CN-121579658-B - Intelligent proposition method, system, equipment and storage medium for simulating expert proposition

CN121579658BCN 121579658 BCN121579658 BCN 121579658BCN-121579658-B

Abstract

The invention relates to the technical field of artificial intelligence, in particular to an intelligent proposition method, system, equipment and storage medium for simulating expert propositions, which comprises the steps of constructing a structured knowledge network integrating knowledge points, cognition layers and propositions specifications; the method comprises the steps of establishing a material vector database associated with a knowledge point system, extracting proposition constraint information from a knowledge network according to target knowledge points, retrieving relevant materials from the vector database, combining the information into a generated prompt input large language model to generate a test question initial draft, inputting the test question initial draft into at least one large language model serving as a simulation answer agent, evaluating test question quality by analyzing the answer process and result, and outputting standard test questions. The invention can obviously improve the proposition efficiency and consistency, reduce absolute dependence on expert experience and realize the automatic production of high-quality test questions.

Inventors

MA LEI
YUAN FENG
JIANG PENGMIN
XING JINBAO
XUE YONG
GUO CHENGFENG
ZHAO RUIRUI

Assignees

山东山大鸥玛软件股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260127

Claims (7)

1. An intelligent proposition method for simulating expert propositions, comprising: based on various propositions, constructing a structured knowledge network containing a knowledge point system and attribute association thereof; Collecting propositional materials and carrying out vectorization representation, and establishing a material vector database associated with the knowledge point system; acquiring proposition constraint information from the structured knowledge network according to the target knowledge points, and retrieving related materials from the material vector database; combining the proposition constraint information and the related materials into a generated prompt, and inputting the generated prompt into a large language model to generate a test question manuscript; Inputting the test question manuscript into at least one large language model as a simulation answer agent to obtain the answer process and result of the test question; Based on various propositions, constructing a structured knowledge network containing a knowledge point system and attribute association thereof, comprising: acquiring various proposition data, wherein the proposition data at least comprises examination outline, electronic teaching materials and calendar year questions; Analyzing the examination outline, extracting knowledge points, cognition levels and capability requirements, and dynamically calculating knowledge point weights based on semantic association; Analyzing the electronic teaching materials, and extracting core content paragraphs corresponding to knowledge points; Analyzing the calendar year true questions to construct structural data for association among the characterization test questions, knowledge points, cognition levels and difficulties; Constructing and storing a knowledge graph comprising nodes, attributes and association relations based on the extracted knowledge points, weights, core content paragraphs and structured data, wherein the logic dependency relations among the knowledge points are inferred and completed by using a graph inference model; Based on the extracted knowledge points, weights, core content paragraphs and structured data, constructing and storing a knowledge graph comprising nodes, attributes and association relations, wherein the knowledge graph comprises: Defining multiple types of nodes, wherein the nodes at least comprise KnowledgePoint nodes representing knowledge points, textbookChapter nodes representing the content of chapters of teaching materials, pastQuestion nodes representing historical test questions, propositionRule nodes representing proposition specifications and CognitiveLevel nodes representing cognitive ability levels; defining a relation type for connecting the nodes, wherein the relation type at least comprises BelongsTo relation between a connection knowledge point and a section of a teaching material, examines relation between a connection history test question and the knowledge point, requires relation between the connection knowledge point and a cognitive ability level, follows relation between the connection knowledge point and a proposition standard, and Prerequisite relation for representing logic dependence among the knowledge points; Deducing and completing Prerequisite relations among knowledge points through a graph reasoning model; The KnowledgePoint nodes are provided with dynamic weight attributes, weight values of the dynamic weight attributes are obtained based on multi-factor weighted calculation, and the factors at least comprise frequency factors obtained based on statistics of the historical test question data, association factors obtained based on semantic similarity calculation and hierarchy factors determined based on the cognitive ability hierarchy; Deducing and complementing the Prerequisite relation among knowledge points through a graph reasoning model, wherein the deducing and complementing comprises the following steps: based on the existing nodes and relations in the knowledge graph, extracting observation features between at least one pair of knowledge points, and calculating a first probability value of a pre-knowledge relation between the knowledge point pairs by using a probability reasoning model; carrying out graph representation learning on the knowledge graph, transmitting neighbor information of aggregation nodes through a message to update vector representation of the aggregation nodes, and calculating a second probability value of a pre-knowledge relationship between the knowledge point pairs through a link prediction model based on the updated node vector representation; And when the comprehensive confidence coefficient exceeds a preset threshold value and the relation does not exist in the map, automatically creating a corresponding Prerequisite relation.
2. The method of claim 1, wherein capturing and vectorizing propositional material, creating a material vector database associated with the knowledge point hierarchy, comprises: collecting original proposition material texts from at least one material source, and performing topic identification and abstract generation on the original proposition material texts to obtain structured material fragments; Encoding the structured material segments into a vector representation of fixed dimensions using a pre-trained text embedding model; Based on at least one pre-association strategy, determining a preliminary association relation between the structured material segment and one or more knowledge points in the knowledge point system, and storing knowledge point identifications representing the preliminary association relation as metadata in correspondence with the vector representations.
3. The method of claim 1, wherein obtaining proposition constraint information from the structured knowledge network and retrieving relevant stories from the story vector database based on target knowledge points, combining the proposition constraint information with the relevant stories into a generative hint, and inputting to a large language model to generate a test draft, comprising: Extracting proposition constraint information from the structured knowledge network according to a target knowledge point, wherein the proposition constraint information at least comprises a cognition level, a difficulty requirement and an applicable problem type rule of the target knowledge point; Retrieving at least one semantically related material segment from the material vector database based on the target knowledge points; Formatting and combining the proposition constraint information, the at least one material fragment and a preset proposition task instruction to construct a generated prompt text; Inputting the generated prompt text into a large language model, and generating test question manuscripts conforming to the proposition constraint information in batches by adjusting the generation parameters of the large language model to control the output characteristics.
4. The method of claim 1, wherein evaluating the proposition quality of the test question manuscript and outputting a standard test question based on the solving process and the result comprises: analyzing a solution process and a result generated by the simulation answer agent according to a plurality of preset quality dimensions, and generating scores of the test question manuscript in each quality dimension; calculating the comprehensive quality score of the test question manuscript based on the scores in all quality dimensions; and comparing the comprehensive quality score with a preset recording threshold value to judge whether the test question initial draft is a standard test question.
5. An intelligent proposition system for simulating expert propositions, comprising: The map construction module is used for constructing a structured knowledge network containing a knowledge point system and attribute association thereof based on various propositions according to data; The material processing module is used for collecting the propositional materials and carrying out vectorization representation, and establishing a material vector database associated with the knowledge point system; The test question generation module is used for acquiring proposition constraint information from the structured knowledge network according to a target knowledge point and retrieving related materials from the material vector database; The test question evaluation module is used for inputting the test question initial draft into at least one large language model serving as a simulation answer agent to acquire the answer process and result of the test question, evaluating the proposition quality of the test question initial draft based on the answer process and result, and outputting the standard test questions; Based on various propositions, constructing a structured knowledge network containing a knowledge point system and attribute association thereof, comprising: acquiring various proposition data, wherein the proposition data at least comprises examination outline, electronic teaching materials and calendar year questions; Analyzing the examination outline, extracting knowledge points, cognition levels and capability requirements, and dynamically calculating knowledge point weights based on semantic association; Analyzing the electronic teaching materials, and extracting core content paragraphs corresponding to knowledge points; Analyzing the calendar year true questions to construct structural data for association among the characterization test questions, knowledge points, cognition levels and difficulties; Constructing and storing a knowledge graph comprising nodes, attributes and association relations based on the extracted knowledge points, weights, core content paragraphs and structured data, wherein the logic dependency relations among the knowledge points are inferred and completed by using a graph inference model; Based on the extracted knowledge points, weights, core content paragraphs and structured data, constructing and storing a knowledge graph comprising nodes, attributes and association relations, wherein the knowledge graph comprises: Defining multiple types of nodes, wherein the nodes at least comprise KnowledgePoint nodes representing knowledge points, textbookChapter nodes representing the content of chapters of teaching materials, pastQuestion nodes representing historical test questions, propositionRule nodes representing proposition specifications and CognitiveLevel nodes representing cognitive ability levels; defining a relation type for connecting the nodes, wherein the relation type at least comprises BelongsTo relation between a connection knowledge point and a section of a teaching material, examines relation between a connection history test question and the knowledge point, requires relation between the connection knowledge point and a cognitive ability level, follows relation between the connection knowledge point and a proposition standard, and Prerequisite relation for representing logic dependence among the knowledge points; Deducing and completing Prerequisite relations among knowledge points through a graph reasoning model; The KnowledgePoint nodes are provided with dynamic weight attributes, weight values of the dynamic weight attributes are obtained based on multi-factor weighted calculation, and the factors at least comprise frequency factors obtained based on statistics of the historical test question data, association factors obtained based on semantic similarity calculation and hierarchy factors determined based on the cognitive ability hierarchy; Deducing and complementing the Prerequisite relation among knowledge points through a graph reasoning model, wherein the deducing and complementing comprises the following steps: based on the existing nodes and relations in the knowledge graph, extracting observation features between at least one pair of knowledge points, and calculating a first probability value of a pre-knowledge relation between the knowledge point pairs by using a probability reasoning model; carrying out graph representation learning on the knowledge graph, transmitting neighbor information of aggregation nodes through a message to update vector representation of the aggregation nodes, and calculating a second probability value of a pre-knowledge relationship between the knowledge point pairs through a link prediction model based on the updated node vector representation; And when the comprehensive confidence coefficient exceeds a preset threshold value and the relation does not exist in the map, automatically creating a corresponding Prerequisite relation.
6. An intelligent proposition device for simulating expert propositions, comprising: the memory is used for storing an intelligent proposition program simulating expert propositions; A processor for implementing the steps of the intelligent proposition method for simulating expert propositions according to any one of claims 1-4 when executing the intelligent propositions program for simulating expert propositions.
7. A computer readable storage medium storing a computer program, wherein the readable storage medium has stored thereon an intelligent proposition program simulating an expert proposition, the intelligent proposition program simulating an expert proposition, when executed by a processor, implementing the steps of the intelligent proposition method simulating an expert proposition as claimed in any of claims 1-4.

Description

Intelligent proposition method, system, equipment and storage medium for simulating expert proposition Technical Field The invention belongs to the technical field of artificial intelligence, and particularly relates to an intelligent proposition method, system, equipment and storage medium for simulating expert propositions. Background The traditional test questions mainly depend on expert experience, and have the problems of low efficiency, high cost, difficult quality consistency assurance and the like. Currently, although automated proposition technology based on a large language model has appeared, the technology generally adopts an end-to-end generation mode, lacks of deep combination of education measurement rules and a subject knowledge system, and is easy to generate phenomena such as content "illusion", logic contradiction, superclass or difficulty out of control. Meanwhile, the existing method focuses on generation, lacks an automatic and simulation type evaluation mechanism for test question quality, and is difficult to ensure the rigor and effectiveness of the test questions. Therefore, an automatic proposition method capable of simulating expert proposition logic, deeply fusing structural knowledge and performing intelligent quality check on a generated result is needed. Disclosure of Invention Aiming at the defects in the prior art, the invention provides an intelligent proposition method, system, equipment and storage medium for simulating expert propositions so as to solve the technical problems. In a first aspect, the present invention provides an intelligent proposition method for simulating expert propositions, comprising: based on various propositions, constructing a structured knowledge network containing a knowledge point system and attribute association thereof; Collecting propositional materials and carrying out vectorization representation, and establishing a material vector database associated with the knowledge point system; acquiring proposition constraint information from the structured knowledge network according to the target knowledge points, and retrieving related materials from the material vector database; combining the proposition constraint information and the related materials into a generated prompt, and inputting the generated prompt into a large language model to generate a test question manuscript; Inputting the test question manuscript into at least one large language model as a simulation answer agent to obtain the answer process and result of the test question, evaluating the proposition quality of the test question manuscript based on the answer process and result, and outputting the standard test question. In an alternative embodiment, based on a plurality of propositional basis data, constructing a structured knowledge network comprising a knowledge point system and attribute association thereof, comprising: acquiring various proposition data, wherein the proposition data at least comprises examination outline, electronic teaching materials and calendar year questions; Analyzing the examination outline, extracting knowledge points, cognition levels and capability requirements, and dynamically calculating knowledge point weights based on semantic association; Analyzing the electronic teaching materials, and extracting core content paragraphs corresponding to knowledge points; Analyzing the calendar year true questions to construct structural data for association among the characterization test questions, knowledge points, cognition levels and difficulties; based on the extracted knowledge points, weights, core content paragraphs and structured data, constructing and storing a knowledge graph comprising nodes, attributes and association relations, wherein the logical dependency relations among the knowledge points are inferred and completed by using a graph inference model. In an optional implementation manner, based on the extracted knowledge points, weights, core content paragraphs and structured data, a knowledge graph including nodes, attributes and association relations is constructed and stored, and the method includes: Defining multiple types of nodes, wherein the nodes at least comprise KnowledgePoint nodes representing knowledge points, textbookChapter nodes representing the content of chapters of teaching materials, pastQuestion nodes representing historical test questions, propositionRule nodes representing proposition specifications and CognitiveLevel nodes representing cognitive ability levels; defining a relation type for connecting the nodes, wherein the relation type at least comprises BelongsTo relation between a connection knowledge point and a section of a teaching material, examines relation between a connection history test question and the knowledge point, requires relation between the connection knowledge point and a cognitive ability level, follows relation between the connection knowledge point and a proposition standard, and Prerequisi