Individualized question practice scheme generation method for intelligently integrating wrong questions

Abstract

The invention discloses a personalized question practice scheme generation method for intelligent integration of wrong questions, and relates to the technical field of personalized education, and the method comprises the steps: receiving a user wrong question record set, analyzing question stems, options and answers, and mapping knowledge points to a multi-level knowledge system tree; the occurrence frequency of the knowledge points is calculated, and statistics is carried out on error answer option distribution modes of the associated wrong questions; determining a weak knowledge node set and a typical error understanding mode corresponding to the weak knowledge node set; on the basis, matched correction exercise questions are retrieved in the question bank, difficulty adaptation adjustment is carried out, and a preliminary personalized exercise question sequence is generated; according to the method, a deep cognitive misunderstanding area is recognized by analyzing a wrong answer distribution mode, and structural weaknesses are positioned by utilizing a knowledge system tree, so that a personalized exercise scheme capable of accurately correcting wrong understanding and consolidating a knowledge system is generated.

Description

Technical Field

[0001] This invention belongs to the field of personalized education technology, specifically a method for generating personalized question practice schemes that intelligently integrates incorrect questions. Background Technology

[0002] Current personalized learning systems primarily rely on frequency statistics of the knowledge points involved in incorrect answers to generate practice exercises. The system identifies weaknesses based on a threshold for the number of incorrect answers for a single knowledge point and then extracts questions on the same knowledge point from the question bank for repeated training. This method attributes user errors to isolated knowledge point weaknesses, and its technical logic is limited to mechanical matching based on quantity.

[0003] Existing technical solutions have shortcomings. They fail to perform pattern analysis on the options used in users' incorrect answers, thus failing to distinguish the different cognitive causes behind the errors. This results in generated exercises lacking relevance and failing to effectively address users' deeper misunderstandings; they only focus on discrete knowledge points without placing them within a pre-defined multi-level knowledge system for correlation analysis. Users' weaknesses often manifest as systemic defects in knowledge chains or clusters, and existing methods cannot identify such sets of weaknesses with inherent structural connections, leading to fragmented practice questions that fail to help users build a complete and coherent knowledge structure.

[0004] A method is needed to process incorrect answers by analyzing patterns of wrong answers to identify typical misunderstandings and locating sets of weakly related knowledge nodes based on a structured knowledge system. This method should be based on the analysis of the distribution patterns of wrong answers and the structured computation of a multi-level knowledge system, achieving a shift from correcting surface errors to correcting erroneous cognitive patterns, and from training isolated knowledge points to consolidating the entire knowledge system. Summary of the Invention

[0005] This invention aims to solve at least one of the technical problems existing in the prior art; Therefore, this invention proposes a method for generating personalized practice schemes that intelligently integrates incorrect answers, comprising: Receive a set of incorrect question records submitted by the user through the terminal. The set of incorrect question records includes the user identifier, the original content of the incorrect question, the corresponding knowledge point tag, and the incorrect answer option when the answer is wrong. The set of incorrect questions is subjected to structured parsing to extract the question stem text, option text, and answer text from the original content of the incorrect questions, and the knowledge point tags are mapped to a preset multi-level knowledge system tree; Based on the multi-level knowledge system tree, the frequency of occurrence of each knowledge point tag in the wrong question record set is calculated, and the distribution pattern of wrong answer options in the wrong questions associated with each knowledge point tag is statistically analyzed. Based on the frequency of occurrence and the distribution pattern, determine the set of weak knowledge nodes reflected by the set of incorrect question records and the typical misunderstanding pattern corresponding to each weak knowledge node; Based on the set of weak knowledge nodes and the typical misunderstanding patterns, corrective practice questions that match them are retrieved from the preset question bank resource pool; The difficulty of the retrieved corrective practice questions is adjusted to generate a preliminary personalized practice question sequence.

[0006] Furthermore, the structured parsing of the incorrect question record set, extracting the question stem text, option text, and answer text from the original content of the incorrect questions, and mapping the knowledge point tags to a preset multi-level knowledge system tree, specifically involves: Read the original content of each incorrect question from the collection of incorrect question records. The original content of the incorrect question includes, but is not limited to, plain text format and rich text format containing formatting tags. The text segmentation rules in natural language processing are used to identify the question stem and answer options in the original content of the incorrect questions. The text segmentation rules are based on keywords, punctuation marks and format features. Using regular expression matching or a deep learning-based sequence labeling model, the individual option texts are separated from the answer options section, and the answer text marked as the correct answer is identified. Keyword extraction and entity recognition are performed on the extracted question stem text to generate a summary feature vector representing the topic of the question; The abstract feature vector is matched with a pre-constructed inverted index based on subject knowledge structure to supplement or correct the corresponding knowledge point tags for the question text; The multi-level knowledge system tree includes a root node, chapter nodes, and knowledge point leaf nodes, and each knowledge point leaf node has a unique path code. Use the knowledge point tags in each wrong question record as the query key, and perform matching or fuzzy matching in the multi-level knowledge system tree to locate the corresponding knowledge point leaf node or its parent chapter node. The path code of the leaf node or chapter node of the successfully matched knowledge point is used as the standardized knowledge point identifier of the wrong question record and written into the extended information of the wrong question record set.

[0007] Furthermore, based on the multi-level knowledge system tree, the frequency of occurrence of each knowledge point tag in the set of incorrect question records is calculated, and the distribution pattern of incorrect answer options in the incorrect questions associated with each knowledge point tag is statistically analyzed, specifically as follows: Traverse the set of incorrect question records and read the standardized knowledge point identifier in each incorrect question record, that is, the path code in the multi-level knowledge system tree; Count the number of times each unique path code appears in the set of incorrect question records, divide this number by the total number of incorrect question records, and obtain the relative error frequency of each knowledge point corresponding to each path code; For a subset of incorrect questions with the same path code, further analyze the incorrect answer options for each incorrect question record under it; The text content of incorrect answer options, or the standardized option identifiers, are used as the basic unit of analysis. The frequency distribution table of incorrect answer options is formed by counting the number of times each basic unit appears in the aforementioned subset of incorrect questions. Based on the question type, analyze the characteristics of the frequency distribution table, including: whether it is concentrated on a specific interference item, whether it is evenly distributed among multiple interference items, and whether there are some interference items that have never been selected; The relative error frequency and the frequency distribution characteristics of incorrect answer options are used together as quantitative indicators to describe the user's mastery of the knowledge points corresponding to the path encoding.

[0008] Furthermore, determining the set of weak knowledge nodes reflected in the set of incorrect question records and the typical misunderstanding pattern corresponding to each weak knowledge node based on the frequency of occurrence and the distribution pattern specifically involves: Set a threshold for relative error frequency, and encode the knowledge points corresponding to paths whose relative error frequency exceeds the threshold, and initially mark them as suspected weak knowledge nodes; For each suspected weak knowledge point, analyze the frequency distribution table of its corresponding incorrect answer options; If the frequency distribution table of incorrect answer options shows that the errors are highly concentrated in a specific distractor item, it is determined that the user has a typical misunderstanding pattern of the concept represented by the specific distractor item, and this pattern is marked as "concept confusion type". If the frequency distribution table of incorrect answer options shows that the errors are evenly distributed among multiple distractors and the error rate exceeds a preset threshold, it is determined that the user's overall understanding of the knowledge point is vague or there are knowledge gaps. This mode is marked as "vague understanding". If the frequency distribution table of incorrect answer options shows that the incorrect options are scattered, but the non-answer rate exceeds the preset first threshold or the average answering time is lower than the preset second threshold, then it is determined that the user has problems with problem-solving skills or time allocation, and this mode is marked as "strategy deficiency type". Based on the hierarchy and relationship of suspected weak knowledge nodes in the multi-level knowledge system tree, suspected weak knowledge nodes that are closely related and exhibit similar misunderstanding patterns are merged to form a comprehensive weak knowledge node. Each identified weak knowledge node and its corresponding typical misunderstanding pattern are grouped into a binary tuple. The set of these binary tuples is the set of weak knowledge nodes and the typical misunderstanding pattern corresponding to each weak knowledge node.

[0009] Furthermore, the step of retrieving matching corrective practice questions from a pre-defined question bank resource pool based on the set of weak knowledge nodes and the typical misunderstanding patterns specifically involves: Take a pair from the set of weak knowledge nodes, which contains the path encoding of the weak knowledge nodes and typical misunderstanding patterns; Based on the path code, search for all questions marked with the same or similar path codes in the question knowledge point index of the question bank resource pool to form an initial candidate question set; The initial candidate question set is filtered based on the typical misunderstanding patterns described above; If the typical error misunderstanding pattern is "conceptual confusion type", then prioritize questions whose stems or options explicitly involve distinguishing the confused concepts identified by the frequency distribution of the incorrect answer options, or questions with distractors that target the confused concepts. If the typical misunderstanding pattern is "vague understanding", then prioritize questions that test the basic definitions and core principles of the knowledge point, or a combination of questions that test the same knowledge point from multiple perspectives; If the typical misunderstanding pattern is "strategy deficiency", then prioritize questions that are typical in terms of problem-solving approach or require specific problem-solving skills; For the filtered questions, calculate their topic relevance scores to the weak knowledge nodes and typical misunderstanding patterns; The questions are sorted according to the topic relevance score, and the top-ranked questions are selected as corrective practice questions that match the binary pair. For each pair in the set of weak knowledge nodes, repeat the above retrieval process to finally obtain a set of corrective practice questions that match all weak knowledge nodes and misunderstanding patterns.

[0010] Furthermore, the step of adjusting the difficulty of the retrieved corrective practice questions to generate a preliminary personalized practice question sequence specifically involves: Obtain the user's historical ability level assessment data, which includes a comprehensive score of the user's accuracy and answering time in previous practice on similar knowledge points; For each question in the retrieved set of corrective practice questions, its preset baseline difficulty coefficient is read from the question bank resource pool; Establish a difficulty adjustment function. The input of the difficulty adjustment function is the user's historical ability level assessment data and the baseline difficulty coefficient of the question. The output is the adjusted expected difficulty value suitable for the user. The difficulty adjustment function follows the following logic: for users whose historical ability level assessment data is greater than or equal to the preset ability threshold, the expected difficulty value after adjustment is increased based on the baseline difficulty coefficient; for users whose historical ability level assessment data is less than the preset ability threshold, the expected difficulty value after adjustment is decreased based on the baseline difficulty coefficient. Based on the calculated adjusted expected difficulty value, the questions in the set of corrective practice questions are filtered out, and those questions whose adjusted expected difficulty value exceeds the user's current ability range are removed. The selected questions are grouped according to the weak knowledge points they are associated with; Within each group, the questions are arranged in ascending order of difficulty based on the adjusted expected difficulty value, forming a progressive structure from easy to difficult. Sort all groups according to the relative error frequency of their corresponding weak knowledge points, and arrange the question groups corresponding to the weak knowledge points with higher relative error frequency first; Connect all the sorted question groups in sequence to form a linear, preliminary personalized practice question sequence that is first ordered by the importance of the weak points and then by the increasing difficulty of the questions.

[0011] Furthermore, the method also includes: obtaining the user's historical practice performance records, and optimizing the quantity and order of the preliminary personalized practice question sequence based on the historical practice performance records, specifically: Obtain the user's historical practice performance records, which include the user's average accuracy rate, average reaction time, and the proportion of questions abandoned for different difficulties and types; By analyzing the historical practice performance records, a practice fatigue curve for the user is fitted. The practice fatigue curve describes the trend of the user's accuracy rate changing with the increase of the number of practice questions during continuous practice. Based on the practice fatigue curve, a recommended number of questions is determined to ensure that the user can maintain a level of accuracy and attention in a single effective practice session that is not lower than the preset accuracy threshold and attention level threshold. The total number of questions in the initial personalized practice question sequence is compared with the suggested question number threshold. If the total number of questions exceeds the recommended question number threshold, then starting from the end of the sequence, delete some questions in ascending order of the relative error frequency of the associated weak knowledge nodes, until the total number of questions equals the recommended question number threshold. If the total number of questions does not exceed the recommended question number threshold, the sequence remains unchanged; In the adjusted sequence, check if there are multiple consecutive questions of the same question type; If present, then based on the user's average reaction time and abandonment rate for the question types of the consecutive questions in the historical practice performance record, it is determined whether the consecutive questions meet the preset type fatigue trigger condition. If it is determined that there is type fatigue, then while ensuring the increasing order of difficulty within the question group of the same weak knowledge node, the questions of the same type should be arranged with appropriate intervals, or a small number of questions of other types should be inserted as a buffer. After adjusting the quantity and order, an optimized personalized practice question sequence is formed.

[0012] Furthermore, the method also includes: encapsulating the optimized personalized practice question sequence with corresponding weak knowledge node tags and correction target descriptions to generate an executable personalized practice plan data package, specifically: For each question in the optimized personalized practice question sequence, generate a unique practice entry identifier; Associate the following information with each exercise entry identifier: complete question content, standard answer and explanation of the question, path code of the weak knowledge node targeted by the question, and brief description of the correction target generated based on typical error comprehension patterns; The corrective goal description text is generated based on the typical error comprehension pattern template and is used to remind users of the key points of this exercise. The optimized personalized practice question sequence is organized according to a predetermined data exchange format, including the logical order of each practice item identifier and all associated information. The data exchange format defines the storage structure of the scheme header information, the list of exercise items, and the scheme metadata. Record the generation time, target user identifier, and estimated completion time of this exercise plan in the plan header information; The complete data for each exercise item is stored sequentially in the exercise item list; The solution metadata records a summary of the weak knowledge nodes covered by this solution and an overview of the main corrective objectives; The organized data is serialized into a single, executable, personalized practice program data package containing all necessary information and instructions.

[0013] Furthermore, the method also includes: pushing the personalized practice plan data package to the user terminal and initiating practice process monitoring, specifically: The executable personalized practice scheme data packet is sent to the user terminal application bound to the user identifier via a message queue or network interface. After receiving the executable personalized practice scheme data packet, the user terminal application parses the data packet and presents the practice questions in sequence on the user interface. When a user begins answering the first question, the user's terminal application sends a practice start signal to the server. The server records the practice start timestamp and starts a monitoring timer for this practice. The user terminal application records the following behavioral data about each question: selected answer options, history of answer modification, duration of time spent on each question page, and the time of final submission of the answer; The user terminal application transmits the behavioral data back to the server in real time or near real time. The server receives and stores the behavioral data, and associates it with the corresponding exercise item identifier and the standard answer to the question; The server calculates the elapsed practice time based on the current time and the practice start timestamp, and compares it with the estimated completion time in the plan header information; If the server detects that a user spends an unusually long time on a question that exceeds the historical average reaction time for that question type, or that the overall practice progress is significantly behind schedule, the server can generate a notification message to gently remind the user to pay attention to the time via the user's terminal application.

[0014] Furthermore, the method also includes: collecting the user's answers and answering times for each question in the personalized practice plan data package during practice, and updating the user's set of incorrect answer records and historical practice performance records, specifically: After the user completes or terminates all the questions in the personalized practice program data package, the server integrates the final answers to all questions collected from the user's terminal application. The final answers are compared with the standard answers for each question, and the questions with incorrect answers are marked. For questions answered incorrectly, extract the practice entry identifier and trace back the associated question content, the user's submitted incorrect answer, and the corresponding path code of the weak knowledge node; The information obtained from the backtracking is added as a new wrong question record to the user's wrong question record set, and the context identifier of this exercise is also recorded. For all questions, regardless of whether they are correct or incorrect, record the time the user spends answering them; The answer to each question, the answer time, the question type, and the expected difficulty value after the question adjustment will be recorded as a new practice performance record and updated to the user's historical practice performance record. Based on the updated set of incorrect answers and historical practice performance records, the trend of changes in the user's mastery of relevant knowledge points is recalculated. The updated set of incorrect questions becomes the input data when the next round of personalized question practice plan generation method is executed, realizing a closed loop of practice feedback and plan optimization.

[0015] Compared with the prior art, the beneficial effects of the present invention are: By statistically analyzing the distribution patterns of incorrect answer choices in incorrect questions, the system can identify typical misunderstandings of specific knowledge points by users. By quantifying the frequency and combination patterns of different incorrect options, the system distinguishes between different cognitive causes such as conceptual confusion, reasoning errors, or memory lapses. Corrective practice questions retrieved and generated based on these identification results are designed to directly target these quantified error patterns. This allows the intervention to focus on specific erroneous thought processes rather than the general notion of "unfamiliarity with knowledge points," thus avoiding ineffective repetitive training due to inaccurate attribution and improving the accuracy and efficiency of error correction training.

[0016] By mapping knowledge point tags to a pre-defined multi-level knowledge tree and performing structured parsing, the system can calculate the degree of weakness based on the hierarchy and logical connections of knowledge points. When multiple related knowledge points in the knowledge tree exhibit high-frequency errors, this method can identify their common upper-level knowledge nodes or related knowledge clusters as a holistic set of weak nodes. The generated sequence of practice questions is logically organized around these structural knowledge weaknesses. The questions may cover multiple related knowledge points under the same knowledge cluster or test the comprehensive application of knowledge points. This type of practice helps users systematically fill structural gaps in their knowledge network, establish effective connections between knowledge points, and thus consolidate the stability and coherence of their overall knowledge framework. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of the personalized question practice scheme generation method for intelligent integration of incorrect questions as described in this invention. Figure 2 A flowchart for structured parsing of the incorrect question record set; Figure 3 A flowchart for statistical analysis of knowledge point frequency and error distribution patterns; Figure 4 A comparison chart of core performance indicators for different question types; Figure 5 Heatmap of weak knowledge points and typical error patterns. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] See Figure 1 The system receives a set of incorrect question records submitted by users through their terminals. This set includes the user identifier, the original content of the incorrect question, the corresponding knowledge point tag, and the incorrect answer option when the answer was wrong. The system performs structured parsing on the incorrect question record set, extracting the question stem text, option text, and answer text from the original content of the incorrect question, and mapping the knowledge point tags to a pre-defined multi-level knowledge system tree. Based on the multi-level knowledge system tree, the system calculates the frequency of occurrence of each knowledge point tag in the incorrect question record set and statistically analyzes the distribution pattern of incorrect answer options in the incorrect questions associated with each knowledge point tag. Based on the frequency of occurrence and distribution pattern, the system determines the set of weak knowledge nodes reflected in the incorrect question record set and the typical misunderstanding patterns corresponding to each weak knowledge node. Based on the set of weak knowledge nodes and typical misunderstanding patterns, the system retrieves matching corrective practice questions from a pre-defined question bank resource pool. The retrieved corrective practice questions are adjusted for difficulty to generate a preliminary personalized practice question sequence.

[0020] In one embodiment of the present invention, see [reference] Figure 2 The system reads the original content of each incorrect question from the incorrect question record set. This original content includes plain text and rich text with formatting tags. It uses text segmentation rules from natural language processing to identify the question stem and answer options. These rules are based on keywords, punctuation, and formatting features. In some embodiments, the rules also utilize formatting features such as bolding, indentation, or line breaks to help locate the boundary between the question stem and answer options, ensuring accurate parsing. Regular expression matching or a deep learning-based sequence labeling model is used to separate individual option text from the answer options and identify the correct answer text. Regular expression matching extracts options using preset patterns such as letter numbers followed by content or specific delimiters, while the deep learning-based sequence labeling model is trained to recognize the start and end positions of options to achieve more complex text structure parsing. In practice, the system prioritizes using the deep learning-based sequence labeling model to handle rich text content with irregular formats to improve robustness.

[0021] The extracted question text undergoes keyword extraction and entity recognition to generate a summary feature vector representing the question's theme. Keyword extraction is achieved using a term frequency-inverse document frequency algorithm or a sequence labeling method based on a pre-trained language model. Entity recognition identifies specific subject-specific terms in the question stem, such as formulas, names, or concept names. The summary feature vector is obtained by mapping the extracted keywords and entities to a predefined semantic space. In essence, the summary feature vector captures the core semantic information of the question text, providing a foundation for subsequent knowledge point tag matching. The summary feature vector is then matched against a pre-constructed inverted index based on subject knowledge structures to supplement or correct the corresponding knowledge point tags in the question text. The inverted index stores the mapping relationship between knowledge point tags and related feature vectors. Similarity matching is achieved by calculating cosine similarity or Euclidean distance. The matching process returns the most relevant knowledge point tags as candidates. In practice, the system sets a similarity threshold; supplementary or corrected results are only adopted when the matching score exceeds the threshold; otherwise, the original knowledge point tags are retained.

[0022] The multi-level knowledge system tree contains root nodes, chapter nodes, and knowledge point leaf nodes. Each knowledge point leaf node has a unique path code, represented by a hierarchical string such as "subject-chapter-knowledge point". The knowledge point tag in each incorrect question record is used as the query key for matching or fuzzy matching within the multi-level knowledge system tree to locate the corresponding knowledge point leaf node or its parent chapter node. The matching process is implemented using string exact matching or edit distance algorithms. For fuzzy matching, the system allows for partial character differences to handle tag variations. In some embodiments, the matching also considers synonym expansion of knowledge point tags, improving coverage through a thesaurus. The path code of the successfully matched knowledge point leaf node or chapter node is used as a standardized knowledge point identifier for the incorrect question record and written into the extended information of the incorrect question record set. The standardized knowledge point identifier ensures that subsequent analysis is based on a unified knowledge structure. Optionally, the system records the matching confidence when writing the extended information for subsequent weight calculation.

[0023] It is understandable that the process of generating summary feature vectors involves mathematical representation, such as forming a vector representation through a weighted average of word embedding vectors. The formula is calculated from the set of word embedding vectors in the question text and is expressed as follows: in: It is the feature vector of the summary. It is the number of valid words in the question stem text after processing. It is the first The pre-trained embedding vectors of each word, It is a weighting coefficient based on word importance allocation. Calculated using term frequency-inverse document frequency values ​​or an attention mechanism, this formula ensures that the summary feature vector focuses on key semantic elements in the question stem text.

[0024] In one embodiment of the present invention, see [reference] Figure 3 The system iterates through the set of incorrect question records, reads the standardized knowledge point identifier from each record, and recognizes the path code in the multi-level knowledge tree. It counts the number of times each unique path code appears in the set of incorrect questions and divides this count by the total number of records to obtain the relative error frequency of each path code corresponding to the knowledge point. For example, in a set of 100 incorrect questions, if the path code "Mathematics - Algebra - Finding the root of a quadratic equation" appears 15 times, then the relative error frequency of the knowledge point "Mathematics - Algebra - Finding the root of a quadratic equation" is 0.15. For subsets of incorrect questions with the same path code, the system further analyzes the incorrect answer options for each record. The text content of the incorrect answer options, or the standardized option identifier, is used as the basic unit of analysis. In some embodiments, the option identifier is generated by hashing the option text or mapping it to a predefined distractor code library to ensure that incorrect options with different expressions but the same semantics can be uniformly counted. The system counts the number of times each basic unit appears in the subset of incorrect questions to form a frequency distribution table of incorrect answer options.

[0025] By analyzing the frequency distribution table of incorrect answer options in conjunction with the question types, we can identify characteristics such as whether the frequency distribution table is concentrated on a specific distractor, whether it is evenly distributed across multiple distractors, and whether some distractors are never selected. For example, for a subset of 10 multiple-choice questions with incorrect answers, the frequency distribution table might show that option "B" was selected 7 times, option "C" was selected 3 times, while options "A" and "D" were not selected. This demonstrates that the errors are concentrated on the specific distractor "B". The relative error frequency and the frequency distribution characteristics of incorrect answer options are used together as quantitative indicators to describe the user's mastery of the knowledge points corresponding to the path encoding. This quantitative indicator data provides a calculable basis for subsequent identification of weaknesses.

[0026] A threshold for relative error frequency is set. Paths with a relative error frequency exceeding this threshold are encoded as corresponding knowledge points and initially marked as potential weak knowledge nodes. For example, if the threshold is set to 0.1, then the knowledge point "Mathematics - Algebra - Finding the Root of a Quadratic Equation" with a relative error frequency of 0.15 is initially marked as a potential weak knowledge node. For each potential weak knowledge node, the frequency distribution table of its corresponding incorrect answer options is analyzed. If the table shows that errors are highly concentrated on a specific distractor, the user is determined to have a typical misunderstanding pattern regarding the concept represented by that distractor, and this pattern is marked as "Conceptual Confusion Type." If the table shows that errors are evenly distributed across multiple distractors, and the error rate exceeds a preset threshold, the user's overall understanding of the knowledge point is unclear or there are knowledge gaps, and this pattern is marked as "Unclear Understanding Type." In some embodiments, the preset threshold is dynamically calculated based on the total number of question options and the evenness of error distribution. If the frequency distribution table of incorrect answer options shows that the incorrect options are scattered, but the non-answer rate exceeds the preset first threshold or the average answering time is lower than the preset second threshold, it is determined that the user has a problem with problem-solving skills or time allocation. This mode is marked as "strategy deficiency type". The non-answer rate refers to the proportion of questions in which the user did not submit any answers. The average answering time is calculated from the user's historical behavior data.

[0027] Based on the hierarchy and relationships of suspected weak knowledge nodes within the multi-level knowledge system tree, closely related nodes exhibiting similar error patterns are merged into a single comprehensive weak knowledge node. For example, the suspected weak knowledge nodes "Mathematics - Geometry - Triangle Area Calculation" and "Mathematics - Geometry - Triangle Perimeter Calculation" both exhibit "conceptual confusion" errors and belong to the same "Mathematics - Geometry - Triangle Properties" chapter node in the multi-level knowledge system tree. Therefore, the system merges these two nodes into the comprehensive weak knowledge node "Basic Triangle Measurement Calculation". Each confirmed weak knowledge node and its corresponding typical error pattern are then grouped into a binary tuple. The set of these binary tuples constitutes the set of weak knowledge nodes and the typical error pattern for each weak knowledge node.

[0028] Optionally, when analyzing the distribution pattern of incorrect answer options, the system may introduce a concentration index. To quantify whether the frequency distribution of incorrect answer options is concentrated on a specific distractor, a concentration index is used. The calculation formula is: in: It represents the total number of all uniquely incorrect answer choices that have appeared within a subset of incorrect questions for a specific knowledge point. It is the first Frequency of occurrence of incorrect options, concentration index The closer the value is to 1, the more concentrated the errors are among a few options; concentration index The closer the value is This indicates that the error distribution is more uniform.

[0029] In one embodiment of the present invention, the system extracts a binary tuple from the set of weak knowledge nodes, which includes the path code of the weak knowledge node and the typical misunderstanding pattern. Based on the path code, the system searches for all questions labeled with the same or similar path codes in the question knowledge point index of the question bank resource pool to form an initial candidate question set. For example, the path code of the weak knowledge node is "physics-mechanics-Newton's second law", and the typical misunderstanding pattern is "conceptual confusion type" (the specific confusion point is confusing the concepts of mass and gravity). The system searches for all questions in the question bank index whose knowledge point tags contain "physics-mechanics-Newton's second law" or its child node paths to obtain an initial candidate question set containing 50 questions. Based on typical error comprehension patterns, the initial candidate question set is filtered. If the typical error comprehension pattern is "conceptual confusion," questions that explicitly involve clarifying confusing concepts identified by the frequency distribution of incorrect answer options, or questions with distractors targeting confusing concepts, are prioritized. Continuing the previous example, the system selects 10 questions from the initial 50 candidate questions whose options explicitly include a comparison of "mass" and "gravity" or are likely to cause confusion between the two. If the typical error comprehension pattern is "vague understanding," questions that test basic definitions and core principles of knowledge points, or combinations of questions that test the same knowledge point from multiple perspectives, are prioritized. If the typical error comprehension pattern is "strategy deficiency," questions that are typical in their problem-solving approach or require specific problem-solving skills are prioritized. In some embodiments, the system identifies questions with typical problem-solving approaches by using predefined "skill tags" or analyzing keywords in the question's explanation text.

[0030] For the filtered questions, calculate their topic relevance score to weak knowledge points and typical error comprehension patterns. The calculation formula is: in: It is the final topic relevance score. It is the semantic similarity score between the content of the question and the core concepts of the weak knowledge points. It is the score of the match between the characteristics of the question and the typical misunderstanding patterns. and It is a preset weighting coefficient and .

[0031] The system acquires the user's historical ability level assessment data, which includes a comprehensive score of the user's accuracy rate and response time in past practice on similar knowledge points. For example, user "U001" has a historical average accuracy rate of 70% on "Mechanics" related knowledge points, and the ratio of average response time to the system's baseline time is 1.2, with a comprehensive score of a scalar value of 65 (range 0-100). For each question in the retrieved set of corrective practice questions, the system reads its preset baseline difficulty coefficient from the question bank resource pool. The baseline difficulty coefficient is a standardized value, for example, ranging from 1 to 10. A difficulty adjustment function is established. The input of the difficulty adjustment function is the user's historical ability level assessment data and the baseline difficulty coefficient of the question, and the output is the adjusted expected difficulty value suitable for the user. In other words, the difficulty adjustment function achieves a personalized match between the difficulty of the question and the user's current ability. The difficulty adjustment function follows this logic: For users whose historical ability assessment data is greater than or equal to the preset ability threshold, the expected difficulty value after adjustment will be increased based on the baseline difficulty coefficient; for users whose historical ability assessment data is less than the preset ability threshold, the expected difficulty value after adjustment will be decreased based on the baseline difficulty coefficient. For example, if the preset ability threshold is 70 and the user's historical ability assessment data is 65, a question with a baseline difficulty coefficient of 5 may have an expected difficulty value of 4.2 after adjustment.

[0032] Based on the calculated adjusted expected difficulty value, the questions in the corrective practice question set are filtered out, removing those whose adjusted expected difficulty value exceeds the user's current acceptable range. For example, if the user's current acceptable range is set to 2.0 to 6.5, then questions with an adjusted expected difficulty value lower than 2.0 or higher than 6.5 are removed. The filtered questions are then grouped according to their associated weak knowledge nodes. Within each group, the questions are arranged in ascending order of adjusted expected difficulty value, forming a progressive structure from easy to difficult. All groups are then sorted according to the relative error frequency of their corresponding weak knowledge nodes, with question groups corresponding to weak knowledge nodes with higher relative error frequencies placed first. For example, if the relative error frequency of weak knowledge node A is 0.2 and that of node B is 0.15, then the question group corresponding to node A is placed before the question group corresponding to node B. All sorted question groups are sequentially connected to form a linear, preliminary personalized practice question sequence, first ordered by the importance of weaknesses, then by increasing question difficulty. Optionally, a lower-difficulty transition question is inserted between different question groups during connection to smooth the transition. In some embodiments, the system also calculates the total estimated completion time of the preliminary personalized practice question sequence and compares it with the user-defined available practice time. If the time limit is exceeded, the last questions are removed in descending order of adjusted expected difficulty until the time constraint is met. It can be understood that the generation logic of the preliminary personalized practice question sequence focuses on targeting and gradual progression.

[0033] In one embodiment of the present invention, the system acquires the user's historical practice performance records. These records include the user's average accuracy rate, average reaction time, and the percentage of questions abandoned for different difficulties and types. For example, the historical practice performance record for user "U002" might show an average accuracy rate of 75% and an average reaction time of 90 seconds for multiple-choice questions, with a 5% abandonment rate; and an average accuracy rate of 60% and an average reaction time of 180 seconds for calculation questions, with a 15% abandonment rate. Analyzing the historical practice performance records, the system fits a practice fatigue curve for the user. This curve describes the trend of accuracy rate as the number of practice questions increases during continuous practice. It can be understood that the practice fatigue curve reflects the decay pattern of the user's attention and cognitive resources. In some embodiments, the system uses piecewise linear regression or an exponential decay model to fit the user's historical accuracy rate sequence for continuously completing questions to obtain the practice fatigue curve. See Table 1.

[0034] Table 1: Sample Data of User's Historical Practice Performance Records Question types Average accuracy (%) Average reaction time (seconds) Percentage of respondents who gave up answering the question (%) Multiple choice questions 75 90 5 Fill in the blanks 70 120 8 Calculation problems 60 180 15 Proof question 55 240 20 Based on the practice fatigue curve, a suggested question number threshold is determined to ensure that the user maintains a level no lower than the preset accuracy threshold and attention level threshold in a single effective practice session. The preset accuracy threshold is set, for example, to 65%. The attention level threshold can be indirectly measured through reaction time stability. The system identifies the inflection point on the practice fatigue curve where the accuracy rate first consistently falls below the preset accuracy threshold, and uses the number of questions corresponding to this inflection point as a reference benchmark for the suggested question number threshold. For example, if the fitted curve shows that user "U002's" accuracy rate starts to consistently fall below 65% after completing the 18th consecutive question, then the initial suggested question number threshold can be set to 18. The total number of questions in the initial personalized practice question sequence is compared with the suggested question number threshold. If the total number of questions exceeds the suggested question number threshold, starting from the end of the sequence, questions are deleted in ascending order of relative error frequency related to weak knowledge nodes, until the total number of questions equals the suggested question number threshold. For example, if the initial personalized practice question sequence contains 25 questions and the suggested question number threshold is 20, the system identifies and deletes the 5 questions with the lowest relative error frequency related to weak knowledge nodes from the end of the sequence. If the total number of questions does not exceed the suggested question number threshold, the sequence remains unchanged.

[0035] In the adjusted sequence, check if there are multiple consecutive questions of the same type, such as more than three consecutive multiple-choice questions. If so, based on the user's average reaction time and abandonment rate for the question type in the historical practice performance record, determine if the multiple consecutive questions meet the preset type fatigue trigger condition. The type fatigue trigger condition can be that the number of consecutive questions of the same type exceeds a threshold and the abandonment rate of that type is higher than the global average. For example, if there are four consecutive multiple-choice questions and the abandonment rate (5%) is not high, the system rule sets the threshold for the number of consecutive questions of the same type to 3, so it is still determined to meet the trigger condition. If it is determined that type fatigue has occurred, while ensuring the increasing order of difficulty within the question group of the same weak knowledge node, arrange the questions of the same type with appropriate intervals, or insert a small number of other types of questions as a buffer. For example, swap the position of the fourth consecutive multiple-choice question with a fill-in-the-blank question in another weak knowledge node group.

[0036] After adjusting the quantity and order, an optimized personalized practice question sequence is formed. It can be understood that the practice fatigue curve can be characterized by a mathematical model; the mathematical formula for the practice fatigue curve is: in: Indicates that the user has completed the first step. The accuracy rate of predictions when answering questions. This indicates the user's initial accuracy level. It is the fatigue attenuation coefficient ( ), These are the sequence numbers of the completed questions. It is the asymptotic or baseline value of the accuracy, parameter , and This model is obtained by performing nonlinear regression fitting on users' historical continuous practice data. The system uses this model to predict when the accuracy drops to a preset accuracy threshold. value, and this The value serves as the primary basis for determining the threshold for the number of recommended questions. Optionally, the system will dynamically update the model parameters of the practice fatigue curve based on the user's recent practice data.

[0037] See Figure 4 This chart compares key performance indicators for different question types, clearly showing the correlation between average accuracy and average reaction time. As the complexity of the question type increases from "multiple choice" to "proof," the average accuracy rate shows a significant downward trend. Simultaneously, the average reaction time shows a significant upward trend. This pattern reflects the correlation between cognitive load and performance: the more complex the question, the longer the user needs to think, and the lower the accuracy rate. This chart directly supports the scenario you mentioned earlier regarding intelligent integration of incorrect answers and generation of practice plans. Based on the accuracy and time taken for different question types, it can generate personalized mixed-question practice sequences for users, avoiding fatigue caused by consecutive high-load question types. For question types like "proof," which have low accuracy and high time consumption, more corrective questions can be prioritized in the practice plan, and the number of consecutive occurrences can be appropriately controlled.

[0038] In one embodiment of the present invention, the system generates a unique practice item identifier for each question in the optimized personalized practice question sequence. The practice item identifier is generated using a globally unique identifier algorithm. Each practice item identifier is associated with the following information: complete question content, the standard answer and explanation of the question, the path code of the weak knowledge node targeted by the question, and a brief corrective goal description generated based on typical error misunderstanding patterns. For example, for a "conceptual confusion" weakness, the corrective goal description might be "Focus on distinguishing the difference between mass and gravity." The corrective goal description is generated based on a typical error misunderstanding pattern template, used to remind the user of the key points of this practice question. The pattern template pre-sets descriptive statement frames for different typical error misunderstanding patterns such as "conceptual confusion," "fuzzy understanding," and "strategy deficiency." The system fills the specific confused concepts or knowledge point names into the frames to generate the final text. The logical order of the optimized personalized practice question sequence, each practice item identifier, and all associated information are organized according to a predetermined data exchange format. The data exchange format defines the storage structure of the scheme header information, the practice item list, and the scheme metadata.

[0039] The header information records the generation time, target user identifier, and estimated completion time of this practice plan. The estimated completion time is obtained by summing the baseline answering time of each question in the optimized personalized practice question sequence and adding a certain percentage upwards. In the practice item list, the complete data of each practice item is stored sequentially. Each practice item is a structured data object containing a practice item identifier, question content field, answer explanation field, weak knowledge node path code field, and correction target description text field. The plan metadata records a summary of the weak knowledge nodes covered by this plan and an overview of the main correction targets. The weak knowledge node summary is a collection of all associated weak knowledge node path codes, and the main correction target overview is the extraction result of high-frequency keywords from all correction target description texts. The organized data is serialized into an independent, executable personalized practice plan data package containing all necessary information and instructions. This executable personalized practice plan data package can be understood as a self-contained data entity that can be parsed and executed by the terminal application.

[0040] The executable personalized practice plan data packet is sent to the user terminal application bound to the user identifier via a message queue or network interface. Upon receiving the data packet, the user terminal application parses it and presents the practice questions sequentially on the user interface. When the user begins answering the first question, the user terminal application sends a practice start signal to the server. The server records the practice start timestamp and starts a monitoring timer for this practice session. The user terminal application records the following behavioral data for each question: selected answer options, history of answer modifications, duration of time spent on each question page, and the final submission time. The history of answer modifications is saved as timestamps and a sequence of option changes. The user terminal application transmits the behavioral data back to the server in real-time or near real-time. In some embodiments, to save bandwidth, the behavioral data is cached locally and transmitted in batches every three questions completed. The server receives and stores the behavioral data, associating it with the corresponding practice item identifier and the standard answer to the question. A unique association index is established with the current practice plan instance when storing the behavioral data.

[0041] The server calculates the elapsed practice time based on the current time and the practice start timestamp, and compares it with the estimated completion time in the solution header information. If it finds that the user's dwell time on a certain question abnormally exceeds the historical average reaction time for that question type, or that the overall practice progress is significantly behind the estimated timeline, the server can generate a prompt message to gently remind the user of the time via the user's terminal application. For example, if the dwell time exceeds twice the historical average reaction time, the message may say, "The current question has taken a lot of time; you may consider skipping it temporarily or viewing the prompt." In some embodiments, the monitoring logic also includes detecting whether the user has been offline for an extended period or whether the application is running in the background, and pausing the monitoring timer accordingly.

[0042] After a user completes or terminates all questions in their personalized practice plan data package, the server integrates the final answers collected from the user's terminal application. It compares these final answers with the standard answers for each question, marking incorrectly answered questions. For incorrectly answered questions, the server extracts the practice entry identifier and traces back the associated question content, the user's submitted incorrect answer, and the corresponding weak knowledge node path code. This traced information is added as a new incorrect question record to the user's incorrect question record set. Simultaneously, the context identifier for this practice session is recorded; this context identifier is used to associate the original practice plan that generated the incorrect question record. For all questions, regardless of correctness, the server records the user's answering time, calculated from the moment the question page is displayed until the final answer is submitted. The answer, answering time, question type, and the adjusted expected difficulty value for each question are added as a new practice performance record to the user's historical practice performance record.

[0043] Based on the updated set of incorrect answers and historical practice performance records, the user's mastery trend in relevant knowledge points is recalculated. The calculation formula is: in: Indicates knowledge points The trend value of the degree of mastery. The current mastery score is calculated based on the latest practice performance (such as accuracy and reaction time in the most recent related practice). It is a smoothed value of the historical mastery score for this knowledge point. It is the weighting coefficient assigned to the latest performance. ).

[0044] See Figure 5 This is a heatmap showing the correlation between weak knowledge points and typical error patterns, clearly illustrating the strength of the association between different weak knowledge points and typical error patterns. The function concept shows the highest correlation (≈0.7) among "conceptual confusion" errors, indicating that users are extremely prone to confusion regarding basic function-related concepts, making it the most pressing issue to address. The root cause of errors for each weak knowledge point can be clearly identified; for example, the problem with the "function concept" lies in "conceptual confusion," not "lack of strategy." When generating practice plans, the most suitable questions can be matched to different error patterns. For instance, concept clarification questions can be prioritized for "conceptual confusion" errors, while typical problem-solving technique questions can be prioritized for "lack of strategy." Based on the correlation strength (color intensity), the two most prominent weaknesses—"function concept" and "derivative application"—can be addressed first, ensuring the efficiency of practice resource allocation.

[0045] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for generating personalized practice solutions by intelligently integrating incorrect answers, characterized in that: include: Receive a set of incorrect question records submitted by the user through the terminal. The set of incorrect question records includes the user identifier, the original content of the incorrect question, the corresponding knowledge point tag, and the incorrect answer option when the answer is wrong. The set of incorrect questions is subjected to structured parsing to extract the question stem text, option text, and answer text from the original content of the incorrect questions, and the knowledge point tags are mapped to a preset multi-level knowledge system tree; Based on the multi-level knowledge system tree, the frequency of occurrence of each knowledge point tag in the wrong question record set is calculated, and the distribution pattern of wrong answer options in the wrong questions associated with each knowledge point tag is statistically analyzed. Based on the frequency of occurrence and the distribution pattern, determine the set of weak knowledge nodes reflected by the set of incorrect question records and the typical misunderstanding pattern corresponding to each weak knowledge node; Based on the set of weak knowledge nodes and the typical misunderstanding patterns, corrective practice questions that match them are retrieved from the preset question bank resource pool; The difficulty of the retrieved corrective practice questions is adjusted to generate a preliminary personalized practice question sequence.

2. The method for generating a personalized practice scheme by intelligently integrating incorrect questions according to claim 1, characterized in that, The process involves performing structured parsing on the set of incorrect question records, extracting the question stem text, option text, and answer text from the original content of the incorrect questions, and mapping the knowledge point tags to a preset multi-level knowledge system tree. Specifically: Read the original content of each incorrect question from the collection of incorrect question records. The original content of the incorrect question includes, but is not limited to, plain text format and rich text format containing formatting tags. The text segmentation rules in natural language processing are used to identify the question stem and answer options in the original content of the incorrect questions. The text segmentation rules are based on keywords, punctuation marks and format features. Using regular expression matching or a deep learning-based sequence labeling model, the individual option texts are separated from the answer options section, and the answer text marked as the correct answer is identified. Keyword extraction and entity recognition are performed on the extracted question stem text to generate a summary feature vector representing the topic of the question; The abstract feature vector is matched with a pre-constructed inverted index based on subject knowledge structure to supplement or correct the corresponding knowledge point tags for the question text; The multi-level knowledge system tree includes a root node, chapter nodes, and knowledge point leaf nodes, and each knowledge point leaf node has a unique path code. Use the knowledge point tags in each wrong question record as the query key, and perform matching or fuzzy matching in the multi-level knowledge system tree to locate the corresponding knowledge point leaf node or its parent chapter node. The path code of the leaf node or chapter node of the successfully matched knowledge point is used as the standardized knowledge point identifier of the wrong question record and written into the extended information of the wrong question record set.

3. The method for generating a personalized practice scheme by intelligently integrating incorrect questions according to claim 1, characterized in that, Based on the multi-level knowledge system tree, the frequency of occurrence of each knowledge point tag in the set of incorrect question records is calculated, and the distribution pattern of incorrect answer options in the incorrect questions associated with each knowledge point tag is statistically analyzed, specifically as follows: Traverse the set of incorrect question records and read the standardized knowledge point identifier in each incorrect question record, that is, the path code in the multi-level knowledge system tree; Count the number of times each unique path code appears in the set of incorrect question records, divide this number by the total number of incorrect question records, and obtain the relative error frequency of each knowledge point corresponding to each path code; For a subset of incorrect questions with the same path code, further analyze the incorrect answer options for each incorrect question record under it; The text content of incorrect answer options, or the standardized option identifiers, are used as the basic unit of analysis. The frequency distribution table of incorrect answer options is formed by counting the number of times each basic unit appears in the aforementioned subset of incorrect questions. Based on the question type, analyze the characteristics of the frequency distribution table, including: whether it is concentrated on a specific interference item, whether it is evenly distributed among multiple interference items, and whether there are some interference items that have never been selected; The relative error frequency and the frequency distribution characteristics of incorrect answer options are used together as quantitative indicators to describe the user's mastery of the knowledge points corresponding to the path encoding.

4. The method for generating a personalized practice scheme by intelligently integrating incorrect questions according to claim 1, characterized in that, The step of determining the set of weak knowledge nodes reflected in the set of incorrect question records and the typical misunderstanding pattern corresponding to each weak knowledge node based on the frequency of occurrence and the distribution pattern is as follows: Set a threshold for relative error frequency, and encode the knowledge points corresponding to paths whose relative error frequency exceeds the threshold, and initially mark them as suspected weak knowledge nodes; For each suspected weak knowledge point, analyze the frequency distribution table of its corresponding incorrect answer options; If the frequency distribution table of incorrect answer options shows that errors are highly concentrated in a specific distractor item, it is determined that the user has a typical misunderstanding pattern of the concept represented by the specific distractor item, and this pattern is marked as "concept confusion type". If the frequency distribution table of incorrect answer options shows that the errors are evenly distributed among multiple distractors and the error rate exceeds a preset threshold, it is determined that the user's overall understanding of the knowledge point is vague or there are knowledge gaps. This mode is marked as "vague understanding". If the frequency distribution table of incorrect answer options shows that the incorrect options are scattered, but the non-answer rate exceeds the preset first threshold or the average answering time is lower than the preset second threshold, then it is determined that the user has problems with problem-solving skills or time allocation, and this mode is marked as "strategy deficiency type". Based on the hierarchy and relationship of suspected weak knowledge nodes in the multi-level knowledge system tree, suspected weak knowledge nodes that are closely related and exhibit similar misunderstanding patterns are merged to form a comprehensive weak knowledge node. Each identified weak knowledge node and its corresponding typical misunderstanding pattern are grouped into a binary tuple. The set of these binary tuples is the set of weak knowledge nodes and the typical misunderstanding pattern corresponding to each weak knowledge node.

5. The method for generating a personalized practice scheme by intelligently integrating incorrect questions according to claim 1, characterized in that, The step of retrieving matching corrective practice questions from a pre-set question bank resource pool based on the set of weak knowledge points and the typical misunderstanding patterns specifically involves: Take a pair from the set of weak knowledge nodes, which contains the path encoding of the weak knowledge nodes and typical misunderstanding patterns; Based on the path code, search for all questions marked with the same or similar path codes in the question knowledge point index of the question bank resource pool to form an initial candidate question set; The initial candidate question set is filtered based on the typical misunderstanding patterns described above; If the typical error misunderstanding pattern is "conceptual confusion type", then prioritize questions whose stems or options explicitly involve distinguishing the confused concepts identified by the frequency distribution of the incorrect answer options, or questions with distractors that target the confused concepts. If the typical misunderstanding pattern is "vague understanding", then prioritize questions that test the basic definitions and core principles of the knowledge point, or a combination of questions that test the same knowledge point from multiple perspectives; If the typical misunderstanding pattern is "strategy deficiency", then prioritize questions that are typical in terms of problem-solving approach or require specific problem-solving skills; For the filtered questions, calculate their topic relevance scores to the weak knowledge nodes and typical misunderstanding patterns; The questions are sorted according to the topic relevance score, and the top-ranked questions are selected as corrective practice questions that match the binary pair. For each pair in the set of weak knowledge nodes, repeat the above retrieval process to finally obtain a set of corrective practice questions that match all weak knowledge nodes and misunderstanding patterns.

6. The method for generating a personalized practice scheme by intelligently integrating incorrect questions according to claim 1, characterized in that, The process of adjusting the difficulty of the retrieved corrective practice questions to generate a preliminary personalized practice question sequence is as follows: Obtain the user's historical ability level assessment data, which includes a comprehensive score of the user's accuracy and answering time in previous practice on similar knowledge points; For each question in the retrieved set of corrective practice questions, its preset baseline difficulty coefficient is read from the question bank resource pool; Establish a difficulty adjustment function. The input of the difficulty adjustment function is the user's historical ability level assessment data and the baseline difficulty coefficient of the question. The output is the adjusted expected difficulty value suitable for the user. The difficulty adjustment function follows the following logic: for users whose historical ability level assessment data is greater than or equal to the preset ability threshold, the expected difficulty value after adjustment is increased based on the baseline difficulty coefficient; for users whose historical ability level assessment data is less than the preset ability threshold, the expected difficulty value after adjustment is decreased based on the baseline difficulty coefficient. Based on the calculated adjusted expected difficulty value, the questions in the set of corrective practice questions are filtered out, and those questions whose adjusted expected difficulty value exceeds the user's current ability range are removed. The selected questions are grouped according to the weak knowledge points they are associated with; Within each group, the questions are arranged in ascending order of difficulty based on the adjusted expected difficulty value, forming a progressive structure from easy to difficult. Sort all groups according to the relative error frequency of their corresponding weak knowledge points, and arrange the question groups corresponding to the weak knowledge points with higher relative error frequency first; Connect all the sorted question groups in sequence to form a linear, preliminary personalized practice question sequence that is first ordered by the importance of the weak points and then by the increasing difficulty of the questions.

7. The method for generating a personalized practice scheme by intelligently integrating incorrect questions according to claim 1, characterized in that, The method further includes: obtaining the user's historical practice performance records, and optimizing the quantity and order of the initial personalized practice question sequence based on the historical practice performance records, specifically: Obtain the user's historical practice performance records, which include the user's average accuracy rate, average reaction time, and the proportion of questions abandoned for different difficulties and types; By analyzing the historical practice performance records, a practice fatigue curve for the user is fitted. The practice fatigue curve describes the trend of the user's accuracy rate changing with the increase of the number of practice questions during continuous practice. Based on the practice fatigue curve, a recommended number of questions is determined to ensure that the user can maintain a level of accuracy and attention in a single effective practice session that is not lower than the preset accuracy threshold and attention level threshold. The total number of questions in the initial personalized practice question sequence is compared with the suggested question number threshold. If the total number of questions exceeds the recommended question number threshold, then starting from the end of the sequence, delete some questions in ascending order of the relative error frequency of the associated weak knowledge nodes, until the total number of questions equals the recommended question number threshold. If the total number of questions does not exceed the recommended question number threshold, the sequence remains unchanged; In the adjusted sequence, check if there are multiple consecutive questions of the same question type; If present, then based on the user's average reaction time and abandonment rate for the question types of the consecutive questions in the historical practice performance record, it is determined whether the consecutive questions meet the preset type fatigue trigger condition. If it is determined that there is type fatigue, then while ensuring the increasing order of difficulty within the question group of the same weak knowledge node, the questions of the same type should be arranged with appropriate intervals, or a small number of questions of other types should be inserted as a buffer. After adjusting the quantity and order, an optimized personalized practice question sequence is formed.

8. The method for generating a personalized practice scheme by intelligently integrating incorrect questions according to claim 1, characterized in that, The method further includes: encapsulating the optimized personalized practice question sequence with corresponding weak knowledge node tags and correction target descriptions to generate an executable personalized practice plan data package, specifically: For each question in the optimized personalized practice question sequence, generate a unique practice entry identifier; Associate the following information with each exercise entry identifier: complete question content, standard answer and explanation of the question, path code of the weak knowledge node targeted by the question, and brief description of the correction target generated based on typical error comprehension patterns; The corrective goal description text is generated based on the typical error comprehension pattern template and is used to remind users of the key points of this exercise. The optimized personalized practice question sequence is organized according to a predetermined data exchange format, including the logical order of each practice item identifier and all associated information. The data exchange format defines the storage structure of the scheme header information, the list of exercise items, and the scheme metadata. Record the generation time, target user identifier, and estimated completion time of this exercise plan in the plan header information; The complete data for each exercise item is stored sequentially in the exercise item list; The solution metadata records a summary of the weak knowledge nodes covered by this solution and an overview of the main corrective objectives; The organized data is serialized into a single, executable, personalized practice program data package containing all necessary information and instructions.

9. The method for generating a personalized practice scheme by intelligently integrating incorrect questions according to claim 1, characterized in that, The method further includes: pushing the personalized practice plan data package to the user terminal and starting practice process monitoring, specifically: The executable personalized practice scheme data packet is sent to the user terminal application bound to the user identifier via a message queue or network interface. After receiving the executable personalized practice scheme data packet, the user terminal application parses the data packet and presents the practice questions in sequence on the user interface. When a user begins answering the first question, the user's terminal application sends a practice start signal to the server. The server records the practice start timestamp and starts a monitoring timer for this practice. The user terminal application records the following behavioral data about each question: selected answer options, history of answer modification, duration of time spent on each question page, and the time of final submission of the answer; The user terminal application transmits the behavioral data back to the server in real time or near real time. The server receives and stores the behavioral data, and associates it with the corresponding exercise item identifier and the standard answer to the question; The server calculates the elapsed practice time based on the current time and the practice start timestamp, and compares it with the estimated completion time in the plan header information; If the server detects that a user spends an unusually long time on a question that exceeds the historical average reaction time for that question type, or that the overall practice progress is significantly behind schedule, the server can generate a notification message to gently remind the user to pay attention to the time via the user's terminal application.

10. The method for generating a personalized practice scheme by intelligently integrating incorrect questions according to claim 1, characterized in that, The method further includes: collecting the user's answers and answering time for each question in the personalized practice plan data package during practice, and updating the user's set of incorrect answer records and historical practice performance records, specifically: After the user completes or terminates all the questions in the personalized practice program data package, the server integrates the final answers to all questions collected from the user's terminal application. The final answers are compared with the standard answers for each question, and the questions with incorrect answers are marked. For questions answered incorrectly, extract the practice entry identifier and trace back the associated question content, the user's submitted incorrect answer, and the corresponding path code of the weak knowledge node; The information obtained from the backtracking is added as a new wrong question record to the user's wrong question record set, and the context identifier of this exercise is also recorded. For all questions, regardless of whether they are correct or incorrect, record the time the user spends answering them; The answer to each question, the answer time, the question type, and the expected difficulty value after the question adjustment will be recorded as a new practice performance record and updated to the user's historical practice performance record. Based on the updated set of incorrect answers and historical practice performance records, the trend of changes in the user's mastery of relevant knowledge points is recalculated. The updated set of incorrect questions becomes the input data when the next round of personalized question practice plan generation method is executed, realizing a closed loop of practice feedback and plan optimization.

Images