Knowledge graph-driven textbook automatic generation method and system

Abstract

The invention relates to the technical field of automatic generation, in particular to a knowledge graph driven teaching material automatic generation method and system. The method comprises the following steps: receiving teaching target input, analyzing the teaching target input into a multi-layer structure based on a preset teaching intention meta-model, and mapping the multi-layer structure to a knowledge graph node; generating an optimal teaching path through a rule engine in combination with the node topology position and the teaching intention level; after the topological fingerprint of the target textbook system is recognized, compatibility analysis is executed, and if a structural gap exists, a transition module is automatically inserted to reconstruct a path; calling the bound multi-modal content unit according to the reconstruction path node, and completing modal coordination according to a cross-modal dependency graph; and finally, performing structured rendering on the integrated content, and outputting a textbook body adaptive to the terminal. The method has the advantages of teaching intention deep understanding, multi-modal content consistency control and cross-textbook system adaptation, and the intelligence level and generalization ability of textbook generation are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of teaching material automatic generation, in particular to a knowledge graph driven teaching material automatic generation method and system. BACKGROUND

[0002] With the rapid development of education informatization, the digitalization and individualization of teaching material content have become an important direction of intelligent teaching. How to automatically generate teaching material resources with reasonable structure, coordinated content and multi-terminal adaptation according to teaching goals has become a key link in the intelligent upgrading of teaching systems.

[0003] However, the existing teaching material automatic generation methods generally have three technical defects: first, the teaching goal analysis is superficial, often relying on keyword matching, lacking modeling and execution path support for high-order cognitive intentions; second, the content organization structure is rigid, lacking semantic coordination mechanism between text and image modalities, which is easy to cause cognitive discontinuity or expression redundancy; third, lacking cross-teaching system adaptation ability, it is difficult to cope with the differences in knowledge topology structure and cognitive path between different versions of teaching materials, resulting in high content migration reconstruction cost, weak generalization ability, and limiting the practical application range of teaching material automatic system.

[0004] Therefore, a knowledge graph driven teaching material automatic generation method and system are proposed. SUMMARY

[0005] The present application provides a knowledge graph driven teaching material automatic generation method and system, which analyzes teaching goals through a teaching intention meta-model, combines cross-modal dependency graph and teaching system adaptation technology, and automatically generates teaching resources with reasonable structure and coordinated content, solving the technical problems of path planning, modal coordination and cross-teaching system adaptation in traditional teaching material generation, and improving the intelligence and adaptability of teaching material generation.

[0006] To achieve the above purpose, the present application provides the following technical solutions: The present application proposes a knowledge graph driven teaching material automatic generation method, comprising the following steps: Receiving teaching goal input, analyzing the teaching goal into a multi-layer structure including high-order goal, core ability label and observable behavior index according to the preset teaching intention meta-model, and mapping the analyzed teaching goal into the starting node of the knowledge graph to build the starting point of the teaching path; Based on the topological position of the starting node and the teaching intention level, calling a rule engine supporting dynamic reasoning and multi-objective optimization to generate an optimal teaching path; A topological fingerprint of the target teaching material system is generated, and compatibility analysis is performed with the knowledge graph structure. If there is a structural conflict in one of the knowledge density, cognitive gradient, pre-dependence depth, and core path consistency between the target topological fingerprint and the source graph, a transition module is dynamically inserted, and the teaching path is reconstructed to meet the topological consistency requirement; Based on the reconstructed path node, the bound multi-modal content atomic unit is called, and according to the multi-dimensional modal coordination constraint rule defined by the cross-modal dependency graph, the semantic and visual consistency index is combined to complete the content binding, consistency detection and logic integration; The integrated teaching content is structured and rendered, and the teaching material ontology suitable for the terminal is output.

[0007] As a preferred technical solution of the application, the teaching target analysis step includes: From the course standard text, teaching evaluation framework and typical teaching case, the teaching target expression and its related behavior verbs, cognitive levels and ability labels are extracted; Through expert rule setting, combined with a dynamic adjustment mechanism based on teaching feedback, the teaching target expression, behavior verbs, cognitive levels and ability labels are organized into a three-layer structure model, including a high-order target set, core ability labels and observable behavior indicators; Semantic mapping rules are established for each layer structure to form a rule chain from target terminology to ability labels to knowledge graph anchor points; The structure model and mapping rules are stored in the teaching intent rule engine in a structured format, and template expansion and dynamic updating are supported.

[0008] As a preferred technical solution of the application, the step of mapping the teaching target to the starting node of the knowledge graph includes: The verb phrases, ability verbs and object nouns in the teaching target input are extracted through semantic analysis; The extracted verb phrases are mapped to the high-order target set in the preset teaching intent meta-model to identify the type of teaching target they belong to; According to the target type, a set of core ability labels associated with it is selected, and the corresponding observable behavior indicator template is further matched; Based on the behavior indicators, the teaching intent rule engine is called to execute a semantic anchoring algorithm to dynamically map and optimize the node set in the knowledge graph, forming a candidate set of starting points of the teaching path.

[0009] As a preferred technical solution of the application, the optimal path generation process includes: The graph structure position of the target knowledge node in the knowledge graph is identified, including its predecessor nodes, successor nodes and adjacent branch nodes; According to the level of the teaching intention, the rule engine is called to combine the heuristic search and the historical teaching data to obtain a teaching strategy template; A restricted path search operation is performed in the graph to screen all node sequences that are reachable from the target node in the topology and meet the policy constraints; All candidate paths are input into a path scoring function for evaluation, and the scoring indicators include path length, policy adaptation degree, and node knowledge weight; The path with the highest score is selected as the optimal teaching path, and the nodes in the path are sequentially bound to corresponding teaching activity types for subsequent content scheduling.

[0010] As a preferred technical solution of the application, the specific steps of the compatibility analysis of the knowledge graph structure include: The structural information including the core path, the pre-requisite, the hierarchical span, the node density, and the cognitive gradient in the target topology fingerprint is extracted; The target topology fingerprint is mapped to the corresponding nodes and paths in the source graph structure; The differences in path connectivity and pre-knowledge coverage between the two are compared; If it is detected that there is a necessary pre-node in the target topology that is not covered by the source graph, the structural gap is marked and the path adjustment engine is triggered.

[0011] As a preferred technical solution of the application, the reconstructed teaching path includes: For the detected structural gap, a module matching the corresponding teaching target and knowledge point label is called from a preset transition module library; If there is no matching module in the library, a new transition module is constructed by calling an external knowledge unit generation component; The supplemented transition module is inserted into the breakpoint position in the source graph path, and the node order and dependency edge are updated; The finally output reconstructed path needs to meet the pre-logic constraints of the target topology fingerprint in the topology structure.

[0012] As a preferred technical solution of the application, the calling steps of the multi-modal content atomic unit include: The content unit identifier associated with the selected knowledge node is queried from the knowledge graph; The content resource index corresponding to the identifier is traversed, and the multi-modal data bound thereto, including text, image, formula, simulation animation, and interactive question components, are loaded in sequence; The metadata description of each content unit is extracted, including the content type, the difficulty label, and the applicable terminal type, and is uniformly packaged as a multi-modal content package; The multi-modal content package is mounted under the corresponding knowledge node for subsequent teaching material rendering and output module calling.

[0013] As a preferred technical solution of the application, the content binding, consistency detection and logical integration include: A set of cross-modal dependency rules is maintained for each type of knowledge node in the knowledge graph, and the rules record the semantic dependency and presentation order between content units in the form of directed edges; After loading the multi-modal content package, the dependency chain recorded in the graph is read to establish the logical binding relationship between the elements of each modality; The threshold value is set based on the semantic vector similarity and visual-text matching degree of the modal content to judge the conflict degree; The conflict modal is executed to eliminate, replace or prompt operation, and the multi-modal content output in the final output is ensured to be consistent in semantic and cognitive level.

[0014] As a preferred technical solution of the application, the step of outputting the textbook ontology of the terminal includes: The multi-modal content and path structure are divided and assembled according to the preset teaching module unit, and each module includes a main knowledge node and its associated content; According to the content density, modality distribution and logical level of the teaching module, a matched layout template is automatically selected for layout; For modules containing graphs, texts, questions and interactive controls, the presentation order of the elements is determined according to the priority of the cross-modal dependency graph; According to the terminal type identifier, a matched content presentation strategy set and style template layout scheme are selected; The rendered textbook content is structured and packaged as a textbook ontology object, and output as a format file that supports terminal rendering.

[0015] The application also proposes a knowledge graph driven automatic textbook generation system, comprising: A teaching goal analysis module is used to receive teaching goal input, analyze the teaching goal into a multi-layer structure including high-level goal, core ability label and observable behavior index according to the preset teaching intention meta-model, and map the analyzed teaching goal to the starting node of the knowledge graph to construct the starting point of the teaching path; A teaching path generation module generates an optimal teaching path based on the topological position of the starting node and the teaching intention level, and calls a rule engine supporting dynamic reasoning and multi-objective optimization; A textbook system adaptation module generates the topological fingerprint of the target textbook system, and performs compatibility analysis with the knowledge graph structure. If there is a structural conflict in one of the knowledge density, cognitive gradient, pre-dependence depth and core path consistency between the target topological fingerprint and the source graph, a transition module is dynamically inserted to reconstruct the teaching path to meet the topological consistency requirement; A multi-modal content scheduling and coordinating module, based on the reconstructed path node, calls the bound multi-modal content atomic unit, and according to the multi-dimensional modal coordination constraint rule defined by the cross-modal dependency graph, combines the semantic and visual consistency index to complete the content binding, consistency detection and logical integration; A teaching material rendering and output module performs structured rendering on the integrated teaching content, and outputs the teaching material ontology adapted to the terminal.

[0016] Compared with the prior art, the beneficial effects of the present application are: 1. The present application proposes a target analysis method based on a teaching intention meta-model, which can structure the abstract teaching target into a multi-layer expression structure of "high-order cognitive target-ability label-observable behavior index", and automatically map the behavior index to the knowledge graph node through the semantic anchoring mechanism, has good interpretability and expandability, and significantly improves the intelligence and intention adaptation of the teaching material generation.

[0017] 2. The present application designs a cross-modal dependency graph for realizing automatic binding and logical coordination of modal content. The system establishes a modal dependency graph with directionality and priority through the modal unit extraction, dependency template construction and graph structure generation process, and combines the semantic vector similarity to detect modal conflicts, realizes multi-modal coordinated generation and structure self-consistent output.

[0018] 3. The present application constructs a teaching material system topology fingerprint model, introduces knowledge density, cognitive gradient, pre-requisite dependency depth and core path consistency, and objectively depicts the differences in the underlying structure of different teaching material versions. On this basis, the system supports compatibility comparison and transition module filling mechanism, and can realize path migration from one teaching material system to another without manual reconstruction, has strong structure understanding ability and dynamic reconstruction ability, and significantly improves the cross-version and cross-region generalization ability of the teaching material generation system. DETAILED DESCRIPTION

[0019] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation of the present application. In the drawings: Figure 1 is a flowchart of a knowledge graph driven teaching material automatic generation method provided by the present application; Figure 2 is a structural schematic diagram of a knowledge graph driven teaching material automatic generation system of the present application. DETAILED DESCRIPTION

[0020] The preferred embodiments of the present application are described below in conjunction with the drawings, and it should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and do not limit the present application.

[0021] Example 1: The present invention provides a knowledge graph driven teaching material automatic generation method, such as Figure 1 Shown, including: Receive teaching goal input, parse the teaching goal into a multi-layer structure including high-level goals, core competency labels, and observable behavior indicators according to the preset teaching intention meta-model, and map the parsed teaching goal to the starting node of the knowledge graph to construct the starting point of the teaching path; Furthermore, the teaching objective analysis step includes: Extract teaching objective expressions and their related action verbs, cognitive levels, and ability labels from curriculum standards, teaching evaluation frameworks, and typical teaching cases; Through expert rule setting and a dynamic adjustment mechanism based on teaching feedback, the teaching goal expressions, behavioral verbs, cognitive levels, and ability labels are organized into a three-layer structure model, including a high-level goal set, core ability labels, and observable behavior indicators; Establish semantic mapping rules for each layer of the structure, forming a rule chain from target terms to capability labels and then to knowledge graph anchor points; The structural model and mapping rules are stored in a teaching intention rule engine in a structured format, and template expansion and dynamic update are supported.

[0022] Specifically, first, the system constructs a teaching intention meta-model based on the preset curriculum standard text, teaching evaluation framework, and typical teaching case corpus. The specific steps include: Use natural language processing tools 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.

[0023] The above elements are formed into a three-layer structure model according to the teaching objectives: High-level goal set: represents goals at different cognitive levels; Core competency label: defines the core literacy or professional competencies required to be achieved by the teaching objectives; Observable behavior indicators: record the semantic expression of observable behavior patterns.

[0024] Construct a semantic mapping rule chain from teaching target terms to knowledge graph anchors. Each rule chain defines the following structure: target term → cognitive level classification → ability label → behavior indicator template → knowledge graph node label.

[0025] The above model and rule chain are stored in a structured form in the teaching intention rule engine, and a version control mechanism is established to support template expansion and expert maintenance. The system provides a rule editing interface for experts with teaching design qualifications to adjust the structure template and matching logic after logging in. All editing behaviors are recorded in the database operation log and generate version hash.

[0026] Further, the step of mapping the teaching goal as a starting node of the knowledge graph comprises: performing semantic analysis on the teaching goal input to extract verb phrases, ability verbs and object nouns therefrom; mapping the extracted verb phrases to a set of high-order target in the preset teaching intention meta-model to identify the teaching goal type to which the verb phrases belong; According to the target type, a set of core ability labels associated with the target type is selected, and a corresponding observable behavior index template is further matched; Based on the behavior index, the teaching intention rule engine is called to perform a semantic anchoring algorithm to dynamically map and optimize the node set in the knowledge graph to form a candidate set of starting points of the teaching path.

[0027] Specifically, in actual operation, the user inputs the teaching goal expressed in natural language through the terminal interface. The system performs the following processing on the input text: Chinese word segmentation and syntactic dependency analysis are performed on the input text to identify verb phrases, ability verbs and object nouns. The verb phrases are mapped to a set of high-order targets to determine their corresponding cognitive levels. Further, the object nouns and context are combined to determine the matching ability labels. The behavior index templates matching the target level and the ability labels are retrieved from the teaching intention rule engine to form a structured teaching goal representation.

[0028] The above behavior index is mapped to a node in the knowledge graph as input for the next stage to determine the starting point of the teaching path. The mapping process uses a semantic anchoring matching method, which includes the following steps: The behavior index text is encoded using a pre-trained BERT model to generate a semantic embedding vector , and the node labels of the knowledge graph are preprocessed to generate semantic vectors .

[0029] The semantic similarity between the generated semantic vector and the semantic vector of the first node in the knowledge graph is calculated , which is defined as follows: ; On the basis of calculating semantic similarity, the structural importance of nodes in the graph and the keyword weight are introduced to form the final matching score function: ; Wherein: parameter , and satisfies ; The normalized PageRank value of the node represents the graph structure importance of the node in the knowledge graph structure, which is calculated by the classical PageRank algorithm: ; Wherein: is the damping coefficient; is the total number of nodes; is the set of all nodes pointing to node i; is the out-degree of node . After all the node PageRank values are calculated, they are normalized by the maximum value to obtain .

[0030] The term frequency-inverse document frequency value of the node is calculated based on the course teaching corpus used in the knowledge graph construction stage, and its definition is as follows: ; Wherein: is the term frequency of the node label in the current corpus; is the number of documents in which the label appears; is the total number of corpus documents; add 1 for smoothing processing to avoid division by zero. The obtained is also normalized to be uniformly weighted and integrated with semantic similarity in the final scoring function.

[0031] The system sets a matching threshold , and selects the top starting nodes of the teaching path from the candidate nodes with a score greater than the threshold: ; Finally, the system constructs the starting point structure of the teaching path, and each starting point contains the following fields: teaching goal ID; knowledge graph node ID; matching score ; belonging to high-order goal and ability label; and dependency relationship description with upper and lower nodes. This starting point structure is used as the input of the subsequent teaching path generation and content binding module, and enters the path reasoning stage.

[0032] Based on the topological position and teaching intention level of the starting node, a rule engine supporting dynamic reasoning and multi-objective optimization is called to generate the optimal teaching path; Further, the optimal path generation process includes: Identifying the graph structure position of the target knowledge node in the knowledge graph, including its predecessor nodes, successor nodes, and adjacent branch nodes; According to the level of the teaching intention, calling the teaching strategy template in the rule engine combined with heuristic search and historical teaching data; Performing a restricted path search operation in the graph to filter all node sequences that are topologically reachable from the target node and meet the strategy constraints; Inputting all candidate paths into the path scoring function for evaluation, with scoring indicators including path length, strategy adaptation degree, and node knowledge weight; Selecting the highest scoring path as the optimal teaching path, and binding the nodes in sequence to the corresponding teaching activity types for subsequent content scheduling.

[0033] Specifically, the system first reads the graph structure information of each node in the set of teaching path starting points For each node, extract the following topological properties: direct predecessor node set ; direct successor node set ; adjacent node set connected by edges ; the level depth of the subgraph where the node is located , used to determine the knowledge dependency depth. Each starting node

[0034] corresponding teaching goal contains an explicit cognitive level, and the system selects a matching teaching strategy template according to this level. The template is called from the rule engine and contains the following three elements: Target node quantity range; allows knowledge depth span, but cannot exceed 2 layers; teaching activity type sequence, bound to the knowledge node sequence. The system uses an improved heuristic path search algorithm to expand from each starting node as the source point in the knowledge graph, generating a path candidate set

[0035] . The path expansion needs to meet the following constraints: all paths are directed paths, and there are explicit dependency edges between nodes; the level depth difference of each node in the path does not exceed the allowed span in the strategy template; the path length cannot exceed the set upper limit; must cover the number of teaching activity types specified in the strategy template.

[0036] For all candidate paths that meet the constraint conditions, the system introduces the following path scoring function for evaluation: ; Where: ​ : Path structure coherence, defined as the average of the dependency strength between adjacent nodes in the path in the graph; for path , the structure coherence is: ; where is the semantic similarity of node , calculated using the cosine similarity of embedding vectors: ; : Activity sequence fitness, calculated as the matching degree between the teaching activity type and the template : ; where represents the current teaching activity sequence of the path, is the edit distance.

[0037] : The average knowledge importance of each node in the path: ; Parameters , , are empirical values, derived from historical course analysis and optimization.

[0038] After scoring all paths, the system selects the highest-scoring path as the optimal teaching path.

[0039] The system binds each knowledge node in the optimal path according to its index position and the one-to-one correspondence of the teaching activity type defined in the strategy template, and outputs it in the form of a structured path object, including: an ordered node list and its activity label; a set of topological dependency edges; the total score of the path; the ID and associated label of the teaching intent meta-model.

[0040] This structure serves as the input for subsequent multi-modal content calling and teaching material rendering modules, laying the foundation for automatically generating teaching materials adapted to the terminal.

[0041] Generate the topological fingerprint of the target teaching material system and perform compatibility analysis with the knowledge graph structure. If there is a structural conflict in one of the indicators such as knowledge density, cognitive gradient, pre-dependence depth, and core path consistency between the target topological fingerprint and the source graph, dynamically insert a transition module to reconstruct the teaching path to meet the topological consistency requirements. Further, the specific steps of the compatibility analysis of the knowledge graph structure include: The structural information in the target topology fingerprint includes core paths, pre-requisites, hierarchical span, node density, and cognitive gradient. Map the target topology fingerprint to the corresponding nodes and paths in the source graph structure. Compare the differences in path connectivity and pre-knowledge coverage. If necessary pre-requisite nodes that are not covered by the source graph are detected in the target topology, mark it as a structural gap and trigger the path adjustment engine.

[0042] Specifically, the system generates an optimal teaching path based on the current generation Construct the topology structure fingerprint of the path, including the following structural information: core path set: the main dependency edge sequence between each pair of nodes in the path; pre-requisite mapping table: records the pre-requisite node set required for each knowledge node; hierarchical span parameter: the hierarchical difference between the highest and lowest nodes in the path; node density parameter: the number of nodes per unit path length, used to measure the knowledge intensity; cognitive gradient: a trend function based on the teaching intent hierarchical value of the node binding, used to evaluate the smoothness of cognitive requirements in the path. The system combines the above structural information into a structured vector.

[0043] The system extracts the current path nodes from the knowledge graph and performs compatibility comparison, with the following steps: Map each node in the target topology fingerprint to the actual node set in the knowledge graph, and establish a mapping relationship table . Compare the following four indicators to see if they are consistent with the target topology fingerprint: path connectivity consistency: judge whether all core edges in the path exist in ; pre-requisite integrity: judge whether each node in contains all pre-requisite nodes specified in its pre-requisite mapping table; hierarchical span deviation: if the actual path hierarchical span deviates from the hierarchical span defined in the target topology fingerprint by more than a threshold , mark it as abnormal; cognitive gradient consistency: calculate the fitting degree of the change trend of the node teaching level in the path and the cognitive gradient, if the fitting degree is less than the minimum preset acceptable fitting degree threshold, it is considered inconsistent. If any of the above does not meet the requirements, the system marks it as a structural gap and starts the path adjustment engine.

[0044] Further, the reconstructed teaching path includes: For the detected structural gap, call the module matching the teaching target and knowledge point label from the preset transition module library; If there is no matching module in the library, call the external knowledge unit generation component to construct a new transition module. The supplemented transition module is inserted into a breakpoint position in the source graph path, and the node order and dependent edge are updated; The final output reconstruction path needs to satisfy the pre-logic constraints of the target topology fingerprint in the topology structure.

[0045] Specifically, according to the pre-dependence analysis and path breakpoint detection, the breakpoint position is identified , and the transition target knowledge label set that needs to be inserted is determined . The system queries the local transition module library to try to match the label set ; if a matching module is hit , the structure and binding content are directly extracted; if not, an external knowledge generation component is called, and the label set is input ; a semantic and visual consistency transition module is constructed using a micro text generation model and a structure template, and the module structure is: ; Wherein is an inserted node, is a dependent edge set, is a default teaching activity type.

[0046] Insert into the path breakpoint, update the node order list and dependent edge relationship, re-evaluate the target topology fingerprint of the complete path, confirm that the pre-logic and hierarchical constraints are satisfied, and finally reconstruct the path as , which is used as the basis path for subsequent multi-modal content binding and teaching material rendering.

[0047] Based on the reconstructed path node, the bound multi-modal content atomic unit is called, and according to the multi-dimensional modal coordination constraint rules defined by the cross-modal dependence graph, combined with the semantic and visual consistency indicators, the content binding, consistency detection and logic integration are completed; Further, the calling step of the multi-modal content atomic unit includes: Query 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 bound multi-modal data one by one, including text, image, formula, simulation animation and interactive question components; Extract the metadata description of each content unit, including content type, difficulty label, and terminal type suitable for uniform packaging as a multi-modal content package; The multi-modal content package is mounted under the corresponding knowledge node for subsequent teaching material rendering and output module calling.

[0048] Specifically, for each node in the reconstructed path , the system first retrieves its bound content unit identifier set from the knowledge graph .

[0049] The system traverses each content identifier , loads its corresponding multi-modal resource item from the content index database, including: text, image, mathematical formula, simulation animation, and interactive question components.

[0050] For each resource file, the system extracts its content metadata fields: modal type, difficulty level label, applicable terminal type, presentation duration or data length. Then the above resources are packaged into structured multi-modal content packages , which are uniformly mounted on the corresponding knowledge nodes for subsequent content coordination calls.

[0051] Further, the specific steps of completing content binding, consistency detection and logical integration include: Maintain a set of cross-modal dependency rules for each type of knowledge node in the knowledge graph, which records the semantic dependency and presentation order between content units in the form of directed edges; After loading the multi-modal content package, read the dependency chain recorded in the graph to establish the logical binding relationship between modal elements; Set threshold values based on semantic vector similarity and visual-text matching degree indicators to judge the degree of conflict; Perform elimination, replacement or prompt operations on conflicting modalities to ensure that the final output multi-modal content remains consistent in terms of semantics and cognition.

[0052] 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 presented before , and has a priority . The rules are derived from statistical analysis of modal co-occurrence order in typical textbook samples and summarized as dependency templates by expert rule library.

[0053] For example, in the function image analysis class node, the system automatically learns that the modal order pattern is "image → formula → interactive control".

[0054] The cross-modal dependency graph is organized in the form of a directed graph, with nodes representing specific modal units and edges representing "predecessor dependency" or "display trigger" logical relationships, and with edge weights as priority weights.

[0055] After loading , the system reads the dependency chain recorded in the cross-modal dependency graph to dynamically establish the logical binding relationship between modal units and form a constrained presentation path structure.

[0056] The system makes consistency judgments between each set of logically associated modalities: Semantic similarity calculation: Use the BERT model to extract semantic embedding vectors for each modality, and calculate the semantic cosine similarity between modalities ; Visual-text matching degree calculation: For images and text, simulations and descriptions, calculate the matching degree between visual embedding and text embedding ; The system sets semantic thresholds and visual thresholds to determine whether the content is in conflict.

[0057] If the similarity between any pair of modality elements is lower than the threshold, one of the following operations is performed: elimination: remove the current modality content; replacement: preferentially select a modality resource with higher matching degree from the alternative content set to replace; prompt: generate a system warning prompt for manual confirmation or supplementary replacement content by an editor. All coordinated multi-modal content is reorganized into a teaching path logical teaching content package and attached to the path structure for final rendering and output.

[0058] The integrated teaching content is structured and rendered, and the output is adapted to the terminal textbook ontology.

[0059] Further, the step of outputting the textbook ontology adapted to the terminal includes: The multi-modal content and the path structure are assembled in blocks according to the preset teaching module unit, and each module includes a main knowledge node and its associated content; According to the content density, modality distribution and logical level of the teaching module, a matching layout template is automatically selected for layout; For modules containing graphs, texts, questions, and interactive controls, the presentation order of elements is determined according to the priority of the cross-modal dependency graph; According to the terminal type identifier, a matching set of content presentation strategies and style template layout schemes are selected; The rendered teaching content is structured and packaged as a textbook ontology object, and output as a format file that supports terminal rendering.

[0060] Specifically, the system first traverses the reconstructed teaching path, and encapsulates each knowledge node and its mounted multi-modal content package as a teaching module unit. Each teaching module includes: main knowledge point label; associated content resources; teaching activity type; level label. The system forms an ordered module sequence according to the structure of the teaching path, as the basic unit for subsequent rendering processing.

[0061] The system automatically matches the corresponding layout template according to the modal content composition and density parameters of each module. The template types include: single main modal highlighting type; double modal parallel type; multi-modal partition type; interactive special template; The selection criteria mainly depend on: modal type combination; content presentation length; preset cognitive difficulty level; terminal screen size and interaction support capability.

[0062] The system performs regional mapping and spatial arrangement on the content elements in the teaching module according to the structure preset of the template, and generates a preliminary rendering sketch.

[0063] In order to ensure that the user reading process is clear and the cognitive curve is smooth, the system determines the display order of each modal content in each module according to the priority rules of the cross-modal dependency graph in the knowledge graph.

[0064] After the layout is completed, the system performs style adaptation operation according to the target terminal type set by the user. The adaptation strategies include: font and scaling adjustment, interactive component reconstruction, image and animation size compression, module grid reorganization and breakpoint response control. All style schemes are preset as a template set, and the system matches and calls according to the target terminal.

[0065] Finally, the system encapsulates all the laid-out, sorted and style-adapted teaching modules into a textbook ontology object. The object is a structured document structure, and the internal format supports: Teaching module list structure; multi-modal content reference path and binding relationship; inter-module logical connection and jump index; metadata. The output format supports multiple terminal adaptation standards, including: HTML5+CSS for web-based display; EPUB for e-book readers; PDF for printed textbooks; JSON structured output for platform access.

[0066] At this point, the structured rendering and terminal adaptation output process of the textbook is completed, and the system returns the textbook ontology file for the user to download or the platform to call.

[0067] Embodiment two: The knowledge graph driven textbook automatic generation system of the present application comprises: Figure 2 As shown in the figure, it comprises: This embodiment takes the teaching of the "Newton's Second Law" chapter in the junior high school physics course as the background, and details the module composition and implementation steps of a knowledge graph driven textbook automatic generation system, covering the whole process from teaching target analysis, teaching path construction, textbook system adaptation, to multi-modal content scheduling and terminal output.

[0068] The teaching goal analysis module is configured to receive a teaching goal input, analyze the teaching goal into a multi-layer structure including a high-order goal, a core ability label, and an observable behavior index according to a preset teaching intention meta-model, and map the analyzed teaching goal into a starting node of a knowledge graph to construct a starting point of a teaching path. Specifically, first, the teacher inputs a teaching goal at the front end of the system: enabling students to master Newton's second law and to design experiments to verify the formula .

[0069] The system calls the teaching intention meta-model to analyze the input goal: The high-order goal is extracted: "mastering + application"; the core ability label is matched: "experimental design ability"; and the behavior index is extracted: "being able to construct a variable control experiment and analyze the results".

[0070] The semantic anchoring algorithm is used to match the behavior index embedding vector with the nodes in the knowledge graph, and the most semantically similar node "Newton's second law experiment design" is selected as the starting point of the path.

[0071] The teaching path generation module generates an optimal teaching path based on the topological position of the starting node and the teaching intention hierarchy by calling a rule engine supporting dynamic reasoning and multi-objective optimization. Specifically, it is analyzed that the teaching intention of this physics course is "experimental exploration type", and the system selects the corresponding activity path template from the teaching strategy library: "asking questions → making hypotheses → experimental verification → result analysis → reflection and improvement".

[0072] The system performs a restricted path search based on the starting node in the knowledge graph to generate several candidate teaching paths. For each path, the following is calculated: : path coherence; : teaching activity matching degree; : average weight of knowledge nodes; The optimal path is selected after sorting the comprehensive score function: Newton's second law definition → force and acceleration relationship derivation → car experiment design → data error analysis.

[0073] The teaching material system adaptation module generates a topological fingerprint of the target teaching material system and performs compatibility analysis with the knowledge graph structure. If there is a structural conflict in one of the knowledge density, cognitive gradient, pre-dependence depth, and core path consistency between the target topological fingerprint and the source graph, a transition module is dynamically inserted to reconstruct the teaching path to meet the topological consistency requirement. Specifically, the teacher selects the target output system, the system loads its corresponding topological fingerprint structure, including the structure information of core path, pre-dependence, hierarchical span, node density and cognitive gradient, and the system compares the generated path topology with the target system topology: it is found that the target system requires "data modeling foundation" before "experimental design"; the current path lacks this pre-node, forming a structural gap.

[0074] The system automatically calls the module library and inserts the linear fitting and residual analysis content block; if no matching module is found, the knowledge generation interface is called to generate the corresponding content and fill in the gap; the new path is updated to: definition → derivation → fitting analysis → experimental design → data reflection.

[0075] The multi-modal content scheduling and coordination module, based on the reconstructed path nodes, calls the bound multi-modal content atomic units, and according to the multi-dimensional modal coordination constraint rules defined by the cross-modal dependence graph, combines the semantic and visual consistency indicators to complete content binding, consistency detection and logical integration; Specifically, the system sequentially extracts the content units bound to the path nodes from the knowledge graph, including: text explanation; experimental device diagram; formula ; interactive questions.

[0076] For the experimental design node, the system constructs its modal dependence graph as follows: image → text explanation → formula derivation → interactive control; each modal node is composed of content ID, modal type and semantic label; the edge weight represents the dependence strength.

[0077] The system performs semantic conflict detection on the loaded modal content according to the cross-modal dependence graph: compare whether the image and text semantic vectors are consistent; judge whether there is a case of formula without graph or graph without text; if the interactive question expression is found to be inconsistent with the formula, prompt the developer to correct and replace the modal.

[0078] The textbook rendering and output module performs structured rendering on the integrated teaching content and outputs the textbook ontology adapted to the terminal.

[0079] Specifically, the path is divided into blocks by module unit: each block includes main knowledge points + modal content + teaching activity explanation; the system selects a presentation template according to the target terminal: PC: side-by-side text and image + interactive control on the right; mobile: text below image + interactive folding; print version: insert two-dimensional code link dynamic resources.

[0080] All content blocks are combined into a structured textbook object, and the rendering engine outputs a PDF format file to form a complete textbook.

[0081] Finally, it should be noted that the above only describes the preferred embodiments of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art will appreciate that the technical solutions described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalent ones. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A knowledge graph driven textbook automatic generation method, characterized in that, The method comprises the following steps: Receiving teaching target input, parsing the teaching target into a multi-layer structure including high-level goals, core competency labels, and observable behavior indicators according to a preset teaching intention meta-model, and mapping the parsed teaching target to a starting node of a knowledge graph to construct a starting point of a teaching path; Based on the topological position of the starting node and the level of teaching intention, a rule engine supporting dynamic reasoning and multi-objective optimization is called to generate an optimal teaching path; A topological fingerprint of the target teaching material system is generated, and compatibility analysis is performed with the knowledge graph structure. If there is a structural conflict in one of the knowledge density, cognitive gradient, pre-dependence depth, and core path consistency between the target topological fingerprint and the source graph, a transition module is dynamically inserted to reconstruct the teaching path to meet the topological consistency requirement; Based on the reconstructed path nodes, the bound multi-modal content atomic units are called, and according to the multi-dimensional modal coordination constraint rules defined by the cross-modal dependency graph, the semantic and visual consistency indicators are combined to complete content binding, consistency detection, and logical integration; The integrated teaching content is structured and rendered, and the teaching material ontology suitable for the terminal is output. 2.The knowledge graph driven textbook automatic generation method of claim 1, wherein, The teaching target parsing step comprises: Extracting teaching target expressions and their related behavior verbs, cognitive levels, and competency labels from course standard texts, teaching evaluation frameworks, and typical teaching cases; Through expert rule setting and combining dynamic adjustment mechanisms based on teaching feedback, the teaching target expressions, behavior verbs, cognitive levels, and competency labels are organized into a three-layer structure model, including a high-level goal set, core competency labels, and observable behavior indicators; Semantic mapping rules are established for each layer structure to form a rule chain from target terminology to competency labels to knowledge graph anchor points; The structure model and mapping rules are stored in the teaching intention rule engine in a structured format, and template expansion and dynamic updating are supported.

3. The method for automatically generating teaching materials driven by knowledge graph according to claim 1, characterized in that: The step of mapping the teaching target to the starting node of the knowledge graph comprises: Performing semantic analysis on the teaching target input to extract verb phrases, competency verbs, and object nouns; Mapping the extracted verb phrases to the high-level goal set in the preset teaching intention meta-model to identify the teaching target type to which they belong; According to the target type, select the associated core competency label set and further match the corresponding observable behavior indicator template; Based on the behavior indicators, call the teaching intention rule engine to perform semantic anchoring algorithm, dynamically map and optimize the node set in the knowledge graph to form a candidate set of starting points of the teaching path. 4.The knowledge graph driven textbook automatic generation method of claim 1, wherein, The optimal path generation process comprises: Identifying the graph structure position of the target knowledge node in the knowledge graph, including its predecessor nodes, successor nodes, and adjacent branch nodes; According to the level of teaching intention, call the teaching strategy template in the rule engine combined with heuristic search and historical teaching data; Perform a restricted path search operation in the graph to filter all node sequences that are topologically reachable from the target node and meet the strategy constraints; Input all candidate paths into the path scoring function for evaluation, and the scoring indicators include path length, strategy adaptation degree, and node knowledge weight. Select the highest score path as the optimal teaching path, and sequentially bind the nodes to the corresponding teaching activity type for subsequent content scheduling. 5.The knowledge graph driven textbook automatic generation method of claim 1, wherein The specific steps of compatibility analysis of the knowledge graph structure include: Extracting structural information including core paths, pre-requisites, hierarchical span, node density, and cognitive gradient from the target topology fingerprint; Mapping the target topology fingerprint to the corresponding nodes and paths in the source graph structure; Comparing the differences in path connectivity and pre-knowledge coverage between the two; If the necessary pre-node that is not covered by the source graph is detected in the target topology, mark it as a structural gap and trigger the path adjustment engine. 6.The knowledge graph driven textbook automatic generation method of claim 1, wherein, The reconstructed teaching path includes: For the detected structural gap, call the module matching the corresponding teaching goal and knowledge point label from the preset transition module library; If there is no matching module in the library, call the external knowledge unit generation component to construct a new transition module; Insert the supplemented transition module into the breakpoint position in the source graph path, and update the node order and dependency edge; The final output of the reconstructed path needs to satisfy the pre-logic constraints of the target topology fingerprint in the topology structure.

7. The knowledge graph driven textbook automatic generation method according to claim 1, wherein, The calling steps of the multi-modal content atomic unit include: Query 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 multi-modal data bound to it in turn, including text, image, formula, simulation animation, and interactive question components; Extract the metadata description of each content unit, including content type, difficulty label, and applicable terminal type, and uniformly package it as a multi-modal content package; Mount the multi-modal content package under the corresponding knowledge node for subsequent teaching material rendering and output module calling. 8.The knowledge graph driven textbook automatic generation method of claim 1, wherein, The content binding, consistency detection, and logic integration include: Maintain a set of cross-modal dependency rules for each type of knowledge node in the knowledge graph, which records the semantic dependency and presentation order between content units in the form of directed edges; After loading the multi-modal content package, read the dependency chain recorded in the graph to establish the logical binding relationship between the modal elements; Set threshold values based on the semantic vector similarity and visual-text matching degree of the modal content to judge the conflict degree; Perform elimination, replacement, or prompt operations on conflicting modalities to ensure that the final output of the multi-modal content is consistent in semantics and cognition. 9.The knowledge graph driven textbook automatic generation method of claim 1, wherein, The steps of outputting the teaching material ontology of the adaptive terminal include: Block and assemble multi-modal content and path structure according to the preset teaching module unit, each module including a main knowledge node and its associated content; According to the content density, modal distribution, and logical level of the teaching module, automatically select the matching layout template for layout; For modules containing graphs, texts, questions, and interactive controls, determine the presentation order of elements according to the priority of the cross-modal dependency graph; According to the terminal type identifier, select the matching content presentation strategy set and style template layout scheme; Structurally package the rendered teaching material content into a teaching material ontology object and output it as a format file that supports terminal rendering.

10. A knowledge graph driven textbook automatic generation system, characterized in that, ​ The teaching goal analysis module is configured to receive a teaching goal input, analyze the teaching goal into a multi-layer structure including a high-order goal, a core competence label, and an observable behavior index according to a preset teaching intention meta-model, and map the analyzed teaching goal into a starting node of a knowledge graph to construct a starting point of a teaching path. The teaching path generation module is configured to generate an optimal teaching path by calling a rule engine supporting dynamic reasoning and multi-objective optimization based on a topological position of the starting node and a teaching intention level. The teaching material system adaptation module is configured to generate a topological fingerprint of a target teaching material system, and perform compatibility analysis on the topological fingerprint and a knowledge graph structure. The multi-modal content scheduling and coordination module is configured to call bound multi-modal content atomic units based on the reconstructed path nodes, and complete content binding, consistency detection, and logical integration according to multi-dimensional modal coordination constraint rules defined by a cross-modal dependency graph, and in combination with semantic and visual consistency indexes. The teaching material rendering and output module is configured to perform structured rendering on the integrated teaching content, and output a teaching material ontology suitable for an adaptation terminal.

Images