Search

CN-121542417-B - Intelligent wrong question data management method and device

CN121542417BCN 121542417 BCN121542417 BCN 121542417BCN-121542417-B

Abstract

The application relates to the technical field of computers, in particular to an intelligent wrong question data management method and device, wherein the method comprises the steps of obtaining historical wrong question data and historical answer behavior data of students, and constructing a discipline knowledge graph and a student cognitive ability image; the method comprises the steps of performing semantic coding on a wrong text through a pre-training transducer model, associating the wrong text to a corresponding knowledge point node in a subject knowledge graph based on coded vector representation, identifying the error type corresponding to the wrong text, updating a student cognitive ability portrait according to the attribute of the knowledge point node of the subject knowledge graph associated with the wrong text and the identified error type, dynamically updating the mastery weight of the associated knowledge point node in the subject knowledge graph, and optimizing the cognitive state based on the updated student cognitive ability portrait and the subject knowledge graph through LEMMA-type jeopardy training to generate a personalized learning path. The application is helpful for providing accurate and personalized study guidance for students.

Inventors

  • TAN WEIYONG
  • SUN JIAWANG
  • Li Manju

Assignees

  • 湖南达美策略信息技术服务有限公司

Dates

Publication Date
20260505
Application Date
20260116

Claims (6)

  1. 1. An intelligent wrong question data management method is characterized by comprising the following steps: Acquiring historical wrong question data and historical answer behavior data of students, and constructing a subject knowledge graph and a student cognitive ability image based on the historical wrong question data and the historical answer behavior data; acquiring wrong text in a preset latest period through a plurality of acquisition modes, and carrying out semantic coding on the wrong text through a pre-training transducer model; based on the coded vector representation, associating the wrong questions to corresponding knowledge point nodes in the subject knowledge graph, and identifying the wrong types corresponding to the wrong questions through a multi-label classification model; updating the student cognitive ability portraits according to the attribute of the knowledge point node of the subject knowledge graph associated with the wrong problem and the identified error type, and dynamically updating the mastery weight of the associated knowledge point node in the subject knowledge graph based on the updated student cognitive ability portraits; based on the updated student cognitive ability image and the subject knowledge graph, optimizing the cognitive state through LEMMA type negative training, and generating a personalized learning path; the step of associating the error questions to the corresponding knowledge point nodes in the subject knowledge graph based on the encoded vector representation, and identifying the error types corresponding to the error questions through the multi-label classification model comprises the following steps: Calculating the similarity between the wrong question vector and each knowledge point node vector in the subject knowledge graph through a cosine similarity algorithm based on the coded vector representation, and automatically associating the wrong question to the knowledge node with high similarity by combining a graph matching algorithm; calling a multi-label classification model, inputting the vector representation and the error question context information, and identifying an error type corresponding to the error question, wherein the error type comprises a single error and a compound error; The method comprises the steps of calling feature vectors of all knowledge point nodes in a subject knowledge graph in a link of association of the wrong questions and the knowledge point nodes, calculating the similarity of the wrong question vectors and the knowledge point node vectors in the subject knowledge graph through a cosine similarity algorithm, screening candidate association nodes with the similarity higher than a preset threshold value, and primarily locking a knowledge point range to which the wrong questions belong, wherein in order to further improve association accuracy, combining a graph matching algorithm, carrying out secondary screening on the candidate association nodes by utilizing logical dependency relation of the knowledge points in the subject knowledge graph, subject and subject type information of the knowledge points to which the wrong questions belong, and realizing accurate association of the wrong questions and the knowledge points, and avoiding error association problems caused by traditional keyword matching; the step of obtaining the history wrong question data and the history answering behavior data of the students and constructing the subject knowledge graph and the student cognitive ability image based on the history wrong question data and the history answering behavior data comprises the following steps: Acquiring historical wrong question data and historical answer behavior data of students; Based on the historical wrong question data, extracting a mapping relation between the questions and the knowledge points, and constructing a subject knowledge graph by adopting a graph structure modeling tool; According to the historical wrong question data and the historical answer behavior data, different cognition dimension characteristics are obtained through characteristic extraction and screening, wherein the different cognition dimension characteristics at least comprise knowledge mastering dimension characteristics, thinking ability dimension characteristics and comprehensive application dimension characteristics; Selecting an adaptive algorithm model for calculation aiming at the different cognitive dimension characteristics, converting the characteristics into a quantized result, and constructing a student cognitive ability image according to the quantized result; The knowledge mastering dimension features comprise answer accuracy, error rate and error repetition frequency of each knowledge point, the thinking capability dimension features comprise sequence patterns of similar errors, error type distribution and answer hesitation duration distribution, and the comprehensive application capability dimension features comprise score rate crossing the purpose of the knowledge bring out the theme, answer step integrity and knowledge point migration application representation; The method comprises the steps of selecting an adaptive algorithm model for calculating different cognitive dimension characteristics, converting the characteristics into quantized results, and constructing a student cognitive ability image according to the quantized results, wherein the step of selecting the adaptive algorithm model for carrying out quantized calculation for the different cognitive dimension characteristics, calculating knowledge grasping degree probability by adopting a Bayesian knowledge tracking model, analyzing thinking potential strength by using an FP-Growth sequence pattern mining algorithm, evaluating comprehensive application ability grades by using a graph neural network model, integrating quantized numerical values or grade results of conversion of each dimension characteristic, constructing a comprehensive, accurate and dynamic updated student cognitive ability image, and clearly presenting knowledge grasping state, thinking characteristics and ability level of students; Based on the updated student cognitive ability image and the subject knowledge graph, optimizing the cognitive state through LEMMA's negative training, generating the personalized learning path comprises: determining target knowledge points to be enhanced by combining the updated student cognitive ability image, and generating a customized error example containing targeted errors based on the target knowledge points, the front knowledge dependency relationship and the historical high-frequency error types in the subject knowledge graph; According to the selected thinking-back mode and the customized error example, completing error correction training through LEMMA-type thinking-back training, and correcting corresponding cognitive dimension parameters in the student cognitive ability image based on mastered knowledge points after the student successfully corrects errors, so as to realize cognitive state optimization, wherein the thinking-back mode comprises a step error correction mode and an overall error correction mode; Based on the updated student cognitive ability image and the subject knowledge graph, generating a personalized learning path through a genetic algorithm; The LEMMA-type thinking-back training is to guide a learner to accurately position errors and deep thinking-back factors and efficiently correct the errors and the deep thinking-back factors by constructing error data with targets, so that a complete learning thinking-back closed loop is formed, a learning main body can clearly grasp where and how to change the errors, the current concrete problem can be solved, and the autonomous thinking-back and error correction capability can be improved; The step error correction mode supports backtracking of students according to the problem solving step, and after the first error node is positioned, the system can give out error cause analysis and correct deduction reference of the corresponding step to help the students to comb the problem link by link; the system judges whether the error correction is effective or not by analyzing error correction duration, prompting dependent times and error correction accuracy data after the error correction training is completed by the students based on the selected mode and the customized error examples, and corrects corresponding cognitive dimension parameters in the student cognitive ability image based on the mastered knowledge points if the error correction is successful; Obtaining the wrong text in a preset latest period through a plurality of acquisition modes, and carrying out semantic coding on the wrong text through a pre-training transducer model comprises the following steps: acquiring wrong question text in a preset latest period through a plurality of acquisition modes; Loading a corpus special for the subject field, performing fine tuning training on a pre-training transducer model, and optimizing the semantic understanding capability of the transducer model on subject specific texts; Dividing the wrong text into a plurality of text fragments, inputting the text fragments into a fine-tuned pre-training transducer model, capturing semantic association in the wrong text through a multi-head attention mechanism, and generating a wrong semantic vector representation with fixed dimension; The subject field special corpus comprises subject technical terms, formula expression, typical problem description and problem solving step specification targeted content, the pre-training transformation model is subjected to fine tuning training based on the subject field special corpus, and the model parameters are optimized through multiple rounds of iteration, so that the model can accurately capture semantic features and logic association of subject texts.
  2. 2. The method of claim 1, wherein updating the student cognitive ability representation according to the attribute of the knowledge point node of the subject knowledge graph associated with the wrong problem and the identified wrong type, and dynamically updating the grasping degree weight of the associated knowledge point node in the subject knowledge graph based on the updated student cognitive ability representation comprises: Extracting knowledge point node attributes of a subject knowledge graph associated with the wrong questions, wherein the attributes comprise knowledge point difficulty coefficients, prepositioned knowledge dependency relationships and knowledge point IDs; Updating the student cognitive ability image according to a preset rule by combining the identified error types according to the subject knowledge graph knowledge point node attribute associated with the error questions; Based on the updated student cognitive ability image, the latest mastery degree weight of the corresponding knowledge point is extracted, and the mastery degree weights of the related knowledge point nodes in the subject knowledge map are synchronously adjusted.
  3. 3. The intelligent fault data management method of claim 1, further comprising: Performing hash operation on the wrong text, the coded vector representation, the subject knowledge graph knowledge point nodes related to the wrong text, the identified error type, the student cognitive ability portrait and the generated personalized learning path to generate a unique data hash value; and storing the data hash value to a blockchain distributed account book through a preset blockchain certification contract to ensure that the data cannot be tampered.
  4. 4. An intelligent mistopic data management device, comprising: The data acquisition module is used for acquiring historical wrong question data and historical answer behavior data of students and constructing a subject knowledge graph and a student cognitive ability image based on the historical wrong question data and the historical answer behavior data; The semantic coding module is used for acquiring the wrong text in a preset latest period in a plurality of acquisition modes and carrying out semantic coding on the wrong text by pre-training a transducer model; The error type identification module is used for associating the error questions to the corresponding knowledge point nodes in the subject knowledge graph based on the coded vector representation, and identifying the error types corresponding to the error questions through the multi-label classification model; the data dynamic updating module is used for updating the student cognitive ability portrait according to the knowledge point node attribute of the subject knowledge map associated with the wrong problem and the identified wrong type, and dynamically updating the mastery weight of the associated knowledge point node in the subject knowledge map based on the updated student cognitive ability portrait; the learning path generation module is used for generating a personalized learning path based on the updated student cognitive ability image and the subject knowledge graph through LEMMA type negative training to optimize the cognitive state; The error type identification module is also used for calculating the similarity between the error question vector and each knowledge point node vector in the subject knowledge graph through a cosine similarity algorithm based on the coded vector representation, and automatically associating the error question to the knowledge node with high similarity through combining a graph matching algorithm; The method comprises the steps of calling feature vectors of all knowledge point nodes in a subject knowledge graph in a link of association of the wrong questions and the knowledge point nodes, calculating the similarity of the wrong question vectors and the knowledge point node vectors in the subject knowledge graph through a cosine similarity algorithm, screening candidate association nodes with the similarity higher than a preset threshold value, and primarily locking a knowledge point range to which the wrong questions belong, wherein in order to further improve association accuracy, combining a graph matching algorithm, carrying out secondary screening on the candidate association nodes by utilizing logical dependency relation of the knowledge points in the subject knowledge graph, subject and subject type information of the knowledge points to which the wrong questions belong, and realizing accurate association of the wrong questions and the knowledge points, and avoiding error association problems caused by traditional keyword matching; The data acquisition module is also used for acquiring historical wrong question data and historical answer behavior data of students, extracting a mapping relation between questions and knowledge points based on the historical wrong question data, constructing a subject knowledge graph by adopting a graph structure modeling tool, acquiring different cognition dimension characteristics by characteristic extraction and screening according to the historical wrong question data and the historical answer behavior data, wherein the different cognition dimension characteristics at least comprise knowledge grasping dimension characteristics, thinking capability dimension characteristics and comprehensive application dimension characteristics; The knowledge mastering dimension features comprise answer accuracy, error rate and error repetition frequency of each knowledge point, the thinking capability dimension features comprise sequence patterns of similar errors, error type distribution and answer hesitation duration distribution, and the comprehensive application capability dimension features comprise score rate crossing the purpose of the knowledge bring out the theme, answer step integrity and knowledge point migration application representation; The method comprises the steps of selecting an adaptive algorithm model for calculating different cognitive dimension characteristics, converting the characteristics into quantized results, and constructing a student cognitive ability image according to the quantized results, wherein the step of selecting the adaptive algorithm model for carrying out quantized calculation for the different cognitive dimension characteristics, calculating knowledge grasping degree probability by adopting a Bayesian knowledge tracking model, analyzing thinking potential strength by using an FP-Growth sequence pattern mining algorithm, evaluating comprehensive application ability grades by using a graph neural network model, integrating quantized numerical values or grade results of conversion of each dimension characteristic, constructing a comprehensive, accurate and dynamic updated student cognitive ability image, and clearly presenting knowledge grasping state, thinking characteristics and ability level of students; the learning path generation module is also used for combining the updated student cognitive ability image, determining target knowledge points to be enhanced, generating a customized error example containing a targeted error based on the target knowledge points, the front knowledge dependency relationship and the history high-frequency error type in the subject knowledge graph, completing error correction training through LEMMA type of the anti-thinking training according to the selected anti-thinking mode and the customized error example, and correcting corresponding cognitive dimension parameters in the student cognitive ability image based on the mastered knowledge points after the student successfully corrects the errors, so as to realize cognitive state optimization, wherein the anti-thinking mode comprises a step error correction mode and an integral error correction mode; The LEMMA-type thinking-back training is to guide a learner to accurately position errors and deep thinking-back factors and efficiently correct the errors and the deep thinking-back factors by constructing error data with targets, so that a complete learning thinking-back closed loop is formed, a learning main body can clearly grasp where and how to change the errors, the current concrete problem can be solved, and the autonomous thinking-back and error correction capability can be improved; The step error correction mode supports backtracking of students according to the problem solving step, and after the first error node is positioned, the system can give out error cause analysis and correct deduction reference of the corresponding step to help the students to comb the problem link by link; the system judges whether the error correction is effective or not by analyzing error correction duration, prompting dependent times and error correction accuracy data after the error correction training is completed by the students based on the selected mode and the customized error examples, and corrects corresponding cognitive dimension parameters in the student cognitive ability image based on the mastered knowledge points if the error correction is successful; The semantic coding module is also used for acquiring wrong text in a preset latest period through various acquisition modes, loading a special corpus in the subject field, carrying out fine tuning training on a pre-trained transducer model, optimizing the semantic understanding capability of the transducer model on the subject specific text, dividing the wrong text into a plurality of text fragments, inputting the text fragments into the pre-trained transducer model after fine tuning, capturing semantic association in the wrong text through a multi-head attention mechanism, and generating a wrong semantic vector representation with fixed dimension; The subject field special corpus comprises subject technical terms, formula expression, typical problem description and problem solving step specification targeted content, the pre-training transformation model is subjected to fine tuning training based on the subject field special corpus, and the model parameters are optimized through multiple rounds of iteration, so that the model can accurately capture semantic features and logic association of subject texts.
  5. 5. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any one of claims 1 to 3 when the computer program is executed.
  6. 6. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 3.

Description

Intelligent wrong question data management method and device Technical Field The application relates to the technical field of computers, in particular to an intelligent wrong question data management method and device. Background In the current education digitization process, wrong question management is used as a key link for checking leakage and repairing defects of students and improving learning efficiency, and the prior art still has a plurality of defects, for example, a traditional wrong question management mode is mostly dependent on manual arrangement, the acquisition dimension is single, the association of wrong questions and knowledge points mostly adopts keyword matching, the semantic understanding precision is insufficient, the wrong matching of knowledge points is easy to occur, the modeling of the cognitive ability of students is static, and the generated learning path is not specific because the latest wrong question data cannot be updated in real time. Therefore, there is a need for an intelligent wrong question data management method capable of providing accurate and personalized learning guidance for students and improving learning efficiency of the students. Disclosure of Invention Based on the above, it is necessary to provide an intelligent wrong question data management method and device capable of providing accurate and personalized learning guidance for students and improving learning efficiency of students. In a first aspect, the present application provides a method for managing intelligent wrong question data, the method comprising: Acquiring historical wrong question data and historical answer behavior data of students, and constructing a subject knowledge graph and a student cognitive ability image based on the historical wrong question data and the historical answer behavior data; acquiring wrong text in a preset latest period through a plurality of acquisition modes, and carrying out semantic coding on the wrong text through a pre-training transducer model; based on the coded vector representation, associating the wrong questions to corresponding knowledge point nodes in the subject knowledge graph, and identifying the wrong types corresponding to the wrong questions through a multi-label classification model; updating the student cognitive ability portraits according to the attribute of the knowledge point node of the subject knowledge graph associated with the wrong problem and the identified error type, and dynamically updating the mastery weight of the associated knowledge point node in the subject knowledge graph based on the updated student cognitive ability portraits; based on the updated student cognitive ability image and the subject knowledge graph, optimizing the cognitive state through LEMMA type negative training, and generating a personalized learning path. In one embodiment, the generating the personalized learning path based on the updated student cognitive ability image and the subject knowledge graph through LEMMA's negative training to optimize the cognitive state comprises: determining target knowledge points to be enhanced by combining the updated student cognitive ability image, and generating a customized error example containing targeted errors based on the target knowledge points, the front knowledge dependency relationship and the historical high-frequency error types in the subject knowledge graph; According to the selected thinking-back mode and the customized error example, completing error correction training through LEMMA-type thinking-back training, and correcting corresponding cognitive dimension parameters in the student cognitive ability image based on mastered knowledge points after the student successfully corrects errors, so as to realize cognitive state optimization, wherein the thinking-back mode comprises a step error correction mode and an overall error correction mode; Based on the updated student cognitive ability image and the subject knowledge graph, a personalized learning path is generated through a genetic algorithm. In one embodiment, the associating the fault question to the corresponding knowledge point node in the subject knowledge graph based on the encoded vector representation, and identifying the fault type corresponding to the fault question through the multi-label classification model includes: Calculating the similarity between the wrong question vector and each knowledge point node vector in the subject knowledge graph through a cosine similarity algorithm based on the coded vector representation, and automatically associating the wrong question to the knowledge node with high similarity by combining a graph matching algorithm; and calling a multi-label classification model, inputting the vector representation and the error question context information, and identifying an error type corresponding to the error question, wherein the error type comprises a single error and a compound error. In one embodiment, the upda