A multi-level heuristic problem-solving guidance method and system based on error cause tracing

Abstract

The application discloses a kind of multi-level heuristic problem solving guidance method and system based on error cause tracing, and the application is based on subject AI model in combination with whole subject knowledge graph, realize concept confusion, error cause such as calculation error is accurately identified;According to error cause type, match exclusive targeted guidance strategy, execute error cause counterquestion, knowledge point association and other multi-level step-by-step prompt, do not output final answer throughout, support user voice / text bidirectional interrogation to form interactive closed loop.The guidance intensity of the application is self-adaptively adjusted according to grade, subject and knowledge weakness degree;After user corrects error, same error cause variant question is pushed, and still not answered correctly, then open problem solving thought framework;The system continuously iterates and optimizes model and strategy, and also provides lightweight adaptation version for low-computing-power equipment.The application realizes personalized thinking training, cultivates user's independent problem solving ability, and solves knowledge gap from the root.

Description

Technical Field

[0001] This invention relates to the fields of AI education, subject-based problem-solving guidance, and personalized learning technology, specifically a multi-level heuristic problem-solving guidance method and system based on error cause tracing. Background Technology

[0002] Existing AI-assisted problem-solving technologies mostly output answers or standardized explanations directly, without accurately identifying and targeting the root causes of user errors, making it difficult to cultivate users' independent thinking and problem-solving abilities.

[0003] Some problem-solving guidance methods only use single-step prompts and do not match specific guidance strategies according to the type of error. The guidance effect is poor, and it cannot solve the user's knowledge gaps at the root. It is easy for users to become dependent on the answer and it is difficult to achieve personalized thinking training. Summary of the Invention

[0004] To address the aforementioned shortcomings of existing technologies, this invention provides a multi-level heuristic problem-solving guidance method and system based on error cause tracing. By combining subject-specific AI models with knowledge graphs, it achieves automatic and accurate identification of error causes. Based on error cause type matching targeted guidance strategies, multi-level step-by-step heuristic prompts, and two-way interactive guidance, it achieves precise teaching that only guides problem-solving approaches without outputting the final answer.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A multi-level heuristic problem-solving guidance method based on error cause tracing includes: acquiring question information and user-submitted incorrect answers; identifying error causes in the question information and incorrect answers using a subject-specific AI model pre-trained on subject corpora combined with a subject-specific knowledge graph, and outputting the error cause type; matching corresponding targeted guidance strategies according to the identified error cause type, and executing multi-level step-by-step prompts in a preset order, wherein the multi-level step-by-step prompts do not output the final answer or complete explanation; providing targeted answers in response to user follow-up questions on the guidance content, forming a two-way interactive guidance closed loop; and pushing variant questions related to the current error cause in response to the user's self-correction and submission of the correct answer.

[0006] As a further aspect of the present invention, the error types include at least two of the following: conceptual confusion, calculation error, misreading the question, missing ideas, and omission of steps.

[0007] As a further aspect of the present invention: the multi-level step-by-step prompts include at least two of the following prompt forms: questioning the cause of the error, associating knowledge points, decomposing logic, reviewing conditions, and decomposing steps; the targeted guidance strategy matches different combinations of prompt forms and prompt order according to different error types.

[0008] As a further solution of the present invention: in response to the error type being conceptual confusion, a combination of knowledge point association and error cause question is matched; in response to the error type being misreading the question, a combination of condition review and error cause question is matched; in response to the error type being missing ideas, a combination of logical decomposition and knowledge point association is matched; in response to the error type being calculation error, a combination of step decomposition and error cause question is matched; in response to the error type being missing steps, a combination of logical decomposition and condition review is matched.

[0009] As a further aspect of the present invention, the guidance intensity is adaptively adjusted based on at least one of the following factors: user grade level, subject type, and degree of weakness in knowledge points; the adjustment of guidance intensity includes adjusting the level of detail of step-by-step prompts and / or the number of opportunities for follow-up questions.

[0010] As a further solution of the present invention: in response to the confidence level of error identification being lower than the first preset threshold, the current wrong question is marked as pending manual confirmation, and a review request is sent to the teacher's or parent's end; in response to the confidence level of error identification being lower than the second preset threshold, a general problem-solving framework is directly output.

[0011] As a further aspect of the present invention: record the user's interaction history with AI; and provide a summary of the historical dialogue in response to repeated follow-up questions on the same knowledge point.

[0012] As a further aspect of the present invention: in response to the user still not answering correctly after multiple rounds of guidance, a problem-solving framework is output, wherein the problem-solving framework includes core problem-solving steps or methods, but does not include specific calculation content and the final answer.

[0013] As a further aspect of the present invention: collect error identification data, guidance response data, and variant question answering data; and iteratively optimize the subject-specific AI model and targeted guidance strategy based on the collected data.

[0014] The present invention also provides a multi-level heuristic problem-solving guidance system based on error cause tracing, including a memory and a processor; the memory is used to store executable instructions and a subject knowledge graph; the processor is used to execute the executable instructions to implement the above method.

[0015] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above method.

[0016] The present invention also provides a method for error guidance adapted to low-computing-power devices, which is a lightweight adaptation based on the above method, including simplifying the error identification types, simplifying the prompt format of multi-level step-by-step prompts, and disabling the two-way interactive follow-up question function.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention uses a subject-specific AI model combined with a subject-specific knowledge graph to identify the causes of users' incorrect answers and matches different targeted guidance strategies according to the type of error, thereby improving the pertinence and accuracy of answer guidance.

[0018] This invention does not output the final answer or complete analysis during the multi-level step-by-step prompting process. Instead, it inspires users to think independently by asking questions about the reasons for errors, associating knowledge points, breaking down logic, reviewing conditions, and breaking down steps. It also forms a two-way interactive guidance loop through voice / text follow-up questions.

[0019] The guidance intensity of this invention can be adaptively adjusted according to the user's grade, subject type, and knowledge weakness. After the user corrects the error, it pushes the same error variation questions. When the confidence level is low or multiple rounds of guidance are unsuccessful, it provides a general or fallback framework. At the same time, it takes into account the iterative optimization of the model and strategy and the adaptation to low computing power devices, and has good personalization, stability and applicability. Attached Figure Description

[0020] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is an overall flowchart of the error tracing guidance method of the present invention.

[0022] Figure 2 This is a schematic diagram of the error cause identification model architecture of the present invention.

[0023] Figure 3 This is a schematic diagram illustrating the targeted guidance strategy matching of the present invention.

[0024] Figure 4 This is a schematic diagram of the bidirectional interactive guidance closed loop of the present invention. Detailed Implementation

[0025] In the description of this invention, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0026] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "configuration" should be interpreted broadly. For example, they can refer to a fixed connection or configuration, a detachable connection or configuration, or an integral connection or configuration. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] After a user submits an incorrect answer, the system first identifies the cause of the error, then implements multi-level step-by-step prompts based on the type of error and a targeted guidance strategy; it provides targeted answers when the user asks follow-up questions; it pushes a variation of the same error after the user corrects the error and submits the correct answer; and it outputs a general or fallback framework when there is low confidence or multiple rounds of guidance fail.

[0029] Example 1: Construction of a Model for Accurate Error Cause Identification After a user submits an incorrect answer, the system analyzes the incorrect answer using a subject-specific AI model combined with a knowledge graph covering all subjects, enabling automatic and accurate identification of the cause of the error.

[0030] In this embodiment, the error identification model is built based on the subject pre-trained model of the BERT architecture. The training data is taken from real student answer data, which includes question text, incorrect answers and manually annotated error reasons. The data is manually annotated by front-line teachers and desensitized, covering five major subjects: mathematics, physics, chemistry, Chinese, and English.

[0031] The model input consists of the question text and the user's incorrect answers, and the output consists of the probability distribution of error cause types and confidence levels. The error cause types are divided into five categories: conceptual confusion, calculation error, misreading the question, missing solution approach, and missing steps. Specifically, conceptual confusion corresponds to misunderstanding of basic subject concepts or theorems; calculation error corresponds to errors in the calculation process; misreading the question corresponds to omitting key conditions of the question or a misunderstanding of the question's meaning; missing solution approach corresponds to failing to find the core solution strategy or method; and missing steps correspond to incomplete solution steps.

[0032] Example 2: Confidence Threshold Mechanism During the error identification process, if the confidence level is below 85%, the system automatically marks it as "awaiting manual confirmation" and sends a review request to the teacher or parent.

[0033] If the confidence level is below 60%, the system will directly enter the fallback guidance process and output a general problem-solving framework to avoid incorrect guidance.

[0034] Example 3: Targeted Guidance Strategy Matching Based on the identified error type, the system automatically matches a dedicated targeted guidance strategy and executes multi-level step-by-step prompts in a preset fixed order, without outputting the final answer or complete analysis throughout the process.

[0035] In this embodiment, the multi-level step-by-step prompts include error reasoning questions, knowledge point associations, logical breakdowns, condition reviews, and step breakdowns. The specific matching rules are as follows: concept confusion matches "knowledge point association + error reasoning questions", misreading the question matches "condition reviews + error reasoning questions", missing ideas match "logical breakdowns + knowledge point associations", calculation errors match "step breakdowns + error reasoning questions", and missing steps match "logical breakdowns + condition reviews".

[0036] Example 4: Two-way interactive guided closed loop and historical memory Users can ask follow-up questions via voice or text to the system's output guidance content. The system will then provide targeted answers to the user's questions, forming a two-way interactive guidance loop of "system guidance - user follow-up questions - accurate response" until the user independently sorts out the solution.

[0037] For example, if a user asks, "I still don't quite understand this concept," the system will respond, "Okay, let's explain this concept in more detail. [Detailed explanation of the concept] Now think about how to apply this concept to this question?"

[0038] The system records the user's interaction history with the AI. When the same knowledge point is repeatedly asked, it automatically identifies and provides a summary of the historical dialogue to avoid repeated explanations. At the same time, it dynamically adjusts the guidance intensity based on historical interaction data, increasing the guidance intensity for knowledge points that are repeatedly misunderstood and reducing intervention for knowledge points that have been mastered.

[0039] Example 5: Adaptive Adjustment of Guiding Intensity The system has built-in rules for matching grade level, subject, and weakness level, and achieves adaptive adjustment through guidance strength logic.

[0040] • Grade level: Primary school focuses on concrete step-by-step breakdown and condition review, providing more detailed guidance, more detailed steps, and more examples; middle school, high school, and university focus on guiding abstract thinking and connecting knowledge points, reducing examples and providing only the core direction.

[0041] • Subject-specific dimensions: Science subjects such as mathematics, physics, and chemistry emphasize logical decomposition and formula derivation; humanities subjects such as Chinese, English, and history emphasize the connection between knowledge points and contextual understanding.

[0042] • Knowledge gaps: Based on users' historical wrong answers, the error rate of each knowledge point is calculated. Knowledge points with an error rate higher than 60% are considered weak knowledge points, and the guidance coefficient of the corresponding questions is increased to provide more detailed step-by-step prompts and more opportunities for follow-up questions.

[0043] Example 6: Enhanced Training and Backup Guidance for Variation Problems If a user corrects their mistake and submits the correct answer after being guided through multiple levels of system prompts, the system will immediately push variant questions with the same cause of error, the same knowledge point, but different question stems, to enhance the user's ability to solve problems with this type of error.

[0044] If a user still fails to answer correctly after multiple rounds of guidance, the system will provide a framework for solving the problem, outlining only the core steps or methods, without providing specific calculations, written content, or the final answer.

[0045] Example 7: Iterative Optimization of Model and Policy The system continuously collects user error identification data, guidance response data, and variant question answering data. Based on big data analysis, it iteratively optimizes the error identification model and targeted guidance strategy to improve the accuracy of error identification and guidance effect, and achieve personalized and precise problem-solving guidance.

[0046] Example 8: Lightweight Adaptation for Low-Computing-Power Devices This embodiment provides a lightweight adaptation version for low-computing-power educational hardware, which is compatible with simple educational devices without independent AI chips and with limited storage, reducing the computational load and data storage requirements of the algorithm while retaining the core heuristic guidance effect.

[0047] • Only identify the two core error causes: misreading the question and calculation errors, and remove the identification of error causes such as conceptual confusion, missing ideas, and missing steps.

[0048] • Step-by-step prompts retain only two basic forms: error reasoning and condition review, while complex prompts such as logical breakdown and knowledge point association are removed.

[0049] • Disable the user's two-way interactive follow-up question function, and the system will complete the guidance according to a fixed process.

[0050] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0051] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A multi-level heuristic problem-solving guidance method based on error root cause tracing, characterized in that, include: (1) Obtain question information and incorrect answers submitted by users; (2) By combining a subject AI model pre-trained based on subject corpus with a subject knowledge graph, the error causes of the question information and the incorrect answers are identified, and the error cause type is output. (3) Based on the identified error type, match the corresponding targeted guidance strategy and execute multi-level step-by-step prompts in a preset order. The multi-level step-by-step prompts do not output the final answer or complete analysis. (4) Respond to users' follow-up questions about the guidance content, provide targeted answers, and form a two-way interactive guidance loop; (5) In response to users correcting their mistakes and submitting the correct answers, push variant questions related to the current error.

2. The multi-level heuristic problem-solving guidance method based on error cause tracing as described in claim 1, characterized in that, The error types include at least two of the following: conceptual confusion, calculation error, misreading the question, missing ideas, and omission of steps.

3. The multi-level heuristic problem-solving guidance method based on error tracing as described in claim 1, characterized in that, The multi-level step-by-step prompts include at least two of the following prompt formats: asking questions about the cause of the error, associating knowledge points, breaking down logic, reviewing conditions, and breaking down steps; the targeted guidance strategy matches different combinations of prompt formats and prompt order according to different error types.

4. The multi-level heuristic problem-solving guidance method based on error tracing according to claim 3, characterized in that, The matching includes: In response to the error type being conceptual confusion, a combination of knowledge point association and rhetorical questions about the error is matched; In response to the error type being misreading the question, a combination of reviewing the matching conditions and asking questions about the error cause is used. In response to the error type being a lack of thought process, a combination of logical decomposition and knowledge point association is used; In response to the error type being a calculation error, a combination of step breakdown and error cause questioning is matched; In response to the error type of step omission, a combination of matching logic decomposition and condition review is used.

5. The multi-level heuristic problem-solving guidance method based on error root cause tracing as described in claim 1, characterized in that, Also includes: The guidance intensity is adaptively adjusted based on at least one of the following factors: user's grade level, subject type, and level of weakness in knowledge points; the guidance intensity adjustment includes adjusting the level of detail of step-by-step prompts and / or the number of opportunities to ask follow-up questions.

6. The multi-level heuristic problem-solving guidance method based on error root cause tracing as described in claim 1, characterized in that, Also includes: If the confidence level of error identification is lower than the first preset threshold, the current incorrect question is marked as requiring manual confirmation, and a review request is sent to the teacher's or parent's end. If the confidence level of error cause identification is lower than the second preset threshold, a general problem-solving framework is directly output.

7. The multi-level heuristic problem-solving guidance method based on error tracing according to claim 1, characterized in that, Also includes: Record the history of user interactions with AI; In response to repeated follow-up questions on the same knowledge point, a summary of historical dialogues is provided.

8. The multi-level heuristic problem-solving guidance method based on error cause tracing as described in claim 1, characterized in that, Also includes: If the user still fails to answer correctly after multiple rounds of guidance, a problem-solving framework is output. The problem-solving framework includes core problem-solving steps or methods, but does not include specific calculations or the final answer.

9. A multi-level heuristic problem-solving guidance method based on error root cause tracing as described in claim 1, characterized in that, Also includes: Collect error identification data, guidance response data, and variation question answering data; The subject-specific AI model and targeted guidance strategy are iteratively optimized based on the collected data.

10. A multi-level heuristic problem-solving guidance system based on error root cause tracing, characterized in that, include: Memory, used to store executable instructions and subject knowledge graphs; A processor for executing executable instructions stored in the memory to implement the method according to any one of claims 1 to 9.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 9.

12. A fault-based guidance method adapted for low-computing-power devices, characterized in that, Lightweight adaptation based on the method of claim 1 includes: Simplify error cause identification types; Simplify the prompt format for multi-level step-by-step suggestions; Disable the two-way interactive follow-up question function.

Images