Artificial intelligence proposition method and device, equipment and storage medium

Abstract

The invention discloses an artificial intelligence proposition method, device and equipment and a storage medium, and the method comprises the steps: constructing a multidisciplinary knowledge base based on an education normative document and a subject knowledge document and according to a multidisciplinary knowledge ontology framework; based on the structured semantics of the multidisciplinary knowledge base and the hierarchical semantics of the general semantic dictionary, calculating the comprehensive semantic similarity between the interdisciplinary knowledge points, and screening candidate knowledge point pairs according to the comprehensive semantic similarity; based on the comprehensive semantic similarity and a graph topological structure in the multidisciplinary knowledge base, performing association strength evaluation on the candidate knowledge point pairs through multi-hop path reasoning and graph neural network model conjoint analysis, and determining target knowledge point pairs meeting interdisciplinary proposition availability requirements; and returning the target knowledge point pair. Therefore, a complete technical path is provided for high-quality, explainable, compliant and controllable interdisciplinary intelligent propositions by constructing a multi-disciplinary knowledge base for education alignment and fusing structured semantics and map reasoning.

Description

Technical Field

[0001] This application relates to the technical field of intelligent question setting, and more specifically, to an artificial intelligence question setting method, apparatus, device, and storage medium. Background Technology

[0002] The application of artificial intelligence technology in the field of educational assessment is becoming increasingly in-depth. In particular, in the automatic generation of test questions, large language models (LLMs) are widely used for the automated construction of research questions or test content due to their powerful text generation and semantic understanding capabilities.

[0003] However, existing technologies mainly focus on the generation of questions within a single discipline, and their knowledge sources and semantic spaces are limited to specific fields, making it difficult to effectively support the intelligent generation of interdisciplinary integrated test questions. Summary of the Invention

[0004] In view of the above problems, this application proposes an artificial intelligence proposition method, apparatus, device and storage medium that can solve the above problems.

[0005] In a first aspect, embodiments of this application provide an artificial intelligence-based question-setting method. This method includes: constructing a multidisciplinary knowledge base by structurally extracting, semantically verifying, and graph-based storing knowledge content based on educational normative documents and subject knowledge literature, and according to a multidisciplinary knowledge ontology framework; calculating the comprehensive semantic similarity between various interdisciplinary knowledge points based on the structured semantics of the multidisciplinary knowledge base and the hierarchical semantics of a general semantic dictionary, and selecting candidate knowledge point pairs based on the comprehensive semantic similarity; evaluating the association strength of candidate knowledge point pairs through multi-hop path reasoning and a graph neural network model, based on the comprehensive semantic similarity and the graph topology in the multidisciplinary knowledge base, and determining target knowledge point pairs that meet the interdisciplinary question-setting usability requirements; and responding to an interdisciplinary question-setting request by returning the target knowledge point pairs to drive automatic question generation.

[0006] Secondly, embodiments of this application also provide an artificial intelligence-based question-setting device, comprising: a construction module for constructing a multidisciplinary knowledge base based on educational normative documents and subject knowledge literature, and according to a multidisciplinary knowledge ontology framework, performing structured extraction, semantic verification, and graph-based storage of knowledge content; a filtering module for calculating the comprehensive semantic similarity between interdisciplinary knowledge points based on the structured semantics of the multidisciplinary knowledge base and the hierarchical semantics of a general semantic dictionary, and filtering candidate knowledge point pairs based on the comprehensive semantic similarity; an evaluation module for evaluating the association strength of candidate knowledge point pairs based on the comprehensive semantic similarity and the graph topology in the multidisciplinary knowledge base, through multi-hop path reasoning and graph neural network model joint analysis, and determining target knowledge point pairs that meet the interdisciplinary question-setting usability requirements; and a response module for responding to interdisciplinary question-setting requests and returning target knowledge point pairs to drive automatic question generation.

[0007] Thirdly, embodiments of this application also provide a cooking device, including a processor, a memory, and one or more applications; the one or more applications are stored in the memory and configured to be executed by the processor to implement the aforementioned artificial intelligence proposition method.

[0008] Fourthly, embodiments of this application also provide a computer-readable storage medium storing program code, wherein the above-mentioned artificial intelligence proposition method is executed when the program code is run by a processor.

[0009] The technical solution provided in this application includes the following method: Based on educational normative documents and subject knowledge literature, and according to a multidisciplinary knowledge ontology framework, structured extraction, semantic verification, and graph-based storage of knowledge content are performed to construct a multidisciplinary knowledge base; based on the structured semantics of the multidisciplinary knowledge base and the hierarchical semantics of a general semantic dictionary, the comprehensive semantic similarity between various interdisciplinary knowledge points is calculated, and candidate knowledge point pairs are selected based on the comprehensive semantic similarity; based on the comprehensive semantic similarity and the graph topology in the multidisciplinary knowledge base, the association strength of candidate knowledge point pairs is evaluated through joint analysis of multi-hop path reasoning and graph neural network models to determine target knowledge point pairs that meet the usability requirements of interdisciplinary question setting; responding to interdisciplinary question setting requests, the target knowledge point pairs are returned to drive automatic question generation. Thus, by constructing an educationally aligned multidisciplinary knowledge base and integrating structured semantics and graph reasoning, a reliable transformation from "fragmented knowledge" to "propositional interdisciplinary relationships" is achieved, providing a complete technical path for high-quality, interpretable, compliant, and controllable interdisciplinary intelligent question setting. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments and drawings obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0011] Figure 1 A flowchart illustrating an artificial intelligence proposition method provided in an embodiment of this application is shown.

[0012] Figure 2 A schematic diagram of the structure of an artificial intelligence proposition device provided in an embodiment of this application is shown.

[0013] Figure 3 A schematic diagram of the structure of an artificial intelligence proposition device provided in an embodiment of this application is shown.

[0014] Figure 4 This illustration shows a schematic diagram of the structure of a computer-readable storage medium provided in an embodiment of this application. Detailed Implementation

[0015] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0016] The application of artificial intelligence technology in the field of educational assessment is becoming increasingly in-depth. In particular, in the automatic generation of test questions, large language models (LLMs) are widely used for the automated construction of research questions or test content due to their powerful text generation and semantic understanding capabilities.

[0017] However, existing technologies mainly focus on the generation of questions within a single discipline, and their knowledge sources and semantic spaces are limited to specific fields, making it difficult to effectively support the intelligent generation of interdisciplinary integrated test questions.

[0018] To address the aforementioned issues, this application provides an artificial intelligence-based question-setting method, apparatus, device, and storage medium. The method includes: constructing a multidisciplinary knowledge base by structurally extracting, semantically verifying, and graph-based storing knowledge content based on educational normative documents and subject-specific knowledge literature, and according to a multidisciplinary knowledge ontology framework; calculating the comprehensive semantic similarity between interdisciplinary knowledge points based on the structured semantics of the multidisciplinary knowledge base and the hierarchical semantics of a general semantic dictionary, and selecting candidate knowledge point pairs based on the comprehensive semantic similarity; evaluating the association strength of candidate knowledge point pairs through multi-hop path reasoning and a graph neural network model, based on the comprehensive semantic similarity and the graph topology in the multidisciplinary knowledge base, to determine target knowledge point pairs that meet the interdisciplinary question-setting usability requirements; and responding to interdisciplinary question-setting requests by returning target knowledge point pairs to drive automatic question generation.

[0019] Thus, by constructing an educationally aligned multidisciplinary knowledge base and integrating structured semantics and graph reasoning, a reliable transformation from "fragmented knowledge" to "propositional interdisciplinary relationships" is achieved, providing a complete technical path for high-quality, explainable, compliant and controllable interdisciplinary intelligent propositions.

[0020] Please see Figure 1 , Figure 1 A flowchart illustrating an artificial intelligence proposition method provided in an embodiment of this application is shown. Figure 1 As shown, the method may include steps 110 to 140.

[0021] In step 110, based on educational normative documents and subject knowledge literature, and in accordance with the multidisciplinary knowledge ontology framework, the knowledge content is structurally extracted, semantically verified, and stored in a graph format to construct a multidisciplinary knowledge base.

[0022] In some implementations, educational normative documents and subject matter literature include textbooks, curriculum standards, examination syllabi, and academic papers.

[0023] In some implementations, a multidisciplinary knowledge ontology framework can be a structured semantic model designed to meet the needs of interdisciplinary teaching and intelligent question setting in basic education, used to uniformly describe the core concepts, attributes and their interrelationships in different disciplines.

[0024] In one specific implementation, the multidisciplinary knowledge ontology framework is constructed using a combination of top-down and bottom-up approaches.

[0025] The top-down approach is based on national curriculum standards, subject core competency systems, and educational theories, predefining top-level subject classifications (such as physics, biology, and mathematics), knowledge dimensions (such as concepts, principles, and methods), relationship types (such as "applied to," "dependent on," and "analogous to"), and interdisciplinary mapping rules.

[0026] The bottom-up approach extracts high-frequency knowledge points and relationship patterns from authentic educational texts such as textbooks and academic papers, and then adds them to the ontology after expert verification to ensure that it covers the actual teaching content.

[0027] The multidisciplinary knowledge ontology framework not only standardizes entity types (e.g., "function" belongs to mathematics, "photosynthesis" belongs to biology) and relational semantics (e.g., "the law of conservation of energy applied to collision problems"), but also explicitly defines the legitimacy constraints of cross-disciplinary associations (e.g., "propositional associations" are only allowed when two knowledge points overlap in cognitive level or application scenario). Therefore, the multidisciplinary knowledge ontology framework serves as the semantic constraint foundation for the entire system.

[0028] In some implementations, the multidisciplinary knowledge base can be a structured collection of knowledge extracted and verified from educational normative documents (such as curriculum standards and examination outlines) and subject knowledge literature (such as textbooks, teaching aids, and academic papers) based on the aforementioned multidisciplinary knowledge ontology framework, and stored in a graph database in the form of a knowledge graph.

[0029] In one specific implementation, the multidisciplinary knowledge base includes entity nodes, relation edges, and metadata attributes.

[0030] Entity nodes include effective knowledge points (such as "Newton's Second Law" and "exponential functions"), ability requirement nodes (from the curriculum standards, such as "scientific inquiry" and "mathematical modeling"), and high-frequency test points (from the examination syllabus).

[0031] Relationship edges include intradisciplinary logical relationships between knowledge points (e.g., "contains", "derives"), interdisciplinary semantic links (e.g., "applies to", "analogous to"), and mapping relationships between knowledge points and abilities / exam points.

[0032] Metadata attributes include the subject, education stage, cognitive level, and interdisciplinary tags for each entity node, and the relationship type, confidence level, and source basis for each relationship edge.

[0033] Knowledge points scattered across diverse and heterogeneous texts such as textbooks, curriculum standards, examination syllabi, and academic papers are uniformly incorporated into a structured representation constrained by a multidisciplinary knowledge ontology, forming an organic knowledge network covering multiple disciplines, levels, and sources. Then, through structured extraction, semantic verification, and graph-based storage of knowledge content, the construction of a multidisciplinary knowledge base is completed. Specifically, in some implementations, the step "based on educational normative documents and subject knowledge literature, and according to the multidisciplinary knowledge ontology framework, structured extraction, semantic verification, and graph-based storage of knowledge content to construct a multidisciplinary knowledge base" may include the following steps:

[0034] (1) Based on educational normative documents and subject knowledge literature, and combined with the semantic modeling requirements of interdisciplinary intelligent propositions, a multidisciplinary knowledge ontology framework is constructed using a combination of top-down and bottom-up approaches.

[0035] (2) Based on the multidisciplinary knowledge ontology framework, structured knowledge units that conform to predefined entity types and relation types are extracted from educational normative documents and subject knowledge literature. After semantic standardization verification, the structured knowledge units are stored in the graph database in the form of nodes and edges to construct a multidisciplinary knowledge graph, so as to complete the construction of the multidisciplinary knowledge base.

[0036] When generating interdisciplinary test questions, the system needs to ensure that the combined knowledge points are semantically related, logically integrated, and cognitively understandable. Therefore, when constructing a multidisciplinary knowledge ontology framework, the semantic modeling requirements are internalized into the structural elements and constraint rules of the ontology, so that the multidisciplinary knowledge ontology framework is not only a knowledge classification system, but also a semantic infrastructure that supports the rationality, controllability, and interpretability of intelligent test questions.

[0037] In some implementations, the semantic modeling requirements for interdisciplinary intelligent question generation can address the structural and semantic requirements on the underlying knowledge representation needed to achieve the automatic generation of high-quality, interpretable, and pedagogically compliant interdisciplinary test questions. These requirements collectively determine the design goals and structural characteristics of the multidisciplinary knowledge ontology framework. This may include:

[0038] (1) It can explicitly distinguish the subject, cognitive level and core competency dimension of the knowledge point;

[0039] (2) Support the definition of interdisciplinary semantic relation types with educational significance (such as “applied to”, “analogous to”, “reflecting common literacy”), rather than just weak associations based on word co-occurrence;

[0040] (3) Constrain the legality of knowledge point combinations to ensure that the generated interdisciplinary connections are based on the curriculum standards and are logically valid in teaching;

[0041] (4) Provide a traceable reasoning path so that the system can explain "why these two knowledge points can be combined to create questions".

[0042] In some implementations, a structured knowledge unit can be the smallest unit of knowledge representation extracted from the original educational text, conforming to a predefined semantic type within a multidisciplinary knowledge ontology framework. It takes the form of a triple or attribute-value pair with a clearly defined semantic role. Each structured knowledge unit includes entity nodes, relation edges, and attribute information, as detailed above.

[0043] In some implementations, semantic specification verification refers to the process of automatically and manually verifying whether structured knowledge units conform to the semantic rules of a multidisciplinary knowledge ontology framework after knowledge extraction and before being stored in the database. This aims to ensure the educational compliance and logical consistency of knowledge.

[0044] In some implementations, a multidisciplinary knowledge graph can be a directed semantic network that organizes structured knowledge units that have been semantically validated, using a graph database (such as Neo4j or NebulaGraph) as the storage medium.

[0045] The nodes of the multidisciplinary knowledge graph represent various educational entities (including knowledge points, literacy, test points, etc.), and the edges represent predefined semantic relationships, with attached type labels, directions, and confidence weights.

[0046] Multidisciplinary knowledge graphs not only fully preserve the knowledge systems within each discipline (such as the evolution path of "mechanics - electromagnetism" in physics), but also explicitly model cross-disciplinary semantic links (such as "population growth" in biology - "differential equations" in mathematics), forming a heterogeneous knowledge network that supports efficient graph traversal, multi-hop reasoning, and neighborhood aggregation.

[0047] As the underlying data structure of a multidisciplinary knowledge base, the multidisciplinary knowledge graph provides a unified, dynamic, and scalable knowledge engine for semantic similarity calculation, cross-disciplinary relationship discovery, and intelligent test question generation.

[0048] So how do we construct a multidisciplinary knowledge graph? In some implementations, the step "based on a multidisciplinary knowledge ontology framework, extracting structured knowledge units that conform to predefined entity types and relation types from educational normative documents and subject knowledge literature, and after semantic standardization verification, storing the structured knowledge units in the form of nodes and edges into a graph database to construct a multidisciplinary knowledge graph" may include the following steps:

[0049] (1) Use pre-trained language models to perform word segmentation, part-of-speech tagging and named entity recognition on unstructured texts in textbooks and academic papers, and extract candidate knowledge points that conform to the predefined entity types in the multidisciplinary knowledge ontology framework as knowledge points to be annotated.

[0050] (2) The BIO sequence annotation method is used to annotate the knowledge points to be annotated, and the annotation results are obtained;

[0051] (3) If the annotation result conforms to the semantic specification, the corresponding knowledge point to be annotated shall be determined as a valid knowledge point;

[0052] (4) Parse the curriculum standards into a structured format, and establish semantic associations between the various ability requirement nodes in the curriculum standards and effective knowledge points based on the predefined subject attributes and literacy mapping rules in the multidisciplinary knowledge ontology framework.

[0053] (5) Identify and extract high-frequency test points and their corresponding question types from the examination syllabus, and establish semantic links between high-frequency test points and effective knowledge points based on the predefined relationship types in the multidisciplinary knowledge ontology framework.

[0054] (6) Valid knowledge points are used as entity nodes, ability requirement nodes and high-frequency test points are used as auxiliary nodes, and semantic associations and semantic links are used as relation edges. These are stored in a graph database to construct a multidisciplinary knowledge graph.

[0055] In some implementations, a pre-trained language model can refer to a neural network language model that is pre-trained on a large-scale general corpus and possesses deep contextual semantic understanding capabilities. Pre-trained language models learn distributed semantic representations of words, phrases, and sentences through self-supervised learning (such as masked language modeling and next-sentence prediction tasks), effectively capturing the contextual features and subject affiliation of knowledge points in educational texts.

[0056] In one specific implementation, the pre-trained language model may be a pre-trained model based on the Transformer architecture (such as BERT, RoBERTa, or their educational-domain fine-tuned versions), but is not limited thereto; any pre-trained language model capable of achieving context-aware semantic representation and sequence labeling functions is applicable to the technical solution of this invention.

[0057] The BIO (Begin-Inside-Outside) sequence labeling method is used to perform fine-grained boundary and type annotation on the knowledge points output by the pre-trained language model, generating structured annotation results. This annotation process not only determines the start and end positions of the knowledge points, but also clarifies their semantic categories in the ontology, ensuring consistency in subsequent processing.

[0058] Furthermore, if the type, contextual semantics, or subject affiliation of a knowledge point to be labeled is consistent with the definition in the multidisciplinary knowledge ontology framework (for example, "exponential function" is correctly classified as a mathematical "concept" rather than a physical "phenomenon"), then it is confirmed as a valid knowledge point; otherwise, it is removed or submitted to experts for review.

[0059] Simultaneously, the system performs structured parsing of the curriculum standard document, extracting competency requirements (such as "being able to use function models to describe real-world problems" and "possessing scientific inquiry awareness"), and establishing semantic associations between these competency requirement nodes and confirmed valid knowledge points based on predefined subject attributes and competency mapping rules in the multidisciplinary knowledge ontology framework. For example, the "mathematical modeling" competency node is linked to knowledge points such as "linear function" and "exponential growth model," forming a "competency-knowledge" mapping network.

[0060] In addition, the system automatically identifies high-frequency test points (such as "application of Newton's laws of motion" and "calculation of genetic laws") and their corresponding question types (such as multiple choice questions and comprehensive questions) from the examination syllabus. Based on the predefined relationship types in the ontology (such as "examined in" and "key coverage"), it establishes semantic links between high-frequency test points and related effective knowledge points, thereby integrating test-oriented information into the knowledge system.

[0061] Ultimately, effective knowledge points are used as core entity nodes, ability requirement nodes in the curriculum standards and high-frequency test points in the examination syllabus are used as auxiliary nodes, and the above-mentioned "competency-knowledge" associations and "test point-knowledge" links are used as type-labeled relation edges. All of these are uniformly stored in a graph database (such as Neo4j) to construct a multidisciplinary knowledge graph that supports interdisciplinary reasoning and intelligent question setting.

[0062] Therefore, it can be seen that multidisciplinary knowledge graphs not only include the knowledge structure within a discipline, but also integrate competency goals and examination syllabus requirements, becoming an integrated knowledge infrastructure that connects teaching content, ability development, and assessment needs.

[0063] In other implementations, the step "based on a multidisciplinary knowledge ontology framework, extracting structured knowledge units that conform to predefined entity types and relation types from educational normative documents and subject knowledge literature, and after semantic standardization verification, storing the structured knowledge units in the form of nodes and edges into a graph database to construct a multidisciplinary knowledge graph" may include the following steps:

[0064] (1) Using a pre-defined joint extraction model, entities and relationships between entities are extracted from structured and unstructured text constrained by a multidisciplinary knowledge ontology framework, generating triples that conform to the predefined entity types and relationship types in the multidisciplinary knowledge ontology framework.

[0065] (2) Verify the triples, and store the entities in the verified triples as nodes and their corresponding relationships as edges in the graph database to construct a multidisciplinary knowledge graph.

[0066] In some implementations, the pre-defined joint extraction model can be an end-to-end deep neural network architecture (such as a Span-based or sequence-to-sequence joint extraction model), whose training objective is to simultaneously identify entities and their relationships in the text in a single inference, avoiding the error propagation problem in traditional pipeline methods.

[0067] The pre-defined joint extraction model incorporates the constraints of a multidisciplinary knowledge ontology framework during the training phase, using predefined entity types (such as "concept", "law", "literacy") and relation types (such as "applied to", "embodies literacy", "depends on") in the ontology as a label space to ensure that the output results naturally conform to educational semantic norms.

[0068] The input text for the pre-defined joint extraction model includes both structured and unstructured text. Structured text can be content organized by chapter or item in curriculum standards or examination outlines, which, although formatted, still requires semantic parsing; unstructured text can be free text such as textbook text, academic papers, and supplementary teaching materials.

[0069] The pre-defined joint extraction model automatically extracts triples in the form of (head entity, relation, tail entity) from structured and unstructured text. For example, ("photosynthesis", "depends on", "light intensity"); or ("exponential function", "applies to", "population growth model"), and all entities and relations strictly match the predefined types in the ontology framework.

[0070] In some implementations, multi-dimensional verification includes type validity verification, semantic consistency verification, and educational rationality review.

[0071] For example, for type validity verification, we check whether the entity category and relationship are within the range of combinations allowed by the ontology (e.g., the "competency" node cannot be used as the subject of "applied to"); for semantic consistency verification, we use a pre-trained language model to calculate the semantic similarity between the context of the triple and the original sentence to filter out extraction biases; for educational rationality review, we perform rule matching or expert sampling review on interdisciplinary triples (e.g., physics × biology) to ensure that they conform to teaching logic.

[0072] Only triples that pass all verification stages are retained: entities are created as nodes in a graph database, labeled with attributes such as type, discipline, and origin; relations are created as edges with direction and type, and confidence weights can be attached. Ultimately, all valid triples together constitute a multidisciplinary knowledge graph, serving as the knowledge infrastructure supporting interdisciplinary intelligent propositions.

[0073] The two methods for constructing multidisciplinary knowledge graphs provided in this application both use a multidisciplinary knowledge ontology framework as the core of semantic constraints, ensuring that the constructed knowledge graphs meet the needs of interdisciplinary intelligent question setting in terms of type system, relationship legitimacy, and pedagogical rationality. The system can flexibly select or combine the two methods based on data sources, accuracy requirements, and computing resources, achieving an organic unity of accuracy, completeness, and scalability in knowledge graph construction.

[0074] Therefore, by constructing a multidisciplinary knowledge base, the problem of fragmented educational knowledge can be solved, multi-source integration can be achieved, and intelligent discovery and verification of interdisciplinary relationships can be supported. This also ensures the educational standardization and authority of the test content, thus providing the system with an efficient and reasonable knowledge engine.

[0075] In step 120, based on the structured semantics of the multidisciplinary knowledge base and the hierarchical semantics of the general semantic dictionary, the comprehensive semantic similarity between each interdisciplinary knowledge point is calculated, and candidate knowledge point pairs are selected according to the comprehensive semantic similarity.

[0076] In some implementations, the structured semantics of a multidisciplinary knowledge base can refer to the semantic information contained in knowledge stored in the form of a multidisciplinary knowledge graph, which is explicitly defined by the ontology framework.

[0077] In some implementations, the hierarchical semantics of a general semantic dictionary can refer to the hierarchical semantic relationships such as superordinate, synonym, and antonym embodied in a lexical semantic network organized from external authoritative language resources (such as WordNet, HowNet, or Chinese thesaurus).

[0078] For example, in WordNet, "mammal" is a hypernym of "dog," and "photosynthesis" and "respiration" may belong to the same semantic field. These dictionaries encode the semantic distance and abstraction level between words through tree-like or graph-like structures, providing cross-domain semantic alignment capabilities for interdisciplinary knowledge points. This is particularly suitable for situations where different disciplines use different terms but express similar concepts (such as "rate" and "speed," "energy conversion" and "work").

[0079] In some implementations, comprehensive semantic similarity can refer to a unified similarity measure between cross-disciplinary knowledge point pairs calculated by integrating the structured semantics of a multidisciplinary knowledge base and the hierarchical semantics of a general semantic dictionary. Comprehensive semantic similarity is obtained by weighted fusion of two components: graph semantic similarity and dictionary semantic similarity.

[0080] In some implementations, candidate knowledge point pairs can refer to binary pairs (such as <photosynthesis, energy conservation> and <function graph, enzyme activity curve>) that are retained after preliminary screening from all possible interdisciplinary knowledge point combinations and have potential propositional value.

[0081] By initially screening out candidate knowledge point pairs with potential propositional value, high-quality input is provided for subsequent deep correlation evaluation. Specifically, in some implementations, the step "based on comprehensive semantic similarity and the graph topology in a multidisciplinary knowledge base, through joint analysis of multi-hop path reasoning and graph neural network models, the correlation strength of candidate knowledge point pairs is evaluated to determine target knowledge point pairs that meet the requirements of interdisciplinary propositional usability" may include the following steps:

[0082] (1) Train the embedding model based on the entity relationship triples in the multidisciplinary knowledge base to obtain the vector representation of each interdisciplinary knowledge point, and calculate the graph similarity between any two vector representations;

[0083] (2) Based on a general semantic dictionary, semantic analysis is performed on each interdisciplinary knowledge point to obtain its corresponding hierarchical semantic representation, and the hierarchical similarity between any two hierarchical semantic representations is calculated.

[0084] (3) Multi-source fusion of graph similarity and hierarchical similarity to generate comprehensive semantic similarity;

[0085] (4) Based on the comprehensive semantic similarity, select target knowledge point pairs that meet the requirements of cross-disciplinary proposition association.

[0086] In some implementations, the entity relation triple training embedding model can utilize structured knowledge triples (i.e., semantic units of the form <head entity, relation, tail entity>, such as <photosynthesis, depends on, light intensity>, <Newton's second law, applied to, inclined plane problem>) stored in a multidisciplinary knowledge base. Through Knowledge Graph Embedding (KGE) technology, discrete knowledge point symbols are mapped into continuous low-dimensional vector representations (i.e., embedding vectors), so that the semantic relations in the original graph can be approximately expressed in the vector space through geometric operations.

[0087] In one specific implementation, the core idea of ​​knowledge graph embedding is to represent relations as translation operations of entity vectors, and the corresponding loss function can be defined as:

[0088] L=max(0,γ+||h+rt||-||h ′ +rt ′ ||)

[0089] Where γ is a hyperparameter, h is the head entity vector, r is the relation vector, t is the tail entity vector, and h ′ These are negative sample entities.

[0090] The system employs a pre-defined knowledge graph embedding model (such as TransE, DistMult, RotatE, or GraphSAGE) and uses all valid triples from a multidisciplinary knowledge base as training data. By minimizing a loss function (such as negative sampling cross-entropy or ranking loss), it learns the vector representation of each knowledge point's entity and relation type. After training, the graph similarity between any two knowledge points can be calculated using methods such as cosine similarity, Euclidean distance, or inner product between their vectors. This similarity effectively reflects the structural proximity, functional relevance, and potential interdisciplinary connections between the two knowledge points within the educational knowledge system.

[0091] The system invokes a general semantic dictionary (such as WordNet, HowNet, or a Chinese thesaurus) to perform semantic parsing on the textual representations of various interdisciplinary knowledge points. For example, it maps "power" to its synonym set and superordinate concepts in the dictionary (such as "physical quantity" → "energy conversion rate"). Based on the hierarchical structure defined in the dictionary (such as superordinate / hyperordinate, whole-part, and synonym relationships), it uses path length, information content (IC), or vector averaging methods to generate hierarchical semantic representations for each knowledge point and calculates the hierarchical similarity between any two representations.

[0092] Hierarchical similarity captures the semantic generalization ability of interdisciplinary terms at the level of a common language, and is particularly applicable to situations where different disciplines use different terms to express similar concepts (such as "metabolic rate" and "power").

[0093] The graph similarity and hierarchical similarity are weighted and fused to generate a unified comprehensive semantic similarity. The comprehensive semantic similarity value ranges from 0 to 1. The higher the value, the more likely the two knowledge points are to form an interdisciplinary connection with pedagogical significance.

[0094] For example, the final association score = α * hierarchical similarity + β * graph similarity. Here, α and β are weighting parameters. In one specific implementation, α is set to 0.6 and β to 0.4.

[0095] Finally, based on comprehensive semantic similarity, the system selects knowledge point pairs with comprehensive semantic similarity greater than or equal to a preset threshold (e.g., 0.65) and that meet the requirements for cross-disciplinary proposition association as target knowledge point pairs.

[0096] The requirements for question association can include: the subjects belong to a permissible cross-disciplinary combination (such as physics-biology, mathematics-geography), at least one subject covers high-frequency test points in the syllabus, and the combination constraints in the ontology are not violated. The screening results constitute the input set for subsequent in-depth evaluation of association strength, significantly reducing the search space and improving the efficiency and quality of question creation.

[0097] However, while screening mechanisms based on comprehensive semantic similarity can effectively identify cross-disciplinary knowledge point pairs with similar semantics, they mainly rely on static semantic representations (such as vector similarity and dictionary hierarchy) and have not fully explored the dynamic structural information and multi-level reasoning paths contained in the knowledge graph. For example, two knowledge points may not have high semantic similarity, but they can form a deep teaching connection through a multi-hop path of "physics → mathematics → biology"; conversely, some highly similar pairs may lack real teaching logic support only because of word co-occurrence.

[0098] In step 130, based on the comprehensive semantic similarity and the graph topology in the multidisciplinary knowledge base, the candidate knowledge point pairs are evaluated for correlation strength through joint analysis of multi-hop path reasoning and graph neural network model, and the target knowledge point pairs that meet the requirements of interdisciplinary proposition usability are determined.

[0099] Based on the initial screening, the system comprehensively considers semantic similarity and the graph topology in the multidisciplinary knowledge base. Through joint analysis of multi-hop path reasoning and graph neural network models, it quantitatively evaluates the correlation strength of candidate knowledge point pairs.

[0100] Among them, multi-hop path reasoning is used to discover the teaching logic links between two points (such as interdisciplinary bridges formed through intermediate concepts), while graph neural networks are used to capture the contextual semantic representation of nodes in the global knowledge network; the two work together to effectively identify high-quality associations that have solid teaching reasoning paths but have moderate semantic similarity.

[0101] Finally, based on the evaluation results, pairs of target knowledge points that meet the requirements for interdisciplinary question usability (such as logical coherence, integration of literacy, and support for curriculum standards) are selected as reliable input for automatic question generation.

[0102] Specifically, in some implementations, the step "based on comprehensive semantic similarity and the graph topology in a multidisciplinary knowledge base, through joint analysis of multi-hop path reasoning and graph neural network models, to evaluate the correlation strength of candidate knowledge point pairs and determine target knowledge point pairs that meet the requirements of interdisciplinary proposition usability" may include the following steps:

[0103] (1) Based on the knowledge graph topology, perform multi-hop path search on candidate knowledge point pairs, identify potential reasoning paths connecting two candidate knowledge point pairs, and calculate the path weight of each reasoning path.

[0104] (2) Input the candidate knowledge point pairs into the graph neural network model, generate node representations by aggregating neighborhood information, and calculate the structure-aware similarity based on the node representations;

[0105] (3) The comprehensive semantic similarity and structurally perceived similarity are fused to obtain the final association score;

[0106] (4) When the final association score exceeds the preset threshold, it is determined that the candidate knowledge point pair has an implicit cross-disciplinary relationship that can be propositioned, and it is identified as the target knowledge point pair.

[0107] Based on the graph topology of a multidisciplinary knowledge base, a multi-hop path search (such as 2-hop or 3-hop) is performed on each pair of candidate knowledge points to identify all potential reasoning paths connecting the two (e.g., "photosynthesis → energy conversion → power").

[0108] Each path consists of a series of intermediate nodes and relational edges. The path weight is calculated based on factors such as path length, the pedagogical rationality of relation type (e.g., "applied to" has a higher weight than "related to"), and the authority of intermediate nodes (e.g., whether they come from core concepts in the curriculum standards). This is used to measure the credibility of the path as an interdisciplinary teaching logic chain.

[0109] Candidate knowledge point pairs are input into a pre-trained graph neural network model (such as GraphSAGE or GAT). Through multi-layer neighborhood information aggregation, node vector representations that integrate local and global contexts are generated. Based on this, the structure-aware similarity (such as cosine similarity) between two node representations is calculated to capture their structural associations in the knowledge network, even if there is no direct edge connection between them.

[0110] The aforementioned comprehensive semantic similarity (derived from semantic embedding and dictionary level) is weighted and fused with the currently obtained structure-aware similarity (using fixed weights or adaptive gating mechanisms) to generate a final association score that reflects multidimensional evidence.

[0111] If the final association score exceeds the preset value (e.g., 0.75), and the corresponding reasoning path meets the requirements for interdisciplinary question availability (e.g., does not cross unrelated disciplines, reflects common core competencies, and is supported by curriculum standards or examination outlines), then it is determined that the candidate knowledge point pair has an implicit interdisciplinary relationship that can be used for question setting, and it is identified as the target knowledge point pair for automatic generation of subsequent test questions.

[0112] Thus, the system has constructed a high-quality set of target knowledge point pairs, each of which has undergone triple verification in terms of semantics, structure, and pedagogical rationality, demonstrating clear potential for interdisciplinary question setting. This set, as a reusable knowledge asset, is persistently stored in the knowledge base for subsequent intelligent question setting tasks.

[0113] When the system receives a user-initiated interdisciplinary question request (e.g., specifying "integration of high school physics and biology," "assessing scientific inquiry literacy," or a limited theme such as "energy conversion"), it retrieves matching items from the set of target knowledge point pairs. The system then encapsulates the matching target knowledge point pairs and their associated metadata (including reasoning paths, associated scores, and relevant core competencies) into a structured response and returns it to the question generation engine. This response directly drives the construction of the question stem logic, the design of the question format, and the generation of reference answers, thereby achieving automated generation of interdisciplinary questions with knowledge relationships at their core and educational goals as their guide.

[0114] In step 140, in response to the interdisciplinary question-setting request, the target knowledge point pair is returned to drive the automatic generation of test questions.

[0115] After constructing and validating the target knowledge point pairs, the system enters the service phase. When it receives an externally initiated interdisciplinary question request (such as specifying a subject combination, competency dimension, topic scope, or difficulty level), the system retrieves matching items from the determined set of target knowledge point pairs based on the request parameters, and returns the selected knowledge point pairs and their associated metadata (including reasoning path, correlation strength, core competencies involved, and curriculum standards) in a structured form.

[0116] The returned result is directly used as input to the automatic question generation module to drive the construction of question stem logic, scenario design, question generation, and answer derivation, thereby achieving efficient, compliant, and interpretable interdisciplinary intelligent question creation. Specifically, in some implementations, the step "responding to the interdisciplinary question creation request and returning the target knowledge point pair to drive automatic question generation" may include the following steps:

[0117] (1) Responding to interdisciplinary question requests, retrieve matching knowledge point pairs from the target knowledge point pairs;

[0118] (2) Based on the matched knowledge point pairs and their associated metadata, the test questions are automatically generated.

[0119] When receiving a request for an interdisciplinary question from a user or upper-level application (e.g., specifying "integration of high school physics and biology," "examining mathematical modeling and scientific inquiry literacy," or a limited theme such as "energy and ecosystems"), the system analyzes the constraints in the request (including subject combination, educational stage, core competencies, theme keywords, difficulty level, etc.) and efficiently retrieves matching knowledge point pairs that meet the conditions from the pre-constructed set of target knowledge point pairs. Each pair in this set has undergone triple verification through semantic similarity, graph topology, and pedagogical rationality to ensure that it possesses genuine interdisciplinary question value.

[0120] The system takes the matched knowledge point pairs and their associated metadata (including reasoning paths, relationship types, association strength scores, relevant curriculum standard items, core competency mappings, and typical application scenarios) as structured input and passes them to the test question generation engine. Based on preset question rule templates, natural language generation models, or large language models, and combined with the teaching logic links in the metadata, the engine automatically generates the question stem, question format, options (if applicable), and reference answer, thereby outputting a semantically coherent, logically sound, and educationally compliant interdisciplinary test question.

[0121] Please see Figure 2 , Figure 2 This illustration shows a structural diagram of an artificial intelligence question-posing device 200 provided in an embodiment of this application. The artificial intelligence question-posing device 200 includes: a construction module 210, a screening module 220, an evaluation module 230, and a response module 240. Specifically:

[0122] Module 210 is used to construct a multidisciplinary knowledge base based on educational normative documents and subject knowledge literature, and in accordance with a multidisciplinary knowledge ontology framework, to perform structured extraction, semantic verification, and graph-based storage of knowledge content.

[0123] The filtering module 220 is used to calculate the comprehensive semantic similarity between various interdisciplinary knowledge points based on the structured semantics of the multidisciplinary knowledge base and the hierarchical semantics of the general semantic dictionary, and to filter candidate knowledge point pairs according to the comprehensive semantic similarity.

[0124] Evaluation module 230 is used to evaluate the correlation strength of the candidate knowledge point pairs based on the comprehensive semantic similarity and the graph topology in the multidisciplinary knowledge base, through joint analysis of multi-hop path reasoning and graph neural network model, and to determine the target knowledge point pairs that meet the requirements of interdisciplinary proposition usability.

[0125] The response module 240 is used to respond to interdisciplinary question requests and return the target knowledge point pair to drive the automatic generation of test questions.

[0126] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described device and module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0127] In the several embodiments provided in this application, the coupling or direct coupling or communication connection between the modules shown or discussed may be an indirect coupling or communication connection through some interface, device or module, and may be electrical, mechanical or other forms.

[0128] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0129] Please see Figure 3 , Figure 3 The diagram shows a structural schematic of an artificial intelligence proposition device provided in an embodiment of this application. The artificial intelligence proposition device 300 in this application may include one or more of the following components: processor 310, memory 320, and one or more application programs. The one or more application programs may be stored in memory 320 and configured to be executed by one or more processors 310. The one or more programs are configured to execute the artificial intelligence proposition method as described in the foregoing method embodiments.

[0130] Processor 310 may include one or more processing cores. Processor 310 connects to various parts within the entire artificial intelligence testing device 300 using various interfaces and lines. It executes various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 320, and by calling data stored in memory 520. Optionally, processor 310 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). Processor 310 may integrate one or more of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the displayed content; and the modem handles wireless communication. It is understood that the modem may also not be integrated into processor 310 and may be implemented separately using a communication chip.

[0131] The memory 320 may include random access memory (RAM) or read-only memory (ROM). The memory 320 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 320 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for implementing at least one function, instructions for implementing the various method embodiments described below, etc. The data storage area may also store data created by the artificial intelligence problem-solving device 300 during use.

[0132] Please see Figure 4 , Figure 4 The diagram illustrates the structure of a computer-readable storage medium 400 provided in an embodiment of this application. The computer-readable storage medium 400 stores program code, which can be called by a processor to execute the artificial intelligence proposition method described in the above method embodiment.

[0133] The computer-readable storage medium 400 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Optionally, the computer-readable storage medium 400 includes a non-transitory computer-readable storage medium. The computer-readable storage medium 400 has storage space for program code 410 that performs any of the method steps described above. This program code can be read from or written to one or more computer program devices. The program code 410 may be compressed, for example, in a suitable form.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An artificial intelligence proposition method, characterized in that, The method includes: Based on educational normative documents and subject knowledge literature, and in accordance with the multidisciplinary knowledge ontology framework, the knowledge content is extracted in a structured manner, semantically verified, and stored in a graph-like manner to construct a multidisciplinary knowledge base; Based on the structured semantics of the multidisciplinary knowledge base and the hierarchical semantics of the general semantic dictionary, the comprehensive semantic similarity between each interdisciplinary knowledge point is calculated, and candidate knowledge point pairs are selected according to the comprehensive semantic similarity. Based on the comprehensive semantic similarity and the graph topology in the multidisciplinary knowledge base, the candidate knowledge point pairs are evaluated for correlation strength through joint analysis of multi-hop path reasoning and graph neural network model, and target knowledge point pairs that meet the requirements of interdisciplinary proposition usability are determined. In response to interdisciplinary question requests, the system returns the target knowledge point pairs to drive the automatic generation of test questions.

2. The artificial intelligence proposition method according to claim 1, characterized in that, The aforementioned method, based on educational normative documents and subject-specific knowledge literature, and in accordance with a multidisciplinary knowledge ontology framework, performs structured extraction, semantic verification, and graph-based storage of knowledge content to construct a multidisciplinary knowledge base, including: Based on the aforementioned educational normative documents and subject knowledge literature, and combined with the semantic modeling requirements of interdisciplinary intelligent propositions, a multidisciplinary knowledge ontology framework is constructed using a combination of top-down and bottom-up approaches. Based on the multidisciplinary knowledge ontology framework, structured knowledge units conforming to predefined entity types and relation types are extracted from the educational normative documents and subject knowledge literature. After semantic standardization verification, the structured knowledge units are stored in a graph database in the form of nodes and edges to construct a multidisciplinary knowledge graph, thereby completing the construction of the multidisciplinary knowledge base.

3. The artificial intelligence proposition method according to claim 2, characterized in that, The educational normative documents and subject knowledge literature include textbooks, curriculum standards, examination syllabi, and academic papers; Based on the multidisciplinary knowledge ontology framework, structured knowledge units conforming to predefined entity types and relation types are extracted from the educational normative documents and subject knowledge literature. After semantic standardization verification, the structured knowledge units are stored in a graph database in the form of nodes and edges to construct a multidisciplinary knowledge graph, including: Using a pre-trained language model, unstructured text in the textbooks and academic papers is segmented, part-of-speech tagging is performed, and named entity recognition is performed. Candidate knowledge points that conform to the predefined entity types in the multidisciplinary knowledge ontology framework are extracted as knowledge points to be annotated. The BIO sequence labeling method was used to label the knowledge points to be labeled, and the labeling results were obtained. If the annotation result conforms to the semantic specification, then the corresponding knowledge point to be annotated is determined as a valid knowledge point; The curriculum standards are parsed into a structured format, and semantic associations are established between each ability requirement node in the curriculum standards and the effective knowledge points based on the predefined subject attributes and literacy mapping rules in the multidisciplinary knowledge ontology framework. Identify and extract high-frequency test points and their corresponding question types from the examination syllabus, and establish semantic links between the high-frequency test points and the effective knowledge points based on the predefined relationship types in the multidisciplinary knowledge ontology framework. The effective knowledge points are used as entity nodes, the ability requirement nodes and the high-frequency test points are used as auxiliary nodes, and the semantic associations and semantic links are used as relation edges. These are stored in a graph database to construct the multidisciplinary knowledge graph.

4. The artificial intelligence proposition method according to claim 2, characterized in that, Based on the multidisciplinary knowledge ontology framework, structured knowledge units conforming to predefined entity types and relation types are extracted from the educational normative documents and subject knowledge literature. After semantic standardization verification, the structured knowledge units are stored in a graph database in the form of nodes and edges to construct the multidisciplinary knowledge graph, including: A pre-defined joint extraction model is used to extract entities and the relationships between entities from structured and unstructured text constrained by the multidisciplinary knowledge ontology framework, generating triples that conform to the predefined entity types and relationship types in the multidisciplinary knowledge ontology framework. The triples are verified, and the entities in the verified triples are stored as nodes and their corresponding relationships as edges in the graph database to construct the multidisciplinary knowledge graph.

5. The artificial intelligence proposition method according to claim 1, characterized in that, The step of calculating the comprehensive semantic similarity between interdisciplinary knowledge points based on the structured semantics of the multidisciplinary knowledge base and the hierarchical semantics of the general semantic dictionary, and selecting candidate knowledge point pairs based on the comprehensive semantic similarity, includes: The embedding model is trained based on the entity relationship triples in the multidisciplinary knowledge base to obtain the vector representation of each interdisciplinary knowledge point, and the graph similarity between any two vector representations is calculated. Based on the general semantic dictionary, semantic parsing is performed on each interdisciplinary knowledge point to obtain its corresponding hierarchical semantic representation, and the hierarchical similarity between any two hierarchical semantic representations is calculated. The graph similarity and the hierarchical similarity are fused from multiple sources to generate the comprehensive semantic similarity. Based on the comprehensive semantic similarity, the target knowledge point pairs that meet the requirements for cross-disciplinary proposition association are selected.

6. The artificial intelligence proposition method according to claim 1, characterized in that, Based on the comprehensive semantic similarity and the graph topology in the multidisciplinary knowledge base, the association strength of the candidate knowledge point pairs is evaluated through joint analysis of multi-hop path reasoning and graph neural network models to determine the target knowledge point pairs that meet the requirements for interdisciplinary proposition usability, including: Based on the knowledge graph topology, a multi-hop path search is performed on the candidate knowledge point pairs to identify potential inference paths connecting two candidate knowledge point pairs, and the path weight of each inference path is calculated. The candidate knowledge point pairs are input into a graph neural network model, and node representations are generated by aggregating neighborhood information. The structure-aware similarity is then calculated based on the node representations. The comprehensive semantic similarity and the structure-aware similarity are fused to obtain the final association score; When the final association score exceeds a preset threshold, it is determined that the candidate knowledge point pair has a propositional interdisciplinary implicit relationship, and it is identified as the target knowledge point pair.

7. The artificial intelligence proposition method according to claim 1, characterized in that, The response to the interdisciplinary question-setting request returns the target knowledge point pair, which is used to drive the automatic generation of test questions, including: In response to the interdisciplinary question request, retrieve matching knowledge point pairs from the target knowledge point pairs; Based on the matched knowledge point pairs and their associated metadata, test questions are automatically generated.

8. An artificial intelligence proposition device, characterized in that, The device includes: The module is used to construct a multidisciplinary knowledge base based on educational normative documents and subject knowledge literature, and in accordance with a multidisciplinary knowledge ontology framework, to perform structured extraction, semantic verification, and graph-based storage of knowledge content. The filtering module is used to calculate the comprehensive semantic similarity between various interdisciplinary knowledge points based on the structured semantics of the multidisciplinary knowledge base and the hierarchical semantics of the general semantic dictionary, and to filter candidate knowledge point pairs according to the comprehensive semantic similarity. The evaluation module is used to evaluate the correlation strength of the candidate knowledge point pairs based on the comprehensive semantic similarity and the graph topology in the multi-disciplinary knowledge base, through joint analysis of multi-hop path reasoning and graph neural network model, and to determine the target knowledge point pairs that meet the requirements of interdisciplinary proposition usability. The response module is used to respond to interdisciplinary question requests and return the target knowledge point pairs to drive the automatic generation of test questions.

9. An artificial intelligence-based question-asking device, characterized in that, include: One or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications being configured to perform the artificial intelligence propositional method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code, which can be invoked by a processor to execute the artificial intelligence proposition method as described in any one of claims 1-7.

Images