A method and device for generating similar problems of wrong problems, an electronic device, and a storage medium

Abstract

Embodiments of the present application provide a method and device for generating similar questions of wrong questions, an electronic device and a storage medium. The method comprises: obtaining text information of a target wrong question, the text information comprising question information of the target wrong question; determining a target similar question related to the target wrong question according to the text information, the similar question being consistent with a structured feature of the target wrong question, the structured feature comprising at least one of a stem structure, a knowledge point, a difficulty level and a logical relationship. The above technical solution generates and determines a target similar question related to the target wrong question, generates a brand-new similar question based on the structural feature of the target wrong question, does not need to rely on a preset question bank, solves the problem of no questions to push or repeated recommendations caused by insufficient question banks in the traditional solution, further, the consistency of the structured feature enhances the expansibility and adaptability of the content, improves the accuracy and controllability of the similar question matching, and improves the review efficiency and learning effect of the wrong questions.

Description

Technical Field

[0001] This application relates to the field of intelligent learning technology, and in particular to a method, apparatus, electronic device, and storage medium for generating similar questions to incorrect ones. Background Technology

[0002] In existing technologies, error correction notebook products generally adopt a text similarity matching scheme based on a question bank: the system calculates the text similarity between the original question and questions in a pre-set question bank, and then selects questions with high similarity to push to the user. However, since similar questions are entirely derived from a pre-set question bank, if the question bank is not updated in a timely manner, there may be situations where no similar questions can be pushed, affecting the user experience. Therefore, how to accurately generate similar questions for incorrect answers is an urgent problem to be solved. Summary of the Invention

[0003] This application provides a method, apparatus, electronic device, and storage medium for generating similar questions to incorrect questions, which can achieve accurate generation of similar questions to incorrect questions and improve user experience.

[0004] The first aspect of this application provides a method for generating similar questions to incorrect ones, the method comprising: Obtain the text information of the target incorrect question, wherein the text information includes the question information of the target incorrect question; Based on the text information, target similar questions related to the target wrong question are identified. The similar questions have the same structured features as the target wrong question. The structured features include at least one of the following: question stem structure, knowledge points, difficulty level, and logical relationship.

[0005] In some embodiments, determining the target similar questions related to the target incorrect question based on the text information includes: The text information is parsed to extract the parsing results of the target incorrect question. The parsing results include the structured features of the target incorrect question and the replacement items of the target incorrect question. Based on the analysis results of the target incorrect question, the generation constraint rules for the similar question and the replacement items for the similar question are determined. The generation constraint rules are used to indicate that the structural features of the similar question are consistent with the structural features of the target incorrect question. Based on the generation constraints of the similar questions and the replacement terms of the similar questions, the target similar questions related to the target wrong question are determined.

[0006] In some embodiments, determining the target similar question related to the target incorrect question based on the generation constraint rules of the similar question and the replacement terms of the similar question includes: Determine the replacement content of the replacement items for the similar questions, wherein the replacement items include at least one of the question scenario, object name, and life situation; Under the constraints of the generation constraint rules, the target similar question is determined based on the replacement content of the replacement item.

[0007] In some embodiments, determining the replacement content for the replacement item of the similar question includes: According to the preset reasonable range of replacement content, the replacement content of the replacement item is checked for reasonableness; the reasonableness check includes at least one of numerical reasonableness check, digital combination logic reasonableness check, and scenario description reasonableness check. If the replacement content of the replacement item is within a reasonable range of the replacement content, the replacement content of the replacement item is determined to be the replacement content of the similar question.

[0008] In some embodiments, determining the target similar questions related to the target incorrect question based on the text information includes: Based on the text information, at least one similar question related to the target incorrect question is generated; The similarity consistency is verified between the structured features of the target incorrect question and the structured features of each similar question; Similar questions with a similarity score greater than the similarity threshold are identified as target similar questions to the target incorrect question.

[0009] In some embodiments, the text information may also include at least one of the user's textbook information, the user's educational progress, and the user's age.

[0010] In some embodiments, determining the target similar questions related to the target incorrect question based on the text information includes: The text information is input into the question generation model to determine the target similar questions related to the target wrong question. The question generation model is trained based on the sample wrong questions and the sample similar questions of the sample wrong questions.

[0011] A second aspect of this application provides an apparatus for generating similar questions to incorrect ones, the apparatus comprising: The acquisition module is used to acquire the text information of the target incorrect question, wherein the text information includes the question information of the target incorrect question; The determination module is used to determine, based on text information, target similar questions related to the target wrong question. The similar questions have the same structural features as the target wrong question. The structural features include at least one of the following: question stem structure, knowledge points, difficulty level, and logical relationship.

[0012] A third aspect of this application provides an electronic device, including a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described in the first aspect of this application.

[0013] A third aspect of this application provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the steps of any of the methods described in the first aspect of this application.

[0014] The technical solutions provided in this application have at least the following beneficial effects: This application proposes a method, apparatus, electronic device, and storage medium for generating similar questions to incorrect answers. This solution acquires the text information of a target incorrect answer and determines similar questions related to it based on that text information. It can generate entirely new similar questions based on the structural features of the target incorrect answer, without relying on a pre-set question bank. This solves the problem of insufficient question banks leading to a lack of questions or duplicate recommendations in traditional solutions. Furthermore, unlike simple keyword or vector similarity retrieval methods, the similar questions generated in this application have the same structural features as the target incorrect answer. These structural features include at least one of the following: question stem structure, knowledge points, difficulty level, and logical relationships. This effectively avoids recommending similar questions that are too easy or too difficult, enhancing the scalability and adaptability of the content, and improving the accuracy and controllability of similar question matching. This provides users with continuous, accurate, and personalized error correction training, improving the efficiency and effectiveness of error review. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a method for generating similar questions to incorrect questions according to an embodiment of this application; Figure 2 This is a flowchart illustrating the process of determining the target similar question based on text information in a method for generating similar questions based on incorrect questions proposed in an embodiment of this application. Figure 3 This is a flowchart illustrating the process of determining the target similar question based on the rule for generating the number of sentences and the replacement item in a method for generating similar questions based on incorrect questions proposed in an embodiment of this application. Figure 4 This is a schematic diagram illustrating the process of verifying the rationality of the replacement content of the similar question replacement item in a method for generating similar questions based on incorrect questions proposed in an embodiment of this application. Figure 5 This is a schematic diagram illustrating the process of verifying similar questions generated in a method for generating similar questions to incorrect questions proposed in an embodiment of this application. Figure 6 This is a schematic diagram of the layer relationship of the neural network proposed in an embodiment of this application; Figure 7This is a schematic diagram of one iteration process of the neural network proposed in the embodiments of this application; Figure 8 This is a flowchart illustrating another method for generating similar incorrect questions according to an embodiment of this application. Figure 9 A schematic diagram of the interface for displaying the target incorrect answers; Figure 10 This is a schematic diagram of the interface displaying similar questions related to the target incorrect questions; Figure 11 This is a schematic diagram of the structure of a device for generating similar questions to incorrect questions according to an embodiment of this application; Figure 12 This is a schematic diagram of the structure of the electronic device proposed in the embodiments of this application. Detailed Implementation

[0016] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0017] To facilitate a clear description of the technical solutions in the embodiments of this application, the terms "first" and "second" are used in the embodiments of this application to distinguish identical or similar items with essentially the same function and effect. For example, "first instruction" and "second instruction" are used to distinguish different user instructions and do not limit their order. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0018] It should be noted that in the embodiments of this application, the words "exemplarily" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplarily" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplarily" or "for example" is intended to present the relevant concepts in a specific manner.

[0019] Furthermore, "at least one" refers to one or more, while "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, and c can mean: a, or b, or c, or a and b, or a and c, or b and c, or a, b, and c, where a, b, and c can be single or multiple.

[0020] It should be noted that, in the embodiments of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0021] In existing technologies, error correction notebook products generally adopt a text similarity matching scheme based on a question bank: the system calculates the text similarity between the original question and questions in a pre-set question bank, and then selects questions with high similarity to push to the user. However, since similar questions are entirely derived from a pre-set question bank, if the question bank is not updated in a timely manner, there may be situations where no similar questions can be pushed, affecting the user experience. Therefore, how to accurately generate similar questions for incorrect answers is an urgent problem to be solved.

[0022] In view of this, embodiments of this application propose a method, apparatus, electronic device, and storage medium for generating similar questions to incorrect questions. The method includes: acquiring text information of a target incorrect question, the text information including question information of the target incorrect question; and determining target similar questions related to the target incorrect question based on the text information, wherein the similar questions have the same structural features as the target incorrect question, and the structural features include at least one of question stem structure, knowledge points, difficulty level, and logical relationship. The above technical solution, by acquiring the text information of the target incorrect question and determining the target similar questions related to the target incorrect question based on the text information, can generate entirely new similar questions based on the structural features of the target incorrect question, without relying on a preset question bank. This solves the problem of traditional solutions having no questions to recommend or recommending duplicates due to insufficient question banks. Furthermore, unlike simple keyword or vector similarity retrieval methods, the similar questions generated in this application embodiment are consistent with the structural features of the target incorrect question. The structural features include at least one of the following: question stem structure, knowledge points, difficulty level, and logical relationship. This can effectively avoid the problem of recommending similar questions that are too easy or too difficult, enhance the scalability and adaptability of the content, and improve the accuracy and controllability of similar question matching. As a result, it provides users with continuous, accurate, and personalized error consolidation training, improving the review efficiency and learning effect of incorrect questions.

[0023] In some embodiments, a method for generating similar questions to incorrect questions can be applied to a terminal device, which may be a smart learning device with text processing and display functions, such as a tutoring machine, learning tablet, smartphone, smart learning notebook, dictionary pen, AI learning desk lamp, or student computer.

[0024] Taking a tutoring device as an example, when a student encounters a wrong question during the learning process, they can input the wrong question into the tutoring device by taking a picture, handwriting input, or voice-to-text conversion. After obtaining the text information of the target wrong question, the tutoring device generates one or more questions that are highly similar to the target wrong question and pushes them to the screen for the student to practice.

[0025] For example, the tutoring machine can also dynamically adjust the difficulty level of subsequent questions based on the student's answers to similar questions: if the student answers correctly, the difficulty is appropriately increased; if the student answers incorrectly, similar questions with slightly lower difficulty or simpler structure are generated, achieving truly personalized error consolidation training.

[0026] The following details the execution steps of a method for generating similar incorrect questions according to embodiments of this application, such as... Figure 1 As shown, it includes: Step 101: Obtain the text information of the target incorrect question. The text information includes the question information of the target incorrect question.

[0027] For example, the text information includes the question information of the target incorrect question.

[0028] The system receives target incorrect questions uploaded by users to the tutoring machine. These target incorrect questions can be in the form of images or text. If the target incorrect question is uploaded as an image, it can be recognized by Optical Character Recognition (OCR) to obtain the question information of the target incorrect question in image format.

[0029] For example, the target incorrect question can be directly input into the terminal device in text form, and the question information of the target incorrect question can be obtained directly through the text information.

[0030] In this embodiment, the text information also includes at least one of the following: the user's textbook information, the user's educational progress, and the user's age.

[0031] Among these, the user's textbook information can be the subject and publisher selected by the user; the user's educational progress can refer to the grade level currently being studied, for example, if the user is a fourth-grade student, the user's educational progress is fourth grade; the user's age can be the grade level directly selected by the user, and the terminal device infers the user's current educational progress or textbook information based on the user's age, etc.

[0032] For example, when a user (student) uses the tutoring device for the first time, they are prompted to select their current grade level, such as the second semester of fifth grade. After obtaining this grade information, the terminal device automatically infers that the user should be following the second semester of fifth grade. It further combines the device's built-in regional settings, such as the People's Education Press version, Beijing Normal University Press version, or Jiangsu Education Press version, or the textbook version manually confirmed by the user, to determine the scope of core knowledge points in subjects such as mathematics, Chinese, and English covered at the current stage of that grade, such as addition and subtraction of fractions, and the volume of cuboids and cubes.

[0033] Based on this, when the tutoring machine generates similar questions to the user's mistakes, it will prioritize ensuring that the knowledge points of the generated questions are strictly limited to the textbook syllabus of the user's current grade, avoiding outdated or excessive content, thereby achieving precise synchronization with the school's teaching progress.

[0034] Step 102: Based on the text information, identify the target similar questions related to the target wrong question. The similar questions have the same structural features as the target wrong question. The structural features include at least one of the following: question stem structure, knowledge points, difficulty level, and logical relationship.

[0035] The following is a detailed elaboration based on the content of step 102 you provided, presented in a continuous and coherent paragraph format: Based on the obtained text information of the target wrong question, identify the target similar questions related to the target wrong question, and the identified similar questions are consistent with the target wrong question in terms of structural features, wherein: the structural features include at least one of the following: question stem structure, knowledge points, difficulty level, and logical relationship.

[0036] In the embodiment, the question stem structure refers to the question type and layout, such as whether the question type is a fill-in-the-blank question, an application question, or a multiple-choice question, the number of blanks in a fill-in-the-blank question, and the segmented structure of the question stem: whether it is a single-segment description or presented in segments based on conditions. Maintaining a consistent question stem structure helps students quickly understand the meaning of the question in a familiar information arrangement.

[0037] In this embodiment, the knowledge point refers to the subject content category tested by the question. For example, if the knowledge point is fraction multiplication, the formulas involved are multiplying the numerators and the denominators, as well as the operation rules, to ensure that similar questions and target wrong questions are located in the same weak knowledge area.

[0038] In this embodiment, the difficulty level can be quantified from multiple dimensions, including the range of numbers, the number of calculation steps, and the complexity of the question. For example, the range of numbers includes integer operations within 10, 20, or 100; the number of calculation steps includes one-step calculation, two-step calculation, or multi-step recursive calculation; and the complexity of the question includes simple descriptive questions, questions containing redundant information, questions requiring conditional filtering, etc.

[0039] For example, the difficulty level can be obtained by summing the weighted product of the number range, the calculation steps, and the complexity of the problem.

[0040] For example, each dimension can be assigned a quantitative score, and a corresponding weight can be allocated according to the importance of that dimension in the overall difficulty of the question. By multiplying the scores of each dimension by their weights and summing the results, the overall difficulty level of the question can be obtained. This quantitative method can transform the originally abstract subjective judgment of difficulty into a calculable and comparable objective indicator, providing a precise data foundation for matching similar questions and controlling difficulty.

[0041] For example, suppose the formula for calculating the difficulty level is as follows: Difficulty level = Numeric range weight coefficient × Numeric range score + Calculation step weight coefficient × Calculation step score + Question complexity weight coefficient × Question complexity score.

[0042] The weights of each dimension can be dynamically adjusted according to the characteristics of the subject and the requirements of the grade level. For example, in lower elementary school mathematics, the range of numbers may be given higher weight because the main cognitive bottleneck for students at this stage is the perception of the size of numbers and their proficiency in calculation. In higher grade word problems, the weights of calculation steps and problem complexity will be increased accordingly to reflect the focus of the assessment on students' logical reasoning and information extraction abilities.

[0043] For example, a primary school math problem is: Xiaoming has 15 apples. He gives 5 to Xiaohong and 3 to Xiaogang. How many apples does Xiaoming have left? For this question, assuming a score of 0.5 for numbers within 10, 1 for numbers within 20, and 2 for numbers within 100, the range of numbers is within 20, so its score can be set to 1. Assuming one-step calculation is 1, two-step is 2, and three or more steps is 3, the calculation steps include two subtraction operations, totaling two steps, so its score can be set to 2. Assuming the complexity of the question stem is set as 1 for simple, straightforward, 2 for redundant information, and 3 for conditional filtering, then the complexity of this question stem belongs to the simple, straightforward type, containing one main character and three consecutive actions, without redundant or distracting information, so its score can be set to 1. If the weights are set as 0.2 for the number range, 0.5 for the calculation steps, and 0.3 for the question stem complexity, then the difficulty level = 0.2 × + 0.5 × 2 + 0.3 × 1 = 1.5. Therefore, the difficulty level can be 1.5. By controlling the consistency or gradual adjustment of the difficulty level, students can be matched with practice questions of appropriate difficulty.

[0044] In the embodiment, the logical relationship includes the main conditions and quantity given by the target incorrect question, including the number of known conditions, the type of each condition (numerical conditions, textual description conditions, graphical implicit conditions, etc.), and the logical organization method between conditions, such as parallel relationship (multiple conditions are true at the same time and do not depend on each other), progressive relationship (the conclusion of the previous condition is a known premise of the next condition), causal relationship (B is true because A is true), or condition limiting relationship (the problem is discussed within a certain limited scope).

[0045] Based on the text information, target similar questions related to the target wrong questions are identified. The similar questions have the same structural features as the target wrong questions, thereby ensuring that the generated similar questions truly meet the students' needs for consolidation and training.

[0046] The above technical solution, by acquiring the text information of the target incorrect question and determining the target similar questions related to the target incorrect question based on the text information, can generate entirely new similar questions based on the structural features of the target incorrect question, without relying on a preset question bank. This solves the problem of traditional solutions having no questions to recommend or recommending duplicates due to insufficient question banks. Furthermore, unlike simple keyword or vector similarity retrieval methods, the similar questions generated in this application embodiment are consistent with the structural features of the target incorrect question. The structural features include at least one of the following: question stem structure, knowledge points, difficulty level, and logical relationship. This can effectively avoid the problem of recommending similar questions that are too easy or too difficult, enhance the scalability and adaptability of the content, and improve the accuracy and controllability of similar question matching. As a result, it provides users with continuous, accurate, and personalized error consolidation training, improving the review efficiency and learning effect of incorrect questions.

[0047] In some implementation examples, based on text information, target similar questions related to the target incorrect question are identified, such as... Figure 2 As shown, it includes the following steps: Step 201: Parse the text information and extract the parsing results of the target incorrect questions. The parsing results include the structured features of the target incorrect questions and the replacement items of the target incorrect questions.

[0048] The text information of the acquired target incorrect questions is analyzed in depth to extract the analysis results. These results include two parts: first, the structured features of the target incorrect question, namely the multi-dimensional structured information such as the question stem structure, knowledge points, difficulty level, and logical relationships mentioned above; second, the replacement items of the target incorrect question. Replacement items refer to specific elements in the question that can be replaced or changed while maintaining the core logic, such as specific numerical values, character names, item names, scene settings, unit symbols, and other variable details.

[0049] For example, taking a typical elementary school math mistake as an example, the target mistake is "A rectangular flower bed is 8 meters long and 5 meters wide. How long is the fence to be built around the flower bed?" After the model analyzes the problem, it outputs the following structured features.

[0050] In terms of question type and layout, the model identified the question as an application problem. The question stem has a single-segment layout, the entire question contains only one question, there are no fill-in-the-blank options, and the sentence structure is a standard structure of stating known conditions and solving the problem. This layout is clear and direct, making it easy for students to quickly locate the known information and the solution objective.

[0051] In terms of the main conditions and quantity, the model extracts two explicit known conditions from the problem: the rectangle is 8 meters long and 5 meters wide. It also identifies one implicit condition: the actual scenario of "building a fence" corresponds to calculating the perimeter of the rectangle, not its area or other quantities. Correctly identifying this implicit condition is crucial for maintaining the logical premise of the original problem when generating similar problems later. In terms of the logical relationships between conditions, the model determines that the conditions in the problem form a causal relationship: because the problem explicitly provides the length and width of the rectangle as necessary and sufficient conditions, the length of the fence can be calculated using the perimeter formula. This causal chain from the known to the unknown is the core framework for maintaining the logical consistency of the problem. In terms of the knowledge points tested, the model identifies the core knowledge point of the problem as calculating the perimeter of a rectangle, involving the core formula (length + width) multiplied by 2. Any transformation or substitution unrelated to this formula will change the essential testing intent of the problem. In terms of difficulty level, the model performs quantitative evaluation from multiple sub-dimensions: Regarding the range of numbers, both the length and width are single digits (8 and 5), and the calculated result of 26 is a two-digit number, placing the overall value within 100; regarding the number of calculation steps, the question requires only one step, namely, summing and then multiplying by 2, which is a single compound operation; from the perspective of grade level suitability, the question corresponds to the basic difficulty level of third grade, matching the cognitive level of students at this grade level regarding the concept of perimeter. Through the above multi-dimensional analysis, the model transforms the original natural language question into a structured feature set, where question type and layout, number of conditions and logical relationships, core knowledge points and formulas, and difficulty level are all marked as fixed items, while specific numerical values ​​(8, 5) and scene names (flower bed, fence) can be used as subsequent replacement items for variation.

[0052] For example, parsing the text information of target incorrect questions and extracting structured features and replacement items can be achieved using various methods. The first is a rule-based and template-based parsing method. This method predefines parsing rules or regular expressions for specific question types and subjects, extracting information through keyword matching, sentence template matching, etc. Another example is a parsing method based on traditional natural language processing, which comprehensively utilizes techniques such as word segmentation, part-of-speech tagging, named entity recognition, and dependency parsing for deep parsing of the question text. Yet another example is a parsing method based on deep learning and large models, which uses a pre-trained large language model for end-to-end semantic understanding and information extraction. For example, by designing prompt words, the target incorrect question text is input into a large model, and the model is required to output the parsing results in a structured format (such as JSON).

[0053] It should be noted that the above three parsing methods are merely illustrative examples and do not constitute a limitation of the embodiments of this application. Those skilled in the art can also adopt other suitable parsing methods according to actual needs, such as semantic parsing based on graph neural networks, reasoning parsing based on knowledge graphs, etc., which will not be listed in detail here.

[0054] Step 202: Based on the analysis results of the target incorrect question, determine the generation constraint rules for similar questions and the replacement items for similar questions. The generation constraint rules are used to indicate that the structured features of similar questions are consistent with the structured features of the target incorrect question.

[0055] Based on the analysis results of the target incorrect questions obtained in step 201, two key types of information needed to generate similar questions are further determined: the generation constraint rules for similar questions and the replacement items for similar questions. The core function of the generation constraint rules is to strictly instruct that the generated similar questions maintain a high degree of consistency with the target incorrect questions in terms of structural features. Specifically, these rules transform various structural features in the analysis results—including question type and layout, main conditions and quantity, logical relationships between conditions, knowledge points tested, and difficulty level—into fixed generation constraint rules.

[0056] Taking the aforementioned rectangular fence problem as an example, the analysis results identify the problem type as an application problem, the problem stem as a single-segment known condition and problem, the number of conditions as two explicit conditions (length and width), the logical relationship as causal logic, the knowledge point as the rectangle perimeter formula (length + width) × 2, and the difficulty level as "number range within 100, one-step calculation, third-grade basic difficulty". These characteristics will be encapsulated into generation constraint rules, requiring that every similar problem generated subsequently must meet the following requirements: it must be an application problem that directly solves the perimeter using known conditions, it must contain a scenario premise of exactly two explicit conditions, it must follow causal logic, it must use the rectangle perimeter formula as the core solution method, and the number of numbers and calculation steps must be at the same difficulty level as the original problem.

[0057] Simultaneously, the system extracts or automatically generates replacements for similar questions from the analysis results. These replacements are specific contents that can be freely varied while strictly adhering to the aforementioned generation constraints. For example, the length values ​​8 meters and 5 meters in the original question can be replaced with other single-digit combinations such as 7 meters and 4 meters, or 9 meters and 6 meters; the flower bed in the scene can be replaced with a playground, vegetable garden, or picture frame; and the expression "fence," which implies perimeter, can be replaced with semantically equivalent but different terms such as "lace," "fenced with colored lights," or "framed." By determining the generation constraints and replacements, the system ensures strict consistency between similar questions and the original incorrect questions in terms of essential logic, knowledge points, and difficulty level, while also giving the question generation process sufficient diversity and flexibility. This allows multiple similar questions to be derived from the same incorrect question, providing students with rich and precise resources for reinforcement exercises.

[0058] Step 203: Based on the generation constraints of similar questions and the replacement terms of similar questions, determine the target similar questions related to the target wrong question.

[0059] Based on the determined generation constraint rules and replacement items, step 203 generates new similar questions in real time (instead of extracting them from the question bank) by replacing the scenario and specific numbers, under the premise of strictly implementing the constraint rules, and finally determines the target similar questions related to the target wrong questions.

[0060] The aforementioned technical solution generates constraint rules requiring similar questions to maintain strict consistency with the target incorrect question in terms of structured features such as question stem structure, knowledge points, difficulty level, and logical relationships. This overcomes the problems of poor matching accuracy and uncontrollable similarity caused by traditional keyword or vector similarity-based retrieval, effectively avoiding the phenomenon of recommended questions being too easy or too difficult, and improving the accuracy and controllability of similar question matching. Secondly, by using structured features as a fixed constraint framework and treating question scenarios, numerical values, person names, and object names as flexible replacement items, it is possible to diversify the questions while maintaining the core problem-solving logic and knowledge points. This ensures the relevance of practice and enriches the presentation of questions, avoiding aesthetic fatigue caused by mechanical repetition. Thirdly, this method does not rely on a pre-set static question bank but generates entirely new questions in real time based on the analysis results and constraint rules. This completely solves the problem of insufficient question bank coverage leading to a lack of questions to recommend or duplicate recommendations in traditional solutions, greatly enhancing the scalability and adaptability of the learning system.

[0061] The following examples detail how to determine target similar questions related to the target wrong question based on the similar question generation constraints and similar question replacement items.

[0062] In some embodiments, target similar questions related to the target incorrect question are determined based on the similar question generation constraints and the similar question replacement terms, such as... Figure 3As shown, it includes the following steps: Step 301: Determine the replacement content for similar questions. Replacement items include at least one of the following: question scenario, object name, and life situation.

[0063] Determine the specific replacement content for similar questions. For example, the replacement includes at least one of the following: question scenario, object name, and life situation, which are elements that can be flexibly modified based on the original question.

[0064] The question scenario refers to the background environment set in the question, such as flower bed, playground, study, supermarket, etc.; the name of the object refers to the specific object or object involved in the question, such as fence, border, tape, fence, colored lights, etc.; the life situation refers to the daily life segment reflected in the question, such as building a fence, embroidering lace, applying tape, wrapping colored lights, framing, etc.

[0065] These replaceable elements are extracted from the analysis results, and appropriate replacement content is generated in real time using a pre-defined replacement library or a large model. Taking the aforementioned rectangular fence problem as an example, the original problem's scenario is a flower bed, the object name is a fence, and the life situation is building a fence. In step 301, it can be determined that the problem scenario can be replaced with a desk, poster, or vegetable garden; the object name can be replaced with a wooden frame, tape, or fence; and the life situation can be replaced with a wooden frame, a ring of tape, or a fence.

[0066] For example, the selection of replacement content needs to maintain semantic rationality and consistency. For instance, it is reasonable for a desk to correspond to a frame, but it is not logical for a desk to correspond to a fence. The system will perform semantic matching verification during the replacement to ensure that the generated question description is natural and fluent.

[0067] Step 302: Under the constraints of the generated constraint rules, determine the target similar question based on the replacement content of the replacement item.

[0068] Under the strict constraints of the generation rules, the target similar questions are finally determined. The generation rules stipulate the structured characteristics that similar questions must maintain, including fixed items such as question type and layout, number of conditions and logical relationships, knowledge points tested, and difficulty level. These fixed items are used as the framework boundaries, and then the replacement content determined in step 301 is filled into the corresponding positions in the framework one by one to form a complete target similar question.

[0069] The following will illustrate this in detail with examples: Based on the analysis of the aforementioned rectangular fence problem, the generation constraint rules explicitly stipulate that: the problem type must be an application problem; the problem stem must adopt a single-segment structure from "statement of known conditions to problem solution"; it must include scenario conditions with two explicit side length conditions; the logical relationship must maintain a causal relationship; the core knowledge point must be the formula for the perimeter of a rectangle (length + width) × 2; and the difficulty level must be controlled within the numerical range of 100 and be a one-step calculation. Replacements include: the length value can be replaced with other single-digit or two-digit combinations (but must remain within 100 and be within the range that can be completed in one step); scenario terms can be replaced with "desk," "poster," "vegetable garden," etc.; and the implicit perimeter description can be replaced with semantically equivalent terms such as "bordering," "tapping," "fence," etc.

[0070] Based on the generation constraints and replacement terms of similar questions, the following three brand-new similar questions can be generated in real time: For example, Question 1: A rectangular desk is 9 decimeters long and 4 decimeters wide. If a wooden frame is added to the desk, how long will the frame be in decimeters? This question strictly maintains the question type and layout structure of the original question, containing two explicit conditions (9 decimeters long and 4 decimeters wide), with a cause-and-effect relationship. The knowledge point is still the formula for the perimeter of a rectangle, and the calculation process is (9+4)×2=26 decimeters. The range of numbers is controlled within 100, and the number of calculation steps is one step. The difficulty level is highly consistent with the original question.

[0071] For example, Question 2: A rectangular poster is 12 cm long and 5 cm wide. If a strip of tape is applied around the poster, how long will the tape be? This question also follows the generation constraint rules. Although the numbers 12 and 5 are slightly larger, they are still within 100, and the calculation result (12+5)×2=34 cm is still a two-digit number. The number of calculation steps is still one. The expression "applying tape" is semantically equivalent to "fence" and "bordering" and is a concept of calculating perimeter. The overall difficulty is in line with the basic difficulty of the third grade set in the original question.

[0072] For example, question 3: A rectangular vegetable garden is 10 meters long and 6 meters wide. If a fence is built around the garden, how long will the fence be? This question also satisfies all the constraints; the length of 10 meters and the width of 6 meters are both within 100. The calculation result is (10+6)×2=32 meters. The scenario of building a fence is very similar to building a wall, and the logical relationship and knowledge points are completely consistent.

[0073] Through the above process, while strictly adhering to the generation constraints, replacement items (numbers, scenarios, and implicit expressions) are flexibly filled into the structural framework, generating three entirely new similar questions that are highly consistent with the target incorrect question in terms of structural features in real time. These three questions are not extracted from any pre-set question bank, but are created in real time based on the constraints and replacement items. This fully demonstrates the technical advantages of this method, which does not rely on a fixed question bank and can achieve unlimited dynamic supply, providing students with precise consolidation practice resources with multiple similar questions that are close to the target incorrect question.

[0074] In the embodiment, during the process of generating target similar questions, simple repetition of the content of the replacement items of similar questions should be avoided. For example, each content of the replacement item of the target wrong question can be replaced to obtain the replacement content of the replacement item of the similar question.

[0075] For example, when the replacement items include numbers, item names, people, situations, and pattern elements, all the content of each replacement item for the target incorrect question can be replaced to obtain the replacement content for the similar questions.

[0076] For example, in the process of generating similar questions, in order to avoid simple repetition, a variety of replacement strategies can be adopted to differentiate the questions from multiple dimensions such as numbers, item names, people, situations, and pattern elements, so as to ensure that each new question generates replacement content while maintaining the consistency of the core structured features.

[0077] For example, in terms of number replacement, not only can the size of the value be changed, but also the type and combination of the numbers can be changed. For example, the simple quantity of items can be replaced with values ​​of different units such as amount, time, and length, or numbers with special meaning can be introduced to avoid making simple single-digit or two-digit adjustments every time.

[0078] For example, when it comes to replacing item names, you can cross-select from different item categories, such as from fruits to stationery, household goods, or animals. By changing items across categories, you can make the question scenarios more diverse and interesting.

[0079] For example, in terms of character replacement, different types of characters can be introduced, including professional titles, kinship relationships, nicknames, or anthropomorphic animal characters, to avoid always using fixed and common combinations of names.

[0080] For example, in terms of context substitution, you can rotate between multiple life scenarios such as shopping, travel, sports, cooking, school, family, and nature. Each scenario has unique background vocabulary and logical relationships, making the questions more immersive and the context more novel.

[0081] For example, in terms of pattern elements, it is especially suitable for geometry or graphic problems. The shape, direction, color, division method or labeling method of the graphic can be replaced to enrich the visual presentation and thinking angle of the problem while keeping the core knowledge points unchanged.

[0082] Through the above multi-dimensional differentiated replacement strategy, each similar question uses a combination of multiple replacement methods, so that there are obvious differences between batches of similar questions. This avoids learners' aesthetic fatigue and rote memorization, and ensures that each question can accurately train the same knowledge weakness point, truly achieving a high-quality personalized practice supply.

[0083] In the process of generating similar questions to the target, it should also be ensured that the replacement items for similar questions are within a reasonable range, which will be described in detail in the following examples.

[0084] In some embodiments, the replacement content for the alternatives to similar questions is determined, such as... Figure 4 As shown, it includes the following steps: Step 401: According to the preset reasonable range of replacement content, perform reasonableness verification on the replacement content of the replacement item; reasonableness verification includes at least one of numerical reasonableness verification, digital combination logic reasonableness verification, and scenario description reasonableness verification.

[0085] The reasonableness of the replacement content is verified according to the preset reasonable range of replacement content. This reasonableness verification includes at least one of the following: numerical reasonableness verification, numerical combination logic reasonableness verification, and scene description reasonableness verification. Numerical reasonableness verification ensures that the replaced numbers conform to basic common sense and subject requirements. For example, in lower elementary school questions, the replaced numbers should not be negative or excessively large values ​​beyond the cognitive range of that grade level. Similarly, in age-related questions, the replaced character's age should not exceed the normal human lifespan range; for example, questions with an age of 200 years or more should not be illogical. In length-related questions, the size of the replaced object should conform to the typical size range of such objects in the real world. Numerical combination logic reasonableness verification ensures that the replaced numbers meet necessary logical constraints. For example, in time sequence questions, the start time must be earlier than the end time. Scene description reasonableness verification ensures semantic consistency and reasonableness between the replaced scene, character, object names, and context. For example, a desk with a frame is reasonable, but a desk with a fence is illogical.

[0086] For example, a model-based approach can be used. This involves training the model with a large amount of reasonable and unreasonable sample data for each reasonableness verification item, and then using the trained model to perform compliance checks, thereby improving the intelligence and generalization ability of the verification process. The content of each replacement item is compared one by one.

[0087] Step 402: If the replacement content of the replacement item is within a reasonable range, determine that the replacement content of the replacement item is the replacement content of a similar question.

[0088] If a replacement item is deemed within a reasonable range after a validity check, it is considered a valid replacement for a similar question and proceeds to the subsequent question generation process. Conversely, if a replacement item fails the validity check—for example, if the replaced number exceeds a preset range, the number combination does not meet the original logical constraints, or the replaced scenario description has semantic conflicts—it will be rejected. Other suitable replacement items will then be selected or generated until all replacement items pass the validity check.

[0089] Through the above-mentioned rationality verification mechanism, it can be ensured that each similar question generated in the end not only maintains the same structural features as the target wrong question, but also has basic rationality and authenticity in numerical logic, common sense semantics and scenario matching. This provides students with high-quality, logically error-free consolidation practice resources, thereby improving the learning experience and practice effect.

[0090] After generating at least one similar question based on the text information of the target incorrect question, a similarity check is performed on the similar questions to improve the accuracy of the similar questions. The following example describes the detailed process of performing a similarity check on the generated at least one similar question.

[0091] In some embodiments, based on text information, target similar questions related to the target incorrect question are determined, such as... Figure 5 As shown, it includes the following steps: Step 501: Based on the text information, generate at least one similar question related to the target incorrect question.

[0092] Based on the text information of the target incorrect question, the parsing results are extracted. Generation rules and replacement items are determined based on the parsing results. Then, at least one similar question related to the target incorrect question is generated according to the generation rules and replacement items. The process of generating similar questions has been discussed in detail above and will not be repeated here.

[0093] Step 502: Perform a similarity consistency check on the structured features of the target incorrect question and the structured features of each similar question.

[0094] A similarity consistency check is performed on the structured features of the target incorrect question and the structured features of each similar question generated in step 501. This check aims to ensure that the generated questions are highly consistent with the original incorrect questions in core dimensions, including four key dimensions: structural consistency check, knowledge point consistency check, difficulty level consistency check, and logical relationship consistency check. Among them, the structural consistency check is used to compare the matching degree between the target wrong question and the similar question in terms of question stem structure, question type layout, number of conditions and arrangement, etc. For example, it checks whether the question types of both are application questions, whether the question stems are both single-segment "known conditions and questions" structures, and whether the number of blanks is the same. The knowledge point consistency check is used to confirm whether the core knowledge content tested by the similar question is consistent with the target wrong question. For example, whether both are formulas for calculating the perimeter of a rectangle, and whether both involve the operation rules of fraction multiplication. The difficulty level consistency check is to quantitatively compare from sub-dimensions such as the range of numbers, the number of calculation steps, and the complexity of the question stem to ensure that the difference between the difficulty level of the similar question and the original wrong question is controlled within a preset threshold. For example, it requires that the overall difficulty level score deviation between the two does not exceed ±0.3. The logical relationship consistency check is used to verify whether the organization of conditions in the similar question is consistent with the original wrong question, including whether the causal relationship, progressive relationship, parallel relationship or condition limiting relationship between conditions are the same, and whether the number and type of known conditions match.

[0095] For example, in the actual verification process, each dimension can be quantitatively scored. For instance, the maximum similarity score for each dimension can be set to 100 points. The score for each dimension is calculated based on the comparison results, and then the scores for each dimension are weighted and summed to obtain the comprehensive similarity score. In the case of any dimension being seriously deviated, the system can also set a single-dimensional rejection mechanism, that is, if the similarity of a certain dimension is lower than the set single-item threshold, even if the comprehensive score is high, it is determined to be a failure of verification.

[0096] Step 503: Identify similar questions with a similarity greater than the similarity threshold as target similar questions for the target incorrect question.

[0097] Similar questions with a similarity greater than a preset similarity threshold are identified as the final valid similar questions to the target incorrect question. For example, this similarity can be the comprehensive similarity of each feature integrated in the structured features, the similarity between each feature of the target incorrect question and the similar question in the structured features, or the combined similarity of the comprehensive similarity and the similarity between each feature.

[0098] For example, a similarity threshold can be preset, with a comprehensive similarity score greater than or equal to 85 points, or requiring that the individual similarity score of each dimension of the structured features be no less than 80 points and the comprehensive similarity score be no less than 90 points. Similar questions that pass the verification can be marked as qualified and identified as target similar questions for the target incorrect question.

[0099] In other embodiments, similar questions that fail the verification, i.e., questions with a comprehensive similarity below the threshold or that fail the verification of a certain dimension, will be deleted, and a regeneration mechanism will be triggered as appropriate. For example, other similar questions that pass the verification during the current generation process will be retained, while unqualified similar questions will be corrected and regenerated in a targeted manner according to the reasons for failure, until all planned similar questions pass the verification or the maximum number of retries is reached.

[0100] The system pushes verified similar questions to users for practice. Furthermore, before pushing these questions, duplicates can be removed to avoid sending highly similar questions and ensure the diversity and effectiveness of practice resources.

[0101] Through the aforementioned similarity consistency verification mechanism, the quality of the generated questions can be automatically verified, ensuring that each similar question ultimately pushed to students is highly consistent with the target wrong question in terms of structure, knowledge points, difficulty, and logical relationship, thereby providing students with accurate, reliable, and high-quality resources for consolidating and training wrong questions.

[0102] In some embodiments, while generating similar questions to the target incorrect question, the answers and explanations for the similar questions are also generated and displayed through the learning machine, enabling users to practice applying the knowledge gained from the incorrect questions to other situations. Users can directly answer the questions and view the answers and explanations. If the answers or explanations for the similar questions cannot be generated, the similar questions are considered unqualified, and a complete set of similar questions, including expert explanations, is regenerated.

[0103] In some embodiments, determining target similar questions related to the target incorrect question based on text information includes: The text information is input into the question generation model to determine the target similar questions related to the target wrong question. The question generation model is trained based on the sample wrong questions and sample similar questions of the sample wrong questions.

[0104] Identifying similar questions related to the target incorrect question based on textual information can be achieved using a model-based training approach. Specifically, the acquired textual information of the target incorrect question is input into a pre-trained question generation model, which then directly outputs similar questions related to the target incorrect question.

[0105] This question generation model is trained using a large number of sample incorrect questions and their corresponding similar questions. During the training phase, a massive amount of sample data is first collected. Each training sample includes one sample incorrect question and several sample similar questions that maintain consistent structural features with that sample incorrect question. These sample similar questions can be obtained through manual annotation, semi-automatic generation, or selection from a high-quality question bank, ensuring a high degree of matching with the sample incorrect question in dimensions such as question structure, knowledge points, difficulty level, and logical relationships. Then, a sequence-to-sequence generative model architecture is used for training. The text of the sample incorrect question is taken as input, and the corresponding sample similar questions are taken as the target output. By minimizing the loss function between the generated result and the real sample, the model gradually learns the mapping pattern from incorrect questions to similar questions.

[0106] After thorough training, the question generation model has end-to-end similar question generation capabilities: when a new target wrong question is input, the model can automatically parse the structured information of the wrong question and generate target similar questions related to the target wrong question in real time while keeping these core features unchanged.

[0107] The model-based training approach eliminates the need for manually defining parsing rules, constraint rules, and substitution terms. The model autonomously learns the generation patterns from incorrect to similar questions through extensive data, exhibiting stronger generalization and adaptability. It is particularly suitable for scenarios with diverse question types, complex expressions, and those difficult to exhaustively enumerate through rules. Furthermore, the model can be fine-tuned and optimized according to application needs, such as targeted training on labeled data for specific subjects (mathematics, Chinese, English) or specific educational stages (primary, middle, high school) to improve the generation quality and accuracy within those domains. In this way, the system does not rely on a pre-set question bank or require manually writing complex parsing and generation rules; it can achieve real-time, automated generation of high-quality similar questions solely through a trained question generation model.

[0108] In some embodiments, the question generation model can be a neural network.

[0109] For example, a neural network typically includes multiple layers, each of which may include one or more neurons. Increasing the depth and / or width of a neural network can improve its expressive power, providing more powerful information extraction and abstract modeling capabilities for complex systems. The depth of the neural network can refer to the number of layers it includes, and the number of neurons in each layer can be called the width of that layer. In one implementation, the neural network includes an input layer and an output layer. The input layer processes the received input information through neurons and passes the processing results to the output layer, which then obtains the output of the neural network. In another implementation, the neural network includes an input layer, hidden layers, and an output layer, as described in [reference needed]. Figure 6This diagram illustrates a feedforward neural network (FNN). In an FNN, neurons in adjacent layers are completely connected pairwise. The input layer processes the received input information through neurons and passes the processing results to the hidden layers. The hidden layers perform calculations on the received processing results and pass these results to the output layer or an adjacent hidden layer. Finally, the output layer obtains the output of the neural network. A neural network can include one hidden layer or multiple sequentially connected hidden layers; there is no limitation on this.

[0110] A training dataset is used for model training. The training dataset can include the model's input data, or it can include both the model's input data and the target output data. Specifically, the training dataset includes one or more training data sets, which can be data input to the model or the model's target output data. In this embodiment, the training dataset includes sample incorrect questions and sample similar questions, where the sample incorrect questions are the model's input data and the sample similar questions are the target output data. The target output data can also be referred to as labels, output label data, or output label samples. The training dataset is a crucial part of machine learning. Model training essentially involves learning certain features from the training data to make the model's output data as close as possible to the target output data, minimizing the difference between the model's output data and the target output data. The composition and selection of the training dataset can, to a certain extent, determine the performance of the trained model.

[0111] Furthermore, a loss function can be defined during the training process of a model (such as a neural network). The loss function describes the difference or discrepancy between the model's output value and the target output value. This application does not limit the specific form of the loss function. The model training process involves adjusting the model parameters to make the loss function value less than a threshold, or to make the loss function value meet the target requirements.

[0112] The model parameters can include one or more of the following: structural parameters of the model (e.g., the number of layers, and / or weights), for example, if the model is a neural network, the structural parameters of the neural network include at least one of the following: the number of layers, width, weights of neurons, or parameters in the activation function of neurons; input parameters of the model (e.g., input dimension, number of input ports); and output parameters of the model (e.g., output dimension, number of output ports). It can be understood that the input dimension refers to the size of an input data set; for example, when the input data is a sequence, the input dimension corresponding to that sequence can indicate the length of the sequence. The number of input ports can refer to the quantity of input data. Similarly, the output dimension can refer to the size of an output data set; for example, when the output data is a sequence, the output dimension corresponding to that sequence can indicate the length of the sequence. The number of output ports can refer to the quantity of output data.

[0113] Furthermore, neural networks can process data in batches, enabling parallel computation to accelerate training. For example... Figure 7 This diagram illustrates one iteration of a neural network. Multiple training data sets are selected to form a batch, which is then input into the neural network to obtain output data. The output data and its labels are fed into a loss function to calculate the loss value for this iteration. This loss value is then used to update each structural parameter of the neural network, matching the step size parameter, thus completing one iteration of the training process. The batch size represents the number of training data sets processed in each batch. Figure 7 The diagram illustrates a batch of n training data.

[0114] The above embodiments provide a detailed description of the question generation model proposed in this application, using a neural network as an example. It should be understood that the question generation model can also be other types of models, such as linear regression models, decision tree models, etc., which will not be discussed in detail here.

[0115] In some embodiments, the complete process of generating similar questions to incorrect ones is as follows: Figure 8 As shown.

[0116] Step 801: Receive multiple incorrect questions stored by the user in the tutoring machine. The user selects a target incorrect question from the multiple incorrect questions to generate similar questions for that target incorrect question. Step 802: Obtain the text information of the target incorrect question. Next, execute step 803 to parse the text information and extract the parsing result of the target incorrect question. In step 804, based on the parsing result of the target incorrect question, determine the generation constraint rules of the similar questions and the replacement items of the similar questions. Based on the generation constraint rules of the similar questions and the replacement items of the similar questions, determine the target similar questions related to the target incorrect question.

[0117] It should be understood that the execution process of the steps in this embodiment has been described in detail above, and will not be repeated here.

[0118] The following examples demonstrate the target incorrect questions ( Figure 9 ) and similar questions to the target questions () Figure 10 The display of the home tutoring machine interface.

[0119] For example, Figure 9 The interface displays the target incorrect questions. Users can select a question to correct and see that question. They can then perform calculations on the scratch paper to review the target incorrect question. After selecting or answering the correct answer, they can choose to submit. After submission, they can view the explanation of the incorrect question. Even if a user doesn't know the answer to a question, they can directly view the explanation of the incorrect question.

[0120] For example, Figure 10 This interface displays similar questions related to the target incorrect question. The variant questions are the similar questions mentioned in this embodiment. After reviewing the target incorrect question, the user can continue to answer the variant questions, thereby consolidating the knowledge corresponding to the target incorrect question and improving learning efficiency. There can be one or more similar questions for the target incorrect question.

[0121] It should be understood that although the steps in the flowchart above are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowchart above may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0122] In some embodiments, such as Figure 11 As shown, a device for generating similar questions to incorrect questions is provided. The device includes: an acquisition module 1101 and a determination module 1102, wherein: The acquisition module 1101 is used to acquire the text information of the target incorrect question, which includes the question information of the target incorrect question.

[0123] The determination module 1102 is used to determine the target similar questions related to the target wrong question based on the text information. The similar questions have the same structural features as the target wrong question. The structural features include at least one of the following: question stem structure, knowledge points, difficulty level, and logical relationship.

[0124] Further limitations on the device for generating similar questions based on incorrect answers can be found in the limitations on the method for generating such questions above, and will not be repeated here. Each module in the aforementioned device for generating similar questions based on incorrect answers can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of the terminal device in hardware form or independently of it, or stored in the memory of the terminal device in software form, so that the processor can call and execute the operations corresponding to each module.

[0125] Another embodiment provides a computer-readable storage medium for storing a computer program. This computer program contains instructions for implementing the methods described in the embodiments of this application. By installing this computer program on a computer, the computer can execute the corresponding methods.

[0126] Another embodiment proposes a computer program product that includes computer program code. When this computer program code is run on a computer, it causes the computer to implement the methods proposed in the embodiments of this application. Thus, a user can implement these methods by using this computer program product.

[0127] In some embodiments, Figure 12 This is a schematic block diagram of the electronic device provided in the embodiments of this application.

[0128] Electronic device 1200 may include: a memory 1201 storing executable program code and a processor 1202 coupled to the memory 1201.

[0129] In this embodiment, the processor 1202 calls executable program code stored in memory to execute any of the methods disclosed in the embodiments of this application. Those skilled in the art will understand that... Figure 12 The electronic device structure shown does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0130] The processor 1202 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines, and performs various functions and processes data by running or executing software programs and / or modules stored in the memory, and by calling data stored in the memory, thereby providing overall monitoring of the electronic device. Optionally, the processor may include one or more processing units; preferably, the processor may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into the processor.

[0131] The memory 1201 can be used to store software programs and modules. The processor executes various functional applications and data processing of the electronic device by running the software programs and modules stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0132] It should be understood that, in the embodiments of this application, the processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0133] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The steps of the method disclosed in the embodiments of this application can be directly manifested as execution by a hardware processor, or as a combination of hardware and software modules within the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor executes the instructions in the memory, combining them with its hardware to complete the steps of the above method. To avoid repetition, detailed descriptions are omitted here.

[0134] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of this application.

[0135] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0136] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0137] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0138] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0139] If the aforementioned function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application embodiment, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0140] The above description is merely a specific implementation of the embodiments of this application, but the protection scope of the embodiments of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the embodiments of this application should be included within the protection scope of the embodiments of this application. Therefore, the protection scope of the embodiments of this application should be determined by the protection scope of the claims.

Claims

1. A method for generating similar problems of a mistake problem, characterized by, The method includes: Obtain the text information of the target incorrect question, wherein the text information includes the question information of the target incorrect question; Based on the text information, target similar questions related to the target wrong question are identified. The similar questions have the same structured features as the target wrong question. The structured features include at least one of the following: question stem structure, knowledge points, difficulty level, and logical relationship.

2. The method according to claim 1, characterized in that, The step of determining target similar questions related to the target incorrect question based on the text information includes: The text information is parsed to extract the parsing results of the target incorrect question. The parsing results include the structured features of the target incorrect question and the replacement items of the target incorrect question. Based on the analysis results of the target incorrect question, the generation constraint rules for the similar question and the replacement items for the similar question are determined. The generation constraint rules are used to indicate that the structural features of the similar question are consistent with the structural features of the target incorrect question. Based on the generation constraints of the similar questions and the replacement terms of the similar questions, the target similar questions related to the target wrong question are determined.

3. The method according to claim 2, characterized in that, The step of determining the target similar question related to the target wrong question based on the generation constraint rules of the similar question and the replacement terms of the similar question includes: Determine the replacement content of the replacement items for the similar questions, wherein the replacement items include at least one of the question scenario, object name, and life situation; Under the constraints of the generation constraint rules, the target similar question is determined based on the replacement content of the replacement item.

4. The method according to claim 3, characterized in that, The process of determining the replacement content for the similar question's replacement item includes: According to the preset reasonable range of replacement content, the replacement content of the replacement item is checked for reasonableness; the reasonableness check includes at least one of numerical reasonableness check, digital combination logic reasonableness check, and scenario description reasonableness check. If the replacement content of the replacement item is within a reasonable range of the replacement content, the replacement content of the replacement item is determined to be the replacement content of the similar question.

5. The method according to claim 1, characterized in that, The step of determining target similar questions related to the target incorrect question based on the text information includes: Based on the text information, generate at least one similar question related to the target incorrect question; The similarity consistency is verified between the structured features of the target incorrect question and the structured features of each similar question; Similar questions with a similarity score greater than the similarity threshold are identified as target similar questions to the target incorrect question.

6. The method according to claim 1, characterized in that, The text information also includes at least one of the following: the user's textbook information, the user's educational progress, and the user's age.

7. The method according to any one of claims 1-6, characterized in that, The step of determining target similar questions related to the target incorrect question based on the text information includes: The text information is input into the question generation model to determine the target similar questions related to the target wrong question. The question generation model is trained based on the sample wrong questions and the sample similar questions of the sample wrong questions.

8. A device for generating similar incorrect questions, characterized in that, The device includes: The acquisition module is used to acquire the text information of the target incorrect question, wherein the text information includes the question information of the target incorrect question; The determination module is used to determine, based on the text information, target similar questions related to the target wrong question. The similar questions have the same structural features as the target wrong question. The structural features include at least one of the following: question stem structure, knowledge points, difficulty level, and logical relationship.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The device stores computer instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1-7.

Images