Wrong question analysis and personalized recommendation method and device based on large model and server

Abstract

The invention discloses a wrong question analysis and personalized recommendation method and device based on a large model and a server, and relates to the technical field of learning assistance and artificial intelligence combination, and the method comprises the steps: obtaining input question-answering information, inputting the processed question-answering information into a preset artificial intelligence model, analyzing the question semantics and the question answering process of the question-answering information, and obtaining a question-answering result; error reasons are positioned, and weak knowledge points are analyzed; based on the error reason of the question answering information, extracting error question core features; on the basis of the extracted wrong question core features, questions which reach a preset association degree with wrong question reasons are retrieved from a preset question bank, and / or three questions related to the wrong question reasons are dynamically generated, the questions which reach the preset association degree with the wrong question reasons are output, and the questions are preferentially matched with the wrong question reasons. The invention provides a method and an intelligent system which are combined with a large model technology, can deeply analyze error causes, supports multi-modal data and can dynamically generate questions, and can realize precise and personalized learning assistance.

Description

Technical Field

[0001] This invention relates to the field of learning assistance and artificial intelligence integration technology, and in particular to a method, device, server and storage medium for error analysis and personalized recommendation based on a large model. Background Technology

[0002] With the development of technology and the continuous improvement of people's living standards, the use of various smart terminals such as smartphones is becoming more and more widespread. In learning scenarios, it is sometimes necessary to use smartphones to take pictures and identify whether the answers are correct.

[0003] In existing learning scenarios, incorrect answers are the core carriers reflecting students' knowledge weaknesses; however, existing technologies and traditional methods for handling incorrect answers have significant limitations. Traditional methods lack deep semantic analysis, multimodal processing, and dynamic generation capabilities, making it difficult to achieve precise and personalized learning assistance.

[0004] Therefore, existing technologies still need improvement and development. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method, apparatus, server, and storage medium for error analysis and personalized recommendation based on a large model. This invention utilizes a large model to achieve error recording, in-depth analysis of error causes, accurate recommendation and dynamic generation of questions that can be applied to similar problems. It provides a method and intelligent system that combines large model technology, can deeply analyze error causes, support multimodal data, and dynamically generate questions. This invention can achieve precise and personalized learning assistance.

[0006] The technical solution of this application is as follows: A method for error analysis and personalized recommendation based on a large model, comprising: The system acquires and preprocesses the input question-answering information, which includes the question content and answer process information. The processed question-answering information is input into a preset artificial intelligence model. The preset artificial intelligence model analyzes the semantics of the questions and the answering process, locates the causes of errors, analyzes weak knowledge points, and returns the analysis results. Based on the error reasons in the question-answering information, extract the core features of the wrong questions; based on the extracted core features of the wrong questions, retrieve questions from the preset question bank that have a predetermined correlation with the error reasons, and / or dynamically generate similar questions that are related to the error reasons, and output questions that have a predetermined correlation with the error reasons, prioritizing matching the error reasons; The system collects and organizes the reasons for errors and weak knowledge points in the user's test-taking information, automatically generates a review and study plan, and regularly reminds the user to practice test questions based on the generated plan.

[0007] The aforementioned method for error analysis and personalized recommendation based on a large model includes the following steps: acquiring and inputting question-answering information, and preprocessing the question-answering information. The system controls the capture of images containing information about answering questions, including the question content and answering process information, via a terminal camera. Preprocess the images of the question-answering information, automatically cropping irrelevant areas and correcting the tilt angle; Upload the pre-processed images of the test-taking information to the server.

[0008] The method for error analysis and personalized recommendation based on a large model, wherein the steps of inputting the processed question-answering information into a preset artificial intelligence model, and using the preset artificial intelligence model to analyze the question semantics and answer process of the question-answering information, and to locate the cause of errors and analyze weak knowledge points, include: The processed question-answering information is input into a preset artificial intelligence model. The preset artificial intelligence model analyzes the core information of the question-answering information, including identifying knowledge points, down to sub-concepts, identifying question types and applicable grade levels. The system analyzes the answer steps using the pre-set artificial intelligence model, compares the correct logic step by step, and points out the incorrect steps. The preset artificial intelligence model locates the cause of the error, selects the most matching error type from the following: completely unmastered knowledge blind spots, correct knowledge points but incorrect application in the scenario, careless mistakes, and logical gaps caused by skipping steps, and generates an explanation. The preset artificial intelligence model outputs weak knowledge points, and based on the reasons for the errors, it identifies the specific areas that students need to strengthen.

[0009] The aforementioned method for error analysis and personalized recommendation based on a large model includes the following steps: extracting core features of incorrect questions based on the reasons for errors in the question-answering information; and retrieving questions from a pre-set question bank based on the extracted core features of incorrect questions that have a predetermined correlation with the reasons for errors: Based on the reasons for errors in the question-answering information, the core features of the incorrect questions are extracted. These core features include: weak knowledge points, reasons for errors, question types, difficulty levels, and grade level. Based on the extracted core features of incorrect questions, candidate questions related to the reasons for the incorrect questions are retrieved from the pre-set question bank; The system calls upon a pre-defined artificial intelligence model to sort candidate questions that are related to the reasons for incorrect answers from highest to lowest relevance. Select questions from the pre-sorted data and use them as questions that have a predetermined correlation with the reasons for incorrect answers.

[0010] The aforementioned method for error analysis and personalized recommendation based on a large model includes the step of calling a preset artificial intelligence large model to sort candidate questions retrieved that are associated with the reasons for errors from high to low relevance, which includes: Candidate topics are ranked based on the following criteria, with priority given to topics that meet the criteria; Error reason matching rate ≥ 80%; The knowledge points covered in the questions were 100% consistent with the knowledge points the students were weak in. The difference between the difficulty level and the incorrect answers is ≤ 1 level. Output the sorting results and the reason for matching each question.

[0011] The aforementioned method for error analysis and personalized recommendation based on a large model includes the following steps: Dynamically generating application-based questions related to the causes of errors; Based on the following parameters, dynamically generate similar problems that relate to the reasons for incorrect answers: Based on the weak knowledge points of users who made mistakes, set questions around the reasons for their errors; Reconstruct the scenario for the original incorrect question and change the question title; The difficulty level of the questions should be consistent with the difficulty level of the incorrect questions, and the length of the question stem should not exceed the specified number of characters. Validation is performed to ensure that the question stem is logically consistent, the answer is unique, and the target error is triggered when simulating error-prone paths; Output the question text, the correct answer, and the explanation.

[0012] The method for error analysis and personalized recommendation based on a large model, wherein the steps of collecting and organizing the reasons for errors and weak knowledge points in the user's question-answering information, automatically generating a review and learning plan, and periodically reminding the user to practice questions according to the generated review and learning plan include: Collect and organize the reasons for errors and weak knowledge points in the corresponding users' test-taking information; Based on the collected and organized error reasons and weak knowledge points of the corresponding users' question-solving information, the system automatically searches for expert explanations and practice questions corresponding to the error reasons and weak knowledge points of the user's question-solving information, and automatically generates a review and study plan according to the corresponding user's memory forgetting curve. The generated review and study plan will periodically remind the corresponding users to practice questions.

[0013] A device for error analysis and personalized recommendation based on a large model, characterized in that the device comprises: The question entry module is used to acquire and input question-answering information and preprocess the question-answering information, which includes question content and answer process information. The question feature extraction and error cause analysis module is used to input the processed question-answering information into a preset artificial intelligence model, analyze the question semantics and answering process of the question-answering information through the preset artificial intelligence model, locate the error cause, analyze weak knowledge points, and return the analysis results; The "learn one thing and apply it to other things" question processing module is used to extract the core features of wrong questions based on the error reasons in the question-solving information; based on the extracted core features of wrong questions, it retrieves questions from a preset question bank that reach a predetermined correlation with the error reasons, and / or dynamically generates "learn one thing and apply it to other things" questions that are related to the error reasons, and outputs questions that reach a predetermined correlation with the error reasons, prioritizing matching the error reasons; The review and study reminder module is used to collect and organize the reasons for errors and weak knowledge points of the corresponding user's question-solving information, automatically generate a review and study plan, and regularly remind the corresponding user to do question-solving exercises based on the generated review and study plan.

[0014] A server includes a memory and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by one or more processors, the one or more programs comprising the method for performing any one of the methods.

[0015] A computer-readable storage medium, wherein, when instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform any of the methods described herein.

[0016] As can be seen from the above, the error analysis and personalized recommendation method, device, server, and storage medium provided in this application, based on a large model, leverages the powerful multimodal understanding, deep semantic analysis, and dynamic generation capabilities of the large model to achieve accurate identification of error causes, highly relevant recommendations of similar questions, and dynamic generation, thereby improving learning efficiency and effectiveness and meeting personalized learning needs. Furthermore, by introducing large model technology, this invention has the following significant advantages compared to existing traditional error handling methods: 1) More accurate error analysis: This invention relies on the multimodal understanding capabilities of a large model to analyze images, handwritten formulas, and other content, accurately distinguishing error types and overcoming the limitations of manual analysis; 2) Stronger relevance of recommended questions: Based on error cause matching, this invention significantly improves the relevance between recommended questions and error causes, avoiding inefficient practice (e.g., when errors are caused by "unit confusion", questions containing "unit traps" are recommended first). 3) Multimodal scenario full coverage: This invention supports error processing including pictures and charts, covering multiple disciplines such as physics and geography; 4) Dynamic generation solves the limitations of the question bank: This invention can dynamically generate new questions that match the reasons for errors in scenarios such as extracurricular questions and self-compiled questions, thus meeting personalized needs; 5) More precise identification of weaknesses: This invention is accurate down to the sub-knowledge points (such as "rules for the symbol of the quadratic formula"), providing a clear direction for targeted learning. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the error analysis and personalized recommendation method based on a large model according to Embodiment 1 of the present invention.

[0019] Figure 2 This is a flowchart illustrating the error analysis and personalized recommendation method based on a large model, according to a specific embodiment 2 of the present invention.

[0020] Figure 3 The principle block diagram of the embodiment of the error analysis and personalized recommendation device based on a large model provided by the present invention.

[0021] Figure 4 This is a block diagram illustrating the internal structure of the server provided in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0023] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0024] In existing learning scenarios, incorrect answers are the core carriers reflecting students' knowledge weaknesses. However, existing technologies and traditional methods for handling incorrect answers have significant limitations, including the following problems: 1) Inefficient and inaccurate error analysis: Traditional methods rely on students' self-summarization or teachers' manual judgment, which is limited by students' cognitive level and teachers' energy, and cannot deeply distinguish the essential differences such as "knowledge blind spots", "logical reasoning errors", "omission of steps", and "careless mistakes". For example, math problems may be due to "incorrect formula memorization" or "confusion about the application of formulas", but existing technology and traditional methods cannot accurately identify such differences, resulting in a lack of targeted improvement suggestions.

[0025] 2) The recommendation of questions that require generalization and application is mechanical and lacks relevance: Current recommendation logic is mostly based on "knowledge point tag matching" or "question type text similarity", without considering the specific reasons for incorrect answers. For example, when a student makes a mistake in a geometry problem by "ignoring the conditions of the three sides of a triangle", current technology and traditional methods only recommend similar geometry problems, but cannot specifically recommend "triangle problems with implicit conditions", resulting in insufficient targeted practice and a high proportion of inefficient practice.

[0026] 3) Lack of multimodal error processing capabilities: For errors containing images (such as diagrams of physics experimental devices, geographical topographic maps), handwritten formulas, and complex charts, existing technologies and traditional methods can only store them as images and cannot parse the semantics of the content. This makes error analysis and question recommendation completely ineffective and difficult to cover multidisciplinary scenarios such as science and geography.

[0027] 4) Limited question bank coverage, unable to meet personalized needs: When students enter extracurricular exercises, self-compiled questions or interdisciplinary comprehensive questions, if there are no similar questions in the preset question bank, the existing technology and traditional methods cannot generate new questions, resulting in the interruption of the "learn one thing and apply it to other situations" function, which cannot meet personalized learning needs.

[0028] It is evident that existing traditional error-handling methods lack deep semantic analysis, multimodal processing, and dynamic generation capabilities, making it difficult to achieve precise and personalized learning assistance. Therefore, there is an urgent need for an intelligent system that combines large-scale modeling technology, can deeply analyze the causes of errors, support multimodal data, and dynamically generate questions.

[0029] To address the aforementioned technical problems, this invention provides a method for error analysis and personalized recommendation based on a large model, as detailed in the following embodiments.

[0030] Example 1 like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for error analysis and personalized recommendation based on a large model, which includes the following steps: Step S100: Obtain the input question-answering information and preprocess the question-answering information, which includes the question content and answer process information; In this step of the embodiment, information related to the user's answering of questions is first collected. This includes not only the basic content of the questions themselves, such as the text, images, and question types, but also key process information of the user's answering process, such as the calculation steps on the scratch paper, the answering time, modification marks, and whether the prompts were consulted. Furthermore, the collected information is preprocessed to eliminate invalid data interference. For example, blurry images of answering marks are removed, and the question format is standardized. For instance, users can take photos of their answering information, and irrelevant areas are automatically cropped during preprocessing. Handwritten formulas are converted into standard text formulas, and key logical nodes in the answering process are extracted, such as setting up equations in the first step and substituting values ​​in the second step. This lays a high-quality data foundation for subsequent AI analysis. The benefits of this step are that it can ensure data quality because invalid information is removed through preprocessing, avoiding interference during subsequent AI analysis; it can also improve analysis efficiency because key steps are extracted in advance, reducing the time for AI to interpret the raw information, making subsequent error location and knowledge point analysis faster; and it can ensure information integrity because both the question content and the answer process are collected at the same time, avoiding looking only at the result and ignoring the rationality of the process.

[0031] Step S200: Input the processed question-answering information into a preset artificial intelligence model, analyze the question semantics and answering process of the question-answering information through the preset artificial intelligence model, locate the cause of error, analyze weak knowledge points, and return the analysis results; The pre-set artificial intelligence model (large model) described in this invention needs to be trained in advance with a large amount of question-solving data, covering the semantic features of questions, error patterns in answering questions, and knowledge point association rules of different subjects and question types.

[0032] This step-by-step embodiment relies on a pre-set artificial intelligence model to complete three main tasks: First, it analyzes the semantics of the question to clarify the core objective being tested, such as calculating pricing, which essentially tests the calculation of profit margin; second, it analyzes the answering process, comparing the differences between the standard solution logic and the user's steps to pinpoint the cause of the error; and third, it associates weak knowledge points, matching the corresponding subject knowledge points based on the cause of the error or the loopholes in the answer, such as misremembering a formula corresponding to a weak knowledge point of the profit margin formula. Finally, it outputs an analysis result that includes the question's testing point, the cause of the error (if none, it is marked as correct), and the weak knowledge point. For example, consider word problems involving linear equations in one variable from junior high school mathematics: The question asks for information on how to answer a problem. The problem states that a store purchases 30 shirts at a cost of 120 yuan each. If sold at the set price, each shirt yields a 25% profit. The task is to determine the price of each shirt. The answer process involves the user first writing "Setting price = Cost price + Profit," then calculating "120 × 25% = 30," and finally arriving at "Setting price = 120 + 30 = 150 yuan." The answering time is 8 minutes, with no modifications allowed. The preprocessing steps are as follows: irrelevant and distracting information such as "30 shirts" in the question is marked (because the pricing calculation is not related to the quantity), the user's handwritten formula "price = cost price + profit" is converted into the standard mathematical expression "price = cost price × (1 + profit rate)", and the key steps to answer the question are "determine the formula, calculate the profit, and derive the price".

[0033] The present invention analyzes the semantics of the questions and the answering process of the question information through the preset artificial intelligence model, including: 1) Question semantic analysis step: The preset artificial intelligence model determines the core examination point of the question as pricing calculation based on profitability by comparing the characteristics of profitability-related questions in the training data, and the related knowledge points are the practical application of linear equations in one variable and profit problems. 2) Analysis of the answering process: The preset artificial intelligence model compares the user's steps "Pricing = Purchase Price + Profit", "120 × 25% = 30", and "Pricing = 150 yuan" with the standard steps "Pricing = Purchase Price × (1 + Profit Margin) = 120 × (1 + 25%) = 150 yuan". It finds that although the user's result is correct, the formula is not rigorous and does not directly reflect the "multiplicative relationship between profit margin and pricing". If the profit margin is a complex percentage, it may lead to calculation errors. 3) Error reason and weak knowledge point identification steps: Since the result is correct and there is no substantial error, the error reason is marked as "no obvious error, but the expression of the problem-solving formula is not standardized"; the weak knowledge point is marked as "no core weak point, need to strengthen the awareness of the details of 'standardized expression of profit problem formula'". The analysis results returned by this invention are as follows: The questions test the following areas: practical application of linear equations in one variable, profit problems, and profit rate calculation. The answer is evaluated as follows: The answer is correct and the logic of the answer process is clear, but the expression of the formula is not standardized. It is recommended to directly write the formula as "Price = Purchase Price × (1 + Profit Margin)". Weak knowledge points: There are no core weak knowledge points, and the ability to write formulas in a standardized way needs to be improved. If we consider a different incorrect example, such as a user calculating "120 × 25%" and getting "20", resulting in a final price of "140 yuan": This invention uses the preset artificial intelligence model to locate the cause of the error as "profit margin calculation error (25% corresponds to 0.25, 120×0.25=30≠20)", and marks the weak knowledge points as "conversion between percentage and decimal" and "the calculation rule of multiplying an integer by a percentage". As can be seen, compared with manual grading which only judges the correctness of the result, this invention can delve deeper into the process to find loopholes (such as non-standard formula expression or calculation errors), avoiding the hidden problem of correct results but wrong process, and can accurately diagnose problems; moreover, this invention is not limited to the question being wrong, but locates the underlying weak knowledge points, such as the "percentage conversion" corresponding to the calculation error, upgrading learning from correcting a question to filling in a knowledge point, and can connect the essence of knowledge points; and this invention can ensure the fairness and accuracy of the analysis results.

[0034] Step S300: Based on the error reasons in the question-answering information, extract the core features of the wrong questions; based on the extracted core features of the wrong questions, retrieve questions from the preset question bank that have a predetermined correlation with the error reasons, and / or dynamically generate questions that are related to the error reasons and provide generalizations, and output questions that have a predetermined correlation with the error reasons, prioritizing matching the error reasons; In this embodiment of the invention, practice resources are precisely matched based on the cause of the error. This is implemented in two ways: First, a pre-built question bank is constructed, where questions are stored in advance categorized by "error cause - knowledge point". Then, related questions are retrieved from the pre-built question bank, and questions that match the current incorrect question's error cause and weak knowledge point with a pre-defined standard (e.g., 80% or more). Second, if the pre-built question bank lacks sufficient related questions, dynamic generation technology is used to automatically generate suitable questions based on the problem-solving logic and knowledge point difficulty corresponding to the error cause, supplementing the practice questions. Finally, when outputting questions, priority is given to matching the error cause, avoiding a haphazard approach of endless practice questions. For example, if a user makes a mistake in converting percentages to decimals, such as calculating 120 × 25% as 20, the question bank will filter out questions with the error reason tag = percentage to decimal conversion error and the knowledge point = profit problem / percentage calculation from the preset question bank. For example, if a product has a purchase price of 200 yuan and a profit rate of 15%, to find the profit amount, the 15% needs to be converted to 0.15 before calculation, and the results of 300 × 35% and 180 × 40% are calculated. If the matching degree is higher than 85%, it will be included in the output list. For example, if there are only 2 questions of this type in the question bank, which is less than the preset 5 practice questions, then in this embodiment of the invention, the system will automatically generate 3 questions of the same type, such as a bookstore selling books at an 80% discount (80% discount = 0.8), how much does a book that originally cost 50 yuan now cost? Calculate the results of 250×28% and 160×55%, ensuring that all questions are designed around the error of converting percentages to decimals.

[0035] The output will prioritize 5 questions that closely match the percentage conversion error, rather than randomly recommending other profit-related questions (such as those that do not require percentage conversion). Case 2 illustrates that if a user's formula expression is not standardized but the user's expression is not correct, the present invention can retrieve and generate questions from the question bank that "examine the standardized expression of profit problem formulas". For example, it may require users to directly calculate the price of a product using the formula 'Pricing = Purchase Price × (1 + Profit Margin)'. If the formula is insufficient, similar questions can be generated to reinforce the habit of writing formulas in a standardized manner. As can be seen from the above, this invention prioritizes matching the reasons for errors, allowing users to practice specific areas they don't understand (e.g., if a percentage conversion error occurs, practice the conversion itself, rather than practicing the entire profit problem), reducing ineffective time spent on already mastered knowledge points and avoiding blind practice. Furthermore, this invention strengthens weak areas by combining retrieval and dynamic generation to ensure ample practice resources, and all questions are designed around the reasons for errors, achieving the effect of learning one wrong question and mastering a whole category of questions. Moreover, this invention can adapt to personalized needs, because even if different users get the same question wrong, if the reasons for the error are different (e.g., A's error is in the formula, B's error is in the calculation), the matched questions will also be different, truly achieving personalized practice for each user.

[0036] Step S400: Collect and organize the reasons for errors and weak knowledge points of the corresponding user's question-solving information, automatically generate a review and study plan, and regularly remind the corresponding user to do question-solving exercises according to the generated review and study plan.

[0037] This invention can also achieve long-term learning management. Specifically, it first accumulates information and continuously collects the reasons for users' mistakes and weak knowledge points each time they do a question. It can be classified and organized by time and knowledge point to form a personal error knowledge base, such as October 2024 - profit problem - percentage conversion error, October 2024 - equation application - formula non-standard.

[0038] Then, planning and reminders are implemented. Based on the organized information and combined with the user's learning goals, such as mastering profit issues within one month, and available time, such as 30 minutes of math practice per day, a phased review and learning plan is automatically generated. Specifically, it can clarify what to practice each day, how much to practice, and which knowledge points to review. Timed notifications, such as app push notifications and SMS reminders, are used to urge users to complete the practice exercises according to the plan. At the same time, the plan is dynamically adjusted according to the user's subsequent performance. For example, if a knowledge point has been mastered, the amount of practice is reduced, and if it is still weak, the frequency of review is increased. For example, if the user is a second-year junior high school student, the math learning goal is to master the practical application of linear equations in one variable, such as profit, travel, and work problems, within one month. There is 30 minutes of practice time every day from 19:00 to 19:30, and the user has already accumulated two weaknesses through steps S100-S300: "profit problem - percentage conversion error" and "profit problem - non-standard formula". During information processing, the system categorizes the two weak points into the "Application of Linear Equations in One Variable - Profit Problems" module, records the first occurrence time as "October 1st", and updates the number of errors for this knowledge point after each subsequent practice session. For example, if the percentage conversion is wrong again on October 2nd, the error count is updated to 2. Then, this invention will automatically generate the following review plan: The plan for the first week (October 3-9) is as follows: Practice 5 problems on "Percentage Conversion + Profit Formula" from 19:00 to 19:15 every day, and review the wrong problems and corresponding knowledge points from 19:15 to 19:30. The plan for the second week (October 10-16) is as follows: Introduce new content on travel problems. Practice profit problem weak points (3 questions) from 19:00-19:10 every day, learn basic travel problem questions (4 questions) from 19:10-19:25, and review from 19:25-19:30. The plan for the third and fourth weeks is to gradually add engineering problems and reduce the amount of profit problems practiced (if there are no errors in profit problems after October 12, only 2 problems will be practiced per day for consolidation). Furthermore, this invention will provide regular reminders, for example via smart terminals: a reminder will be pushed at 18:50 every day: "Math review will begin at 19:00. Today's tasks: 3 profit problems (percentage conversion) + 4 basic travel problems." If the user does not complete the tasks on time, a reminder will be sent again at 20:00: "Today's math practice has not been completed. There are 25 minutes left to complete the task." Thus, this invention has the following advantages: 1) It forms a learning loop, extending from problem-analysis-practice to planning-execution-review, avoiding the recurrence of problems caused by not reviewing incorrect answers; 2) It saves planning time because the plan is automatically generated, eliminating the need for users to manually organize incorrect answers and arrange learning content, which is especially suitable for students with limited time or who are not good at planning; 3) It can dynamically adapt to the progress because the plan is adjusted according to the user's mastery (such as reducing practice after mastering weak points), avoiding inefficiency caused by rigid plans; 4) It can strengthen execution: regular reminders can help users develop regular study habits, avoid procrastination, and ensure that the review plan is implemented.

[0039] In a further embodiment of the present invention, the method for error analysis and personalized recommendation based on a large model, wherein step S100 specifically includes: S101. Control the terminal camera to capture images including question information, wherein the question information includes question content and answer process information; In a specific implementation of this invention, a user can use the camera of a terminal, such as a smartphone, to capture images containing information about answering questions. This information includes the question content and the answering process (such as handwritten solution steps and the final answer). The information can include questions the student answered incorrectly; that is, the user can also directly photograph the questions they answered incorrectly.

[0040] S102. Preprocess the images of the question-answering information, automatically cropping irrelevant areas and correcting the tilt angle; In this embodiment, the images of the problem-solving information are preprocessed, automatically cropping irrelevant areas (such as the edges of the test paper) and correcting the tilt angle. For example, users can select the incorrect question areas in the images of the problem-solving information, and the irrelevant areas are automatically cropped and the tilt angle is corrected.

[0041] S103. Upload the preprocessed images of the question-answering information to the server.

[0042] In this invention, the cropped images of the test-taking information are uploaded to the server for subsequent analysis and processing.

[0043] In a further embodiment of the present invention, step S200 specifically includes: S201. Input the processed question-solving information into a preset artificial intelligence model, and use the preset artificial intelligence model to analyze the core information of the question-solving information, including identifying knowledge points, down to sub-concepts, identifying question types and applicable grade levels; In this embodiment of the invention, the processed question-answering information is input into a preset artificial intelligence model. The input content refers to the processed question-answering information. In this invention, it refers to the organized and standardized question-related data, including question text, option content, answer results, and other information.

[0044] The preset artificial intelligence model is a pre-trained artificial intelligence AI model of the present invention, which is specifically used to analyze question information and combines natural language processing (NLP), machine learning and other technologies.

[0045] The core information of the questions is analyzed using the preset artificial intelligence model. The specific analysis process includes: 1) Identify the knowledge points and determine the specific knowledge content tested in this question, such as "quadratic equations in one variable" in mathematics; 2) Identify down to the sub-concept and refine the knowledge points. For example, further locate "quadratic equation in one variable" to "the formula for finding the roots of a quadratic equation in one variable". 3) Identify the question type: Determine the type of question, such as whether it is a multiple-choice question, a fill-in-the-blank question, a calculation question, or an application question; 4) Identify the applicable grade level: Determine which grade or level of students this question is suitable for, such as the second year of junior high school or the first year of senior high school.

[0046] The purpose of this process in this embodiment is to automatically analyze the characteristics of questions using a pre-trained artificial intelligence (AI) model, providing structured data support for subsequent teaching applications (such as personalized practice recommendations, analysis of weak knowledge points, etc.).

[0047] S202. Analyze the answer steps using the preset artificial intelligence model, compare the correct logic step by step, and point out the incorrect steps; In this embodiment, the answering steps refer to the complete process record of the student's solution to the question, including detailed records of multiple stages such as formulating equations, calculations, reasoning, and argumentation, rather than just the final answer.

[0048] In this embodiment of the invention, the step-by-step comparison is performed by comparing each step of the student's answer with the correct logical path of the question through a preset artificial intelligence model. Logical consistency checks not only verify the correctness of calculations or results, but also the logical connections between steps, such as whether theorems are followed, whether formulas are applied appropriately, and whether reasoning is rigorous. Error localization refers to accurately identifying which steps contain errors and the types of errors, such as misuse of concepts, calculation errors, logical jumps, and omissions of conditions.

[0049] This step goes beyond the traditional right-or-wrong judgment model, delving into the middle stages of the answering process to help pinpoint specific errors. This provides a precise basis for subsequent targeted explanations and corrections, and is especially suitable for subjects such as mathematics and physics that emphasize the reasoning process.

[0050] S203. Locate the cause of the error using the preset artificial intelligence model, select the most matching error type from the following: completely unmastered knowledge blind spots, correct knowledge points but incorrect application in the scenario, careless mistakes, and logical gaps caused by skipping steps, and generate an explanation. In this embodiment, based on the erroneous steps identified in the previous step, we further analyze the root cause of the error, rather than simply focusing on the superficial step errors.

[0051] The present invention has the ability to classify and judge error patterns through the preset artificial intelligence model, and can match the most suitable situation from the preset error types and understand the feature differences of different error types.

[0052] The specific reasons for the errors analyzed include: 1) Analyze whether there are any knowledge gaps that students have not mastered. That is, analyze whether students are completely unaware of the core concepts, formulas or theorems required to solve the problem, which are blank areas in the knowledge system.

[0053] 2) Analyze whether the knowledge points are memorized correctly but the application in the scenario is incorrect. That is, analyze whether students can remember the knowledge points, but do not know how to apply them correctly in specific question scenarios, or use the wrong applicable scenario.

[0054] 3) Analyze whether the error was due to carelessness, that is, analyze whether the student has mastered the relevant knowledge and knows how to apply it, but made a mistake due to negligence (such as writing the wrong number in calculation, misreading the conditions of the question, etc.).

[0055] 4) The analysis steps are skipped, resulting in logical discontinuity errors. That is, the analysis shows that the student may have mastered the knowledge points, but the key reasoning steps are omitted in the answer process, resulting in incoherent logic or a lack of support for the conclusion.

[0056] Then output the results. The preset artificial intelligence model described in this invention not only needs to determine the error type, but also needs to generate a corresponding explanation, explaining why the error belongs to this type of reason. For example, it is determined that the knowledge point is correctly remembered but the application of the scenario is incorrect because the square of the hypotenuse is mistakenly used as the sum of the two legs (instead of the sum of squares) when applying the Pythagorean theorem.

[0057] This step achieves a deeper understanding of the nature of errors, from error discovery to error correction, laying the foundation for providing personalized error correction solutions, such as supplementing explanations for knowledge gaps and strengthening scenario-based practice for application errors.

[0058] S204. The preset artificial intelligence model outputs weak knowledge points, and based on the reasons for the errors, it clarifies the specific areas that students need to strengthen.

[0059] This step builds upon the previously identified errors and their causes, further deepening and transforming the analysis of the answers.

[0060] Specifically, the pre-defined artificial intelligence model described in this invention can accurately pinpoint the specific knowledge points in a student's knowledge system that have deficiencies based on the cause of the error, including concepts, formulas, theorems, and methods. For example, if the error is due to correct memorization of the knowledge point but incorrect application in a scenario, the corresponding weakness analysis would be the application of a quadratic equation in a profit problem.

[0061] In this invention, the specific aspects requiring reinforcement are identified by analyzing error types and weak knowledge points to pinpoint the learning content or abilities that student users need to focus on strengthening. For example: If the analysis reveals a complete lack of mastery of a knowledge point, then the basic concepts and principles of that knowledge point need to be strengthened; if the error is due to incorrect application, then the training of combining the knowledge point with specific scenarios needs to be strengthened; if the error is due to logical gaps, then the completeness of the problem-solving steps and the ability to express logic need to be strengthened. In this embodiment of the invention, the preset artificial intelligence model must have the ability to associate error analysis results with the knowledge system, and be able to extract universal learning weaknesses from specific errors, rather than giving conclusions only for a single question.

[0062] In this step of the embodiment, the problem lies in transforming incorrect answers into specific directions for learning improvement, providing a clear basis for subsequent personalized learning plans and targeted practice recommendations, making learning improvement more targeted.

[0063] In a further embodiment of the present invention, the method for error analysis and personalized recommendation based on a large model, wherein step S300 specifically includes: S301. Based on the reasons for errors in the question-answering information, extract the core features of the incorrect questions. The core features of the incorrect questions include: weak knowledge points, reasons for errors, question types, difficulty, and grade level. In this step of the embodiment, based on the error reasons analyzed above, features that can represent the key attributes of this wrong question are extracted, including: the student's weak knowledge points, such as the solution of fractional equations, the specific reasons for the error, such as confusion of application scenarios, the question type, such as "word problem", the question difficulty, such as "medium", and the applicable grade, such as "second year of junior high school".

[0064] By extracting and summarizing these features, the key information of the incorrect question can be clearly outlined, providing a foundation for subsequent error classification and targeted practice recommendations.

[0065] S302. Based on the extracted core features of incorrect questions, retrieve candidate questions that are related to the reasons for the incorrect questions from the preset question bank; In this step of the embodiment, based on the core features of the previously extracted incorrect questions, especially the reasons for the errors, such as "incorrect application of the Pythagorean theorem," questions directly related to the reasons for the errors are retrieved from a pre-built question bank as candidates. For example, if an incorrect question is due to finding the wrong equality relationship when solving word problems with linear equations in one variable, this embodiment of the invention will filter out questions in the question bank that also test the analysis of equality relationships in word problems with linear equations in one variable, in preparation for subsequent targeted practice.

[0066] S303. Call the preset artificial intelligence big data model to sort the retrieved candidate questions related to the reasons for the wrong questions from high to low relevance. In this step of the embodiment, a pre-prepared large-scale artificial intelligence model is invoked to sort the previously retrieved candidate questions related to the reasons for incorrect answers from high to low according to the degree of their correlation with the core features of the incorrect questions, such as the reasons for the errors and the weak knowledge points.

[0067] This sorting method allows questions that most closely relate to the root causes of students' errors to be prioritized, facilitating subsequent recommendations and improving the relevance of practice.

[0068] S304. Select questions from the pre-sorted data and use them as questions that have a predetermined correlation with the reasons for incorrect questions.

[0069] In this embodiment, after sorting the candidate questions by relevance from high to low using a preset AI model, a predetermined number of questions (e.g., the first 5) are selected. These questions are considered to have reached the preset relevance standard to the reasons for the wrong questions and can be used as targeted practice questions recommended to students later.

[0070] For example, if the predetermined data is set to 5, the top 5 questions are selected from the sorted results to ensure that each recommended question can accurately match the root cause of the student's error.

[0071] The step of calling a preset artificial intelligence model to sort candidate questions related to the reasons for incorrect answers from high to low relevance includes: Candidate topics are ranked based on the following criteria, with priority given to topics that meet the criteria; 1) The matching degree of the error reason is ≥80%. For example, if the error reason of the wrong question is "application error (confusing the resistance calculation rules of series and parallel circuits)," then questions with the same error reason will be given priority. 2) The knowledge points covered in the questions are 100% consistent with the knowledge points the students are weak in; 3) The difference between the difficulty level and the incorrect question is ≤ 1 level; for example, if the difficulty level of the incorrect question is medium, the candidate questions can be easy / medium / difficult. Output the sorting results and the reason for matching each question.

[0072] Furthermore, the step of dynamically generating similar problems related to the reasons for incorrect answers includes: Based on the following parameters, dynamically generate similar problems that relate to the reasons for incorrect answers: 1) Based on the weak knowledge points of users who made mistakes, set up questions around the reasons for the mistakes; for example, retain the weak knowledge point "application of Ohm's law in series circuits", and set up questions around the reason for the mistakes "incorrect application (confusing the resistance calculation rules of series and parallel circuits)"; 2) Reconstruct the original incorrect question by changing the scenario and title; for example, change the original question "light bulb series circuit" to "resistor box series circuit"; 3) The difficulty level of the question should be consistent with the difficulty level of the incorrect question, and the length of the question stem should not exceed the specified number of words; for example, the difficulty level should be consistent with the incorrect question (medium), and the length of the question stem should not exceed 200 words.

[0073] 4) Conduct validity verification to ensure that the question stem is logically consistent, the answer is unique, and the target error cause is triggered when simulating error-prone paths; 5) Output content: Question text, correct answer and explanation. For example, output content: Question text (including necessary image description, such as "resistance box R1 and R2 are connected in series, power supply voltage 12V"), ​​correct answer and explanation.

[0074] In a further embodiment of the present invention, the method for error analysis and personalized recommendation based on a large model, wherein step S400 specifically includes: S401. Collect and organize the reasons for errors and weak knowledge points in the corresponding users' question-answering information; S402. Based on the collected and organized error reasons and weak knowledge points of the corresponding user's question-solving information, automatically search for expert explanation materials and practice questions corresponding to the error reasons and weak knowledge points of the user's question-solving information, and automatically generate a review and study plan according to the corresponding user's memory forgetting curve. In this step of the embodiment, based on the reasons for users' mistakes in answering questions (such as incorrect application of knowledge points) and weak knowledge points (such as the properties of quadratic function graphs) collected and organized beforehand, the system automatically searches for matching expert explanation materials (such as video lessons and graphic explanations of the corresponding knowledge points) and practice questions; then, combined with the user's memory forgetting curve (i.e., the law of knowledge forgetting over time, such as the Ebbinghaus forgetting curve), the system automatically formulates a review and learning plan that includes when to review, what content to review, and how to combine explanations and practice.

[0075] For example, if a user's weakness is the formula for solving quadratic equations, this invention will match expert-led videos explaining that knowledge point with specific practice questions, and arrange reviews at key points such as 1 day later and 3 days later according to the forgetting curve to help consolidate knowledge.

[0076] S403. Based on the generated review and study plan, regularly remind the corresponding users to do practice questions.

[0077] In this embodiment of the invention, the system will promptly remind the corresponding user to do practice questions based on the generated review and learning plan. For example, it will arrange review according to key nodes such as 1 day later, 3 days later, etc., based on the forgetting curve, to help consolidate knowledge, strengthen memory, provide convenience for student users to learn, and avoid forgetting.

[0078] The present invention will be further described in detail below through specific application examples: like Figure 2As shown in the second specific application embodiment, a method for error analysis and personalized recommendation based on a large model is provided, which includes the following steps: Step S11: Begin and proceed to step S12; Step S12: Student operates the terminal to take a picture and record the wrong question (including the question content and the answer process) and then proceeds to step S13. In this embodiment of the invention, students are supported in taking photos of their incorrect questions using a terminal. Specifically, students take photos of their incorrect questions using a terminal such as a mobile phone camera. The photos must include both the question content and the student's answer process (such as handwritten solution steps and the final answer). Step S13: The terminal preprocesses the captured images of incorrect questions (denoise removal / cropping / enhancement) and uploads them to the server, then proceeds to step S14; In this embodiment, the mobile phone preprocesses the captured images of incorrect questions, including automatically cropping irrelevant areas (such as the edges of the test paper) and correcting the tilt angle; and uploads the preprocessed image to the server; the server stores and analyzes the questions.

[0079] Step S14: Server calls large model analysis\n1. Extract question features (knowledge points / difficulty, etc.)\n2. Diagnose the causes of misidentification (conceptual confusion / calculation errors, etc.)\n3. Locate knowledge weaknesses; then proceed to step S15; In this embodiment, a pre-set large-scale artificial intelligence model is used to analyze the semantics of the questions and the answering process, accurately locating the causes of errors and weak knowledge points. Specifically, a multimodal large-scale model can be introduced as the core processing unit to construct a fully intelligent processing mechanism that integrates "input-analysis-recommendation / generation".

[0080] In this embodiment, after receiving the multimodal data, the server calls the large model to perform the analysis task. The input includes: Based on the provided questions (including text and images) and the students' answers, please complete the following analysis: 1) Analyze the core information of the question: Identify the knowledge points (down to the sub-concepts, such as "calculation of vertex coordinates of a quadratic function" rather than "quadratic function"), the question type (such as calculation problem / proof problem), and the applicable grade level; 2) Analyze the answer steps: Compare each step with the correct logic and point out the incorrect steps (e.g., "The formula was applied incorrectly in step 3"). 3) Identify the cause of the error: Select the most matching type from "Knowledge blind spot (completely not mastered)," "Application error (knowledge point is remembered correctly but misused in the scenario)," "Careless mistake (such as incorrect calculation symbols)," and "Logical gap (error caused by skipping steps)" and explain the reason; 4) Identify weak knowledge points: Based on the reasons for the errors, identify the specific areas that students need to strengthen (such as "the sign rules in the quadratic equation quadratic formula").

[0081] It includes images of the questions and students' answers, as well as instructions for large-scale model tasks.

[0082] In this embodiment, the analysis results returned by the large model include, for example: Key point: "Application of Ohm's Law in Series Circuits"; Grade: First year of junior high school; Question type: Multiple choice; Error reason: "Incorrect application (student confused the rules for calculating resistance in series and parallel circuits)"; Weak knowledge point: "Formula for calculating the total resistance of a series circuit".

[0083] Step S15: Check if the question bank matches the question. If yes, proceed to step S16; otherwise, proceed to step S17. Step S16: Intelligent ranking and recommendation of questions using a large model (based on the correlation of error causes / difficulty gradient), then proceed to step S18; Step S17: Generate similar problems based on the same knowledge points (with variations), then proceed to step S18. In this embodiment of the invention, through the logic of "retrieval + sorting + generation", questions that are highly related to the wrong questions and their causes are output, and the reasons for the errors are matched first.

[0084] In this embodiment of the invention, the question feature extraction adopts a large model based on the error analysis results to extract the core features of the wrong questions, including: weak knowledge points, reasons for errors, question type, difficulty, and grade.

[0085] Then, the question bank is searched and sorted. After retrieval and sorting, dynamic and analogy-based question generation is performed. This invention generates questions through a large model-driven approach. Specifically, it includes: a. The server retrieves candidate questions from a pre-set question bank based on core features (such as series circuit questions containing errors related to "confusion in resistance calculation rules for series and parallel circuits"). b. Use the large model to sort the candidate questions. The large model task instructions are as follows: Please rank the candidate questions based on the following criteria, and prioritize those that meet the criteria: 1) The matching degree of the error reason is ≥80% (for example, if the error reason of the wrong question is "application error (confusing the resistance calculation rules of series and parallel circuits)", then questions containing the same error reason will be given priority). 2) The knowledge points covered in the questions are 100% consistent with the knowledge points the students are weak in; 3) The difference between the difficulty level and the incorrect question is ≤ 1 level (e.g., if the difficulty level of the incorrect question is medium, the candidate questions can be easy / medium / difficult). 4) Output the sorting results and the matching reason for each question.

[0086] c. Select the top 3-5 questions in the ranking as recommended candidates.

[0087] d. When there are no questions in the question bank that meet the conditions (e.g., fewer than 3 candidate questions), the dynamic generation process is triggered.

[0088] The instructions for large model tasks include the following: Please generate one new question based on the following parameters: (1) Core requirements: retain the weak knowledge point "application of Ohm's law in series circuits", and set questions around the reason for the error "incorrect application (confusing the resistance calculation rules of series and parallel circuits)"; (2) Scenario reconstruction: Change the original problem "light bulb series circuit" to "resistor box series circuit"; (3) Difficulty control: The difficulty level should be consistent with the incorrect questions (medium), and the length of the question stem should not exceed 200 words; (4) Validity verification: Ensure that the question stem is logically consistent, the answer is unique, and the target error can be triggered when simulating the student's common error path; (5) Output content: Question text (including necessary image descriptions, such as "resistance boxes R1 and R2 are connected in series, power supply voltage is 12V"), ​​correct answer and explanation.

[0089] Specifically, after the invention generates the question, the large model self-checks the validity of the question (such as whether the error cause can be effectively triggered), and if it passes, it is used as the recommendation result.

[0090] In this embodiment of the invention, the output and display of results include: The server will package the "error cause analysis + recommended questions / generated questions" and send it back to the user's terminal; The terminal displays the following content: Explanation of the cause of the error (e.g., "Your error stems from confusing the rules for calculating resistance in series and parallel circuits"). Weaknesses highlighted (e.g., "It is recommended to strengthen the 'formula for calculating the total resistance of a series circuit'"). Generate problems that encourage applying the knowledge to other situations.

[0091] Step S18: Terminal display content: - Error analysis report - Recommended question list - Weakness point prompts, then proceed to step S19; Step S19; End.

[0092] As can be seen from the above, this invention, by introducing large-scale model technology, has the following significant advantages compared to traditional error handling methods: 1) More accurate error analysis: Relying on the multimodal understanding capabilities of large models, it can analyze images, handwritten formulas, and other content to accurately distinguish error types, breaking through the limitations of manual analysis; 2) More relevant question recommendations: Based on error cause matching, the recommended questions are significantly more relevant to the error cause, avoiding inefficient practice (e.g., when the error is caused by "unit confusion", questions containing "unit trap" are recommended first). 3) Multimodal scenario full coverage: Supports error processing including images and charts, covering multiple subjects such as physics and geography; 4) Dynamic generation addresses the limitations of the question bank: For scenarios such as extracurricular questions and self-compiled questions, new questions that match the reasons for errors can be dynamically generated to meet personalized needs; 5) More precise identification of weaknesses: down to specific knowledge points and details (such as "rules of symbols for quadratic formulas"), providing a clear direction for targeted learning.

[0093] Exemplary device like Figure 3 As shown, this embodiment of the invention provides a device for error analysis and personalized recommendation based on a large model. The device includes: The question entry module 310 is used to acquire and input question-answering information and preprocess the question-answering information, which includes question content and answer process information. The question feature extraction and error cause analysis module 320 is used to input the processed question-answering information into a preset artificial intelligence model, analyze the question semantics and answering process of the question-answering information through the preset artificial intelligence model, locate the error cause, analyze weak knowledge points, and return the analysis results; The "learn one thing and apply it to other things" question processing module 330 is used to extract the core features of the wrong questions based on the error reasons in the question-answering information; based on the extracted core features of the wrong questions, it retrieves questions from the preset question bank that have a predetermined correlation with the error reasons of the wrong questions, and / or dynamically generates "learn one thing and apply it to other things" questions that are related to the error reasons of the wrong questions, and outputs questions that have a predetermined correlation with the error reasons of the wrong questions, with priority given to matching the error reasons; The review and learning reminder module 340 is used to collect and organize the reasons for errors and weak knowledge points of the corresponding user's question-solving information, automatically generate a review and learning plan, and regularly remind the corresponding user to do question-solving exercises according to the generated review and learning plan, as described above.

[0094] Based on the above embodiments, the present invention also provides a server, the principle block diagram of which can be as follows: Figure 4 As shown. The server includes a processor, memory, network interface, display screen, and database connected via a system bus.

[0095] The memory stores one or more programs configured to be executed by a processor to implement the large-model-based error analysis and personalized recommendation method described in the above embodiments.

[0096] In this context, "server" refers to an intelligent computer or similar device capable of data processing. "Memory" can be internal memory, flash memory, hard disk, or cloud storage, used to store program code, question information, and various data such as questions with predetermined relevance. "Processor" can be a central processing unit (CPU), used to execute the algorithmic logic within the program. The program includes methods for analyzing incorrect questions based on a large model and providing personalized recommendations.

[0097] In a further embodiment, a server of this embodiment includes a memory and one or more programs, wherein one or more programs are stored in the memory and configured to be executed by one or more processors. The one or more programs include instructions for performing the following operations: The system acquires and preprocesses the input question-answering information, which includes the question content and answer process information. The processed question-answering information is input into a preset artificial intelligence model. The preset artificial intelligence model analyzes the semantics of the questions and the answering process, locates the causes of errors, analyzes weak knowledge points, and returns the analysis results. Based on the error reasons in the question-answering information, extract the core features of the wrong questions; based on the extracted core features of the wrong questions, retrieve questions from the preset question bank that have a predetermined correlation with the error reasons, and / or dynamically generate similar questions that are related to the error reasons, and output questions that have a predetermined correlation with the error reasons, prioritizing matching the error reasons; The system collects and organizes the reasons for errors and weak knowledge points in the user's test-taking information, automatically generates a review and study plan, and regularly reminds the user to practice test questions based on the generated review and study plan, as described above.

[0098] The steps of acquiring and preprocessing the question-answering information include: The system controls the capture of images containing information about answering questions, including the question content and answering process information, via a terminal camera. Preprocess the images of the question-answering information, automatically cropping irrelevant areas and correcting the tilt angle; Upload the pre-processed images of the test-taking information to the server.

[0099] The steps of inputting the processed question-answering information into a preset artificial intelligence model, and using the preset artificial intelligence model to analyze the question semantics and answer process of the question-answering information, and to locate the cause of errors and analyze weak knowledge points include: The processed question-answering information is input into a preset artificial intelligence model. The preset artificial intelligence model analyzes the core information of the question-answering information, including identifying knowledge points, down to sub-concepts, identifying question types and applicable grade levels. The system analyzes the answer steps using the pre-set artificial intelligence model, compares the correct logic step by step, and points out the incorrect steps. The preset artificial intelligence model locates the cause of the error, selects the most matching error type from the following: completely unmastered knowledge blind spots, correct knowledge points but incorrect application in the scenario, careless mistakes, and logical gaps caused by skipping steps, and generates an explanation. The preset artificial intelligence model outputs weak knowledge points, and based on the reasons for the errors, it identifies the specific areas that students need to strengthen.

[0100] The steps of extracting core features of incorrect questions based on the error reasons in the question-answering information, and retrieving questions from a preset question bank that have a predetermined correlation with the error reasons based on the extracted core features of incorrect questions, include: Based on the reasons for errors in the question-answering information, the core features of the incorrect questions are extracted. These core features include: weak knowledge points, reasons for errors, question types, difficulty levels, and grade level. Based on the extracted core features of incorrect questions, candidate questions related to the reasons for the incorrect questions are retrieved from the pre-set question bank; The system calls upon a pre-defined artificial intelligence model to sort candidate questions that are related to the reasons for incorrect answers from highest to lowest relevance. Select questions from the pre-sorted data and use them as questions that have a predetermined correlation with the reasons for incorrect answers.

[0101] The step of calling a preset artificial intelligence model to sort candidate questions related to the reasons for incorrect answers from high to low relevance includes: Candidate topics are ranked based on the following criteria, with priority given to topics that meet the criteria; Error reason matching rate ≥ 80%; The knowledge points covered in the questions were 100% consistent with the knowledge points the students were weak in. The difference between the difficulty level and the incorrect answers is ≤ 1 level. Output the sorting results and the reason for matching each question.

[0102] The steps involved in dynamically generating similar problems related to the reasons for incorrect answers include: Based on the following parameters, dynamically generate similar problems that relate to the reasons for incorrect answers: Based on the weak knowledge points of users who made mistakes, set questions around the reasons for their errors; Reconstruct the scenario for the original incorrect question and change the question title; The difficulty level of the questions should be consistent with the difficulty level of the incorrect questions, and the length of the question stem should not exceed the specified number of characters. Validation is performed to ensure that the question stem is logically consistent, the answer is unique, and the target error is triggered when simulating error-prone paths; Output the question text, the correct answer, and the explanation.

[0103] The steps of collecting and organizing the reasons for errors and weak knowledge points in the user's question-answering information, automatically generating a review and study plan, and regularly reminding the user to practice questions according to the generated review and study plan include: Collect and organize the reasons for errors and weak knowledge points in the corresponding users' test-taking information; Based on the collected and organized error reasons and weak knowledge points of the corresponding users' question-solving information, the system automatically searches for expert explanations and practice questions corresponding to the error reasons and weak knowledge points of the user's question-solving information, and automatically generates a review and study plan according to the corresponding user's memory forgetting curve. The generated review and study plan will periodically remind the corresponding users to practice questions, as detailed above.

[0104] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0105] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for error analysis and personalized recommendation based on a large model, characterized in that, include: The system acquires and preprocesses the input question-answering information, which includes the question content and answer process information. The processed question-answering information is input into a preset artificial intelligence model. The preset artificial intelligence model analyzes the semantics of the questions and the answering process of the question-answering information, and locates the cause of errors and analyzes weak knowledge points. And return the analysis results; Based on the error reasons in the question-answering information, extract the core features of the wrong questions; based on the extracted core features of the wrong questions, retrieve questions from the preset question bank that have a predetermined correlation with the error reasons, and / or dynamically generate similar questions that are related to the error reasons, and output questions that have a predetermined correlation with the error reasons, prioritizing matching the error reasons; The system collects and organizes the reasons for errors and weak knowledge points in the user's test-taking information, automatically generates a review and study plan, and regularly reminds the user to practice test questions based on the generated plan.

2. The method for error analysis and personalized recommendation based on a large model according to claim 1, characterized in that, The steps of acquiring and preprocessing the question-answering information include: The system controls the capture of images containing question-answering information via a terminal camera, the question-answering information including question content and answering process information; Preprocess the images containing the question-answering information, automatically cropping irrelevant areas and correcting tilt angles; Upload the pre-processed images of the test-taking information to the server.

3. The method for error analysis and personalized recommendation based on a large model according to claim 1, characterized in that, The steps of inputting the processed question-answering information into a preset artificial intelligence model, and using the preset artificial intelligence model to analyze the question semantics and answer process of the question-answering information, and to locate the cause of errors and analyze weak knowledge points include: The processed question-answering information is input into a preset artificial intelligence model. The preset artificial intelligence model analyzes the core information of the question-answering information, including identifying knowledge points, down to sub-concepts, identifying question types and applicable grade levels. The system analyzes the answer steps using the pre-set artificial intelligence model, compares the correct logic step by step, and points out the incorrect steps. The preset artificial intelligence model locates the cause of the error, selects the most matching error type from the following: completely unmastered knowledge blind spots, correct knowledge points but incorrect application in the scenario, careless mistakes, and logical gaps caused by skipping steps, and generates an explanation. The preset artificial intelligence model outputs weak knowledge points, and based on the reasons for the errors, it identifies the specific areas that students need to strengthen.

4. The method for error analysis and personalized recommendation based on a large model according to claim 1, characterized in that, Based on the error reasons in the question-answering information, the core features of the incorrect questions are extracted; The steps for retrieving questions from a pre-set question bank that are correlated with the reasons for errors based on the extracted core features of incorrect questions include: Based on the reasons for errors in the question-answering information, the core features of the incorrect questions are extracted. These core features include: weak knowledge points, reasons for errors, question types, difficulty levels, and grade level. Based on the extracted core features of incorrect questions, candidate questions related to the reasons for the incorrect questions are retrieved from the pre-set question bank; The system calls a pre-defined artificial intelligence model to sort the retrieved candidate questions that are related to the reasons for the errors in the questions from high to low relevance. Select questions from the pre-sorted data and use them as questions that have a predetermined correlation with the reasons for incorrect answers.

5. The method for error analysis and personalized recommendation based on a large model according to claim 4, characterized in that, The step of calling a preset artificial intelligence model to sort candidate questions related to the reasons for incorrect answers from high to low relevance includes: Candidate topics are ranked based on the following criteria, with priority given to topics that meet the criteria; Error reason matching rate ≥ 80%; The knowledge points covered in the questions were 100% consistent with the knowledge points the students were weak in. The difference between the difficulty level and the number of incorrect answers is ≤ 1 level. Output the sorting results and the reason for matching each question.

6. The method for error analysis and personalized recommendation based on a large model according to claim 1, characterized in that, The steps for dynamically generating similar problems related to the reasons for incorrect answers include: Based on the following parameters, dynamically generate similar problems that relate to the reasons for incorrect answers: Based on the weak knowledge points of users who made mistakes, set questions around the reasons for their errors; Reconstruct the scenario for the original incorrect question and change the question title; The difficulty level of the questions should be consistent with the difficulty level of the incorrect questions, and the length of the question stem should not exceed the specified number of characters. Validation is performed to ensure that the question stem is logically consistent, the answer is unique, and the target error is triggered when simulating error-prone paths; Output the question text, the correct answer, and the explanation.

7. The method for error analysis and personalized recommendation based on a large model according to claim 1, characterized in that, The steps of collecting and organizing the error reasons and weak knowledge points of the corresponding users' question-answering information, automatically generating review and study plans, and regularly reminding the corresponding users to do question-answering practice according to the generated review and study plans include: Collect and organize the reasons for errors and weak knowledge points in the corresponding users' test-taking information; Based on the collected and organized error reasons and weak knowledge points of the corresponding users' question-solving information, the system automatically searches for expert explanations and practice questions corresponding to the error reasons and weak knowledge points of the user's question-solving information, and automatically generates a review and study plan according to the corresponding user's memory forgetting curve. The generated review and study plan will periodically remind the corresponding users to practice questions.

8. A device for error analysis and personalized recommendation based on a large model, characterized in that, The device includes: The question entry module is used to acquire and input question-answering information and preprocess the question-answering information, which includes question content and answer process information. The question feature extraction and error cause analysis module is used to input the processed question-answering information into a preset artificial intelligence model, analyze the question semantics and answering process of the question-answering information through the preset artificial intelligence model, locate the error cause, analyze weak knowledge points, and return the analysis results; The "learn one thing and apply it to other things" question processing module is used to extract the core features of wrong questions based on the error reasons in the question-solving information; based on the extracted core features of wrong questions, it retrieves questions from a preset question bank that reach a predetermined correlation with the error reasons, and / or dynamically generates "learn one thing and apply it to other things" questions that are related to the error reasons, and outputs questions that reach a predetermined correlation with the error reasons, prioritizing matching the error reasons; The review and study reminder module is used to collect and organize the reasons for errors and weak knowledge points of the corresponding user's question-solving information, automatically generate a review and study plan, and regularly remind the corresponding user to do question-solving exercises based on the generated review and study plan.

9. A server, characterized in that, It includes a memory and one or more programs, wherein one or more programs are stored in the memory and configured to be executed by one or more processors, wherein the one or more programs include methods for performing any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to perform the method as described in any one of claims 1-7.

Images