A knowledge graph driven textbook automatic generation method and system

Abstract

The present application relates to the technical field of automatic generation, in particular to a knowledge graph driven textbook automatic generation method and system. The present application receives teaching target input, analyzes it into a multi-layer structure based on a preset teaching intention meta-model and maps it to a knowledge graph node; generates an optimal teaching path through a rule engine combined with node topology position and teaching intention hierarchy; after identifying the target textbook system topology fingerprint, performs compatibility analysis, and if there is a structural gap, automatically inserts a transition module to reconstruct the path; according to the reconstructed path node, calls the bound multi-modal content unit, and completes modal coordination according to the cross-modal dependency graph; finally, the integrated content is structured and rendered, and the textbook ontology adapted to the terminal is output. The present application has the advantages of deep understanding of teaching intention, multi-modal content consistency control and cross-textbook system adaptation, and improves the intelligent level and generalization ability of textbook generation.

Description

Technical Field

[0001] This invention relates to the field of automated textbook generation, specifically to a knowledge graph-driven method and system for automated textbook generation. Background Technology

[0002] With the rapid development of educational informatization, the digitization and personalized generation of textbook content have become important directions for intelligent teaching. How to automatically generate well-structured, content-coordinated, and multi-terminal-compatible textbook resources based on teaching objectives has become a key aspect of the intelligent upgrading of teaching systems.

[0003] However, existing methods for automatically generating textbooks generally suffer from three technical defects: First, the analysis of teaching objectives is superficial, often relying on keyword matching and lacking support for modeling and execution paths of higher-order cognitive intentions; second, the content organization structure is rigid, and there is a lack of semantic coordination mechanisms between text and image modalities, which easily leads to cognitive gaps or redundant expressions; third, there is a lack of adaptability across textbook systems, making it difficult to cope with the differences in knowledge topology and cognitive paths between different versions of textbooks, resulting in high costs for content migration and reconstruction and weak generalization ability, which limits the practical application scope of textbook automation systems.

[0004] Therefore, a knowledge graph-driven method and system for automated textbook generation are proposed. Summary of the Invention

[0005] This invention provides a knowledge graph-driven method and system for automated textbook generation. By analyzing teaching objectives through a teaching intent meta-model and combining cross-modal dependency graphs with textbook system adaptation technology, it automatically generates teaching resources with reasonable structure and coordinated content. This solves the technical problems of path planning, modal coordination, and cross-textbook system adaptation in traditional textbook generation, and improves the intelligence and adaptability of textbook generation.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention proposes a knowledge graph-driven method for automated textbook generation, comprising the following steps:

[0008] Upon receiving the input of teaching objectives, the teaching objectives are parsed into a multi-layered structure including higher-order objectives, core competency tags, and observable behavioral indicators according to the pre-set teaching intention meta-model. The parsed teaching objectives are then mapped to the starting nodes of the knowledge graph to construct the starting point of the teaching path.

[0009] Based on the topological location of the starting node and the teaching intent level, the optimal teaching path is generated by calling a rule engine that supports dynamic reasoning and multi-objective optimization.

[0010] Generate the topological fingerprint of the target textbook system and perform a compatibility analysis with the knowledge graph structure. If there is a structural conflict between the target topological fingerprint and the source graph in terms of knowledge density, cognitive gradient, depth of prerequisite dependencies, or consistency of core paths, then dynamically insert a transition module to reconstruct the teaching path to meet the topological consistency requirements.

[0011] Based on the reconstructed path nodes, the bound multimodal content atomic units are called, and according to the multi-dimensional modal collaborative constraint rules defined by the cross-modal dependency graph, combined with semantic and visual consistency indicators, content binding, consistency detection and logical integration are completed.

[0012] The integrated teaching content is rendered in a structured manner, and the resulting teaching material is output to be adapted to the terminal.

[0013] As a preferred technical solution of the invention, the steps for analyzing the teaching objectives include:

[0014] Extract teaching objective expressions and their related action verbs, cognitive levels, and ability labels from curriculum standard texts, teaching evaluation frameworks, and typical teaching cases;

[0015] By setting expert rules and combining them with a dynamic adjustment mechanism based on teaching feedback, the teaching objective expressions, action verbs, cognitive levels and ability labels are organized into a three-layer structure model, including a set of higher-order objectives, core ability labels and observable behavioral indicators.

[0016] Establish semantic mapping rules for each layer of the structure to form a rule chain from target terms to capability tags and then to knowledge graph anchors;

[0017] The structural model and mapping rules are stored in the teaching intent rule engine in a structured format, and template expansion and dynamic updates are supported.

[0018] As a preferred embodiment of the invention, the step of mapping the teaching objective to the starting node of the knowledge graph includes:

[0019] Semantic analysis is performed on the input of teaching objectives to extract verb phrases, ability verbs, and object nouns;

[0020] The extracted verb phrases are mapped to a set of higher-order objectives in a pre-defined meta-model of teaching intent in order to identify the type of teaching objective they belong to.

[0021] Based on the target type, select the set of core capability tags associated with it, and further match the corresponding observable behavior indicator templates;

[0022] Based on behavioral indicators, the teaching intent rule engine is invoked to execute the semantic anchoring algorithm, dynamically mapping and optimizing the matching of node sets in the knowledge graph to form a candidate set of teaching path starting points.

[0023] As a preferred technical solution of the invention, the optimal path generation process includes:

[0024] Identify the position of the target knowledge node in the graph structure of the knowledge graph, including its predecessor node, successor node and adjacent branch node;

[0025] Based on the level of the teaching intention, the teaching strategy template that combines heuristic search and historical teaching data is invoked from the rule engine;

[0026] Perform a restricted path search operation in the graph to filter all node sequences that are topologically reachable from the target node and satisfy the policy constraints;

[0027] All candidate paths are input into a path scoring function for evaluation. The scoring metrics include path length, strategy suitability, and node knowledge weight.

[0028] The path with the highest score is selected as the optimal teaching path, and its nodes are sequentially bound to the corresponding teaching activity types for subsequent content scheduling.

[0029] As a preferred technical solution of the invention, the specific steps for performing compatibility analysis on the knowledge graph structure include:

[0030] Extract structural information from the target topological fingerprint, including core path, prerequisite dependencies, hierarchical span, node density, and cognitive gradient;

[0031] Map the target topological fingerprint to the corresponding nodes and paths in the source graph structure;

[0032] Compare the differences between the two in terms of path connectivity and prior knowledge coverage;

[0033] If a necessary preceding node not covered by the source graph is detected in the target topology, it is marked as a structural gap and the path adjustment engine is triggered.

[0034] As a preferred technical solution of the invention, the reconstructed teaching path includes:

[0035] For the detected structural gaps, the module that matches the teaching objective and knowledge point tag is called from the preset transition module library;

[0036] If no matching module exists in the library, an external knowledge unit generation component is invoked to construct a new transition module;

[0037] Insert the supplemented transition module into the breakpoint position in the source graph path, and update the node order and dependency edges;

[0038] The final output reconstruction path must satisfy the pre-conditions of the target topology fingerprint in terms of topology structure.

[0039] As a preferred embodiment of the invention, the invocation steps of the multimodal content atomic units include:

[0040] Retrieve the content unit identifier associated with the selected knowledge node from the knowledge graph;

[0041] Traverse the content resource index corresponding to the identifier and load the multimodal data bound to it in sequence, including text, images, formulas, simulation animations and interactive question components;

[0042] Extract metadata descriptions for each content unit, including content type, difficulty tag, and applicable terminal type, and encapsulate them into a unified multimodal content package;

[0043] The multimodal content package is attached to the corresponding knowledge node for subsequent textbook rendering and output modules to call.

[0044] As a preferred technical solution of the invention, the content binding, consistency detection, and logical integration include:

[0045] In the knowledge graph, a set of cross-modal dependency rules is maintained for each type of knowledge node. The rules record the semantic dependencies and presentation order between content units in the form of directed edges.

[0046] After loading the multimodal content package, read the dependency chain recorded in the graph and establish the logical binding relationship between each modal element;

[0047] Thresholds are set to determine the degree of conflict based on the semantic vector similarity and visual-text matching index of modal content.

[0048] Perform elimination, replacement, or prompting operations on conflicting modalities to ensure that the final output of multimodal content remains consistent at the semantic and cognitive levels.

[0049] As a preferred technical solution of the invention, the step of outputting the textbook body adapted to the terminal includes:

[0050] Multimodal content and path structure are divided and assembled into blocks according to preset teaching module units, each module including a main knowledge node and its related content;

[0051] The layout is automatically selected based on the content density, modal distribution, and logical hierarchy of the teaching modules;

[0052] For modules containing images, text, questions, and interactive controls, the presentation order of elements is determined according to the priority of the cross-modal dependency graph;

[0053] Based on the terminal type identifier, select the matching set of content presentation strategies and style template layout scheme;

[0054] The rendered textbook content is structured and encapsulated into a textbook object, and then output as a format file that supports terminal rendering.

[0055] This invention also proposes a knowledge graph-driven automated textbook generation system, comprising:

[0056] The teaching objective parsing module is used to receive teaching objective input, and according to the preset teaching intention meta-model, it parses the teaching objective into a multi-layered structure including higher-order objectives, core competency tags, and observable behavioral indicators, and maps the parsed teaching objective into the starting node of the knowledge graph to construct the starting point of the teaching path.

[0057] The teaching path generation module, based on the topological location of the starting node and the teaching intention level, calls a rule engine that supports dynamic reasoning and multi-objective optimization to generate the optimal teaching path;

[0058] The textbook system adaptation module generates the topological fingerprint of the target textbook system and performs a compatibility analysis with the knowledge graph structure. If there is a structural conflict between the target topological fingerprint and the source graph in terms of knowledge density, cognitive gradient, depth of prerequisite dependencies, or consistency of core paths, a transition module is dynamically inserted to reconstruct the teaching path to meet the topological consistency requirements.

[0059] The multimodal content scheduling and coordination module, based on the reconstructed path nodes, calls the bound multimodal content atomic units, and completes content binding, consistency detection and logical integration according to the multi-dimensional modal collaboration constraint rules defined by the cross-modal dependency graph and semantic and visual consistency indicators.

[0060] The textbook rendering and output module performs structured rendering of the integrated teaching content and outputs the textbook itself adapted to the terminal.

[0061] Compared with the prior art, the beneficial effects of the present invention are:

[0062] 1. This invention proposes a goal parsing method based on the teaching intent meta-model, which can structure abstract teaching goals into a multi-layered expression structure of "higher-order cognitive goals - ability labels - observable behavioral indicators". Through semantic anchoring mechanism, the behavioral indicators are automatically mapped to knowledge graph nodes, which has good interpretability and extensibility, and significantly improves the intelligence and intent adaptation of textbook generation.

[0063] 2. This invention designs a cross-modal dependency graph for automatic binding and logical coordination of modal content. The system establishes a directional and prioritized modal dependency graph through modal unit extraction, dependency template construction, and graph structure generation. It also combines semantic vector similarity to detect modal conflicts, achieving multimodal coordinated generation and structurally self-consistent output.

[0064] 3. This invention constructs a topological fingerprint model of the textbook system, introducing knowledge density, cognitive gradient, depth of prerequisite dependencies, and consistency of core paths to objectively characterize the differences in the underlying structure of different textbook versions. Based on this, the system supports compatibility comparison and transition module completion mechanisms, enabling path migration from one textbook system to another without manual reconstruction. It possesses strong structural understanding and dynamic reconstruction capabilities, significantly improving the cross-version and cross-regional generalization capabilities of the textbook generation system. Attached Figure Description

[0065] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0066] Figure 1 This is a flowchart illustrating a knowledge graph-driven method for automated textbook generation provided by the present invention.

[0067] Figure 2 This is a schematic diagram of the structure of a knowledge graph-driven automated textbook generation system according to the present invention. Detailed Implementation

[0068] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0069] Example 1:

[0070] This invention provides a knowledge graph-driven method for automated textbook generation, such as... Figure 1 As shown, it includes:

[0071] Upon receiving the input of teaching objectives, the teaching objectives are parsed into a multi-layered structure including higher-order objectives, core competency tags, and observable behavioral indicators according to the pre-set teaching intention meta-model. The parsed teaching objectives are then mapped to the starting nodes of the knowledge graph to construct the starting point of the teaching path.

[0072] Furthermore, the steps for analyzing the learning objectives include:

[0073] Extract teaching objective expressions and their related action verbs, cognitive levels, and ability labels from curriculum standard texts, teaching evaluation frameworks, and typical teaching cases;

[0074] By setting expert rules and combining them with a dynamic adjustment mechanism based on teaching feedback, the teaching objective expressions, action verbs, cognitive levels and ability labels are organized into a three-layer structure model, including a set of higher-order objectives, core ability labels and observable behavioral indicators.

[0075] Establish semantic mapping rules for each layer of the structure to form a rule chain from target terms to capability tags and then to knowledge graph anchors;

[0076] The structural model and mapping rules are stored in the teaching intent rule engine in a structured format, and template expansion and dynamic updates are supported.

[0077] Specifically, firstly, the system constructs a meta-model of teaching intent based on pre-set curriculum standard texts, teaching evaluation frameworks, and typical teaching case corpora. This includes the following steps:

[0078] Natural language processing tools are used to perform semantic analysis on texts such as curriculum standards, identify and extract teaching objective expressions, and extract key elements: action verbs, cognitive level terms, and ability label terms.

[0079] The above elements are structured into a three-tiered model according to the teaching objectives:

[0080] Higher-order goal set: represents goals at different cognitive levels;

[0081] Core competency tags: Define the core competencies or professional skills required to be achieved by the teaching objectives;

[0082] Observable behavior indicators: semantic representations that record observable behavior patterns.

[0083] Construct a semantic mapping rule chain from teaching objective terms to knowledge graph anchor points. Each rule chain is defined with the following structure: target term → cognitive level classification → ability tag → behavioral indicator template → knowledge graph node tag.

[0084] The aforementioned models and rule chains are stored in a structured form within the instructional intent rule engine, and a version control mechanism is established to support template expansion and expert maintenance. The system provides a rule editing interface, allowing qualified experts to log in and adjust the structure template and matching logic. All editing actions are logged in the database and version hashes are generated.

[0085] Furthermore, the step of mapping the teaching objectives to the starting nodes of the knowledge graph includes:

[0086] Semantic analysis is performed on the input of teaching objectives to extract verb phrases, ability verbs, and object nouns;

[0087] The extracted verb phrases are mapped to a set of higher-order objectives in a pre-defined meta-model of teaching intent in order to identify the type of teaching objective they belong to.

[0088] Based on the target type, select the set of core capability tags associated with it, and further match the corresponding observable behavior indicator templates;

[0089] Based on behavioral indicators, the teaching intent rule engine is invoked to execute the semantic anchoring algorithm, dynamically mapping and optimizing the matching of node sets in the knowledge graph to form a candidate set of teaching path starting points.

[0090] Specifically, in actual operation, users input teaching objectives expressed in natural language through the terminal interface. The system processes this input text as follows:

[0091] The input text is subjected to Chinese word segmentation and syntactic dependency analysis to identify verb phrases, ability verbs, and object nouns. Verb phrases are mapped to higher-order target sets to determine their corresponding cognitive levels. Furthermore, the object nouns and context are combined to determine their matching ability tags. Behavioral indicator templates matching the target level and ability tags are retrieved from the teaching intention rule engine to form a structured teaching objective representation.

[0092] The aforementioned behavioral indicators serve as input for the next stage, and the system maps them to nodes in the knowledge graph to determine the starting point of the teaching path. The mapping process employs a semantic anchoring matching method, specifically including the following steps:

[0093] Using a pre-trained BERT model to analyze behavioral indicator text Encoding generates semantic embedding vectors And in the preprocessing stage, the knowledge graph node labels Generate semantic vectors .

[0094] calculate With the first in the knowledge graph semantic similarity between nodes Its definition is as follows:

[0095] ;

[0096] Based on semantic similarity calculation, the structural importance of nodes in the graph and keyword weights are introduced to form the final matching scoring function:

[0097] ;

[0098] Where: parameters and satisfy ;

[0099] Normalized PageRank value of a node Represents nodes in a knowledge graph structure The importance of the graph structure is calculated using the classic PageRank algorithm:

[0100] ;

[0101] in: The damping coefficient; The total number of nodes; Let i be the set of all nodes that point to node i; For nodes The number of outgoing edges. After calculating the PageRank values ​​of all nodes, normalize them to the maximum value to obtain... .

[0102] Node term frequency - inverse frequency It is calculated based on the course teaching corpus used in the knowledge graph construction phase, and its definition is as follows:

[0103] ;

[0104] in: The word frequency of the node label in the current corpus; This indicates the number of documents in which the tag has appeared. This represents the total number of documents in the corpus; adding 1 is used for smoothing to avoid division by zero. The resulting... The same normalization process is performed so that it can be uniformly weighted and fused with semantic similarity in the final scoring function.

[0105] System sets matching threshold And select the top from the candidate nodes whose scores are greater than the threshold. A set of nodes that serve as the starting points of the teaching path:

[0106] ;

[0107] Finally, the system constructs a teaching path starting point structure, with each starting point containing the following fields: teaching objective ID; knowledge graph node ID; and matching score. The structure includes the associated higher-level goals and capabilities tags, and a description of the dependencies between the higher and lower-level nodes. This starting point structure serves as the input for subsequent teaching path generation and content binding modules, leading to the path reasoning stage.

[0108] Based on the topological location of the starting node and the teaching intent level, the optimal teaching path is generated by calling a rule engine that supports dynamic reasoning and multi-objective optimization.

[0109] Furthermore, the optimal path generation process includes:

[0110] Identify the position of the target knowledge node in the graph structure of the knowledge graph, including its predecessor node, successor node and adjacent branch node;

[0111] Based on the level of the teaching intention, the teaching strategy template that combines heuristic search and historical teaching data is invoked from the rule engine;

[0112] Perform a restricted path search operation in the graph to filter all node sequences that are topologically reachable from the target node and satisfy the policy constraints;

[0113] All candidate paths are input into a path scoring function for evaluation. The scoring metrics include path length, strategy suitability, and node knowledge weight.

[0114] The path with the highest score is selected as the optimal teaching path, and its nodes are sequentially bound to the corresponding teaching activity types for subsequent content scheduling.

[0115] Specifically, the system first reads the set of starting points of the teaching path. Each node in The graph structure information. For each node, extract the following topological attributes: set of direct predecessor nodes. ; set of direct successor nodes The set of adjacent nodes connected to it by an edge. The depth of the subgraph containing the node. This is used to determine the depth of its knowledge dependency.

[0116] Each starting node The corresponding learning objective contains a clear cognitive level. The system selects a matching teaching strategy template based on this level. The template is retrieved from the rules engine and contains the following three elements:

[0117] The target number of nodes is within a certain range; the depth of knowledge that can be included is allowed, but it cannot span more than two levels; the sequence of teaching activity types is bound to the sequence of knowledge nodes.

[0118] The system employs an improved heuristic pathfinding algorithm, using each starting node as the source point and expanding backwards within the knowledge graph to generate a candidate set of paths. Path expansion must meet the following constraints: all paths must be directed paths, and there must be explicit dependency edges between nodes; the difference in level depth between nodes within a path must not exceed the span allowed by the strategy template; the path length must not exceed the set upper limit; and the number of teaching activity types specified in the strategy template must be covered.

[0119] For all candidate paths that meet the constraints, the system introduces the following path scoring function for evaluation:

[0120] ;

[0121] in:

[0122] Path structural coherence is defined as the average dependency strength of adjacent nodes within a path in the graph; for a path Its structural coherence is as follows:

[0123] ;

[0124] in For nodes The semantic similarity is calculated using the cosine similarity of the embedded vectors:

[0125] ;

[0126] Activity sequence fit: Calculate the type and template of teaching activities. Match degree:

[0127] ;

[0128] in This indicates the current sequence of teaching activities along the path. To edit the distance.

[0129] Mean knowledge importance of each node in the path:

[0130] ;

[0131] parameter , , The set values ​​for experience are derived from historical curriculum analysis and optimization.

[0132] After scoring all paths, the system selects the path with the highest score. As the optimal teaching approach.

[0133] The system selects the optimal path. Each knowledge node in Based on its index position, it is bound one-to-one with the teaching activity type defined in the strategy template, and output as a structured path object, including: an ordered list of nodes and their activity labels; a set of topological dependency edges; the total path score; the teaching intent meta-model ID and associated labels.

[0134] This structure serves as the input for subsequent multimodal content calls and textbook rendering modules, laying the foundation for automatically generating textbook content adapted to the terminal.

[0135] Generate the topological fingerprint of the target textbook system and perform a compatibility analysis with the knowledge graph structure. If there is a structural conflict between the target topological fingerprint and the source graph in terms of knowledge density, cognitive gradient, depth of prerequisite dependencies, or consistency of core paths, then dynamically insert a transition module to reconstruct the teaching path to meet the topological consistency requirements.

[0136] Furthermore, the specific steps for performing compatibility analysis on the knowledge graph structure include:

[0137] Extract structural information from the target topological fingerprint, including core path, prerequisite dependencies, hierarchical span, node density, and cognitive gradient;

[0138] Map the target topological fingerprint to the corresponding nodes and paths in the source graph structure;

[0139] Compare the differences between the two in terms of path connectivity and prior knowledge coverage;

[0140] If a necessary preceding node not covered by the source graph is detected in the target topology, it is marked as a structural gap and the path adjustment engine is triggered.

[0141] Specifically, the system uses the currently generated optimal teaching path The system constructs a topological fingerprint of the path, containing the following structural information: Core path set: the sequence of primary dependency edges between each pair of nodes in the path; Prerequisite dependency mapping table: recording the set of prerequisite nodes required for each knowledge node; Hierarchical span parameter: the hierarchical difference between the highest and lowest level nodes in the path; Node density parameter: the number of nodes per unit path length, used to measure the degree of knowledge density; Cognitive gradient: a trend function constructed based on the hierarchical values ​​of teaching intentions bound to nodes, used to evaluate the smoothness of changes in cognitive requirements in the path. The system combines the above structural information into a structured vector.

[0142] The system extracts the current path node from the knowledge graph. Subgraph And perform a compatibility comparison, the specific steps of which are as follows:

[0143] Map each node in the target topological fingerprint to the actual node set in the knowledge graph and establish a mapping table. ;

[0144] Compare whether the following four types of indicators are consistent with the target topological fingerprint: Path connectivity consistency: Determine whether all core edges in the path exist in the target topological fingerprint. In the middle; Pre-dependency integrity: Determine the condition of each node in the middle; Does it already contain all the preceding nodes specified in its preceding dependency mapping table? Hierarchical span deviation: If the deviation between the actual path hierarchical span and the hierarchical span defined in the target topology fingerprint exceeds the threshold... If any of the above conditions are not met, the system will mark it as an anomaly; Cognitive gradient consistency: the fit between the trend of changes in the teaching level of nodes in the calculation path and the cognitive gradient. If the fit is less than the minimum preset acceptable fit threshold, it is considered inconsistent. If any of the above conditions are not met, the system will mark it as a structural gap and start the path adjustment engine.

[0145] Furthermore, the reconstructed teaching path includes:

[0146] For the detected structural gaps, the module that matches the teaching objective and knowledge point tag is called from the preset transition module library;

[0147] If no matching module exists in the library, an external knowledge unit generation component is invoked to construct a new transition module;

[0148] Insert the supplemented transition module into the breakpoint position in the source graph path, and update the node order and dependency edges;

[0149] The final output reconstruction path must satisfy the pre-conditions of the target topology fingerprint in terms of topology structure.

[0150] Specifically, based on pre-dependency analysis and path breakpoint detection, breakpoint locations are identified. And determine the transition target knowledge tag set that needs to be inserted. The system queries the local transition module library and attempts to match the tag set. If the matching module is hit It directly extracts the structure and bound content; if no match is found, it calls an external knowledge generation component and inputs a tag set. Using a micro-text generation model and structural templates, a semantically and visually consistent transition module is constructed. Module structure:

[0151] ;

[0152] in To insert a node, For the set of dependent edges, This is the default teaching activity type.

[0153] Will Insert at the path breakpoint, update the node order list and dependency relationships, re-evaluate the target topology fingerprint of the complete path, confirm that the prerequisite logic and hierarchical constraints are met, and finally reconstruct the path as follows. This serves as the foundational path for subsequent multimodal content binding and textbook rendering.

[0154] Based on the reconstructed path nodes, the bound multimodal content atomic units are called, and according to the multi-dimensional modal collaborative constraint rules defined by the cross-modal dependency graph, combined with semantic and visual consistency indicators, content binding, consistency detection and logical integration are completed.

[0155] Furthermore, the invocation steps of the multimodal content atomic unit include:

[0156] Retrieve the content unit identifier associated with the selected knowledge node from the knowledge graph;

[0157] Traverse the content resource index corresponding to the identifier and load the multimodal data bound to it in sequence, including text, images, formulas, simulation animations and interactive question components;

[0158] Extract metadata descriptions for each content unit, including content type, difficulty tag, and applicable terminal type, and encapsulate them into a unified multimodal content package;

[0159] The multimodal content package is attached to the corresponding knowledge node for subsequent textbook rendering and output modules to call.

[0160] Specifically, for the refactoring path Each node in The system first retrieves the set of content unit identifiers bound to it from the knowledge graph. .

[0161] The system iterates through each content identifier. It loads the corresponding multimodal resource items from the content index database. The resource modalities include: text, images, mathematical formulas, simulation animations, and interactive question components.

[0162] For each resource file, the system extracts its content metadata fields: modality type, difficulty level tag, applicable terminal type, and presentation duration or data length. These resources are then packaged into a structured multimodal content package. They are uniformly mounted on the corresponding knowledge nodes. Below, for subsequent content coordination and retrieval.

[0163] Furthermore, the specific steps for completing content binding, consistency detection, and logical integration include:

[0164] In the knowledge graph, a set of cross-modal dependency rules is maintained for each type of knowledge node. The rules record the semantic dependencies and presentation order between content units in the form of directed edges.

[0165] After loading the multimodal content package, read the dependency chain recorded in the graph and establish the logical binding relationship between each modal element;

[0166] Thresholds are set to determine the degree of conflict based on the semantic vector similarity and visual-text matching index of modal content.

[0167] Perform elimination, replacement, or prompting operations on conflicting modalities to ensure that the final output of multimodal content remains consistent at the semantic and cognitive levels.

[0168] Specifically, the system maintains a set of cross-modal dependency rules for each type of knowledge node in the knowledge graph. Each rule represents a modal element. Should be in Previously presented and has priority. The rules are derived from statistical analysis of modal co-occurrence order in typical textbook samples and are compiled and summarized into dependency templates by the expert rule base.

[0169] For example, in the function graph parsing class node, the system automatically learns the modal sequence pattern of "image → formula → interactive control".

[0170] The cross-modal dependency graph is organized in the form of a directed graph, where nodes represent specific modal units, edges represent logical relationships of "pre-dependencies" or "display triggers", and edge weights are attached as priority weights.

[0171] load Then, the system reads the dependency chains recorded in the cross-modal dependency graph, dynamically establishes the logical binding relationship of each modal unit, and forms a constrained presentation path structure.

[0172] The system performs consistency checks between each group of logically related modes:

[0173] Semantic similarity calculation: The semantic embedding vector of each modality is extracted using the BERT model, and the semantic cosine similarity between modalities is calculated. ;

[0174] Visual-text matching degree calculation: For images and text, simulations and descriptions, calculate the matching degree between visual embeddings and text embeddings. ;

[0175] System sets semantic thresholds and visual threshold Used to determine if there is a content conflict.

[0176] If the similarity between any pair of modal elements is below a threshold, one of the following actions will be taken: Elimination: Remove the current modal content; Replacement: Prioritize selecting a modal resource with a higher matching degree from the candidate content set to replace it; Hint: Generate a system warning prompt, which will be manually confirmed by the editor or supplemented with replacement content. All reconciled multimodal content is reorganized into a teaching material package that conforms to the teaching path logic and is attached to the path structure for final rendering and output.

[0177] The integrated teaching content is rendered in a structured manner, and the resulting teaching material is output to be adapted to the terminal.

[0178] Furthermore, the step of outputting the textbook body adapted to the terminal includes:

[0179] Multimodal content and path structure are divided and assembled into blocks according to preset teaching module units, each module including a main knowledge node and its related content;

[0180] The layout is automatically selected based on the content density, modal distribution, and logical hierarchy of the teaching modules;

[0181] For modules containing images, text, questions, and interactive controls, the presentation order of elements is determined according to the priority of the cross-modal dependency graph;

[0182] Based on the terminal type identifier, select the matching set of content presentation strategies and style template layout scheme;

[0183] The rendered textbook content is structured and encapsulated into a textbook object, and then output as a format file that supports terminal rendering.

[0184] Specifically, the system first traverses the reconstructed teaching path, identifying each knowledge node. and its mounted multimodal content packages Each teaching module is encapsulated as a unit of instruction. Each module includes: main knowledge point tags; associated content resources; teaching activity type; and hierarchical tags. The system forms an ordered sequence of modules based on the structure of the instructional path. This serves as the basic unit for subsequent rendering processes.

[0185] The system automatically matches the corresponding layout template based on the modal content composition and density parameters of each module. Template types include: single main modality highlighting; dual modality parallel type; multimodal partitioning type; and interactive dedicated templates.

[0186] The selection criteria are mainly based on: modal type combination; content presentation length; preset cognitive difficulty level; terminal screen size and interactive support capabilities.

[0187] Based on the template's pre-defined structure, the system maps and arranges the content elements in the teaching module to different regions, generating a preliminary rendering sketch.

[0188] To ensure a clear reading flow and a smooth cognitive curve for users, the system determines the display order of content from each modality in each module based on the priority rules of cross-modal dependency graphs in the knowledge graph.

[0189] After layout is complete, the system performs style adaptation operations based on the target terminal type set by the user. Adaptation strategies include: font and scaling adjustments, interactive component reconstruction, image and animation size compression, module grid reorganization, and breakpoint response control. All style schemes are preset as a set of templates, which the system matches and calls according to the target terminal.

[0190] Finally, the system encapsulates all laid-out, sorted, and styled teaching modules into a textbook object. This object has a structured document structure, and its internal formatting supports:

[0191] The curriculum includes a list structure of teaching modules; multimodal content reference paths and binding relationships; logical connections and jump indexes between modules; and metadata. Output formats support multiple terminal adaptation standards, including: HTML5+CSS for web page display; EPUB for e-book readers; PDF for printable textbooks; and JSON structured output for platform integration.

[0192] At this point, the structured rendering and terminal adaptation output process of the teaching materials is complete, and the system returns the teaching material file for users to download or for the platform to use.

[0193] Example 2:

[0194] This invention provides a knowledge graph-driven automated textbook generation system, such as... Figure 2 As shown, it includes:

[0195] This embodiment takes the teaching of the "Newton's Second Law" chapter in junior high school physics as an example, and details the module composition and implementation steps of a knowledge graph-driven automated textbook generation system. It covers the entire process from teaching objective analysis, teaching path construction, textbook system adaptation, to multimodal content scheduling and terminal output.

[0196] The teaching objective parsing module is used to receive teaching objective input, and according to the preset teaching intention meta-model, it parses the teaching objective into a multi-layered structure including higher-order objectives, core competency tags, and observable behavioral indicators, and maps the parsed teaching objective into the starting node of the knowledge graph to construct the starting point of the teaching path.

[0197] Specifically, first, the teacher inputs the learning objectives into the system front end: to enable students to master Newton's second law and be able to design experiments to verify the formula. .

[0198] The system calls the instructional intent meta-model to parse the input target:

[0199] Extract higher-order objectives: "Mastery + Application"; match core competency tags: "Experimental design ability"; extract behavioral indicators: "Ability to construct variable-controlled experiments and analyze results".

[0200] Using a semantic anchoring algorithm, behavioral indicators are embedded into vectors and matched with nodes in the knowledge graph. The node with the closest semantics, "Newton's Second Law Experimental Design", is selected as the starting point of the path.

[0201] The teaching path generation module, based on the topological location of the starting node and the teaching intention level, calls a rule engine that supports dynamic reasoning and multi-objective optimization to generate the optimal teaching path;

[0202] Specifically, the analysis revealed that the teaching intention of this physics course was "experimental inquiry type". The system selected the corresponding activity path template from the teaching strategy library: "posing questions → proposing hypotheses → experimental verification → result analysis → reflection and improvement".

[0203] The system performs a restricted path search based on the starting node in the knowledge graph to generate several candidate teaching paths;

[0204] Calculate for each path: Path coherence; : Matching degree of teaching activities; : Average weight of knowledge nodes;

[0205] The optimal path is selected after sorting by the comprehensive scoring function: definition of Newton's second law → derivation of the relationship between force and acceleration → trolley experiment design → data error analysis.

[0206] The textbook system adaptation module generates the topological fingerprint of the target textbook system and performs a compatibility analysis with the knowledge graph structure. If there is a structural conflict between the target topological fingerprint and the source graph in terms of knowledge density, cognitive gradient, depth of prerequisite dependencies, or consistency of core paths, a transition module is dynamically inserted to reconstruct the teaching path to meet the topological consistency requirements.

[0207] Specifically, teachers select the target output system, and the system loads its corresponding topological fingerprint structure, including structural information such as core path, prerequisite dependencies, hierarchical span, node density, and cognitive gradient. The system then compares the generated path topology with the target system topology: it finds that the target system requires "data modeling foundation" to precede "experimental design"; the current path lacks this prerequisite node, forming a structural gap.

[0208] The system automatically calls the module library and inserts content blocks on linear fitting and residual analysis; if no matching module is found, it calls the knowledge generation interface to generate the corresponding content and fill in the gaps; the new path is updated to: Definition → Derivation → Fitting Analysis → Experimental Design → Data Reflection.

[0209] The multimodal content scheduling and coordination module, based on the reconstructed path nodes, calls the bound multimodal content atomic units, and completes content binding, consistency detection and logical integration according to the multi-dimensional modal collaboration constraint rules defined by the cross-modal dependency graph and semantic and visual consistency indicators.

[0210] Specifically, the system sequentially extracts content units bound to path nodes from the knowledge graph, including: text descriptions; experimental setup diagrams; and formulas. Animation can be simulated by dragging a slider; interactive question.

[0211] For the experimental design nodes, the system constructs its modal dependency graph as follows: image → text description → formula derivation → interactive control; each modal node consists of content ID, modality type, and semantic label; edge weights represent dependency strength.

[0212] The system performs semantic conflict detection on loaded modal content based on the cross-modal dependency graph: it compares whether the semantic vectors of images and text are consistent; it determines whether there are cases where formulas are missing graphs or graphs are missing text; if it finds that the expression of interactive questions is inconsistent with the formulas, it prompts the developers to correct and perform modal replacement.

[0213] The textbook rendering and output module performs structured rendering of the integrated teaching content and outputs the textbook itself adapted to the terminal.

[0214] Specifically, the path is divided into modules: each module includes the main knowledge point, modal content, and teaching activity descriptions; the system selects the presentation template according to the target terminal: PC: text and images side by side + interactive controls on the right; mobile: text and images above and interactive folding; print version: insert QR code link to dynamic resources.

[0215] All content blocks are combined into a structured textbook object, and the rendering engine outputs a PDF file to form a complete textbook.

[0216] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A knowledge graph-driven method for automated textbook generation, characterized in that, Includes the following steps: Upon receiving the input of teaching objectives, the teaching objectives are parsed into a multi-layered structure including higher-order objectives, core competency tags, and observable behavioral indicators according to the pre-set teaching intention meta-model. The parsed teaching objectives are then mapped to the starting nodes of the knowledge graph to construct the starting point of the teaching path. Based on the topological location of the starting node and the teaching intent level, the optimal teaching path is generated by calling a rule engine that supports dynamic reasoning and multi-objective optimization. Generate the topological fingerprint of the target textbook system and perform a compatibility analysis with the knowledge graph structure. If there is a structural conflict between the target topological fingerprint and the knowledge graph in terms of knowledge density, cognitive gradient, depth of prerequisite dependencies, or consistency of core paths, then dynamically insert a transition module and reconstruct the teaching path to meet the topological consistency requirements. Based on the reconstructed path nodes, the multimodal content atomic units bound to them are called, and according to the cross-modal dependency rules defined by the cross-modal dependency graph, combined with semantic and visual consistency indicators, the content binding, consistency detection and logical integration are completed. The integrated teaching content is rendered in a structured manner, and the resulting teaching material is output to be adapted to the terminal. The specific steps of content binding, consistency detection, and logical integration include: In the knowledge graph, a set of cross-modal dependency rules is maintained for each type of knowledge node. The rules record the semantic dependencies and presentation order between content units in the form of directed edges. After loading the multimodal content package, read the dependency chain recorded in the graph and establish the logical binding relationship between each modal element; Thresholds are set to determine the degree of conflict based on the semantic vector similarity and visual-text matching index of modal content. Perform elimination, replacement, or prompting operations on conflicting modalities to ensure that the final output of multimodal content remains consistent at the semantic and cognitive levels.

2. The knowledge graph-driven automated textbook generation method according to claim 1, characterized in that, The steps for analyzing the teaching objectives include: Extract teaching objective expressions and their related action verbs, cognitive levels, and ability labels from curriculum standard texts, teaching evaluation frameworks, and typical teaching cases; By setting expert rules and combining them with a dynamic adjustment mechanism based on teaching feedback, the teaching objective expressions, action verbs, cognitive levels and ability labels are organized into a three-layer structure model, including a set of higher-order objectives, core ability labels and observable behavioral indicators. Establish semantic mapping rules for each layer of the structure to form a rule chain from target terms to capability tags and then to knowledge graph anchors; The structural model and mapping rules are stored in the teaching intent rule engine in a structured format, and template expansion and dynamic updates are supported.

3. The knowledge graph-driven automated textbook generation method according to claim 1, characterized in that, The steps for mapping the teaching objectives to the starting nodes of the knowledge graph include: Semantic analysis is performed on the input of teaching objectives to extract verb phrases, ability verbs, and object nouns; The extracted verb phrases are mapped to a set of higher-order objectives in a pre-defined meta-model of teaching intent in order to identify the type of teaching objective they belong to. Based on the target type, select the set of core capability tags associated with it, and further match the corresponding observable behavior indicator templates; Based on behavioral indicators, the teaching intent rule engine is invoked to execute the semantic anchoring algorithm, dynamically mapping and optimizing the matching of node sets in the knowledge graph to form a candidate set of teaching path starting points.

4. The knowledge graph-driven automated textbook generation method according to claim 1, characterized in that, The optimal path generation process includes: Identify the position of the target knowledge node in the graph structure of the knowledge graph, including its predecessor node, successor node and adjacent branch node; Based on the level of the teaching intention, the teaching strategy template that combines heuristic search and historical teaching data is invoked from the rule engine; Perform a restricted path search operation in the graph to filter all node sequences that are topologically reachable from the target node and satisfy the policy constraints; All candidate paths are input into a path scoring function for evaluation. The scoring metrics include path length, strategy suitability, and node knowledge weight. The path with the highest score is selected as the optimal teaching path, and its nodes are sequentially bound to the corresponding teaching activity types.

5. The knowledge graph-driven automated textbook generation method according to claim 1, characterized in that... The specific steps for compatibility analysis of the knowledge graph structure include: Extract structural information from the target topological fingerprint, including core path, prerequisite dependencies, hierarchical span, node density, and cognitive gradient; Map the target topological fingerprint to the corresponding nodes and paths in the knowledge graph structure; Compare the differences between the two in terms of path connectivity and prior knowledge coverage; If a necessary prerequisite node not covered by the knowledge graph is detected in the target topology, it is marked as a structural gap and the path adjustment engine is triggered.

6. The knowledge graph-driven automated textbook generation method according to claim 1, characterized in that, The reconstructed teaching path includes: For the detected structural gaps, the module that matches the teaching objective and knowledge point tag is called from the preset transition module library; If no matching module exists in the library, an external knowledge unit generation component is invoked to construct a new transition module; Insert the supplemented transition module into the breakpoint position in the knowledge graph path, and update the node order and dependency edges; The final output reconstruction path must satisfy the pre-conditions of the target topology fingerprint in terms of topology structure.

7. The knowledge graph-driven automated textbook generation method according to claim 1, characterized in that, The steps for invoking the multimodal content atomic unit include: Retrieve the content unit identifier associated with the selected knowledge node from the knowledge graph; Traverse the content resource index corresponding to the identifier and load the multimodal data bound to it in sequence, including text, images, formulas, simulation animations and interactive question components. Each loaded multimodal data item is a multimodal content atomic unit. Extract metadata descriptions for each multimodal content atom unit, including content type, difficulty tag, and applicable terminal type, and encapsulate multiple multimodal content atom units into a unified multimodal content package; The multimodal content package is attached to the corresponding knowledge node for subsequent textbook rendering and output modules to call.

8. The knowledge graph-driven automated textbook generation method according to claim 1, characterized in that, The steps for outputting the textbook body for the adapter terminal include: Multimodal content and path structure are divided and assembled into blocks according to preset teaching module units, each module including a main knowledge node and its related content; The layout is automatically selected based on the content density, modal distribution, and logical hierarchy of the teaching modules; For modules containing images, text, questions, and interactive controls, the presentation order of elements is determined according to the priority of the cross-modal dependency graph; Based on the terminal type identifier, select the matching set of content presentation strategies and style template layout scheme; The rendered textbook content is structured and encapsulated into a textbook object, and then output as a format file that supports terminal rendering.

9. A knowledge graph-driven automated textbook generation system, characterized in that, The system is used to execute the knowledge graph-driven automated textbook generation method according to any one of claims 1-8, the system comprising: The teaching objective parsing module is used to receive teaching objective input, and according to the preset teaching intention meta-model, it parses the teaching objective into a multi-layered structure including higher-order objectives, core competency tags, and observable behavioral indicators, and maps the parsed teaching objective into the starting node of the knowledge graph to construct the starting point of the teaching path. The teaching path generation module, based on the topological location of the starting node and the teaching intention level, calls a rule engine that supports dynamic reasoning and multi-objective optimization to generate the optimal teaching path; The textbook system adaptation module generates the topological fingerprint of the target textbook system and performs a compatibility analysis with the knowledge graph structure. If there is a structural conflict between the target topological fingerprint and the knowledge graph in terms of knowledge density, cognitive gradient, depth of prerequisite dependencies, or consistency of core paths, a transition module is dynamically inserted to reconstruct the teaching path to meet the topological consistency requirements. The multimodal content scheduling and coordination module, based on the reconstructed path nodes, calls the bound multimodal content atomic units, and completes content binding, consistency detection and logical integration according to the cross-modal dependency rules defined by the cross-modal dependency graph and semantic and visual consistency indicators. The textbook rendering and output module performs structured rendering of the integrated teaching content and outputs the textbook itself adapted to the terminal.

Images