AI intelligent proposition method

Abstract

The invention discloses an AI intelligent proposition method, and belongs to the technical field of intelligent proposition, and the method comprises the steps that a proposition platform end is established by a platform side, and the proposition platform end comprises an industry large model training module, a professional knowledge base construction module and a user analysis module; the platform side establishes a user side for a corresponding user according to the service demand data, wherein the user side comprises a test question generation module, a user professional library and a test question checking module; the user professional library is used for storing industry knowledge information of service industries corresponding to users; the test question generation module is used for generating corresponding first draft test questions according to proposition requirements of a user and a professional library of the user; the test question checking module is used for automatically checking the generated first draft test question; the method comprises the steps of obtaining proposition requirements of a user, analyzing the proposition requirements through a test question generation module, obtaining first draft test questions, checking the first draft test questions through a test question checking module, marking the first draft test questions passing the checking as application test questions, and displaying the application test questions to the user.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent proposition, and specifically is an AI intelligent proposition method. Background Art

[0002] In many fields such as education, examinations, and training, question setting plays a vital role. It is not only a key means to measure students' learning outcomes and teaching effectiveness, but also an important factor in promoting educational reform and development. However, the traditional way of setting questions mainly relies on manual completion, which has obvious limitations in efficiency and quality.

[0003] The process of manual question setting is cumbersome and time-consuming. Teachers or question setting experts need to invest a lot of time and energy to design questions, determine the difficulty, and review the content, which not only increases their workload, but also may lead to low question setting efficiency. At the same time, due to the differences in each person's knowledge background, teaching experience, and question setting style, the quality of manual question setting is often uneven, and it is difficult to ensure the scientificity, accuracy, and consistency of the questions. In addition, the traditional manual question setting method lacks scientific support. The design of the questions, the control of the difficulty, and the distribution of question types often rely on the subjective judgment and experience accumulation of the question setter, and lack objective and quantitative evaluation standards. This not only limits the diversity and innovation of the questions, but also affects the scientificity and fairness of the question setting work.

[0004] At present, there are some intelligent question-setting software based on textbook knowledge graphs and general large models. However, the existing automatic question-setting systems have the following problems: The quality of questions is not high: due to the reliance on the general capabilities of large models and the lack of in-depth understanding and mastery of professional knowledge, the generated test questions often contain errors, biased questions, and inappropriate difficulty, resulting in the test questions being similar in appearance and the quality of the test content being low.

[0005] Narrow scope of questions: Due to the lack of learning and supplementation of professional knowledge, the scope of questions cannot cover all knowledge points, and there is a lack of comprehensive understanding and grasp of the knowledge points.

[0006] Single question type: Due to the lack of professional knowledge and the construction of test question models, only basic question types (such as single-choice questions, multiple-choice questions, true-or-false questions, fill-in-the-blank questions and short-answer questions) can be set, and complex question types (such as reading comprehension, case analysis, material analysis and nested combination questions) cannot be set.

[0007] Low test-setting efficiency: Since test questions are generated based on general test requirements based on knowledge graphs or general large models, they are less stable in terms of facts, logic, grammar, difficulty, etc., and require a lot of manual intervention and proofreading, making it impossible to achieve true automation.

[0008] Based on this, in order to realize intelligent proposition, the present invention provides an AI intelligent proposition method. Summary of the invention

[0009] In order to solve the problems existing in the above scheme, the present invention provides an AI intelligent proposition method.

[0010] The purpose of the present invention can be achieved through the following technical solutions: An AI intelligent proposition method, the method comprising: Step 1: The platform party establishes a proposition platform end, which includes an industry large model training module, a professional knowledge base construction module, and a user analysis module; The professional knowledge base building module is used to establish a professional knowledge base for the corresponding service industry; The industry big model training module is used to establish a corresponding industry big model for the corresponding service industry; The user analysis module is used to collect user information, match the corresponding service industry for the user according to the user information, and generate service demand data according to the service industry and user information.

[0011] Furthermore, the method for establishing the professional knowledge base includes: Obtain the service scope of the platform, and generate various service industries according to the service scope; obtain industry knowledge information of the service industry in real time, and the industry knowledge information includes knowledge points, concepts, standards, theorems, formulas, cases, and sample questions; The industry knowledge information is subjected to knowledge preprocessing, and the knowledge preprocessing includes classification, organization, and structured processing; a professional knowledge base of the service industry is established based on the knowledge preprocessed industry knowledge information.

[0012] Furthermore, the method for establishing the industry big model includes: Collecting text data of the service industry, the text data includes industry books, examination instructions, examination standards, professional reports, industry papers, industry news, and web page knowledge; cleaning and preprocessing the text data; Use cleaned text data to train industry models.

[0013] Step 2: The platform establishes a user terminal for the corresponding user according to the service demand data, and the user terminal includes a test question generation module, a user professional database, and a test question review module; The user professional database is used to store the industry knowledge information of the user's corresponding service industry; The test question generation module is used to generate corresponding first draft test questions according to the user's question setting requirements and the user's professional database; The test question review module is used to automatically review the generated first draft test questions.

[0014] Furthermore, the method for establishing the user professional database includes: Identify the service industry corresponding to the service demand data, identify the professional knowledge base corresponding to the service industry, and establish a user professional database based on the professional knowledge base.

[0015] Further, the test question generation module includes a requirement analysis submodule, a data construction submodule, a test question generation submodule, a quality inspection submodule and a content optimization submodule; The requirement analysis submodule is used to analyze the question requirements and determine key information, including test type, difficulty level, question type requirements, and knowledge point distribution; verify the key information and determine the scope of the question; The data construction submodule is used to design the basic structure of the corresponding question type according to the analyzed question setting requirements, and retrieve the corresponding question related information from the user professional database; The test question generation submodule is used to generate a first draft test question based on the user professional database and the built-in industry large model; The quality inspection submodule is used to inspect the completeness, usability and standardization of the first draft test questions and record the results of the quality inspection; The content optimization submodule is used to adjust and optimize the generated first draft test questions according to the review opinions provided by the test question review module.

[0016] Step three: obtain the user's question requirements, analyze the question requirements through the question generation module, obtain the first draft questions, review the first draft questions through the question review module, mark the reviewed first draft questions as application questions, and display the application questions to the user.

[0017] Furthermore, the method of analyzing the question setting requirements through the question generation module includes: Analyze the proposition requirement, determine the proposition range, and mark the proposition range as a reference range; Extending the requirements based on the reference range to obtain an extended range; integrating the reference range and the extended range into a comprehensive range; generating a knowledge range graph in real time based on the comprehensive range, wherein the knowledge range graph includes an overall range, a reference range and an extended range; and displaying the knowledge range graph to a user; A first draft test question is generated according to the comprehensive scope, and a corresponding requirement tag is added to the first draft test question, wherein the requirement tag includes a reference scope tag and an extended scope tag.

[0018] Furthermore, the method of extending the requirements based on the baseline range includes: Step SA1: obtaining the user's teaching data and marking it as reference material data; setting an initial extension coefficient, which is greater than or equal to 1; When the extension factor is equal to 1, it means that there is no need to conduct demand extension analysis; Step SA2: using the initial extension coefficient as the applied extension coefficient, extending the reference range by the extension coefficient to obtain an extension range; and obtaining answer feedback data corresponding to the extension range in real time; Step SA3: adjusting the extension coefficient according to the answer feedback data to obtain a new extension coefficient, extending the reference range by the new extension coefficient to obtain a new extension range, and acquiring corresponding answer feedback data in real time; Step SA4: Loop step SA3.

[0019] Furthermore, the method for setting the initial extension coefficient includes: Setting the user types in the service industry, setting corresponding initial extension coefficients for the user types, integrating the user types and the corresponding initial extension coefficients to establish an initial extension coefficient matching table; and collecting the initial extension coefficient of each user in the user type in real time, and integrating each of the initial extension coefficients into coefficient correction data of the user type; Adjusting the initial extension coefficient of the corresponding user type in the initial extension coefficient matching table in real time according to the coefficient correction data; A user type corresponding to the user is identified, and a corresponding initial extension coefficient is matched from the initial extension coefficient matching table according to the user type.

[0020] Compared with the prior art, the present invention has the following beneficial effects: Compared with traditional proposition methods and technologies, the present invention is based on a vertical industry domain large model of private training, professional knowledge base construction, and uses RAG knowledge enhancement retrieval and (Agent) workflow automatic collaboration design to provide an intelligent proposition system with significant advantages. The following are the main advantages and beneficial effects brought by the present invention: 1. Highly customized: Through privately trained vertical field models, we can more accurately understand and master the professional knowledge and language expressions of specific industries. This makes the generated test questions more in line with actual needs and improves the professionalism and pertinence of the test questions. Users can customize the settings according to specific needs, including test type, difficulty level, question type requirements, etc., to ensure that the generated test questions fully meet the specific needs of users.

[0021] 2. High-quality output: The RAG (Retriever-Augmented Generation) mechanism combines the advantages of retrieval and generation, and can generate test questions that are both rich and diverse and in line with actual test needs. This not only improves the quality of test questions, but also increases the diversity of test questions. Using industry big models and professional knowledge bases, it can effectively reduce factual errors, logical errors, and grammatical errors in test questions, and improve the accuracy and scientificity of test questions. Through the content optimization sub-module, it ensures that the generated test questions have been reviewed and adjusted many times to reach higher quality standards. This not only improves the accuracy of test questions, but also enhances the scientificity and rationality of test questions.

[0022] 3. Efficient collaboration: From demand analysis to test question generation to quality inspection and content optimization, the entire process is highly automated, reducing the need for manual intervention and improving work efficiency. (Agent) Intelligent design makes the collaboration between modules smoother and improves overall operational efficiency. For example, the agent can transfer data and feedback between different modules to achieve seamless connection.

[0023] 4. Continuous optimization: With the accumulation of more data and the continuous iteration of the model, the system performance will continue to improve. Through continuous optimization and adjustment, the quality of test questions and the stability of the system can be continuously improved. The content optimization submodule records the detailed information of each adjustment. These logs can be used for subsequent tracing and improvement to help the system continuously optimize.

[0024] 5. Comprehensive coverage: The construction of the professional knowledge base ensures the breadth and depth of the scope of questions, covering all necessary knowledge points and assessment points. This makes the generated test questions more comprehensive and can meet the needs of different exams. It can generate a variety of test questions, including basic question types (such as single-choice questions, multiple-choice questions, true-or-false questions, fill-in-the-blank questions, short-answer questions) and complex question types (such as reading comprehension, case analysis, material analysis, and nested combination questions, etc.), to meet the requirements of different exam types.

[0025] 6. Strong adaptability: Evaluate and adjust the difficulty of test questions based on factors such as the age, education, and professional background of the test subjects to ensure that the test questions are suitable for different groups. For example, for primary school students, the test questions should be relatively simple; for college students or professionals, the test questions can be appropriately more difficult. Consider the differences in education levels, teaching resources, and cultural backgrounds in different regions, evaluate and adjust the difficulty of test questions to ensure that the test questions are suitable for specific regions. This helps to improve the fairness and applicability of test questions. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0027] Figure 1 is a flow chart of the method of the present invention; Figure 2 It is a schematic diagram of the proposition platform end of the present invention; Figure 3 It is a schematic diagram of the user terminal of the present invention. DETAILED DESCRIPTION

[0028] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] like Figures 1 to 3 As shown, a method for AI intelligent proposition, the method includes: Step 1: The platform establishes a proposition platform, which includes an industry large model training module, a professional knowledge base construction module, and a user analysis module; Professional knowledge base construction module: obtain the service scope of the platform, such as education assessment, personnel recruitment, professional qualification and other proposition services in the corresponding industry, generate various service industries according to the service scope, and classify them according to whether they are the same professional knowledge base and industry model. Generally, the platform will manually check and adjust them according to the intelligently layered industries to form various service industries; establish corresponding professional knowledge bases for each service industry, which contain all the knowledge points, concepts, theorems, formulas, cases, sample questions, proposition specifications, terminology specifications, expression specifications and other industry knowledge information of the industry, and structure these information to facilitate the understanding and use of subsequent models. For example, extract the knowledge points, concepts, standards, theorems, formulas, cases and sample questions of a specific industry from various teaching materials, examination syllabi, curriculum standards, proposition specifications, terminology specifications, expression specifications, punctuation specifications and sample test questions to form industry knowledge information; classify, organize and structure the industry knowledge information to form a complete knowledge system, and mark the corresponding classification, organization and structured processing as knowledge preprocessing; establish a professional knowledge base based on the processed industry knowledge information. That is, the professional knowledge base is used to store the industry knowledge information of the service industry.

[0030] Industry big model training module: establish corresponding industry big models for each service industry. The industry big model can understand and master the professional knowledge and language expression of the industry, and then intelligently generate corresponding draft test questions based on the question requirements and professional knowledge base.

[0031] The industry large model training module specifically includes: Data collection submodule: used to collect text data of the service industry, including industry books, examination instructions, examination standards, relevant papers, professional reports, relevant news, web page knowledge, etc.

[0032] Data cleaning submodule: used to clean and preprocess the collected data to remove noise, duplicate data, invalid data, etc.

[0033] Model training submodule: used to train the industry model with cleaned data so that it can master the professional knowledge and language expression of a specific industry.

[0034] In one embodiment, the industry big model is established by the platform through professionals to enable it to master the professional knowledge and language expression of the service industry; existing intelligent models can also be applied.

[0035] For example, pre-training is the starting stage of building a large language model. This stage mainly uses a large amount of data to feed the large-scale model for training to obtain the initialized model parameters. This process helps the model learn general language knowledge from a large corpus.

[0036] Fine-tuning: Based on the trained large model, some parameters of the large model are further adjusted based on a specific data set (i.e., cleaned service industry data) to enable the model to better adapt to business scenarios and complete specific tasks accurately and efficiently. Fine-tuning is a key step in the training of industry large models, which enables the model to better grasp the professional knowledge and language expressions of a specific industry.

[0037] Subsequently, based on actual result feedback and other data, the platform can optimize and learn the industry model of the corresponding service industry.

[0038] User analysis module: used to determine the service industry to which the user belongs, collect user information, determine the user's service demand data based on the user information and the service industry, and subsequently establish the corresponding user terminal for the user based on the service demand data; such as purpose, scale, maintenance requirements, etc. to establish demand data related to the user terminal.

[0039] Step 2: The platform establishes a corresponding user terminal for the corresponding user based on the service demand data. The user terminal includes a test question generation module, a user professional database, and a test question review module; User professional database: Match the corresponding professional knowledge database according to the service industry corresponding to the user, establish the user professional database based on the matched professional knowledge database, and make corresponding copies. In order to distinguish them, define different names.

[0040] Test question generation module: It is used to automatically generate the first draft test questions according to the user's test requirements, using the built-in industry model and professional knowledge base. It can generate the corresponding test content, answer analysis and test purpose according to different test types, test subjects, assessment content, difficulty level, question type requirements and other test requirements.

[0041] In one embodiment, the test question generation module specifically includes: Requirements analysis submodule: used to analyze the proposition requirements proposed by users to ensure that the system can accurately understand and meet the user's proposition requirements.

[0042] Requirements reception: Receive the question requirement document submitted by the user through the user interface or API. The requirement document may contain information such as exam type, difficulty level, question type requirements, knowledge point distribution, etc.

[0043] Demand analysis: Structural processing of the received demand documents to extract key information, such as test type (education assessment, personnel recruitment, professional qualification, etc.), difficulty level (easy, medium, difficult), question type requirements (single-choice questions, multiple-choice questions, true or false questions, fill-in-the-blank questions, short-answer questions, writing, reading comprehension, case analysis, material analysis and nested combination questions, etc.), knowledge point distribution (involved knowledge points, chapters, proportions, etc.); Demand verification: Conduct preliminary verification on the parsed key information to ensure that all necessary information has been provided and is within the scope supported by the system; that is, the key information needs to meet the preset format requirements and cannot be missing or omitted; obtain the user's teaching data, such as the use of textbooks, teaching ability, teaching videos, teaching materials and other related data, to understand the user's actual teaching scope, because even for the same knowledge point, the difference in teaching will lead to differences in the corresponding proposition range. For example, a certain knowledge point has 10 question directions, but because of the user's actual teaching situation, 3 of the processing directions are not currently involved, then the scope needs to be screened later; therefore, it is necessary to conduct limited analysis based on the user's teaching data; analyze the key information based on the teaching data to determine the proposition range; that is, first determine the comprehensive question range of each knowledge point based on the user's professional library, and then screen the comprehensive question range based on the teaching data, remove the range outside the teaching data range, and integrate to form the proposition range; the proposition range can also be determined by other means.

[0044] Requirements storage: Store the parsed requirements in the database for use by subsequent modules.

[0045] Data construction submodule: Design the basic structure of each question type based on the analyzed question-setting requirements, and retrieve the corresponding question-related information from the user's professional database.

[0046] Question type design: Design the basic structure of each question type according to the question type requirements in the demand. For example, single-choice questions require corpus, question stem, question options, options, and correct answers; multiple-choice questions require corpus, question stem, multiple options, and correct answer combinations; material analysis questions require material content, question types, and the number of questions.

[0047] Material data retrieval: Retrieve data information that matches the proposition requirements from the user's professional database. This includes relevant knowledge points, concepts, theorems, formulas, cases, etc.

[0048] Knowledge point retrieval: Retrieve detailed information of relevant knowledge points based on the distribution of knowledge points in the question requirements.

[0049] Sample test question retrieval: Retrieve sample test questions similar to current needs for reference.

[0050] Standardized search: Search for relevant question setting standards, terminology standards, expression standards, punctuation standards, etc. to ensure that the generated test questions meet industry standards.

[0051] Data integration: Integrate the retrieved data into the various parts required for the question type. For example, integrate knowledge points, cases, sample questions and other information into materials, corpus, question stems, question options, options, answers, analysis, etc.

[0052] Data preprocessing: preprocess the integrated data to obtain relevant information about test questions and ensure the consistency and accuracy of the data. For example, remove duplicate data, format text, etc.

[0053] Test question generation submodule: Use the trained industry model and user professional library to automatically generate first draft test questions.

[0054] Question template selection: Choose the appropriate template according to the question type requirements. Each question type has a corresponding template, which defines the format of the question stem, options, answers, etc.

[0055] Fill in the test content: fill the data provided by the data construction submodule into the selected template. For example, fill in the question stem, options, answer analysis and other information into the template.

[0056] Natural language generation: Use industry big models to generate fluent and accurate questions and options. Industry big models can understand and master the professional knowledge and language expressions of specific industries, ensuring that the language expression of generated test questions meets professional standards.

[0057] Answer analysis generation: Generate test answers and detailed answer analysis, including problem-solving ideas, steps and final answers. Ensure the accuracy and completeness of the answers.

[0058] Quality check submodule: Check the integrity, usability and standardization of the generated test questions to ensure that the test content units are complete, the question type requirements, knowledge distribution and difficulty level meet the question setting requirements.

[0059] Completeness check: Check whether the test contains all necessary parts, such as the question stem, options, answer explanation, etc. Make sure there is no missing part.

[0060] Usability Check: Check that each section of the test is readable and has no obvious spelling errors or formatting issues. Ensure that the test is readable and understandable.

[0061] Standardization check: Check whether the test questions comply with the question setting standards, wording standards, expression standards, punctuation standards, etc. to ensure the consistency and professionalism of the test questions in form.

[0062] Consistency check: Check whether the content of the test questions is consistent with the requirements of the questions and whether there is any deviation from the topic. Ensure that the test questions cover the necessary knowledge points and assessment points.

[0063] Feedback records: Record the results of quality checks, including problems found and suggested improvements. These records can be used for subsequent content optimization.

[0064] Content optimization submodule: used to adjust and optimize the generated test questions according to the review opinions provided by the test question review module.

[0065] Feedback reception: Receive review comments from the question review module. Review comments may include suggestions on standardization, grammar, facts, logic, etc.

[0066] Content adjustment: Adjust the content of the test questions according to the review opinions. For example, modify the question stem, options, answer analysis, etc. to ensure that the test questions meet all review standards.

[0067] Optimization iteration: If there are still problems after one adjustment, multiple iterations can be performed until all audit requirements are met. After each adjustment, a quality check will be conducted again to ensure the optimization effect.

[0068] Record logs: Record detailed information of each adjustment, including the reason for the adjustment, the comparison before and after the adjustment, etc. These logs can be used for subsequent tracing and improvement, and can also be used as a basis for system optimization.

[0069] Test question review module: used to automatically review the generated first draft test questions to ensure the correctness, rationality and scientificity of the test questions. This module can check the question setting standards, wording standards, expression standards, grammatical errors, factual errors, logical errors, etc., and can evaluate and adjust the difficulty of the test questions.

[0070] In one embodiment, the first draft test questions may be checked based on existing technologies.

[0071] In one embodiment, the test question review module includes: Standards checking submodule: used to check whether the test questions comply with the question setting standards, terminology standards, expression standards and punctuation standards, etc., to ensure the consistency and professionalism of the test questions in form.

[0072] Question standardization check: Check whether the test questions meet the requirements of the examination syllabus, curriculum standards and question setting guidelines, and ensure that the test questions cover the necessary knowledge points and assessment points.

[0073] Check the terminology: Check whether the terms, vocabulary and expressions in the test questions meet industry standards and academic norms, and avoid using ambiguous or inaccurate language.

[0074] Expression standard check: Check whether the expression of the question stem and options is clear, concise and unambiguous to ensure that candidates can accurately understand the requirements of the question.

[0075] Punctuation standardization check: Check whether the punctuation marks in the test questions are used correctly, including commas, periods, quotation marks, etc., to ensure the standardization of the text format.

[0076] Grammar checking submodule: used to check grammatical errors in test questions and ensure that the grammatical structure of the test questions is correct.

[0077] Sentence structure check: Check whether the subject, predicate and object structure of the sentence is complete to avoid incomplete or redundant sentences.

[0078] Tense consistency check: Check whether the tense of the sentence is consistent to avoid difficulties in understanding caused by tense confusion.

[0079] Pronoun consistency check: Check whether the use of pronouns is consistent with the objects they refer to to avoid unclear reference.

[0080] Modifier position check: Check whether the position of the modifier is correct and ensure that the relationship between the modifier and the modified object is clear.

[0081] Fact-checking submodule: used to check whether the factual information in the test questions is accurate and ensure the authenticity and reliability of the test content.

[0082] Data accuracy check: Check whether the data, statistics, historical events, etc. involved in the test questions are accurate to avoid providing erroneous information.

[0083] Concept accuracy check: Check whether the concepts, definitions, principles, etc. involved in the test questions are accurate to ensure consistency with the textbooks and authoritative materials.

[0084] Background knowledge check: Check whether the background knowledge involved in the test questions is accurate and ensure that the background information provided is true and reliable.

[0085] Logic check submodule: used to check whether the logical relationship of the test questions is reasonable, to ensure that the logical structure of the test questions is rigorous and without loopholes.

[0086] Causal relationship check: Check whether the causal relationship in the test questions is reasonable to avoid situations where causality is reversed or causality is invalid.

[0087] Conditional relationship check: Check whether the conditional relationship in the test question is reasonable and ensure that the logical relationship between the conditions and results is correct.

[0088] Reasoning process check: Check whether the reasoning process in the test questions is rigorous to avoid sudden reasoning or logical breaks.

[0089] Option logic check: Check whether the logical relationship between options is reasonable to ensure that the design of interference items will not mislead candidates.

[0090] Difficulty assessment submodule: used to assess the difficulty of test questions and make adjustments as needed.

[0091] Difficulty classification: Test questions are divided into different difficulty levels (such as easy, medium, and difficult) based on factors such as the test content, question types, knowledge points, and actual measured data of sample questions of the same latitude.

[0092] Difficulty adjustment: If the difficulty of the test questions does not meet the predetermined standards, it can be adjusted by modifying the question stem, options, answer analysis, etc.

[0093] Matching of test subjects: Based on factors such as the age, education, and professional background of the test subjects, assess whether the difficulty of the test questions is suitable for the group. For example, for primary school students, the test questions should be relatively simple; for college students or professionals, the test questions can be appropriately more difficult.

[0094] Use regional matching: Consider the differences in education levels, teaching resources, and cultural backgrounds in different regions, and evaluate whether the difficulty of the test questions is suitable for a specific region. For example, in regions with relatively abundant educational resources, the difficulty of the test questions can be appropriately increased; in regions with relatively scarce educational resources, the difficulty of the test questions should be appropriately reduced to ensure fairness.

[0095] Step 3: Obtain the user's question requirements, analyze the question requirements through the question generation module, obtain the first draft questions, review the first draft questions through the question review module, mark the reviewed first draft questions as application questions, and display the application questions to the user.

[0096] In one embodiment, the question generation module is used to analyze the question setting requirements, and the analysis is performed according to the working mode of the above-mentioned question generation module.

[0097] In one embodiment, the method for analyzing the question setting requirements by the question generation module includes: Analyze the proposition requirements, determine the proposition scope, and mark the obtained proposition scope as the benchmark scope; Extend the demand based on the baseline scope to obtain the extended scope; integrate the baseline scope and the extended scope into a comprehensive scope; generate a knowledge scope map in real time based on the comprehensive scope, and display the knowledge scope map to the user; the knowledge scope map is used to display the unscreened comprehensive proposition scope, baseline scope and extended scope, and mark the unscreened comprehensive proposition scope as the overall scope; that is, the knowledge scope map includes the overall scope, baseline scope and extended scope.

[0098] Generate each draft test question according to the comprehensive scope, and add a corresponding requirement tag to each draft test question, the requirement tag includes a baseline scope tag and an extended scope tag; In one embodiment, the method for performing demand extension based on the baseline range includes: Step SA1: obtaining the user's teaching data, which is the teaching data for demand verification in the demand analysis submodule, that is, the teaching data for determining the application of the benchmark range, and marking it as benchmark material data; Set the initial stretch factor. The stretch factor is greater than or equal to 1. If it is equal to 1, it means no stretching. The initial stretch factor can be directly set by the platform according to the actual situation of the user, such as 1.1, 1.2, etc.; Step SA2: Using the initial extension coefficient as the applied extension coefficient, extending the reference range by the extension coefficient to obtain the extended range, and obtaining in real time the answer feedback data of the first draft test questions corresponding to the extended range label, that is, taking the corresponding first draft test questions as applied test questions, and obtaining subsequent answer feedback data, such as the answer pass rate, answer rate, extended feedback and other related data of the test questions corresponding to the extended range; Step SA3: adjusting the extension coefficient according to the answer feedback data to obtain a new extension coefficient, extending the reference range by the new extension coefficient to obtain a new extension range, and obtaining the corresponding answer feedback data in real time; Step SA4: loop step SA3 to dynamically adjust the extension range.

[0099] In one embodiment, the method for setting the initial extension coefficient includes: Obtaining the user types of each service industry, the teaching data of each user and potential user in the service industry can be classified by the existing classification method to form different teaching data classifications, and then user classification is implemented according to the teaching data classification to obtain each user type; for example, classification is performed using the existing clustering algorithm, or manual classification is performed directly by the platform staff; The platform sets the corresponding initial extension coefficient for each user type, and establishes an initial extension coefficient matching table according to each user type and the initial extension coefficient; for example, the extension coefficient of each user in the user type is counted in real time, and the average value, mode, etc. are taken for real-time adjustment. Because the initial extension coefficient and the extension coefficient will be dynamically adjusted according to subsequent feedback, the initial extension coefficient set for the first time does not need to be particularly accurate, because as the system runs, the accuracy of the initial extension coefficient matching table will be gradually improved; Match the corresponding initial extension coefficient for the user according to the initial extension coefficient matching table.

[0100] In one embodiment, the reference range is extended by the extension coefficient. The reference range can be directly expanded according to the proportional value corresponding to the extension coefficient. The expanded range is used as the extended range. However, the expanded range is within the overall range and will not exceed the overall range.

[0101] In one embodiment, the reference range is extended by an initial extension factor, and the extension may be performed based on an existing method.

[0102] In one embodiment, the extension coefficient is adjusted according to the answer feedback data, that is, whether the extension coefficient is reasonable is determined according to the answer feedback data. If the extension range is known through the answer feedback data that the corresponding test questions exceed the range, etc., the extension coefficient needs to be reduced. The extension coefficient can be reduced according to a preset fixed unit, or it can be reduced according to the degree of excessive feedback. Otherwise, the extension coefficient can be increased. The adjustment of the extension coefficient can be achieved in combination with existing analysis technology in the above manner. Exemplarily, a coefficient adjustment model is established based on a neural network such as a CNN network or a DNN network, and a corresponding training set is established manually for training. The training set includes input data and output data. The input data is the extension coefficient and the answer feedback data, and the output data is the adjusted extension coefficient. The coefficient adjustment model after successful training is used for analysis and adjustment.

[0103] The above formulas are all calculated by removing dimensions and taking their numerical values. The formula is a formula that is closest to the actual situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained by simulating a large amount of data.

[0104] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for AI intelligent proposition, characterized in that: Methods include: Step 1: The platform party establishes a proposition platform end, which includes an industry large model training module, a professional knowledge base construction module, and a user analysis module; The professional knowledge base building module is used to establish a professional knowledge base for the corresponding service industry; The industry big model training module is used to establish a corresponding industry big model for the corresponding service industry; The user analysis module is used to collect user information, match the corresponding service industry for the user according to the user information; and generate service demand data according to the service industry and user information; Step 2: The platform establishes a user terminal for the corresponding user according to the service demand data, and the user terminal includes a test question generation module, a user professional database, and a test question review module; The user professional database is used to store the industry knowledge information of the user's corresponding service industry; The test question generation module is used to generate corresponding first draft test questions according to the user's question setting requirements and the user's professional database; The test question review module is used to automatically review the generated first draft test questions; Step three: obtain the user's question requirements, analyze the question requirements through the question generation module, obtain the first draft questions, review the first draft questions through the question review module, mark the reviewed first draft questions as application questions, and display the application questions to the user.

2. The method of AI intelligent proposition according to claim 1, characterized in that: Methods for establishing a professional knowledge base include: Obtain the service scope of the platform, and generate various service industries according to the service scope; obtain industry knowledge information of the service industry in real time, and the industry knowledge information includes knowledge points, concepts, standards, theorems, formulas, cases, and sample questions; The industry knowledge information is subjected to knowledge preprocessing, and the knowledge preprocessing includes classification, organization, and structured processing; a professional knowledge base of the service industry is established based on the knowledge preprocessed industry knowledge information.

3. The AI ​​intelligent proposition method according to claim 1, characterized in that: The methods for establishing industry big models include: Collecting text data of the service industry, the text data includes industry books, examination instructions, examination standards, professional reports, industry papers, industry news, and web page knowledge; cleaning and preprocessing the text data; Use cleaned text data to train industry models.

4. The AI ​​intelligent proposition method according to claim 1, characterized in that: The methods for establishing a user professional database include: Identify the service industry corresponding to the service demand data, identify the professional knowledge base corresponding to the service industry, and establish a user professional database based on the professional knowledge base.

5. The AI ​​intelligent proposition method according to claim 1, characterized in that: The test question generation module includes a requirement analysis submodule, a data construction submodule, a test question generation submodule, a quality inspection submodule and a content optimization submodule; The requirement analysis submodule is used to analyze the question requirements and determine key information, including test type, difficulty level, question type requirements, and knowledge point distribution; verify the key information and determine the scope of the question; The data construction submodule is used to design the basic structure of the corresponding question type according to the analyzed question setting requirements, and retrieve the corresponding question related information from the user professional database; The test question generation submodule is used to generate a first draft test question based on the user professional database and the built-in industry large model; The quality inspection submodule is used to inspect the completeness, usability and standardization of the first draft test questions and record the results of the quality inspection; The content optimization submodule is used to adjust and optimize the generated first draft test questions according to the review opinions provided by the test question review module.

6. The AI ​​intelligent proposition method according to claim 1, characterized in that: Methods for analyzing question requirements through the question generation module include: Analyze the proposition requirement, determine the proposition range, and mark the proposition range as a reference range; Extending the requirements based on the reference range to obtain an extended range; integrating the reference range and the extended range into a comprehensive range; generating a knowledge range graph in real time based on the comprehensive range, wherein the knowledge range graph includes an overall range, a reference range and an extended range; and displaying the knowledge range graph to a user; A first draft test question is generated according to the comprehensive scope, and a corresponding requirement tag is added to the first draft test question, wherein the requirement tag includes a reference scope tag and an extended scope tag.

7. The AI ​​intelligent proposition method according to claim 6, characterized in that: Methods for extending requirements based on baseline scope include: Step SA1: obtaining the user's teaching data and marking it as reference material data; setting an initial extension coefficient, which is greater than or equal to 1; When the extension factor is equal to 1, it means that there is no need to conduct demand extension analysis; Step SA2: using the initial extension coefficient as the applied extension coefficient, extending the reference range by the extension coefficient to obtain an extension range; and obtaining answer feedback data corresponding to the extension range in real time; Step SA3: adjusting the extension coefficient according to the answer feedback data to obtain a new extension coefficient, extending the reference range by the new extension coefficient to obtain a new extension range, and acquiring corresponding answer feedback data in real time; Step SA4: Loop step SA3.

8. The AI ​​intelligent proposition method according to claim 7, characterized in that: The methods for setting the initial extension factor include: Setting the user types in the service industry, setting corresponding initial extension coefficients for the user types, integrating the user types and the corresponding initial extension coefficients to establish an initial extension coefficient matching table; and collecting the initial extension coefficient of each user in the user type in real time, and integrating each of the initial extension coefficients into coefficient correction data of the user type; Adjusting the initial extension coefficient of the corresponding user type in the initial extension coefficient matching table in real time according to the coefficient correction data; A user type corresponding to the user is identified, and a corresponding initial extension coefficient is matched from the initial extension coefficient matching table according to the user type.

Images