Student wrong question automatic acquisition method based on intelligent correction system

Abstract

The invention provides a student wrong question automatic acquisition method based on an intelligent correction system. The method comprises the following steps: S1, acquiring a homework set of any student and inputting the homework set into the intelligent correction system to obtain a correction result; s2, judging whether the correction result meets a triggering condition or not: if so, turning to S3; if not, returning to the step S1; s3, extracting multi-modal wrong question contents including question contents, student answering contents and correction results from the homework set; s4, based on the knowledge graph and the error classification system, performing structured information labeling on the multi-modal error content to obtain an error mastering result; step S5, obtaining a redo correction result of the student for the wrong questions and updating a mastering result of the wrong questions; and S6, storing the updated error question mastering result, the updated multi-modal error question content and the updated structured information according to a general data format. The method has the beneficial effects that full-process automation can be realized, multi-modal information is covered, and high-quality structured error data is generated.

Description

Technical Field

[0001] This invention relates to the technical field of artificial intelligence and education, and more specifically, to a method for automatically collecting students' wrong answers based on an intelligent grading system. Background Technology

[0002] In existing technologies, the collection and management of students' incorrect answers generally suffer from low automation, fragmented processes, and data silos. Specifically, traditional methods of collecting incorrect answers heavily rely on students manually copying down the questions, taking photos and uploading them, or teachers manually screening and marking them. This method is not only time-consuming and labor-intensive (each question takes an average of 3-5 minutes to collect), but it is also prone to missing key information, such as geometric figures, physical circuit diagrams, or handwritten formulas in the questions, resulting in a serious lack of completeness in the collected data.

[0003] While some technical solutions attempt to explore automation using intelligent grading systems, these solutions often have limited functionality and incomplete processes. For example, some systems can only trigger the collection of completely incorrect questions, ignoring the more valuable type of incorrect questions that are "partially incorrect." Furthermore, their extraction of incorrect question content is mostly limited to plain text, failing to effectively handle multimodal information such as formulas and graphs commonly found in science questions, resulting in low collection coverage.

[0004] Furthermore, the various stages from triggering grading and extracting content to storing data are often fragmented, failing to form a unified and coherent automated process. The collected incorrect answer data also generally lacks in-depth structured annotation and dynamic update mechanisms, making it impossible for the data to support precise personalized review. At the same time, due to the lack of standardized data formats, incorrect answer data between different systems is difficult to share, creating serious data silos.

[0005] Therefore, there is an urgent need in this field for an automatic student error collection method that can achieve full-process automation, cover multimodal information, and produce high-quality structured error data. Summary of the Invention

[0006] The technical problem to be solved by this invention is how to achieve full-process automation, cover multimodal information and produce high-quality structured error data. In order to overcome the defects of the above-mentioned existing technologies (or related technologies), this invention provides an automatic collection method for student error data based on an intelligent grading system.

[0007] This invention provides a method for automatically collecting students' wrong answers based on an intelligent grading system, comprising the following steps: Step S1: Obtain the homework set of any student and input the homework set into the intelligent grading system to obtain the corresponding grading result; Step S2: Determine whether the modification result meets the preset triggering conditions: If so, proceed to step S3; If not, return to step S1; Step S3: Extract multimodal incorrect questions from the assignment set, including the question content, student answers, and the grading results; Step S4: Based on the knowledge graph and error classification system, the multimodal incorrect questions are annotated with structured information to obtain the results of mastering the incorrect questions; Step S5: Obtain the student's redo and correction results for the incorrect questions and update the mastery results of the incorrect questions based on the redo and correction results; Step S6: Standardize and store the updated results of mastering incorrect questions, the multimodal incorrect question content, and the structured information according to a common data format.

[0008] Compared with existing technologies, the automatic collection method for student error questions based on an intelligent grading system of the present invention has the following advantages: This invention achieves a fully automated, end-to-end closed loop for error collection and management, covering multimodal information. Through the continuous steps from obtaining grading results to final data storage, it fundamentally solves the problems of high reliance on manual labor and fragmented processes in traditional error collection. It can systematically produce high-quality, structured error data, providing a solid data foundation for personalized learning analysis, thereby significantly improving students' learning efficiency and the accuracy of teachers' teaching interventions. (Step S1: Obtaining grading results; Step S2: Determining triggering conditions; Step S3: Extracting multimodal errors; Step S4: Analyzing error mastery results; Step S5: Updating error mastery results; Step S6: Data storage)

[0009] In one possible implementation, the triggering conditions in step S2 include a completely wrong triggering condition and a partially wrong triggering condition. The completely wrong triggering condition is that the grading result of the objective question is incorrect, or the grading result of the subjective question is 0 points, or the grading result of the subjective question is less than 30% of the score. The partially wrong triggering condition is that the grading result of the subjective question is not less than 30% of the score, but there is a specific error point marked by the intelligent grading system.

[0010] Compared with existing technologies, the above-mentioned technical solution, by distinguishing between completely wrong trigger conditions and partially wrong trigger conditions, can not only collect questions that were completely wrong, but also capture representative wrong questions that students knew how to do but did not answer correctly, thus avoiding the omission of valuable wrong questions. At the same time, it eliminates false wrong questions caused by non-knowledge factors such as not answering, significantly improving the data quality and review value of the wrong question bank, enabling students to focus on their real knowledge weaknesses.

[0011] In one possible implementation, step S3, the extraction of multimodal incorrect question content includes the extraction of basic question information, the extraction of student answers, and the extraction of intelligent grading results. The extraction of basic question information involves extracting text information, formula information, graphic information, and question meta-information from the question and including them in the multimodal incorrect question content. The extraction of student answers involves extracting the text, formulas, graphics, and answer behavior data of the student's answers and including them in the multimodal incorrect question content. The extraction of intelligent grading results involves extracting the grading details and error marking information from the grading results and including them in the multimodal incorrect question content.

[0012] Compared with existing technologies, the above technical solution can clearly define the scope of multimodal error content extraction. It can not only record the text of questions and answers, but also extract unstructured information such as formulas, graphs, and answering behaviors, thus achieving holographic and multidimensional capture of error information.

[0013] In one possible implementation, in step S3, the formula in the student's answer is identified by MathOCR technology and converted into standard LaTeX format and included in the multimodal incorrect question content.

[0014] Compared with existing technologies, the above technical solution enables the formulas to be understood and retrieved by computers, and also facilitates their direct use in subsequent error correction and variation problem generation, greatly improving the machine readability and application value of science error data.

[0015] In one possible implementation, in step S3, the graphic information in the question is extracted using computer vision technology, and a graphic pixel map and graphic feature description are generated and included in the multimodal incorrect question content.

[0016] Compared with existing technologies, the above-mentioned technical solution can not only store the pixel files of the graphics, but also understand the semantic content of the graphics. This enables the system to perform retrieval and reasoning based on graphic features, breaking through the limitation of traditional methods where graphics are just unanalyzable images. This provides the possibility for intelligent tutoring and question recommendation, and realizes the structured parsing and semantic description of graphic information.

[0017] In one possible implementation, in step S4, the annotation of the structured information includes knowledge point annotation, error type annotation, initial mastery annotation, and personalized tag annotation. The knowledge point annotation is to annotate the main knowledge point and related knowledge point corresponding to the wrong question. The error type annotation is to annotate the corresponding error type according to the error classification system. The initial mastery annotation is to annotate the student's initial mastery according to the score rate represented by the grading results. The personalized tag annotation is to annotate personalized tags based on the answer behavior data.

[0018] Compared with existing technologies, the above-mentioned technical solution can transform raw incorrect question data into semantically rich structured information, enabling precise retrieval by marking knowledge points, error types, etc., and providing students with targeted practice opportunities, thus realizing personalized review path planning.

[0019] In one possible implementation, in step S6, a hybrid storage method combining relational databases and graph databases is used to store the structured information and the association information between incorrect questions, knowledge points, and error types.

[0020] Compared with existing technologies, the above technical solution can balance efficient retrieval and complex relationship analysis. Relational databases can be used to quickly and flexibly query structured information, while graph databases can be used to efficiently analyze and mine the complex network relationships between "wrong questions-knowledge points-error types", providing powerful data insight capabilities for macro-level learning diagnosis and teaching strategy adjustment.

[0021] In one possible implementation, step S5 includes: Step S51: Set redo triggering conditions, including timed triggering or active triggering. When the redo triggering conditions are met, the intelligent grading system pushes redo tasks to students to obtain the redo grading results of students on wrong questions. Step S52: Update the incorrect question mastery result according to the redo and correction result and the preset status update logic, and record the redo and correction result and status change log of each redo to form the incorrect question mastery trajectory, which is included in the incorrect question mastery result.

[0022] Compared with existing technologies, the above technical solution can ensure the timeliness and dynamic adaptability of error data. By tracking the redo and correction results and automatically updating the mastery status of error questions, it avoids the problem of a large amount of invalid information accumulating in static error notebooks, and greatly improves review efficiency.

[0023] In one possible implementation, in step S6, the general data format is defined to use JSON-LD format to unify the multimodal error content and the structured information.

[0024] Compared with existing technologies, the above technical solution can clearly express the semantics of data, enabling error data from different intelligent grading systems and platforms to be understood uniformly and seamlessly integrated. Attached Figure Description

[0025] Figure 1 This is a flowchart of the steps of the present invention. Detailed Implementation

[0026] First, those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention. Those skilled in the art can make adjustments as needed to adapt to specific application scenarios.

[0027] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0028] See Figure 1 This invention discloses an automatic method for collecting student error data based on an intelligent grading system. This method achieves high-quality, structured error data collection and management through a fully automated mechanism of "acquisition → triggering → extraction → annotation → updating → storage," specifically including the following steps: Step S1: Obtain the homework set of any student and input the homework set into the intelligent grading system to obtain the corresponding grading results; Step S2: Determine whether the grading result meets the preset triggering conditions: If so, proceed to step S3; If not, return to step S1; Step S3: Extract multimodal incorrect questions from the assignment set, including the question content, student answers, and grading results; Step S4: Based on the knowledge graph and error classification system, the multimodal incorrect questions are annotated with structured information to obtain the results of mastering the incorrect questions; Step S5: Obtain the students' redo and correction results of the wrong questions and update the students' mastery of the wrong questions based on the redo and correction results; Step S6: Standardize and store the updated results of mastering incorrect questions, multimodal incorrect question content, and structured information according to a common data format.

[0029] In this embodiment of the invention, the homework set in step S1 can be derived from various digital learning platforms, including but not limited to online homework systems, school-owned teaching platforms, or third-party examination systems. The intelligent grading system automatically grades the homework set and generates structured grading results, which at least include the score, details of points lost, and error marking information for each question.

[0030] In this embodiment of the invention, step S2 is an intelligent switch for collecting incorrect questions, designed to accurately filter valuable incorrect questions and avoid invalid data from being entered into the database. The triggering conditions include two categories: all-incorrect triggering conditions and partial-incorrect triggering conditions. The all-incorrect triggering condition applies to objective questions (such as multiple choice questions and true / false questions) and subjective questions (such as calculation questions and proof questions). When the grading result of an objective question is marked as incorrect, or the grading result of a subjective question is 0 points, or its score rate is less than 30%, the collection is directly triggered. This all-incorrect triggering condition is used to capture questions that students have not mastered or have seriously misunderstood. The partial-incorrect triggering condition is mainly for subjective questions. When the grading result of a subjective question is a score rate of not less than 30% but there are specific errors clearly marked by the intelligent grading system (e.g., "Step 2 formula error", "Three key points of Chinese reading comprehension question missing"), the collection is triggered. This partial-incorrect triggering condition can capture questions where students have correct thinking but have key omissions, which has extremely high review value. To improve data quality, exclusion conditions can also be set. Grading errors caused by non-knowledge-based factors such as "not answered" or "incorrect filling" will not trigger the collection.

[0031] In this embodiment of the invention, step S3 is the core of achieving completeness of incorrect question information. It comprehensively captures all relevant information of incorrect questions through multimodal technology, mainly including extraction of basic question information, extraction of student answers, and extraction of intelligent grading results.

[0032] In this embodiment of the invention, the extraction of basic question information involves extracting text information, formula information, graphic information, and question meta-information from the question and including them in the multimodal incorrect question content. Specifically, the text information is extracted using Natural Language Processing (NLP) technology to extract the plain text of the question stem, the text of the answer options, and the answer requirements; the formula information is identified using MathOCR technology to recognize printed formulas in the question and convert them into standard LaTeX format to ensure the editability and reusability of the formulas; the graphic information is extracted using computer vision technology (e.g., combining YOLOv8 object detection with GPT-4V multimodal large model) to extract geometric figures, circuit diagrams, etc., from the question and generate graphic pixel images (such as PNG or SVG format) and structured graphic feature descriptions (e.g., "right triangle, AC=3cm, BC=4cm, ∠C is a right angle"); and the question meta-information includes the question type, score, difficulty level, and corresponding textbook chapter information.

[0033] In this embodiment of the invention, the extraction of student answers includes: Extract the answers to objective questions and the handwritten text of subjective questions (through handwriting OCR recognition, supporting cursive writing); The system identifies handwritten formulas in students' answers, converts them to LaTeX format, and preserves the original handwriting trajectory for error detail comparison. Extract the auxiliary lines and force analysis diagrams drawn by students, and compare them with standard graphics; Extract answer behavior data, such as answering time, number of modifications, and number of times prompts were viewed.

[0034] In this embodiment of the invention, the intelligent grading result extraction includes: Extract detailed correction information, including scores, deductions for each step, and system-generated side notes (e.g., "The formula should be a sum of squares, not a sum"). Extract the coordinates of the erroneous location to facilitate subsequent localization.

[0035] In this embodiment of the invention, step S4 transforms the original multimodal incorrect answers into semantically rich structured information, providing data support for precision teaching. The annotation of this structured information includes knowledge point annotation, error type annotation, initial mastery annotation, and personalized tag annotation. Among them, knowledge point annotation is based on the subject knowledge graph, automatically annotating the main knowledge point (such as "application of the Pythagorean theorem") and related knowledge points (such as "definition of a right triangle") corresponding to the incorrect answer; error type annotation is based on a preset subject error classification system, for example, mathematics is annotated as "formula error" and "calculation error"; Chinese is annotated as "deviation from the main idea" and "missing evidence"; physics is annotated as "principle error" and "unit error"; initial mastery annotation is automatically annotated based on the score rate of this answer, such as "score rate < 30%" is annotated as "not mastered", and "30% ≤ score rate < 70%" is annotated as "needs to be consolidated"; personalized tag annotation is generated based on answer behavior data, such as "excessive answering time" is annotated as "weak time management", and "many revisions" is annotated as "insufficient rigor".

[0036] In this embodiment of the invention, step S5 ensures the dynamism and timeliness of the incorrect problem data, enabling the incorrect problem notebook to evolve along with the student's learning progress. Specifically, it includes: Step S51: Set the redo trigger conditions, including timed triggering (such as automatically pushing redo tasks on the 7th and 14th days after the collection of wrong questions according to the Ebbinghaus forgetting curve) and active triggering (students or teachers manually initiate redoing). Step S52: Based on the redo and correction results, automatically update the mastery status of incorrect questions according to the preset status update logic, for example: The original status was "not mastered" or "needs to be consolidated", and the redo score was 100% → it was updated to "mastered" and removed from the review pool; The original status was "Not Mastered", and the redo score rate was ≥70% → it was updated to "Needs Consolidation", and it is planned to redo it again in a week; The original status was "Not Mastered", and the redo score rate was <70% → it was updated to "Still Not Mastered" and marked as "Key Focus", increasing the redo frequency; The system records the correction results and status changes for each redo, forming a visual track of how the user has mastered the incorrect questions.

[0037] In this embodiment of the invention, step S6 aims to solve the problem of cross-platform data silos and achieve efficient data management and retrieval, including: Define a general format for incorrect question data, preferably JSON-LD (linked data) format. This format specifies required fields (such as incorrect question ID, student ID, multimodal content of the question, error type, and mastery level) and optional fields (such as answer behavior data and redo log), ensuring the self-descriptiveness and interoperability of the data. It provides a standardized API interface, enabling intelligent grading systems from different sources to output incorrect question data in this common format, thus achieving cross-platform synchronization of incorrect questions; A hybrid storage scheme using relational databases (such as MySQL) and graph databases (such as Neo4j) is adopted. MySQL is used to store and quickly query the labeled structured information; Neo4j is used to store the complex relationships between "incorrect questions - knowledge points - error types", supporting deep relationship queries such as "retrieving all incorrect questions involving 'application of Pythagorean theorem' and with the error type 'formula error'".

[0038] In the description of this invention, the references to "one embodiment," "some embodiments," "in this embodiment," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0039] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention 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 present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A student mistake question automatic collection method based on an intelligent correction system, characterized in that, The method comprises the following steps: Step S1, obtaining a homework set of any student and inputting the homework set into an intelligent grading system to obtain a corresponding grading result; Step S2, judging whether the grading result meets a preset triggering condition: If yes, go to step S3; If no, return to step S1; Step S3, extracting multi-modal error question content including question content, student answer content and the grading result from the homework set; Step S4, based on a knowledge graph and an error classification system, structuring information annotation of the multi-modal error question content to obtain error question mastering results; Step S5, obtaining a rework grading result of the student on the error question and updating the error question mastering results according to the rework grading result; Step S6, standardizing and storing the updated error question mastering results, the multi-modal error question content and the structured information according to a general data format.

2. The student mistake question automatic collection method according to claim 1, characterized in that, The triggering condition in step S2 includes a full error triggering condition and a partial error triggering condition, the full error triggering condition is that the grading result of an objective question represents an error or the grading result of a subjective question is 0 points or the grading result of a subjective question represents a score rate lower than 30%, and the partial error triggering condition is that the grading result of a subjective question represents a score rate not lower than 30% but there is a specific error point marked by the intelligent grading system.

3. The student mistake question automatic collection method according to claim 1, characterized in that, In step S3, the extraction of the multi-modal error question content includes question basic information extraction, student answer content extraction and intelligent grading result extraction, the question basic information extraction is to extract text information, formula information, graph information and question meta information in the question and include them into the multi-modal error question content, the student answer content extraction is to extract text, formula, graph and answer behavior data of the student answer and include them into the multi-modal error question content, and the intelligent grading result extraction is to extract grading details and error marking information in the grading result and include them into the multi-modal error question content.

4. The student mistake question automatic collection method according to claim 3, characterized in that, In step S3, the formula of the student answer is recognized by MathOCR technology and converted into a standard LaTeX format and included into the multi-modal error question content.

5. The student mistake question automatic collection method according to claim 3, characterized in that, In step S3, the graph information in the question is extracted by computer vision technology and a graph pixel graph and a graph feature description are generated and included into the multi-modal error question content.

6. The student mistake question automatic collection method according to claim 3, characterized in that, In step S4, the annotation of the structured information includes knowledge point annotation, error type annotation, initial mastering degree annotation and personalized label annotation, the knowledge point annotation is to annotate the main knowledge point and the associated knowledge point corresponding to the error question, the error type annotation is to annotate the corresponding error type according to the error classification system, the initial mastering degree annotation is to annotate the initial mastering degree of the student according to the score rate represented by the grading result, and the personalized label annotation is to annotate the personalized label based on the answer behavior data.

7. The student mistake question automatic collection method according to claim 6, characterized in that, In step S6, a hybrid storage method of a relational database and a graph database is adopted to respectively store the structured information and the association relationship information between error questions, knowledge points and error types.

8. The method of claim 1, wherein, Step S5 comprises: Step S51, setting a redo trigger condition including a timing trigger or an active trigger, when the redo trigger condition is met, the intelligent correction system pushes the redo task to the student to obtain the redo correction result of the student on the wrong question; Step S52, according to the redo correction result, updating the wrong question mastering result according to the preset state update logic, and recording the redo correction result and state change log of each redo to form a wrong question mastering track into the wrong question mastering result.

9. The method of claim 1, wherein, In the step S6, the general data format is defined as JSON-LD format to unify the multi-modal wrong question content and the structured information.

Images