Error question tutoring method and system based on generative artificial intelligence and electronic equipment

Abstract

The disclosure provides a mistake question tutoring method and system based on generative artificial intelligence, and an electronic device. The method comprises obtaining mistake question data and determining a target knowledge point corresponding to the mistake question data; determining a corresponding teaching chain structure according to the target knowledge point, and adjusting the thinking prompt information of the guide question in the teaching chain structure based on the question type information in the mistake question data to generate a guide question chain corresponding to the target knowledge point; verifying the guide question chain and sending the guide question chain that passes the verification to the client to collect corresponding answer behavior data; determining a mistake cause result corresponding to the answer behavior data; and generating a corresponding mistake question training scheme based on the mistake cause result. By analyzing the mistake question data of the student, the knowledge point corresponding to the mistake question is identified to provide accurate direction for subsequent guidance and consolidation. The generated guide question chain guides the student to think independently, gradually builds the correct problem solving method, avoids directly giving the answer, and guides the student to establish the correct answer thinking.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a method, system, and electronic device for tutoring students on incorrect questions based on generative artificial intelligence. Background Technology

[0002] With the advancement of digital transformation in education, the application of artificial intelligence (AI) technology in error correction tutoring is becoming increasingly widespread. To a certain extent, it has replaced the repetitive work of traditional human tutoring, providing students with a convenient way to learn independently. Currently, most AI teaching systems can quickly provide corresponding answers and explanations based on students' incorrect answers. While this allows students to quickly understand the answers and helps them correct their mistakes, it ultimately fails to actually correct students' flawed thinking, leading them into a false sense of understanding. Furthermore, current AI teaching often suffers from the illusion of accuracy; the content generated by AI teaching systems is prone to distortion, outputting seemingly reasonable but actually inaccurate explanations or conclusions, which can easily mislead students during application, making it difficult to guarantee the accuracy of knowledge transfer. Summary of the Invention

[0003] This disclosure provides a method, system, and electronic device for tutoring students on incorrect questions based on generative artificial intelligence, in order to at least solve the above-mentioned technical problems existing in the prior art.

[0004] The first aspect of this disclosure provides a method for tutoring students on incorrect answers based on generative artificial intelligence, the method comprising: Obtain incorrect question data and determine the target knowledge points corresponding to the incorrect question data; Based on the target knowledge point, determine the corresponding teaching chain structure, and adjust the thinking prompts for guiding questions in the teaching chain structure based on the question type information in the wrong question data, to obtain the guiding question chain corresponding to the target knowledge point; The guided question chain is verified, and the verified guided question chain is sent to the client to collect the corresponding answer behavior data; The corresponding error causes are determined based on the answer behavior data; Based on the error causes, a corresponding error training plan is generated.

[0005] In one possible implementation, acquiring incorrect question data and determining the target knowledge points corresponding to the incorrect question data includes: Obtain incorrect question data and extract the question stem information from the incorrect question data; The question stem information is compared with the knowledge nodes in the knowledge graph to determine the corresponding comprehensive similarity. In response to the overall similarity exceeding the similarity threshold, the corresponding knowledge node is determined as the target knowledge node, and the knowledge point corresponding to the target knowledge node is used as the target knowledge point of the incorrect question data.

[0006] In one possible implementation, the step of determining the corresponding teaching chain structure based on the target knowledge point, and adjusting the thinking prompts for guiding questions in the teaching chain structure based on the question type information in the incorrect question data, to obtain the guiding question chain corresponding to the target knowledge point, includes: The corresponding teaching chain structure is determined based on the target knowledge points. The teaching chain structure is set with multiple guiding questions that represent the problem-solving ideas. Each guiding question includes a step number, question text information, thinking prompts and expected answer information. Based on the question stem information in the incorrect question data, the corresponding question type information is determined, and the thinking prompt information in the guiding questions is adjusted according to the question type feature rules corresponding to the question type information. The adjusted teaching chain structure is then used as the guiding question chain for the target knowledge point.

[0007] In one possible implementation, the step of determining the corresponding teaching chain structure based on the target knowledge point, and adjusting the thinking prompts for guiding questions in the teaching chain structure based on the question type information in the incorrect question data, to obtain the guiding question chain corresponding to the target knowledge point, further includes: Obtain context adjustment parameters, which include user identity information, incorrect question difficulty information, and user's historical incorrect question information; The number and difficulty level of the guiding questions in the guiding question chain are adjusted according to the context adjustment parameters.

[0008] In one possible implementation, the step of verifying the guiding question chain and sending the verified guiding question chain to the client to collect corresponding question-answering behavior data includes: The question text information, thinking prompt information, and expected answer information of each guiding question in the guiding question chain are compared with the pre-built target knowledge base to verify whether the question text information, the thinking prompt information, and the expected answer information are correct; Based on the question text information of each guiding question, determine whether the guiding question involves calculation content information, and based on the simulation calculation results, determine whether the calculation content information is correct; If the question text information, the thinking prompt information, the expected answer information, and the calculation content information are all correct, a verified guiding question chain is obtained; The verified guiding question chain is sent to the client to collect the corresponding answering behavior data.

[0009] In one possible implementation, determining the corresponding error cause based on the answer behavior data includes: Determine the behavior sequence vector corresponding to the answer behavior data; The behavior sequence vector and the incorrect question data are initially matched with the pre-constructed error cause labeling system to determine the candidate error cause results; The target error result is determined from the candidate error results based on the user's historical learning records and similar wrong question data.

[0010] In one possible implementation, generating a corresponding error-correction training scheme based on the error cause results includes: Based on the error causes, a corresponding set of error correction exercises is generated. The set of error correction exercises includes multiple exercises with different difficulty levels. The difficulty level of the first exercise matches the difficulty level of the error data. The difficulty level of each remaining exercise is determined based on the answer to the previous exercise.

[0011] In one possible implementation, after generating the corresponding error training scheme based on the error cause results, the method further includes: Collect training data corresponding to the incorrect question training scheme, and generate a corresponding learning report based on the training data.

[0012] A second aspect of this disclosure provides a problem-solving tutoring system based on generative artificial intelligence, the system comprising: The error analysis module is used to acquire error data and determine the target knowledge points corresponding to the error data; The guided task generation module is used to determine the corresponding teaching chain structure based on the target knowledge point, and adjust the thinking prompts of the guided questions in the teaching chain structure based on the question type information in the wrong question data, so as to generate a guided question chain corresponding to the target knowledge point. The verification module is used to verify the guiding question chain and send the verified guiding question chain to the client to collect the corresponding answer behavior data; The error analysis module is used to determine the corresponding error results based on the answer behavior data; The dynamic reinforcement module is used to generate corresponding error training schemes based on the error causes.

[0013] A third aspect of this disclosure provides an electronic device including at least one processor and a memory connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the error tutoring method based on generative artificial intelligence described in any of the preceding claims.

[0014] This disclosure discloses a method for error correction tutoring based on generative artificial intelligence, comprising: acquiring error data and identifying the target knowledge points corresponding to the error data; determining the corresponding teaching chain structure based on the target knowledge points, and adjusting the thinking prompts of the guiding questions in the teaching chain structure based on the question type information in the error data to generate a guiding question chain corresponding to the target knowledge points; verifying the guiding question chain and sending the verified guiding question chain to the client to collect corresponding answer behavior data; determining the corresponding error cause based on the answer behavior data; and generating a corresponding error training plan based on the error cause. By analyzing student error data, the knowledge points corresponding to the errors are identified, providing precise direction for subsequent guidance and consolidation. The generated guiding question chain guides students to maintain independent thinking and gradually build correct problem-solving strategies, rather than directly providing the answer, avoiding students falling into the illusion of understanding. Specifically, a verification step is added before sending the guiding question chain to the client for student use to ensure the reliability of the guiding questions and the accuracy of knowledge transmission. The answer behavior data collected accordingly can serve as a reliable basis for error cause judgment, thereby identifying the corresponding root cause of the error. Furthermore, it can provide corresponding error correction training programs targeting the root causes of errors, guide students in correcting their mistakes, and match different students' cognitive levels and learning needs.

[0015] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0016] The above and other objects, features, and advantages of this disclosure will become readily apparent from the following detailed description of exemplary embodiments, taken in conjunction with the accompanying drawings. Several embodiments of this disclosure are illustrated in the drawings by way of example and not limitation, in which: In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

[0017] Figure 1 The diagram illustrates the implementation flow of a method for tutoring incorrect questions based on generative artificial intelligence, according to an embodiment of this disclosure. Figure 2 A flowchart illustrating a specific implementation of a method for tutoring incorrect questions based on generative artificial intelligence, according to an embodiment of this disclosure, is shown. Figure 3 A schematic diagram of the modules of a problem-solving tutoring system based on generative artificial intelligence according to an embodiment of the present disclosure is shown; Figure 4 A schematic diagram of the composition structure of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0018] To make the objectives, features, and advantages of this disclosure more apparent and understandable, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0019] This disclosure provides a method for tutoring incorrect answers based on generative artificial intelligence, primarily applicable to learning scenarios such as educational tutoring. Figure 1 As shown, the method includes: S101. Obtain the data of incorrect questions and determine the target knowledge points corresponding to the data of incorrect questions; In this step, the incorrect question data includes both text-based and image-based data, which are specifically submitted by student users through the client. The appropriate parsing method is selected based on the specific type of incorrect question data to extract the question stem information. This question stem information serves as the basis for determining the target knowledge point, thus identifying the target knowledge point corresponding to the student's incorrect question data. It should be noted that the target knowledge point obtained in this step can provide a precise basis for knowledge localization in the generation of subsequent guiding questions, thereby focusing on the student's true knowledge gaps.

[0020] S102. Determine the corresponding teaching chain structure based on the target knowledge points, and adjust the thinking prompts for guiding questions in the teaching chain structure based on the question type information in the error data, and generate a guiding question chain corresponding to the target knowledge points. In this step, after identifying the target knowledge points corresponding to the incorrect question data, a corresponding chain of guiding questions is generated for each target knowledge point. This allows students to gradually construct the correct problem-solving approach based on the guiding question chain. It should be noted that after obtaining the target knowledge points, each knowledge point can determine its corresponding teaching chain structure from the knowledge structure template. This teaching chain structure includes multiple core steps set according to the thinking path, with each core step corresponding to a guiding question. Furthermore, each guiding question includes a step number, question text information, thinking hints, and expected answer information. Taking the quadratic equation in one variable in mathematics as an example, the corresponding teaching chain structure includes a question guiding the understanding of the problem. The specific content of the question guiding the understanding of the problem is as follows: "Step 1: Which type of algebraic equation does this problem belong to? (Thinking hints: You can judge based on the number of unknowns and the highest power; Expected answer format: Single choice, Option A: Linear equation in one variable, Option B: Quadratic equation in one variable, Option C: Linear equation in two variables)."

[0021] It should also be noted that this step uses the question type information corresponding to the incorrect question data to adjust the thinking prompts in the guiding questions. The question type information is determined based on the question stem information extracted from the incorrect question data. Accordingly, the question prompts in the corresponding guiding questions are determined according to the differences in question types, so as to adjust the corresponding thinking prompts. For example, multiple-choice questions focus on guiding the logic of eliminating options, calculation questions focus on guiding the decomposition of steps, and proof questions focus on guiding the application of theorems, thereby obtaining a guiding question chain adapted to the incorrect question data.

[0022] S103. Verify the guiding question chain and send the verified guiding question chain to the client to collect the corresponding answering behavior data; In this step, before sending the guiding question chain to the client to guide students' problem-solving approaches, the guiding question chain needs to be validated in real time to ensure its accuracy and rationality. Specifically, a pre-built, manually reviewed target knowledge base is used to verify the guiding questions in the chain. For guiding questions involving numerical calculations and formula derivations, the correctness of the calculation results is verified. Furthermore, the rationality of the teaching is verified to ensure the guiding question chain's reliability, addressing situations where guiding questions prematurely reveal the final answer or skip core thinking steps, thus violating teaching principles. The validated guiding question chain is then sent to the client, and data on students' answering behaviors during the process is collected as a basis for subsequent error analysis.

[0023] S104. Determine the corresponding error causes based on the answer behavior data; In this step, after collecting students' answer behavior data, the root causes of students' incorrect answers are determined by analyzing the answer behavior data, providing an accurate basis for generating subsequent error correction training plans. The answer behavior data directly reflects the students' weaknesses in the problem-solving process. Combined with the original error data, candidate error causes are selected from a pre-constructed error cause labeling system, and then the target error cause is selected from the candidate error causes.

[0024] S105. Generate corresponding error training schemes based on the error causes and results.

[0025] In this step, based on the target error analysis results and combined with the student's current knowledge level, a corresponding error-correction training plan is generated to consolidate practice and achieve the goal of personalized error-correction tutoring. That is, by directly linking the error analysis results with the error-correction training, the training plan is tailored to the student's actual learning situation, meeting the error-correction tutoring needs of different users.

[0026] This disclosure presents a method for error correction tutoring based on generative artificial intelligence. By analyzing students' error data, it identifies the corresponding knowledge points for each error, providing precise direction for subsequent guidance and reinforcement. The generated chain of guiding questions encourages students to maintain independent thinking and gradually build correct problem-solving strategies. Before sending the guiding question chain to the client for student use, a verification step is added to ensure the reliability of the guiding questions and the accuracy of knowledge transmission, reducing the generation of misleading or erroneous answers and minimizing the occurrence of illusion problems common in generative artificial intelligence. Illusion problems refer to AI-generated content that appears reasonable but is actually completely inconsistent with objective facts, containing factual errors. This disclosure's added verification step reduces the likelihood of the generated guiding question chain being inconsistent with subject knowledge points, problem-solving patterns, and other facts. The collected answer behavior data can serve as a reliable basis for judging the cause of errors, thereby identifying the corresponding root cause of the error. Furthermore, it can provide corresponding error correction training programs for the root cause of the error, guiding students to correct their erroneous thinking and matching different students' cognitive levels and learning needs. This achieves a closed-loop process, from analyzing incorrect questions to identifying the knowledge points in the incorrect questions, then generating guided learning tasks and verifying the guiding questions, and finally generating targeted reinforcement training based on the causes of the errors, ensuring the accuracy, reliability, and consistency of the teaching output and the teaching logic.

[0027] In one possible implementation, acquiring incorrect question data and determining the target knowledge points corresponding to the incorrect question data includes: Obtain incorrect question data and extract the question stem information from the incorrect question data; The question stem information is compared with the knowledge nodes in the knowledge graph to determine the corresponding comprehensive similarity. In response to a comprehensive similarity exceeding a similarity threshold, the corresponding knowledge node is identified as the target knowledge node, and the knowledge point corresponding to the target knowledge node is used as the target knowledge point for the incorrect question data.

[0028] In this embodiment, Optical Character Recognition (OCR) technology is used to recognize the question stem text in the image for incorrect question data, and key information of the question stem is extracted from the recognized text. For text-based incorrect question data, semantic parsing technology is used to directly extract the corresponding question stem information. It should be noted that the question stem information includes key content such as subject information, question type information, core conditions of the question stem, and target solution term. For example, a student takes a picture of a quadratic equation question that they answered incorrectly and uploads it to the system through a smartphone client. The picture content includes "question stem text: Solve the equation x²-5x+6=0, and the student's original answer: x=2". The system uses OCR technology to parse the question stem text in the image and obtains "Solve the equation x²-5x+6=0". Then, the semantic parsing model extracts the key information of the question stem, which corresponds to the subject information as mathematics, the question type information as calculation question, the target variable (target solution term) as x, and the core condition of the question stem as the quadratic equation x²-5x+6=0. The system initially predicts that the student's error type for this question is an error in the completeness of the result.

[0029] Furthermore, the extracted question stem information is compared with knowledge nodes in the knowledge graph to determine the target knowledge point corresponding to the incorrect question. Specifically, when comparing the question stem information with the knowledge node, the overall similarity is determined by whether it meets a similarity threshold, thus confirming whether the knowledge point corresponding to the knowledge node is the target knowledge point of the incorrect question data. In this embodiment, the overall similarity is determined based on semantic similarity, chapter position matching degree, and question type feature matching degree. Taking the key question stem information determined above as an example, by comparing it with the junior high school mathematics knowledge graph, the similarity between the question stem information and the node "Solving Quadratic Equations by Factorization" in the knowledge graph is calculated to be 0.92 (semantic similarity), 0.95 (chapter position matching degree), and 0.9 (question type feature matching degree). The overall similarity result exceeds the similarity threshold, such as 0.85, thus determining the target knowledge point as "Solving Quadratic Equations by Factorization".

[0030] It should be noted that after identifying the target knowledge points corresponding to the incorrect question data, structured data containing the target knowledge points is constructed as the data foundation for the subsequent generation of guiding question chains. In this embodiment, the structured data includes subject information, chapter information, target knowledge points, question type information, and error type prediction information. For example, the structured data corresponding to the target knowledge point "solving quadratic equations by factorization" is specifically: "Subject: Mathematics; Chapter: Junior High School Mathematics Grade 8, Volume 2 - Quadratic Equations; Knowledge Point: Solving quadratic equations by factorization; Question Type: Calculation Problem; Error Type Prediction: Error in Result Incompleteness (To be verified)". This embodiment identifies accurate target knowledge points by performing structured understanding and knowledge mapping on students' incorrect questions, enabling subsequent guiding question generation and training plan confirmation to focus on students' real knowledge gaps and ensure teaching effectiveness.

[0031] In one implementation, a corresponding teaching chain structure is determined based on the target knowledge point, and the thinking prompts for guiding questions in the teaching chain structure are adjusted based on the question type information in the incorrect question data to generate a guiding question chain corresponding to the target knowledge point, including: The teaching chain structure is determined based on the target knowledge points. The teaching chain structure is set with multiple guiding questions that represent the problem-solving ideas. Each guiding question includes a step number, question text information, thinking prompts and expected answer information. Based on the question stem information in the incorrect question data, the corresponding question type information is determined, and the thinking prompts in the guiding questions are adjusted according to the question type characteristic rules corresponding to the question type information. The adjusted teaching chain structure is then used as the guiding question chain for the target knowledge point.

[0032] In this embodiment, a multi-step guided question chain is generated based on the target knowledge point, adhering to the teaching guidance principle to ensure that students can think independently and gradually build correct problem-solving strategies when answering questions in the guided question chain. Specifically, a knowledge structure template is invoked based on the determined target knowledge point to determine the corresponding teaching chain structure. It should be noted that each knowledge point has a pre-set teaching chain structure. The teaching chain structure for mathematics includes five core stages: understanding the problem, selecting a method, deriving the calculation, verifying the result, and summarizing the knowledge point. Each core stage corresponds to a guided question, and each guided question includes a step number, question text information, thinking prompts, and expected answer information.

[0033] Furthermore, based on the types of incorrect questions corresponding to the incorrect question data, the thinking prompts in the guiding questions are determined, and the prompting method is determined according to the differences in question types. For calculation problems, based on the characteristics and rules of calculation problems, the corresponding thinking prompts are determined, focusing on step-by-step decomposition guidance and method application guidance, thereby determining the corresponding thinking prompt information. After adjusting the thinking prompt information content of the teaching chain structure accordingly, the guiding question chain is obtained.

[0034] Taking the concept of quadratic equations in one variable as an example, the corresponding guiding question chain is as follows: Step 1: What type of math problem is this? (Hint: You can determine this based on the number and degree of the unknowns in the problem; Expected answer format: multiple choice, options include "linear equation in one variable", "quadratic equation in one variable", and "linear equation in two variables"). Step 2: What are some common methods for solving this type of problem? (Thinking prompts: Recall typical problem-solving approaches for this knowledge point, such as the formula method, factorization method, etc.; Expected answer format: written explanation, listing at least one method). Step 3: If you choose one of the methods you mentioned, what operation should you perform in the first step? (Hint: Taking factorization as an example, you need to first transform the equation into its standard form; expected answer format: textual description or formula writing). Step 4: After calculating the preliminary result based on your solution steps, how would you verify whether the result is correct? (Hint: You can substitute the result into the original conditions of the question to verify; Expected answer format: a written description of the verification process). Step 5: Through the solution process of this question, what do you think is the core knowledge point mainly tested in this question? (Thinking prompt: Summarize based on the concepts, formulas or problem-solving methods involved in the question; expected answer format: textual summary).

[0035] This approach guides students through a chain of questions to gradually develop their problem-solving strategies, thereby establishing correct problem-solving thinking, actively constructing problem-solving logic, and improving their ability to apply knowledge.

[0036] In one possible implementation, the corresponding teaching chain structure is determined based on the target knowledge point, and the thinking prompts for guiding questions in the teaching chain structure are adjusted based on the question type information in the incorrect question data to generate a guiding question chain corresponding to the target knowledge point. The implementation also includes: Obtain context adjustment parameters, which include user identity information, difficulty information of incorrect questions, and user's historical incorrect questions; Adjust the number and difficulty level of the guiding questions in the guiding question chain based on the context adjustment parameters.

[0037] In this embodiment, based on adjusting the thinking prompts in the guiding questions, the guiding question chain is further adjusted with reference to the context adjustment parameters, and the difficulty and number of guiding questions in the guiding question chain are adjusted with reference to student-related information, so that students with different cognitive levels can receive appropriate learning guidance and meet the needs of different users.

[0038] It should be noted that the context adjustment parameters in this embodiment include user identity information, error difficulty information, and user historical error information. User identity information includes student grade level information, such as first year of junior high school, second year of junior high school, etc. User historical error information includes the current user's historical accuracy rate and knowledge mastery level for the same knowledge point. For example, the student's past accuracy rate for quadratic equations is 70%, indicating a moderate level of knowledge mastery. Specifically, for lower-grade students or students with lower knowledge mastery, the number of basic guiding questions in the guiding question chain can be increased, while the number of abstract thinking guiding questions can be reduced. For higher-grade students or students with higher knowledge mastery, the basic guiding questions can be simplified, while the number of comprehensive application thinking questions can be increased.

[0039] This embodiment targets specific knowledge points and, in addition to combining knowledge structure templates and question type feature rules, also comprehensively considers context adjustment parameters and adjusts the final guiding question chain based on students' historical data to adapt to different students' learning situations and meet the needs of different users.

[0040] In one possible implementation, the guiding question chain is verified, and the verified guiding question chain is sent to the client to collect corresponding answering behavior data, including: The question text, thinking prompts, and expected answers for each guiding question in the guiding question chain are compared with the pre-built target knowledge base to verify the correctness of the question text, thinking prompts, and expected answers. Based on the question text information of each guiding question, determine whether the guiding question involves calculation content information, and based on the simulation calculation results, determine whether the calculation content information is correct; If the question text, thought prompts, expected answers, and calculations are all correct, a valid guiding question chain is obtained. The verified guide question chain is sent to the client to collect corresponding answer behavior data.

[0041] In this embodiment, before outputting the guidance question chain to the client, the verification module performs factual consistency and calculation correctness checks on the guidance question chain, reducing the risk of factual errors and calculation deviations in the content pushed to the student terminal. The factual consistency check of the guidance question chain is achieved by comparing the guidance questions with a pre-built target knowledge base. It should be noted that the pre-built target knowledge base in this embodiment is a manually reviewed and confirmed knowledge base, including standard content such as standard concept definitions, formulas, theorems, and corresponding normative expressions for each subject. The comparison process uses semantic similarity calculation and logical matching to determine whether the descriptions of the question text information, thinking prompts, and expected answers in the guidance questions are consistent with the standard information content in the target knowledge base. For example, when verifying guidance questions related to the "definition of a quadratic equation," it ensures that the corresponding description of the question does not conflict with the definition in the target knowledge base that "a quadratic equation is an algebraic equation with only one unknown, and the highest power of the unknown is 2," thereby achieving the factual consistency check of the guidance questions.

[0042] Furthermore, the correctness of calculations is verified for guiding questions involving computational processes within the guiding question chain. In this embodiment, guiding questions involving computational processes refer to questions that include numerical calculations, formula derivations, etc., such as equation transformations or geometric formula applications. The accuracy of the computational information in the guiding questions is determined based on the corresponding simulation results. It should be noted that the simulation results in this embodiment are obtained through simulation calculations using a symbolic computation engine and geometric proof auxiliary tools. The standard results (simulation results) output by these tools are compared with the expected calculation results in the guiding questions to determine if any calculation errors exist, thus verifying the correctness of the calculations in the guiding questions. After verifying factual consistency and computational correctness, the risk of factual errors and calculation deviations in the guiding content is effectively reduced, thereby obtaining accurate student answer behavior data as an accurate basis for subsequent error determination.

[0043] In addition, this embodiment also includes: verifying the logical continuity of the thinking path setting logic of the guiding question chain to ensure the logical continuity of each guiding question in the guiding question chain. For example, whether the guiding question of method selection is omitted in the guiding question chain and students are directly guided to perform calculation derivation to answer the guiding question. If so, it is determined that the teaching rationality is not up to standard, indicating that the guiding question chain skips the core thinking steps. Furthermore, this embodiment also includes judging whether the question text information and thinking prompt information of the guiding questions in the guiding question chain contain the target answer. It should be noted that the target answer here refers to the correct answer of the question itself. For example, judging whether the guiding question directly contains the statement "The correct answer to this question is x=2". If so, it is determined that the teaching rationality is not up to standard because the relevant content of the guiding question reveals the final answer too early, deviating from the guiding purpose.

[0044] Therefore, this embodiment comprehensively judges whether the guiding question chain has factual errors, calculation deviations, or unreasonable teaching methods to ensure the reliability of the guiding question chain pushed to the client. When the verification module detects that the guiding question chain has factual errors, calculation deviations, or unreasonable teaching methods, it determines that the verification has failed, automatically discards the output, and sends a regeneration instruction to the guiding task generation module that generated the guiding question chain, clearly indicating the reason for the verification failure. For example, the expected relevant description of a guiding question does not match the standard formula in the knowledge base, or a guiding step skips the core step of method selection, etc.

[0045] In one possible implementation, determining the corresponding error cause based on the answer behavior data includes: Determine the behavior sequence vector corresponding to the answer behavior data; The behavioral sequence vectors and incorrect question data are initially matched with the pre-constructed error cause labeling system to determine the candidate error cause results; The target error result is determined from the candidate error results based on the user's historical learning records and similar wrong question data.

[0046] In this embodiment, the root cause of a student's incorrect answers is identified by comprehensively judging the student's answer behavior data during the process of answering the guided question chain and the student's initial incorrect answer data. It should be noted that the answer behavior data in this embodiment is behavioral data monitored and collected in real time during the student's answering of the guided question chain, including answer time data, number of modifications data, accuracy rate, and confidence score. Specifically, the answer time data is determined based on the time difference between the start time and submission time of each guided question; the number of modifications data is determined based on the number of modifications made after submission for each guided question; the accuracy rate is determined based on the matching degree between the student's actual answer to the guided question and the expected answer information in the guided question; and the confidence score is determined based on the degree of answer information chosen by the student, such as high, medium, or low.

[0047] The above-mentioned answer behavior data is encapsulated into standardized behavior sequence vectors. Through a multi-factor decision-making mechanism, the student's behavior sequence vectors and the types of wrong questions corresponding to the wrong questions are mapped to a pre-constructed error cause labeling system to initially match candidate error cause results.

[0048] It should be noted that the error labeling system in this embodiment includes at least five error types: conceptual errors, calculation errors, method selection errors, comprehension bias errors, and non-cognitive interference errors. Conceptual errors refer to errors caused by students' misunderstanding of core concepts and definitions of knowledge points, such as confusing the definitions of "quadratic equation in one variable" and "linear equation in two variables." Calculation errors refer to errors made by students in numerical calculations, symbolic operations, and formula substitution steps, such as writing the wrong symbol in factorization or substituting the wrong value in the root formula. Method selection errors refer to students not choosing the optimal solution method or applying the method logically incorrectly, such as using the formula method to solve equations that can be easily factored, leading to complex calculations and errors. Comprehension bias errors refer to students' misunderstanding of the conditions in the question stem, the target solution term, etc., such as omitting the square symbol in the question stem or misunderstanding the quantitative relationships in profit calculation problems. Non-cognitive interference errors refer to errors caused by non-knowledge factors such as lack of concentration, time pressure, and nervousness, such as omitting solution steps due to time constraints or carelessly overlooking key information in the options.

[0049] By comparing the behavioral features represented by behavioral sequence vectors with the original incorrect question data and the aforementioned error cause labeling system, candidate error cause results are initially screened. Simultaneously, the target error cause result is finally selected by combining the student's past learning records and similar incorrect question data. In this embodiment, the past learning records include the student's past error cause data for similar questions, knowledge point mastery trajectory, etc., while similar error question data includes common error cause data from other students when answering similar questions. The label weights of the candidate error cause results are adjusted by comprehensively considering the past learning records and similar error question data, thereby determining the final error cause result.

[0050] It should be noted that after obtaining the error cause results in this embodiment, structured data containing the error cause results is constructed as the data foundation for generating subsequent training schemes. Specifically, the structured error cause data in this embodiment includes the error cause result (error cause category), suggested reinforcement directions, and an information index. For example, "Error cause result: Incorrect method selection; Suggested reinforcement direction: Strengthen question type identification training, improve the accuracy of problem-solving method selection; Confidence index: 0.87." It should be noted that the confidence index reflects the reliability of the error cause determination; the higher the confidence index, the more reliable the determination result.

[0051] In one possible implementation, a corresponding error-correction training plan is generated based on the error cause, including: Based on the error causes, a set of corresponding error correction exercises is generated. The set of error correction exercises includes multiple exercises of different difficulty levels. The difficulty level of the first exercise matches the difficulty level of the error data. The difficulty level of each remaining exercise is determined based on the answer to the previous exercise.

[0052] In this embodiment, a corresponding error correction training plan is generated based on the obtained error causes and the student's current level of knowledge mastery. Preferably, this error correction training plan is a set of reinforcement training questions, which includes multiple reinforcement training questions at different difficulty levels, and preferably at least three. It should be noted that the difficulty level of the first reinforcement training question in the set matches the difficulty level of the original error. That is, the first reinforcement training question is a basic repetition question with a lower difficulty level, aiming to help students reproduce the correct problem-solving process and strengthen their basic mastery of core knowledge points and problem-solving methods. For example, if the original error was factoring a quadratic equation with integer coefficients, the first reinforcement training question in the corresponding set would also be set as factoring a quadratic equation with integer coefficients, only changing the coefficient values.

[0053] Furthermore, the second consolidation exercise is an adjusted difficulty level based on the first one. It should be noted that when the first consolidation exercise is answered correctly, the difficulty of the second exercise increases accordingly. This is a moderately challenging exercise designed to test students' ability to flexibly apply knowledge points and avoid rote memorization of solution steps. For example, the question scenario might be adjusted based on a basic repetition question, such as changing integer coefficients to decimal coefficients, or adding distracting conditions, such as adding irrelevant information like "one root of the equation is positive" to the question stem. When the first consolidation exercise is answered incorrectly, the second consolidation exercise maintains the same difficulty level as the first or decreases it to reinforce students' mastery of the basic content.

[0054] Furthermore, the third consolidation exercise is an adjusted difficulty level based on the second consolidation exercise. It should be noted that when the answer to the second consolidation exercise is correct, the difficulty of the third consolidation exercise increases accordingly, becoming a higher-level application problem aimed at cultivating students' comprehensive problem-solving and knowledge transfer abilities. For example, combining the application scenario of quadratic equations with actual profit calculation, a problem could be designed to calculate the profit change after a price adjustment for a certain product, achieving cross-knowledge point or cross-scenario transfer training. When the answer to the second consolidation exercise is incorrect, the third consolidation exercise maintains the same difficulty level as the second exercise or decreases the difficulty level to suit the students' actual ability level.

[0055] Therefore, this embodiment adjusts the question-setting strategy in real time based on students' performance, such as accuracy, to form an adaptive reinforcement path. If a student answers reinforcement questions incorrectly consecutively, the current difficulty level is automatically maintained or reduced, and the number of similar question types is increased. If a student answers reinforcement questions correctly consecutively, the difficulty of subsequent reinforcement questions is gradually increased, adding questions that integrate application and transfer of knowledge to promote advanced training of students' abilities. This achieves full automation from inputting incorrect questions to guiding problem-solving, to error analysis and personalized reinforcement, improving the reliability of error tutoring.

[0056] In one possible implementation, after generating the corresponding error training plan based on the error cause results, it further includes: Collect training data corresponding to the incorrect question training plan, and generate corresponding learning reports based on the training data.

[0057] In this embodiment, a corresponding learning report is generated based on the student's training data from the error correction training program, forming a data loop. The core content of the learning report in this embodiment includes the source of the errors, the distribution of knowledge points, the analysis results of the error causes, the completion status of the reinforcement exercises, and the suggested learning path, so that students can clearly understand their own learning situation and realize a closed loop of learning guidance.

[0058] The method disclosed herein will be described in detail below.

[0059] The overall system architecture includes a client, a task scheduling center, a functional module layer, and a feedback database; The client, or student terminal, provides an interactive entry point for students, enabling them to input incorrect questions (text-based and image-based), view guided tasks, answer reinforcement questions, and view their learning progress and reports. The task scheduling center connects student terminals and various functional modules, and is responsible for tasks such as receiving tasks, arranging modules, and synchronizing status. The functional module layer includes a wrong question analysis module, a guided task generation module, a knowledge point identification module, a verification module, a wrong cause analysis module, and a dynamic consolidation module. Each module works collaboratively according to the instructions of the task scheduling center. The feedback database is used to store interactive data generated during system operation, such as student answer behavior data, guided task verification results, and error analysis results, providing data support and driving continuous iteration of system performance. The implementation process of the error correction tutoring task follows a logic of student-driven needs, module collaborative response, and closed-loop optimization. Specifically, students submit error data through their student terminals. The task scheduling center assigns the errors to the error analysis module, which analyzes the question stem and then synchronizes it to the knowledge point identification module for knowledge point matching. The knowledge point identification results are fed back to the task scheduling center, which then assigns the task to the guidance task generation module, generating a multi-step guidance question chain. This chain is then transmitted to the verification module for verification of factual consistency, calculation correctness, and teaching rationality. Once verified, the task scheduling center pushes the results to the student terminal. Students answer the guidance questions in the chain step by step on their student terminals. The system records the answering behavior data in real time and transmits it to the error cause analysis module for error diagnosis. The error cause diagnosis results are fed back to the task scheduling center, which triggers the dynamic consolidation module to generate consolidation exercises with progressively increasing difficulty. After completing the consolidation exercises, students receive a learning report based on their performance, including the source and distribution of the errors and knowledge points, the error cause analysis results, the completion status of the consolidation exercises, and suggested learning paths. All of the above-mentioned interactive data, including answer behavior data, verification results, error reason results, and completion status of reinforcement questions, are synchronized to the feedback database for system optimization, forming a complete closed loop from generation to verification, then to feedback, and finally to optimization.

[0060] The methods disclosed herein can be applied to the following scenarios, including online tutoring platforms for primary and secondary schools, providing K-12 students with error analysis, guided learning, and dynamic reinforcement services for science subjects such as mathematics and physics, adapting to the cognitive levels of students at different grade levels; smart learning machines and homework apps, integrated into smart learning machines and homework correction apps as core error tutoring functions to help students independently solve errors in their homework and strengthen weak knowledge points; and educational big data analysis systems, providing schools and educational institutions with data analysis services such as the distribution of student errors, mastery of knowledge points, and statistics on error types, assisting teachers in adjusting teaching strategies and optimizing teaching plans.

[0061] The method disclosed herein will be explained in detail using the scenario of "tutoring for incorrect problems in quadratic equations in junior high school mathematics" in the K-12 education field.

[0062] Implement environmental preparation Hardware environment: Student terminals use smartphones (iOS / Android system) or tablets, supporting photo taking and text input functions; the system server uses a cloud server (configured with 8 cores and 16G memory) to ensure concurrent operation of multiple modules and data storage; Software environment: The OCR technology adopts a mature image text recognition interface, which supports accurate recognition of mathematical formulas and handwritten text; the semantic parsing model adopts a fine-tuned model based on BERT, which is adapted to the semantic understanding of mathematical questions; the knowledge graph covers all the knowledge points of junior high school mathematics, and each knowledge point node includes attributes such as concept definition, common solutions, typical error labels, and real question examples; Data preparation: a pre-built, manually reviewed knowledge base (containing junior high school math formulas, theorems, and problem-solving guidelines), an error labeling system (covering five major categories of error causes, including conceptual errors and calculation errors), and an initial reinforcement question bank (each knowledge point corresponds to at least 100 questions of different difficulty).

[0063] like Figure 2 As shown, the implementation steps include: S1. Collection and analysis of incorrect answers, including: S11. Students use their smartphones to select the "Upload Incorrect Question" function to upload a quadratic equation question they answered incorrectly before: "Solve the equation x²-5x+6=0, the student's original answer was x=2". They then select the "Upload Photo" option to upload a picture of the question. S12. The error analysis module uses OCR technology to recognize the question text in the image and obtain the question stem information "x²-5x+6=0". It extracts the key information of the question stem through a semantic parsing model: the subject is mathematics, the question type is a calculation question, the target variable is "x", and the core condition of the question stem is "a quadratic equation x²-5x+6=0". S13. Based on the junior high school mathematics knowledge graph, the similarity between the question stem and the node "Solving quadratic equations in one variable by factorization" in the knowledge graph is calculated. The semantic similarity is 0.92, the chapter position matching degree is 0.95, the question type feature matching degree is 0.9, and the comprehensive similarity exceeds the preset threshold of 0.85. The target knowledge point is determined to be "Solving quadratic equations in one variable by factorization". S14 outputs a structured result, with the following content: "Subject: Mathematics; Chapter: Junior High School Mathematics Grade 8, Volume 2 - Quadratic Equations in One Variable; Knowledge Point: Solving Quadratic Equations in One Variable by Factoring; Question Type: Calculation Problem; Error Type Prediction: Error in Completeness of Result (To be verified)".

[0064] S2, Guiding the generation of the problem chain, including: S21. Based on the knowledge point of "solving quadratic equations by factorization," the knowledge point structure module was invoked to determine the corresponding teaching chain structure as five steps: "understanding the problem—selecting a method—deriving the calculation—verifying the result—summarizing the knowledge point." Combining the characteristics and rules of calculation problems, the guiding questions were determined to emphasize step decomposition and method application hints. Referring to the student's historical data, the student's past accuracy rate on quadratic equation problems was 60%, indicating a moderate level of knowledge mastery. Therefore, four core steps in the five-step teaching chain were retained, while the knowledge point summarization step was omitted and will be added later in the learning report. S22. Call the boot task generation module to generate a multi-step boot issue chain, the details of which are as follows: Step 1: What type of algebraic equation is this? (Hint: You can determine this based on the number of unknowns and the highest power; Expected answer format: multiple choice, Option A: linear equation in one variable, Option B: quadratic equation in one variable, Option C: linear equation in two variables). Step 2: What are some common methods for solving this type of equation? (Hint: Recall typical methods for solving quadratic equations, such as the quadratic formula and factorization; Expected answer format: written explanation, listing at least one method). Step 3: If you choose the factorization method to solve the problem, what form should the equation be transformed into in the first step? (Hint: The core of the factorization method is to transform the equation into the form "(xa)(xb)=0"; expected answer format: textual description or formula writing). Step 4: Based on your factorization results, what are the solutions to the equation? How can you verify that these solutions are correct? (Hint: When the factorization result is "(xa)(xb)=0", the solutions are x=a and x=b. To verify, substitute the solutions into the original equation; expected answer format: a textual description of the solution results and the verification process).

[0065] S23. Output the above-mentioned guidance problem chain to the task scheduling center and wait for the verification module to verify it.

[0066] S3. Verification and knowledge base comparison, including: S31. Compare the generated guiding questions with the standard knowledge base. For example, the definition of a quadratic equation in step 1 of step S22 is consistent with the knowledge base "an algebraic equation containing only one unknown and whose highest power is 2". The statement "factorization is a commonly used solution method for quadratic equations" in step 2 is consistent with the knowledge base content. The factual consistency is determined to be passed. S32. For the "factorization form" problem in step S22, call the symbolic calculation engine, input "x²-5x+6=0", simulate the factorization process, and get "(x-2)(x-3)=0", which is consistent with the expected answer to the problem, and the calculation is judged to be correct. S33. Examining the guiding question chain, no direct provision of the final answer was found, i.e., the solution to the equation is x=2 and x=3. Moreover, the core links of "understanding the problem - choosing the method - derivation of the operation - verification of the result" were covered, and no key steps were skipped. Therefore, the teaching is deemed reasonable. S34. Output the structured results that have passed verification. The task scheduling center will then push the problem chain to the student client.

[0067] S4. Student responses and behavior data collection, including: S41. As students answer the guiding questions step by step on the client side, the system records the following information in real time: Step 1: Select option B (correct), answer time is 12 seconds, no modifications, select "high" for confidence score; Step 2: Fill in "Factorization method, Formula method" (correct), answer time 20 seconds, no corrections, select "High" for confidence score; Step 3: Write "(x-2)(x-3)=0" (correct), answer time: 35 seconds, revised once (originally written as "(x+2)(x+3)=0", then corrected), confidence score: "medium"; Step 4: Write "The solutions are x=2 and x=3. Verification: Substituting x=2 into the original equation, 2²-5×2+6=4-10+6=0, which is true; Substituting x=3 into the equation, 3²-5×3+6=9-15+6=0, which is true" (Correct). Answering time: 45 seconds. No modifications. Confidence score: "High".

[0068] S42. Encapsulate the above behavioral data into a behavioral sequence vector, such as "Question answering time: [12,20,35,45] seconds; Number of modifications: [0,0,1,0]; Accuracy: [100%,100%,100%,100%]; Confidence score: [High, High, Medium, High]", and transmit it to the error analysis module.

[0069] S5. Error analysis and handling, including: S51. After receiving the behavior sequence vector, combine it with the original incorrect question information, such as the student's original answer omitting x=3, and initiate multi-factor decision analysis. S52. By comparing the behavioral characteristics (correct answer in step 4, but missing a solution in the original wrong question) with the error cause labeling system, the "error in completeness of result" is initially screened as a process dimension error and is a candidate error cause. Referring to the student's historical data, the student had missed a solution in two previous quadratic equation problems. The weight of the error in completeness of result label was adjusted to determine that the error cause is "error in completeness of result". S53. Output structured error results, including "Error Category: Error in Completeness of Result (Process Dimension); Error Description: Able to correctly grasp the factorization method to solve quadratic equations, but omitted one solution of the equation in the original answer, indicating incomplete checking of the solution result; Suggested consolidation direction: Strengthen training in verifying the completeness of solutions to quadratic equations, focusing on practicing the 'post-solution verification' step; Confidence Index: 0.89".

[0070] S6. Generation of dynamic reinforcement training question sets, including: S61. Based on the error cause of the result incompleteness, the system generates three reinforcement questions of progressively increasing difficulty: Question 1 (Basic Repetition Question, Difficulty Level 1): "Solve the equation x²-7x+12=0" (The question scenario is the same as the original wrong question, only the coefficients are changed, the goal is to strengthen the "completeness of the solution"); Question 2 (Moderately challenging, Level 2 difficulty): "Solve the equation 2x²-5x-3=0" (Change the coefficients to decimals to increase the difficulty of factorization, and require writing out all solutions and verifying them). Question 3 (Application problem, difficulty level 3): "The area of ​​a rectangle is 12cm², and its length is 1cm more than its width. Find the length and width of the rectangle (round the result to the nearest integer)" (Combining quadratic equations with a geometric scenario, you need to first establish the equation, then solve it and verify the reasonableness of the result). The difficulty level is adjusted based on students' performance during the answering process. The student correctly completed the first question (30 seconds, no corrections), and the system judged that the student had mastered the basics. The second question maintained the difficulty level of 2. The student correctly completed the second question (45 seconds, 1 revision), and the system determined that the student could advance to the next level. The difficulty of the third question was increased to level 3.5 (adding the requirement of "reasonableness analysis of the result", such as "the side length of the rectangle must be a positive number, and negative solutions are excluded"). The student correctly completed question 3 (60 seconds, no modifications). The system determined that the reinforcement effect met the standard and stopped the current reinforcement training.

[0071] S7. Learning report generation, including: S71. A learning report is generated after the consolidation training, and its core content includes: Sources of incorrect answers and distribution of knowledge points: The incorrect answers come from the chapter on "quadratic equations in one variable", corresponding to the knowledge point of "solving by factorization", which is a key knowledge point this semester; Error analysis results: The error was caused by "incomplete results", which stemmed from the failure to verify the completeness of the results after solving the problem. It is necessary to strengthen the habit of "verifying after solving". Consolidation Exercise Completion Status: All 3 consolidation exercises were answered correctly. The basic review questions took 30 seconds (better than the average time of 40 seconds), and the application questions took 60 seconds (in line with the average time). The consolidation effect was rated as "Excellent". Suggested learning path: Complete 5 practice problems on "Complete the integrity verification of solutions to quadratic equations in one variable" within 1 week; review the chapter on "Quadratic equations in one variable and their practical applications" within 2 weeks; and review the questions you got wrong and the reinforcement questions once a month. S72. Synchronize the data from this interaction, including the results of incorrect question analysis, the chain of guiding questions, the verification results, student behavior data, the results of error causes, and the completion status of reinforcement questions, to the feedback database. The task scheduling center will trigger the model optimization task in the early morning of the next day, fine-tune the guiding task generation model based on the feedback data, increase the generation weight of guiding questions related to "completeness of solution", fine-tune the error cause analysis model, optimize the behavioral feature matching rules for "errors in completeness of result", and update the reinforcement question bank, adding two quadratic equation questions related to the application of rectangle area.

[0072] As described in the implementation process above, by structurally understanding and precisely mapping students' incorrect answers to knowledge points, the knowledge point of "solving quadratic equations by factorization" was effectively identified, providing a precise direction for subsequent guidance and consolidation. Furthermore, personalized teaching guidance was achieved through the generation of guided tasks, generating a four-step guided question chain based on students' knowledge mastery. This ensured that answers were not directly leaked while effectively guiding students to independently construct problem-solving strategies. In addition, it is important to emphasize that before transmitting the guided question chain to the student's client, the content was verified from three dimensions: factual, computational, and pedagogical, ensuring the reliability of the output content. Simultaneously, the system could automatically identify the causes of errors and dynamically generate consolidation tasks, accurately locating errors in the completeness of results, generating consolidation questions of progressive difficulty, and adaptively adjusting the difficulty based on the students' performance, achieving personalized reinforcement. The entire implementation process established a continuous learning loop of generation, verification, feedback, and optimization, improving the accuracy and relevance of subsequent services through data feedback models and question banks.

[0073] In summary, this disclosure uses knowledge graph matching to ensure the accuracy of guidance direction and adds a verification step for the guidance question chain to verify the guidance content in real time. Simultaneously, it verifies the guidance effect through student answer feedback, forming a full-process quality control system to ensure the accuracy and rationality of teaching content. Furthermore, it infers cognitive biases based on student answer behavior data (time spent, number of revisions, etc.), accurately identifies the types of errors, and generates progressively more difficult reinforcement questions based on these errors. The difficulty is adaptively adjusted based on answer performance, avoiding a one-size-fits-all reinforcement model and achieving personalized learning. All interaction data is entered into a feedback database and regularly used for model fine-tuning and question bank updates, forming a sustainable AI teaching evolution cycle from service to feedback, to optimization, and back to service, driving continuous improvement in system performance.

[0074] To implement the above method, this disclosure also provides an example of a mistake-correction tutoring system 300 based on generative artificial intelligence, such as... Figure 3 As shown, the system includes: The error analysis module 301 is used to acquire error data and determine the target knowledge points corresponding to the error data; The task generation module 302 is used to determine the corresponding teaching chain structure based on the target knowledge point, and adjust the thinking prompts of the guiding questions in the teaching chain structure based on the question type information in the wrong question data, so as to generate a guiding question chain corresponding to the target knowledge point. The verification module 303 is used to verify the guiding question chain and send the verified guiding question chain to the client to collect the corresponding answer behavior data. Error analysis module 304 is used to determine the corresponding error results based on the answer behavior data; The dynamic consolidation module 305 is used to generate corresponding error training schemes based on the error causes.

[0075] The system in this embodiment also includes a task scheduling center (AI Gateway), which adopts an asynchronous messaging mechanism to achieve task distribution, status synchronization, and multi-task concurrency among various modules within the system, ensuring a low-latency interactive experience. Specifically, in the entire multi-agent collaborative architecture system, the task scheduling center is responsible for receiving requests such as incorrect question input and answer results from student terminals. According to the preset teaching process—namely, incorrect question analysis, knowledge point identification, guided task generation, guided task verification, student answering guided tasks, error analysis, and dynamic consolidation training—the center arranges the execution order of each module to ensure collaborative work between modules, forming a complete teaching loop. Furthermore, based on the request type, such as incorrect question analysis requests, guided task generation requests, and consolidation question answering requests, tasks are assigned to the corresponding functional modules: the incorrect question analysis module, the guided task generation module, and the dynamic consolidation module. Simultaneously, the system synchronizes the task execution status of each module in real time, such as incorrect question analysis in progress, guided task generation completed, verification failed, and regeneration in progress, and feeds the status back to the student terminal so that students understand their current learning progress. It is also used to maintain a unified log system, recording the input and output data of each module, task execution time, error information, etc. The log data is used for teaching quality monitoring, such as analyzing the verification pass rate of guided tasks and the accuracy of error analysis. On the other hand, it is used for subsequent model optimization, providing data support for the fine-tuning of generative AI models and error analysis models.

[0076] In addition, this embodiment achieves controllable output of generative AI through the collaborative operation mechanism of the task generation module and the verification module, effectively avoiding the illusion phenomenon of traditional generative models and improving content reliability. It can also adapt to the cognitive characteristics of different students, automatically identify the causes of errors and adjust learning strategies to form a personalized, progressive learning path, providing students with accurate and safe tutoring services for incorrect answers.

[0077] Therefore, this embodiment provides an AI education system that integrates generation, verification, and consolidation, automating the entire process from student input of incorrect questions to guiding students in solving problems, analyzing the causes of errors, and then to personalized consolidation. Compared with related educational systems, it not only ensures the reliability of the guidance content but also provides targeted personalized learning guidance for students, ensuring a closed-loop learning process for correct question tutoring, and significantly improving the reliability and teaching value of generative AI in the field of education.

[0078] In one embodiment, the error analysis module 301 is used to acquire error data and extract the question stem information from the error data; The question stem information is compared with the knowledge nodes in the knowledge graph to determine the corresponding comprehensive similarity. In response to a comprehensive similarity exceeding a similarity threshold, the corresponding knowledge node is identified as the target knowledge node, and the knowledge point corresponding to the target knowledge node is used as the target knowledge point for the incorrect question data.

[0079] In one embodiment, the guided task generation module 302 is used to determine the corresponding teaching chain structure according to the target knowledge point. The teaching chain structure is set with multiple guided questions that represent the problem-solving ideas in sequence, and each guided question includes a step number, question text information, thinking prompt information and expected answer information. Based on the question stem information in the incorrect question data, the corresponding question type information is determined, and the thinking prompts in the guiding questions are adjusted according to the question type characteristic rules corresponding to the question type information. The adjusted teaching chain structure is then used as the guiding question chain for the target knowledge point.

[0080] In one embodiment, the guided task generation module 302 is further configured to obtain context adjustment parameters, which include user identity information, incorrect question difficulty information, and user historical incorrect question information; Adjust the number and difficulty level of the guiding questions in the guiding question chain based on the context adjustment parameters.

[0081] In this embodiment, to ensure system security and stability, the specific question generation logic of the guidance task generation module is based on the joint decision-making of the internal model template and the teaching graph. It does not involve a fixed algorithm or a publicly available Prompt template. Externally, only the final generated guidance question result can be observed, and the generation logic cannot be deduced in reverse.

[0082] In one embodiment, the verification module 303 is used to compare the question text information, thinking prompt information and expected answer information of each guiding question in the guiding question chain with the pre-built target knowledge base to verify whether the question text information, thinking prompt information and expected answer information are correct; Based on the question text information of each guiding question, determine whether the guiding question involves calculation content information, and based on the simulation calculation results, determine whether the calculation content information is correct; If the question text, thought prompts, expected answers, and calculations are all correct, a valid guiding question chain is obtained. The verified guide question chain is sent to the client to collect corresponding answer behavior data.

[0083] The verification module in this embodiment serves as the reliability assurance unit for the entire system. Operating independently of the aforementioned guided task generation module, it performs real-time verification of the generated guidance content before the guidance task, i.e., the guidance question chain, is output to the student terminal. This ensures the factual consistency, calculation accuracy, and pedagogical rationality of the content. Verification includes comparing with a standard knowledge base to verify factual consistency, calling deterministic solution tools to check results to verify calculation accuracy, and detecting premature revelation of answers or logical inconsistencies to verify pedagogical rationality. This multi-dimensional verification ensures the reliability of the generated content. Through the collaborative operation of the guided task generation module and the verification module, controllable output of generative AI is achieved, thereby avoiding the illusion phenomenon that easily occurs in generative models in related technologies.

[0084] In one possible implementation, the error analysis module 304 is used to determine the behavior sequence vector corresponding to the answer behavior data; perform preliminary matching between the behavior sequence vector and the wrong question data and the pre-constructed error cause label system to determine the candidate error cause results; and determine the target error cause result from the candidate error cause results based on the user's historical learning records and similar wrong question data.

[0085] In this embodiment, to ensure system security, the error determination mechanism is based on the dynamic model results generated by the system's autonomous learning. The specific algorithm and weights are not disclosed; only the logical path of the determination is disclosed, such as "based on the student's choice of the 'formula method' to solve factorization problems, and the number of modifications reaches 3 times, it is determined that the method selection is incorrect."

[0086] In one embodiment, the dynamic consolidation module 305 is used to generate a set of consolidation training questions based on the error cause results. The set of consolidation training questions includes multiple consolidation training questions with different difficulty levels. The difficulty level of the first consolidation training question matches the difficulty level of the error data. The difficulty level of each remaining consolidation training question is determined based on the answer result of the previous consolidation training question.

[0087] In one possible implementation, a report generation module is also included, which is used to collect training data corresponding to the incorrect question training scheme and generate a corresponding learning report based on the training data.

[0088] By way of example, this disclosure also provides an electronic device including at least one processor; and a memory connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the generative artificial intelligence-based error tutoring method described above.

[0089] By way of example, this disclosure also provides a computer-readable storage medium storing a computer program for performing the above-described error tutoring method based on generative artificial intelligence.

[0090] Figure 4 A schematic block diagram of an example electronic device 400 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0091] like Figure 4 As shown, the electronic device 400 includes a computing unit 401, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 402 or a computer program loaded from a storage unit 408 into a random access memory (RAM) 403. The RAM 403 may also store various programs and data required for the operation of the electronic device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0092] Multiple components in electronic device 400 are connected to I / O interface 405, including: input unit 406, such as keyboard, mouse, etc.; output unit 407, such as various types of displays, speakers, etc.; storage unit 408, such as disk, optical disk, etc.; and communication unit 409, such as network card, modem, wireless transceiver, etc. Communication unit 409 allows electronic device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0093] The computing unit 401 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above, such as a generative artificial intelligence-based error correction tutoring method. For example, in some embodiments, a generative artificial intelligence-based error correction tutoring method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by the computing unit 401, one or more steps of the generative artificial intelligence-based error correction tutoring method described above can be performed.

[0094] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0095] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0096] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0097] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means two or more, unless otherwise explicitly specified.

[0098] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A mistake correction method based on generative artificial intelligence, characterized by, The method includes: Obtain incorrect question data and determine the target knowledge points corresponding to the incorrect question data; Based on the target knowledge point, determine the corresponding teaching chain structure, and adjust the thinking prompts of the guiding questions in the teaching chain structure based on the question type information in the wrong question data, thereby generating a guiding question chain corresponding to the target knowledge point; The guided question chain is verified, and the verified guided question chain is sent to the client to collect the corresponding answer behavior data; The corresponding error causes are determined based on the question-answering behavior data; Based on the results of the errors, a corresponding training plan for incorrect questions is generated.

2. The error-correction tutoring method based on generative artificial intelligence according to claim 1, characterized in that, The process of acquiring incorrect question data and determining the target knowledge points corresponding to the incorrect question data includes: Obtain incorrect question data and extract the question stem information from the incorrect question data; The question stem information is compared with the knowledge nodes in the knowledge graph to determine the corresponding comprehensive similarity. In response to the overall similarity exceeding the similarity threshold, the corresponding knowledge node is determined as the target knowledge node, and the knowledge point corresponding to the target knowledge node is used as the target knowledge point of the incorrect question data.

3. The error-correction tutoring method based on generative artificial intelligence according to claim 1, characterized in that, The step of determining the corresponding teaching chain structure based on the target knowledge point, and adjusting the thinking prompts for guiding questions in the teaching chain structure based on the question type information in the incorrect question data, to generate a guiding question chain corresponding to the target knowledge point, includes: The corresponding teaching chain structure is determined based on the target knowledge points. The teaching chain structure is set with multiple guiding questions that represent the problem-solving ideas. Each guiding question includes a step number, question text information, thinking prompts and expected answer information. Based on the question stem information in the incorrect question data, the corresponding question type information is determined, and the thinking prompt information in the guiding questions is adjusted according to the question type feature rules corresponding to the question type information. The adjusted teaching chain structure is then used as the guiding question chain for the target knowledge point.

4. The error-correction tutoring method based on generative artificial intelligence according to claim 3, characterized in that, The step of determining the corresponding teaching chain structure based on the target knowledge point, and adjusting the thinking prompts for guiding questions in the teaching chain structure based on the question type information in the incorrect question data, to generate a guiding question chain corresponding to the target knowledge point, further includes: Obtain context adjustment parameters, which include user identity information, incorrect question difficulty information, and user's historical incorrect question information; The number and difficulty level of the guiding questions in the guiding question chain are adjusted according to the context adjustment parameters.

5. The error-correction tutoring method based on generative artificial intelligence according to claim 1, characterized in that, The process of verifying the guiding question chain and sending the verified guiding question chain to the client to collect corresponding answering behavior data includes: The question text information, thinking prompt information, and expected answer information of each guiding question in the guiding question chain are compared with the pre-built target knowledge base to verify whether the question text information, the thinking prompt information, and the expected answer information are correct; Based on the question text information of each guiding question, determine whether the guiding question involves calculation content information, and based on the simulation calculation results, determine whether the calculation content information is correct; If the question text information, the thinking prompt information, the expected answer information, and the calculation content information are all correct, a verified guiding question chain is obtained; The verified guiding question chain is sent to the client to collect the corresponding answering behavior data.

6. The error-correction tutoring method based on generative artificial intelligence according to claim 1, characterized in that, The step of determining the corresponding error cause based on the answer behavior data includes: Determine the behavior sequence vector corresponding to the answer behavior data; The behavior sequence vector and the incorrect question data are initially matched with the pre-constructed error cause labeling system to determine the candidate error cause results; The target error result is determined from the candidate error results based on the user's historical learning records and similar wrong question data.

7. The error-correction tutoring method based on generative artificial intelligence according to claim 1, characterized in that, The method for generating corresponding error training schemes based on the error causes includes: Based on the error causes, a corresponding set of error correction exercises is generated. The set of error correction exercises includes multiple exercises with different difficulty levels. The difficulty level of the first exercise matches the difficulty level of the error data. The difficulty level of each remaining exercise is determined based on the answer to the previous exercise.

8. The error-correction tutoring method based on generative artificial intelligence according to any one of claims 1-7, characterized in that, After generating the corresponding error training scheme based on the error cause results, it also includes: Collect training data corresponding to the incorrect question training scheme, and generate a corresponding learning report based on the training data.

9. A problem-solving tutoring system based on generative artificial intelligence, used to implement the problem-solving tutoring method based on generative artificial intelligence as described in any one of claims 1-8, characterized in that, The system includes: The error analysis module is used to acquire error data and determine the target knowledge points corresponding to the error data; The guided task generation module is used to determine the corresponding teaching chain structure based on the target knowledge point, and adjust the thinking prompts of the guided questions in the teaching chain structure based on the question type information in the wrong question data, so as to generate a guided question chain corresponding to the target knowledge point. The verification module is used to verify the guiding question chain and send the verified guiding question chain to the client to collect the corresponding answer behavior data; The error analysis module is used to determine the corresponding error results based on the answer behavior data; The dynamic reinforcement module is used to generate corresponding error training schemes based on the error causes.

10. An electronic device, characterized in that, include: At least one processor; And a memory connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the error tutoring method based on generative artificial intelligence as described in any one of claims 1-8.

Images