluster set, and a unified knowledge point name is generated for each cluster to form a knowledge point set. The parent node title of each title in the original textbook table of contents is traced back to the cluster, merged with the single-level knowledge points to form a new title set, and semantic clustering and standardization operations are performed on it; When it is impossible to trace back to a higher-level parent node, stop iterating and output the complete knowledge system.

2. The method according to claim 1, characterized in that, The course list for the target major includes required courses, core courses, and recommended courses.

3. The method according to claim 1, characterized in that, The enhancements and verifications include: Completeness verification: Based on the actual teaching situation of the target major, the completeness of the courses in the course list is checked. If there are any missing courses, the missing courses are added to the course list. Cutting-edge updates: Based on the development trends of the target major, determine whether to add emerging courses; The teaching logic is sequenced by organizing and optimizing the courses of the target major in a hierarchical manner based on the knowledge progression logic, according to the expert's understanding of the target major.

4. The method according to claim 1, characterized in that, The preprocessing includes: for textbooks of different formats, a multimodal document processing workflow is adopted to convert the content of textbooks from different sources into plain text representation in a unified manner, and an intelligent document processing framework is used to finely divide the full text into blocks according to the title level, while retaining the title and content information of the blocks; If the teaching material is in image format or a scanned PDF, an OCR engine is used to recognize the text and obtain the text content; if the teaching material is a PDF with optional Chinese text, a structured parsing tool is used to extract the original text and its layout information to obtain the text content.

5. The method according to claim 4, characterized in that, The automatic location and structured parsing of the table of contents of textbooks from different sources, and the reconstruction of the chapter nesting relationships, include: By using multilingual keyword matching technology, potential directory start positions are located in the preprocessed block sequence; Based on the starting position, the information of subsequent blocks is input into the large language model one by one. Based on the multi-objective collaborative judgment method, it is determined whether it belongs to the directory content, and the directory is extracted to obtain the original directory. If a non-directory judgment result occurs for a preset number of consecutive times, the directory extraction is stopped. The original directory is reconstructed hierarchically to build a complete directory structure; Remove non-knowledge-related noise from the reconstructed hierarchical directory, retain only the knowledge unit titles, and translate these titles into Chinese to obtain a structured directory tree.

6. The method according to claim 5, characterized in that, The multi-target collaborative judgment method is as follows: the large language model judges the block information by judging the judgment targets including text length, format regularity, content semantics and numbering continuity. If the number of correct judgment targets is greater than a preset value, the block is judged to belong to the directory content.

7. The method according to claim 5, characterized in that, The unique course tag that maps the target teaching material to the authoritative course tag library includes: Based on a large language model, the target textbook is subjected to deep semantic understanding. The target textbook is mapped to the single course tag with the highest matching degree in the authoritative course tag library, and a matching confidence score is attached. If the matching confidence score is lower than the preset value, it is confirmed by experts for a second time.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-7.

Images