A group paper writing method and device based on multi-round answering and space constraints, and a medium

Abstract

This application discloses a test paper generation method, device, and medium based on multi-round answering and spatial constraints, relating to the field of educational informatization technology. The method includes: responding to the answering data from the first round, determining the correctness of the first round's answering data, and updating the accuracy rate of the current round based on the determination result; selecting the difficulty level of the second round from a preset difficulty quota profile based on the comparison result of the accuracy rate and a preset accuracy rate threshold; obtaining the corresponding test point identifier set from the test point mapping relationship based on the score improvement space identifier corresponding to the second round; obtaining the set of question identifiers that the current user has answered in the current course from the user's historical answering records, as a global exclusion constraint; and performing graded question selection based on the difficulty level, the test point identifier set, and the global exclusion constraint. This application achieves adaptive test paper generation based on accuracy rate, under score improvement space constraints, and with balanced test point coverage through the above method.

Description

Technical Field

[0001] This application relates to the field of educational informatization technology, and in particular to a test paper generation method, device and medium based on multi-round response and spatial constraints. Background Technology

[0002] Current online assessment and adaptive learning systems mostly rely on single-session fixed test papers or simple random question selection, making it difficult to achieve feedback loops between multiple rounds of testing. Existing solutions suffer from the following shortcomings: There is a lack of test paper generation strategies strongly tied to the overall performance of the previous round, and the difficulty of subsequent rounds cannot be automatically adjusted according to the learner's state; the knowledge organization dimension and the question selection dimension are inconsistent, and random question selection based on only a single dimension can easily lead to a misalignment between the assessment coverage and teaching focus; the difficulty structure lacks an interpretable switching mechanism, often employing a black-box model or simple grade adjustments, resulting in an opaque difficulty trajectory; there is a lack of historical question exclusion mechanisms when the same user takes repeated tests in the same course, weakening assessment validity; and simple random question selection cannot simultaneously meet the requirements of balanced coverage of test points and achieving the target number of questions.

[0003] Based on the above analysis, the problems and shortcomings of the existing technology are as follows: Existing online assessment and adaptive learning systems cannot automatically generate a second round of test papers based on the accuracy of the first round of answers, under the constraint of score improvement space, while ensuring balanced coverage of test points, no repetition of questions between and within rounds, idempotent security of the test paper generation process, and support consistency in the judgment of compound questions. Summary of the Invention

[0004] This application provides a method, device, and medium for generating test papers based on multi-round responses and spatial constraints. It can solve the problems in existing online assessment and adaptive learning systems that cannot automatically generate second-round test papers based on the accuracy of the first round of responses and under the constraint of score improvement space, while ensuring balanced coverage of test points, no repetition of questions between and within rounds, idempotent security of the test paper generation process, and support for consistency in the judgment of compound questions.

[0005] In a first aspect, embodiments of this application provide a test paper generation method based on multi-round answering and spatial constraints. The method includes: responding to the answering data of the first round, determining the correctness of the answering data of the first round, and updating the accuracy rate of the current round based on the determination result; selecting the difficulty level of the second round from a preset difficulty quota profile according to the comparison result of the accuracy rate and a preset accuracy rate threshold, wherein the difficulty level includes multiple difficulty levels arranged in sequence and the corresponding target number of questions; obtaining the corresponding test point identifier set in the test point mapping relationship based on the score improvement space identifier corresponding to the second round; obtaining the set of question identifiers that the current user has answered in the current course from the user's historical answering records, as a global exclusion constraint; and performing graded question extraction according to the difficulty level, the test point identifier set, and the global exclusion constraint.

[0006] In one implementation of this application, questions are selected in tiers based on difficulty level, test point identifier set, and global exclusion constraint. Specifically, this includes: adding the questions already selected in each difficulty level to the inter-level cumulative exclusion set according to the order of difficulty levels; removing the global exclusion constraint and the inter-level cumulative exclusion set to obtain the current available pool; and applying pseudo-random perturbation to the test point identifier set after order preservation and deduplication to obtain the perturbed test point sequence.

[0007] In one implementation of this application, the method further includes: traversing the perturbation test point sequence, extracting questions belonging to the test point and meeting the current difficulty level from the currently available pool; if the number of questions does not reach the target number after traversal, randomly selecting questions with replacement from the currently available pool to fill the gaps by extracting questions that meet the current difficulty level, until the target number of questions is reached or a shutdown condition is triggered; the shutdown condition is that the number of consecutive failed extractions reaches a first threshold or the total number of extraction attempts reaches a second threshold.

[0008] In one implementation of this application, the method further includes: when the extracted question is a compound question, determining the correctness of the parent question based on the correctness of all sub-questions; and using the correctness of the parent question for accuracy calculation.

[0009] In one implementation of this application, the perturbation test point sequence is traversed, and questions belonging to the test point and meeting the current difficulty level are extracted from the currently available pool. Specifically, this includes: obtaining all candidate questions that meet the difficulty level under the current test point; obtaining the historical usage frequency of each candidate question in the question bank; and selecting the question with the lowest historical usage frequency.

[0010] In one implementation of this application, if the number of questions after traversal does not reach the target number, random selection with replacement is performed in the currently available pool. Specifically, this includes: defining the process of traversing the perturbed test point sequence as the first stage, and defining the process of completing the question selection as the second stage; initializing the weight of each test point to 1, and recording the number of times each test point selects a question in the first stage; before each random selection in the second stage, multiplying the weight of the test points with 0 selections in the first stage by a first coefficient, and dividing the weight of the test points with 1 or more successful selections in the first stage by a second coefficient; and performing probabilistic random selection based on the normalized weights of the test points.

[0011] In one implementation of this application, the method further includes: checking whether there are any two questions among the extracted questions that belong to different sub-questions of the same compound question; if so, removing one of the sub-questions; and re-extracting a replacement question from the remaining candidate questions at the same difficulty level and the target number of questions, based on the difficulty level of the removed question and the target number of questions.

[0012] In one implementation of this application, the method further includes: querying questions that have been generated but not answered in the current round; merging the generated but unanswered questions with the questions that the user has answered in the past to form a global exclusion constraint.

[0013] Secondly, embodiments of this application also provide a test paper generation device based on multi-round response and spatial constraints. The device includes at least one processor and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to: perform any step of a test paper generation method based on multi-round response and spatial constraints.

[0014] Thirdly, embodiments of this application also provide a non-volatile computer storage medium for test paper assembly based on multi-round response and spatial constraints, which stores computer-executable instructions, configured to execute any one of the steps of a test paper assembly method based on multi-round response and spatial constraints.

[0015] This application provides a test paper generation method, device, and medium based on multi-round answering and spatial constraints. By comparing the accuracy rate of the first round with a preset threshold, the difficulty level order and target number of questions in the second round are determined within a limited difficulty quota profile, achieving explicit and interpretable switching of the difficulty trajectory and avoiding the opacity of a black box model. By introducing a mapping relationship between the score improvement space and test points, the second round of test paper generation is precisely aligned with the teaching ability advancement space. By constructing global exclusion constraints based on the user's historical answer records, it is ensured that the same user does not repeat questions across rounds in the same course, effectively improving the assessment. Validity assessment: By using a cumulative exclusion set between difficulty levels in the tiered question selection process, mutual exclusion of questions from different difficulty levels within the same round is ensured, further enhancing question diversity. The use of a test point traversal strategy that preserves order and removes duplicates, along with pseudo-random perturbation and a two-stage question selection mechanism, ensures stable achievement of the target number of questions and maintains uniform test point coverage even when test points are sparse or exclusion constraints are strong. Simultaneously, a halting condition is set to control solution complexity and avoid performance degradation caused by invalid searches. Furthermore, the introduction of idempotency detection and compound question judgment rules ensures data consistency and scoring accuracy in retry scenarios. These mechanisms collectively achieve a technical upgrade from single-assessment to a multi-round feedback loop, providing highly robust and highly interpretable intelligent components for online education systems. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a test paper generation method based on multi-round response and spatial constraints, provided for an embodiment of this application; Figure 2 A timing diagram illustrating multi-round answer saving and test paper generation triggering based on multi-round answering and spatial constraints is provided for embodiments of this application. Figure 3 This is a schematic diagram of the internal structure of a test paper generation device based on multi-round response and spatial constraints, provided as an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] This application provides a method, device, and medium for generating test papers based on multi-round answering and spatial constraints. It solves the problems in existing online assessment and adaptive learning systems that cannot automatically generate second-round test papers based on the accuracy of the first round of answering and under the constraint of score improvement space, while ensuring balanced coverage of test points, no repetition of questions between and within rounds, idempotent security of the test paper generation process, and support for consistency in the judgment of compound questions.

[0019] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0020] Figure 1 This document provides a flowchart of a test paper generation method based on multi-round response and spatial constraints, as an embodiment of this application. Figure 1 and Figure 2 As shown in the figure, the present application provides a test paper generation method based on multi-round response and spatial constraints, which specifically includes the following steps: Step 10: In response to the answer data from the first round, determine the correctness of the answer data from the first round, and update the accuracy rate of the current round based on the determination result.

[0021] In this application, the server adopts a layered architecture, including an interaction and transaction layer: receiving structured answer payloads submitted by the client, defining the current round, and ensuring the atomicity and recoverability of scoring, statistics, and test paper assembly within the same database transaction; a cognitive state layer: aggregating batch-level statistics from the test question detail storage unit, the statistics including at least the number of correct answers, the total number of questions, and the accuracy rate, forming the decision input vector for the next round of test paper assembly; a constraint and strategy layer: superimposing constraints such as the score improvement space-test point mapping, global exclusion of user-course history questions, and cumulative exclusion by grade within round, and attaching a difficulty quota strategy library driven by the accuracy rate threshold; and a test paper assembly and solution layer: executing an algorithmic pipeline of test point-level balanced coverage, controlled random completion, and sequential extraction by difficulty grade on the discrete question bank, outputting an ordered question sequence and mapping it to a transmission object, and writing it back to the test question detail storage unit and the question-test point association storage unit.

[0022] In this step, the server receives the answer data submitted by the client for the first round of the current test batch. This data includes at least: user identifier, course identifier, batch identifier, round identifier, question identifier for each question, and the user's answer content. After receiving the above data, the server first determines the correctness of the answer data. For regular question types, the server compares the user's answer content with the standard answer pre-stored in the question bank's question storage unit. If they match, the answer is considered correct; otherwise, it is considered incorrect. For compound questions, the server first obtains the question identifiers of all sub-questions under the compound question and determines the correctness of each sub-question according to the regular question type rules. The parent question of the compound question is determined to be correct only if all sub-questions are determined to be correct; otherwise, it is considered incorrect.

[0023] After determining the correctness of each question, the determination result is written to the test question detail storage unit. Each record must contain at least: batch primary key, question identifier, round identifier, and correctness identifier. If the question is a compound question, the determination result of each sub-question is written to the compound question sub-question detail storage unit, and the compound question mark is marked in the parent question record.

[0024] Subsequently, based on all the judgment results of the current first round, the total number of questions judged in the first round of this batch and the number of questions judged as correct are counted, the first round accuracy rate is calculated, and written back to the corresponding accuracy rate field in the main storage unit of the test batch.

[0025] Meanwhile, the server merges the correctness determination result of this answer into the user's historical answer storage unit. Each record includes at least: user ID, course ID, question ID, and whether it is correct. If the user already has a historical record for the same question in this course, the correctness status is updated according to the current answer timestamp; otherwise, a new record is inserted.

[0026] The operations described above, such as writing to the test question details storage unit, updating the accuracy rate of the test batch main storage unit, and merging and writing to the user's historical answer storage unit, are all completed within the same database transaction to ensure the consistency between detailed data and summary statistics.

[0027] Step 20: Based on the comparison between the accuracy rate and the preset accuracy rate threshold, select the difficulty level for the second round from the preset difficulty quota profile. The difficulty level includes multiple difficulty levels arranged in order and the corresponding target number of questions.

[0028] In this step, upon receiving the second round of test paper assembly instructions, the server first checks the test question details storage unit to see if a second round of question records already exists, based on the batch primary key and round identifier (second round) of the current test batch. If the query result is not empty, the existing record is directly read and returned as the test paper assembly result, without executing subsequent steps such as accuracy comparison, difficulty level selection, and graded question selection; if the record does not exist, the subsequent process continues.

[0029] The accuracy threshold can be pre-configured as a single fixed value or as multiple different values ​​to support multi-level switching. Based on the comparison results, one of the preset difficulty quota profiles is selected as the difficulty configuration for the second round of test paper generation. Different difficulty quota profiles represent different difficulty strategies: for example, when the accuracy in the first round is higher than the threshold, a profile with a higher proportion of low difficulty levels is selected to encourage learners; when the accuracy in the first round is lower than or equal to the threshold, a profile with a higher proportion of medium and high difficulty levels is selected to strengthen reinforcement training.

[0030] Step 30: Based on the score improvement space identifier corresponding to the second round, obtain the corresponding set of test point identifiers in the test point mapping relationship.

[0031] In this step, based on the score improvement space identifier corresponding to the second round, the set of test point identifiers associated with that score improvement space is queried in the test point mapping relationship. The mapping relationship between score improvement spaces and test points is pre-stored in the configuration table on the server. Each score improvement space can be associated with one or more test points, and each test point can belong to one or more score improvement spaces, allowing the same knowledge point to have different examination weights or cognitive levels in different score improvement spaces. Using the score improvement space identifier corresponding to the second round as input, the test point mapping relationship is traversed to collect all test point identifiers associated with that score improvement space. After deduplication, the first set of test points is obtained, which limits the examination scope of the second round of test paper generation, ensuring that question extraction is only carried out within the specified ability advancement space.

[0032] Step 40: In the user's historical answer records, obtain the set of question identifiers that the current user has answered in the current course, and use it as a global exclusion constraint.

[0033] In this step, based on the user's identifier and the course identifier of the current course, the system queries the user's historical answer storage unit for all question identifiers that the user has answered in this course. This is not limited to the current test batch but covers all historical test records of the user in this course. The retrieved set of question identifiers is used as a global exclusion constraint in the subsequent question selection stage: all questions that the user has seen or answered in this course are removed from the question bank to ensure that previously seen questions are not repeatedly selected in the second round of test preparation, thereby improving the validity of the assessment; the system also checks whether there are questions generated in other rounds that the user has not yet answered in the current test batch, such as questions generated in the first round but not yet submitted by the user. If so, these question identifiers are added to the global exclusion constraint to prevent unanswered questions in the same batch from being repeatedly selected in subsequent rounds. The above exclusion operation is implemented through the "question identifier NOT IN (exclusion set)" condition in the database query or an equivalent method.

[0034] Step 50: Based on the difficulty level, the set of test point identifiers, and global exclusion constraints, conduct tiered question selection.

[0035] As an optional implementation, questions are selected in tiers based on difficulty level, test point identifier set, and global exclusion constraints. Specifically, step 501: according to the order of difficulty levels, the questions already selected in each difficulty level are added to the inter-level cumulative exclusion set.

[0036] In this step, for the difficulty level currently being processed, the server first obtains all the question identifiers that have been extracted before this level and adds these question identifiers to the inter-level cumulative exclusion set to prevent duplicate questions from appearing between different difficulty levels in the same round; for the first difficulty level, the inter-level cumulative exclusion set is initially empty.

[0037] Step 502: Exclude global exclusion constraints and inter-level cumulative exclusion sets to obtain the currently available pool.

[0038] In this step, all questions are retrieved from the question bank's question storage unit, while applying two layers of exclusion constraints: the first layer is the global exclusion constraint constructed in step 40 (i.e., all question identifiers that the current user has already answered in the current course), and the second layer is the inter-level cumulative exclusion set updated in step 501 (i.e., the question identifiers that have been extracted from previous levels in the current round). Questions that meet the following conditions are retained to form the current available pool: the question identifier is not in the global exclusion constraint and is not in the inter-level cumulative exclusion set. This current available pool serves as the source data set for extracting questions for this difficulty level.

[0039] Step 503: After preserving the order and removing duplicates from the test point identifier set, apply pseudo-random perturbation to obtain the perturbed test point sequence.

[0040] In this step, a deduplication operation is performed on the test point set while preserving order: the original test point list is traversed, and the unique record of each test point is retained in the order of its first appearance, while duplicate test point identifiers are removed. Then, a pseudo-random perturbation is applied to the deduplicated test point sequence, that is, the order of the test points is shuffled using a preset random seed, resulting in a perturbed test point sequence. This perturbation operation ensures that the order in which test points are traversed is random each time a test paper is generated, thus achieving a statistically balanced test point coverage, while the fixed random seed ensures the reproducibility of the test paper generation results under the same input.

[0041] As an optional embodiment, the method may further include: step 504: traversing the perturbation test point sequence and extracting questions belonging to the test point and meeting the current difficulty level from the currently available pool; As an optional implementation, the perturbation test point sequence is traversed, and questions belonging to the test point and meeting the current difficulty level are extracted from the currently available pool. Specifically, this may include: Step 5041: Obtain all candidate questions that meet the difficulty level under the current test point.

[0042] In this step, based on the currently traversed test point identifier, all questions that meet the following conditions are selected from the currently available pool: the question has a mapping relationship with the current test point identifier in the question-test point association storage unit, and the difficulty scalar value of the question belongs to the difficulty range defined by the current difficulty level; the selection results are used as a candidate question set.

[0043] Step 5042: Obtain the historical usage frequency of each candidate question in the question bank.

[0044] In this step, for each question in the candidate question set obtained in step 5041, the historical usage frequency field of the question in the question bank question storage unit is queried. The statistics are based on the number of times the question is called by all users across all test batches. Each question is extracted and used to assemble a test paper once, and the corresponding frequency field value is incremented by one. This is independent of the current user's personal answer history and reflects the overall exposure of the question in the entire question bank system.

[0045] Step 5043: Select the question with the lowest historical usage frequency.

[0046] In this step, the historical usage frequency of each question in the candidate question set is compared, and the question with the lowest frequency is selected as the extraction result for the current test point. If multiple questions have the same lowest historical usage frequency, one of these questions is randomly selected. After selection, the historical usage frequency of the selected question is incremented by one to update the global statistics, and the question used in this extraction is recorded.

[0047] By following the steps above, while meeting the constraints of test points and difficulty, priority is given to selecting questions that are used less frequently across the entire platform. This balances the exposure rate of questions in the question bank on a macro level, prevents some questions from becoming high-frequency questions due to frequent selection, extends the life cycle of low-exposure questions, and improves the overall utilization efficiency of the question bank.

[0048] Step 505: If the number of questions does not reach the target number after traversal, randomly select questions with replacement from the currently available pool to fill the gap until the target number of questions is reached or the shutdown condition is triggered.

[0049] In this step, after traversing the entire perturbed test point sequence, if the number of questions in the current result set has not yet reached the target number of questions for the current difficulty level, the completion extraction phase is initiated. The completion phase uses a random selection method with replacement: each time, a test point is randomly selected from the test point set, and then questions belonging to that test point and meeting the difficulty requirements of the current level are retrieved from the current available pool. If at least one question is retrieved, one is randomly selected and added to the current result set, and that question is removed from the current available pool. If the search result is empty, the extraction fails, and a consecutive failure is recorded. The above process is repeated until any of the following conditions are met, at which point the algorithm terminates: the number of questions in the current result set reaches the target number of questions; or the number of consecutive failed extractions reaches a preset first threshold; or the total number of extraction attempts reaches a preset second threshold. The first and second thresholds can be dynamically configured according to the target number of questions for the current level and the size of the current available pool to ensure that the algorithm can terminate in a timely manner when the question bank is sparse or the constraints are tightened.

[0050] As an optional implementation, if the number of questions after traversal does not reach the target number, a random address with replacement can be selected from the currently available pool. Specifically, this may include: Step 5051: Defining the process of traversing the perturbed test point sequence as the first stage, and defining the process of completing the question selection as the second stage. Step 5052: Initializing the weight of each test point to 1, and recording the number of times each test point selects questions in the first stage.

[0051] In this step, a weight value is initialized for each test point in the first test point set. The initial weights of all test points are set to equal values. At the same time, during the first stage of execution in step 504, the number of times each test point successfully draws a question is recorded. For test points that have never successfully drawn any questions in the first stage, the number of successful draws is recorded as 0. For test points that have successfully drawn at least one question, the actual number of successful draws is recorded.

[0052] Step 5053: Before each random location selection in the second stage, multiply the weight of the test point that was selected 0 times in the first stage by the first coefficient, and divide the weight of the test point that was successfully selected 1 time or more in the first stage by the second coefficient.

[0053] In this step, before each random site selection in the second stage, the weight of each test point is dynamically adjusted based on the number of successful selections of each test point recorded in the first stage: for test points with 0 successful selections in the first stage, the current weight is multiplied by the first coefficient, which is a value greater than 1, to increase the probability of such test points being selected in the second stage, thereby making up for the shortcomings of not being covered in the first stage. For test points that were successfully selected at least once in the first stage, the current weight is divided by a second coefficient, which is a value greater than 1. This reduces the probability of such test points being selected in the second stage, preventing the same test point from being selected excessively repeatedly. The first and second coefficients can be the same or different values, and can be preset according to the desired compensation strength.

[0054] Step 5054: Perform probabilistic random site selection based on the normalized weights of the test points.

[0055] In this step, after the weight adjustment is completed, the weights of all test points are normalized: the weight of each test point is divided by the sum of the weights of all test points to obtain the probability of each test point being selected; then, random location is selected according to this probability distribution, that is, each time a test point is randomly selected, the probability of the test point being selected is proportional to its normalized weight, and the test point with the higher weight is more likely to be selected in the second stage.

[0056] Through the aforementioned adaptive weight adjustment, the second stage can prioritize compensating for test points that were not covered in the first stage, based on the test points already covered in the first stage. This further approaches the overall goal of balanced test point coverage while avoiding excessive and repetitive extraction of already covered test points.

[0057] Step 506: The shutdown condition is that the number of consecutive failed extractions reaches the first threshold or the total number of extraction attempts reaches the second threshold.

[0058] In this step, the shutdown condition is as follows: the completion process is immediately terminated when the number of consecutive failed extractions reaches the first threshold, or the total number of extraction attempts reaches the second threshold. The number of consecutive failed extractions refers to the number of times, within the same completion phase, multiple random selections fail to extract any questions that meet the difficulty requirements of the current level and belong to the selected test points. The total number of extraction attempts refers to the cumulative number of all random selection attempts within the same completion phase, including both successful and failed attempts. The first and second thresholds can be dynamically configured based on the target number of questions for the current level, the size of the currently available pool, and the overall sparsity of the question bank to prevent infinite loops or excessive consumption of computing resources when the feasible region shrinks drastically.

[0059] After extracting questions for the current difficulty level, the server sorts all questions in the result set from low to high difficulty scalar values, obtaining the first sorted sequence. For multiple questions belonging to the same difficulty level in the first sorted sequence, they are further sorted from low to high according to their question identifiers, resulting in an ordered sequence of questions for that level. All ordered question sequences for all difficulty levels are then concatenated in level order to form the second round of the test. This ordering step ensures that the question order presented to the client is consistent with the internal difficulty trajectory, facilitating learners' perception of the difficulty gradient and simplifying subsequent auditing and result reproduction.

[0060] When the third round of test papers needs to be generated, the server determines the union of the score improvement spaces corresponding to the third round of test paper generation. The union of score improvement spaces must contain at least the first, second, and third score improvement spaces. The server obtains the set of test point identifiers associated with each score improvement space, merges them, removes duplicates, and obtains the second set of test points. The target number of questions in the third round is greater than the target number of questions in the second round.

[0061] The server uses the same tiered question extraction kernel as the second round, performing test paper generation on the second set of test points: First, a global exclusion constraint is constructed based on the user's historical answer storage unit, generating a perturbed test point sequence. Each test point is traversed, attempting to extract at most one question that meets the preset difficulty range. If the number of questions is insufficient, random addressing with replacement is performed on the perturbed test point sequence to complete the sequence until the target number of questions is reached or a halting condition is triggered. After question extraction, the questions are also sorted to obtain the third round of test papers, which are then persisted.

[0062] As an optional embodiment, the method may further include: when the extracted question is a compound question, determining the correctness of the parent question based on the correctness of all sub-questions; and using the correctness of the parent question for accuracy calculation.

[0063] In this step, when the extracted question is a compound question, i.e., a parent question containing multiple sub-questions, the following comprehensive judgment rule is used to determine the correctness of the answers to that question: First, obtain the answers to all sub-questions under the compound question, and judge the correctness of each sub-question according to the judgment logic of conventional question types. Only when all sub-questions are answered and judged as correct is the parent question of the compound question judged as correct; otherwise, if any sub-question is not answered, partially answered, or judged as incorrect, the parent question is judged as incorrect. This judgment result does not depend on the score weight or difficulty level of each sub-question. Subsequently, the correctness result of the parent question is used for the accuracy rate calculation in step 10, that is, the parent question is counted as an independent question in the total number of questions and the number of correct questions.

[0064] As an optional embodiment, the method may further include: checking whether there are any two questions among the extracted questions that belong to different sub-questions of the same compound question; if so, removing one of the sub-questions; and re-extracting a replacement question from the remaining candidate questions at the same difficulty level and the target number of questions, based on the difficulty level of the removed question and the target number of questions.

[0065] In this step, after the question selection is completed, the server performs a consistency check on all selected questions. Specifically, this includes iterating through each question in the overall result set and checking if any two questions belong to different sub-questions of the same composite question. The criterion for "different sub-questions belonging to the same composite question" is that the primary keys of the sub-question rows of the two questions are different, but their associated parent question identifiers are the same. If such a conflict is detected, the server removes one sub-question from the result set (e.g., randomly, or by sub-question number), retaining the other sub-question and its corresponding parent question. Subsequently, based on the difficulty level and target number of questions corresponding to the removed sub-question, a replacement question is re-selected from the remaining candidate questions at that difficulty level. During re-selection, the difficulty requirements of the current level, the latest global exclusion constraints, and the cumulative exclusion set between levels must be reapplied. If the replacement selection fails more than a preset number of times consecutively, forced completion is no longer enforced. The actual number of questions in the current level is recorded as a partial value of the target number of questions, and this test paper assembly is marked as partially successful. The upper-level business can then decide whether to accept or re-trigger the test paper assembly.

[0066] As an optional embodiment, the method may further include: querying questions that have been generated but not answered in the current round; merging the generated but unanswered questions with the questions that the user has answered in the past to form a global exclusion constraint.

[0067] This step not only queries the user's historical answer storage unit for question identifiers already answered in the current course, but also additionally queries the question details for other rounds generated but not yet answered by the user within the current test batch. Specifically, the server retrieves all question records with round identifiers less than the current round (e.g., if the current round is the second, then the first round is retrieved) from the test question detail storage unit based on the batch key and course identifier of the current test batch. If these records have been generated but the user has not yet submitted the corresponding answer data, the question identifiers of these questions are included in the exclusion set. The generated but unanswered questions are merged with the user's historically answered questions to form the final global exclusion constraint. This mechanism prevents the same questions from being repeatedly drawn in subsequent rounds within the same batch due to the user skipping a round or failing to submit due to network delays, further enhancing the diversity of test paper generation and the validity of the assessment.

[0068] During the test paper generation stage, the server retrieves questions from the question bank question storage unit after excluding the question identifiers that the user has already appeared in the course in the user's historical answer storage unit, based on the course identifier and the test point set obtained by parsing the test point set. The questions are then written into the test question details storage unit and the question and test point association storage unit, and the cumulative information such as the number of questions and test points in the test batch main storage unit is updated.

[0069] During the answering phase, the server updates the answer and correct / incorrect fields of the test question details storage unit and sub-question details according to the batch primary key and round identifier, and merges them into the user's historical answer storage unit.

[0070] During the statistical phase, the test question details storage unit aggregates correct answers, incorrect answers, total number of questions, accuracy rate, etc., and writes them back to the test batch main storage unit to provide input for the accuracy rate branch in the next round.

[0071] The following are the core storage units and their field definitions that support the method of this application. In one embodiment, each storage unit may be a data table, document collection or equivalent persistent structure in a relational database. This application does not limit the specific table name and physical naming.

[0072] 1. Question Bank Storage Unit: This unit stores the master data of questions that can be referenced in test paper generation. The logical fields related to this solution include at least: Question Identifier: Uniquely identifies a question within the system; Category or Chapter Identifier: Links the question to a category tree or chapter under the course, used for linkage with test point retrieval; Question Type Identifier: Discrete type encoding; Difficulty Scalar: Used as a grading or range filtering condition when generating test papers; Question Stem and Standard Answer Information: Used for display and scoring; Compound Question Flag: Indicates whether there are sub-questions; Historical Usage Frequency: Records the total number of times the question has been extracted and used in all tests across the platform, used for balanced question exposure control.

[0073] 2. User historical answer storage unit, used to record answers and correct / incorrect answers across batches and rounds at the user-course granularity, supporting the querying and updating of the historical question exclusion set. The logical fields related to this solution include at least: record primary key; user identifier; course identifier; question identifier (consistent with the question bank's question storage unit); correct / incorrect status; valid status.

[0074] In one embodiment, the user identifier and question identifier are merged using the semantics of updating if they exist and inserting if they do not exist, in order to ensure that the same user and the same question have a unique business record in the global dimension.

[0075] 3. The test batch main storage unit represents the aggregate root of a multi-round test session. The logical fields related to this scheme include at least: batch primary key; course identifier; user identifier; total number of questions included in the statistics; number of correct answers; number of incorrect answers; accuracy rate; and cumulative assessment time.

[0076] After the answers to the above statistical fields are stored in the database for each round, they are aggregated and written back in batches by the test question details storage unit to ensure that the main storage unit and the sub-details are consistent.

[0077] 4. Test Question Details Storage Unit (Parent Question Granularity): This unit stores one record for each parent question within a specific batch and round, carrying the display, answering, and scoring results. The logical fields related to this solution include at least: detail row primary key; batch primary key (associated with the main storage unit of the test batch); question identifier (associated with the question bank question storage unit); round identifier (value 1 or 2, indicating the first or second round); compound question identifier; question type and difficulty (consistent with the question selection criteria during test paper generation for easy verification); user's answer content and standard answer information; and whether it is correct (as input for batch-level statistics).

[0078] 5. A question-and-exam-point association storage unit records the exam point number and name associated with each question, used for alignment with the course exam point system and score improvement space configuration. The logical fields related to this solution include at least: associated row primary key; batch primary key; question identifier; exam point number; and exam point name.

[0079] 6. For compound question sub-question details storage units, when the parent question is a compound question, the sub-questions are persisted separately. The logical fields related to this solution include at least: sub-question row primary key; batch primary key, round identifier, parent question identifier; sub-question question type, options, and standard answer information; user's answer content, and whether the sub-question is correct.

[0080] The rule for determining whether a parent question is answered correctly is as follows: if all sub-questions are judged as correct, the parent question is judged as correct; otherwise, it is judged as incorrect. This judgment result is written back to the "whether it is correct" field of the parent question record in the test question details storage unit.

[0081] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide a test paper generation device based on multi-round answering and spatial constraints, the structure of which is as follows: Figure 3 As shown.

[0082] Figure 3 This is a schematic diagram of the internal structure of a test paper generation device based on multi-round response and spatial constraints, provided as an embodiment of this application. Figure 3 As shown, the device includes: At least one processor 301; And a memory 302 that is communicatively connected to at least one processor; The memory 302 stores instructions that can be executed by at least one processor. The instructions are executed by at least one processor 301 to enable at least one processor 301 to: perform any step of a test paper generation method based on multi-round response and spatial constraints.

[0083] Some embodiments of this application provide corresponding to Figure 1 A non-volatile computer storage medium for test paper assembly based on multi-round response and spatial constraints stores computer-executable instructions, which are configured to execute any one of the steps of a test paper assembly method based on multi-round response and spatial constraints.

[0084] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for IoT devices and media are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0085] The systems, media, and methods provided in this application are one-to-one correspondences. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0086] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0087] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0088] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0089] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0090] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0091] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0092] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0093] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0094] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A test paper generation method based on multi-round response and spatial constraints, characterized in that, The method includes: In response to the answer data from the first round, the correctness of the answer data from the first round is determined, and the accuracy rate of the current round is updated based on the determination result; Based on the comparison result between the accuracy rate and the preset accuracy rate threshold, the difficulty level of the second round is selected from the preset difficulty quota profile. The difficulty level includes multiple difficulty levels arranged in sequence and the corresponding target number of questions. Based on the score improvement space identifier corresponding to the second round, obtain the corresponding set of test point identifiers in the test point mapping relationship; In the user's historical answer records, obtain the set of question identifiers that the current user has answered in the current course, and use it as a global exclusion constraint; Based on the aforementioned difficulty levels, test point identifier sets, and global exclusion constraints, questions are selected in tiers.

2. The test paper generation method based on multi-round response and spatial constraints according to claim 1, characterized in that, The step of categorizing and selecting questions based on the difficulty level, the set of test point identifiers, and global exclusion constraints specifically includes: Based on the order of the difficulty levels, the questions that have been extracted from each difficulty level are added to the inter-level cumulative exclusion set. Excluding the global exclusion constraints and the inter-level cumulative exclusion set, the currently available pool is obtained; After preserving the order and removing duplicates from the set of test point identifiers, a pseudo-random perturbation is applied to obtain the perturbed test point sequence.

3. The test paper generation method based on multi-round response and spatial constraints according to claim 2, characterized in that, The method further includes: Traverse the perturbation test point sequence and extract questions belonging to the test point and satisfying the current difficulty level from the currently available pool; If the number of questions does not reach the target number after traversal, a random address with replacement is selected from the current available pool to fill the gap with questions that meet the current difficulty level, until the target number of questions is reached or the shutdown condition is triggered. The shutdown condition is when the number of consecutive failed extractions reaches a first threshold or the total number of extraction attempts reaches a second threshold.

4. The test paper generation method based on multi-round response and spatial constraints according to claim 1, characterized in that, The method further includes: When the selected question is a compound question, the correctness of the parent question is determined by comprehensively considering the correctness of all sub-questions; The correctness of the parent question is used in the accuracy calculation.

5. The test paper generation method based on multi-round response and spatial constraints according to claim 3, characterized in that, The step of traversing the perturbed test point sequence and extracting questions belonging to that test point and meeting the current difficulty level from the currently available pool specifically includes: Get all candidate questions that meet the difficulty level for the current test point; Obtain the historical usage frequency of each candidate question in the question bank; Select the question that was used least frequently in the history.

6. The test paper generation method based on multi-round response and spatial constraints according to claim 3, characterized in that, If the number of questions does not reach the target number after traversal, a random address with replacement is selected from the currently available pool, specifically including: The process of traversing the perturbation test point sequence is defined as the first stage, and the process of completing the question extraction is defined as the second stage. Initialize the weight of each test point to 1, and record the number of times each test point draws questions in the first stage; Before each random location selection in the second stage, the weight of the test point that was selected 0 times in the first stage is multiplied by the first coefficient, and the weight of the test point that was successfully selected 1 time or more in the first stage is divided by the second coefficient. Probabilistic random location selection is performed based on the normalized weights of the test points.

7. The test paper generation method based on multi-round response and spatial constraints according to claim 1, characterized in that, The method further includes: Check if any two questions in the extracted questions are different sub-questions of the same compound question. If so, remove one of the sub-questions. Based on the difficulty level and target number of questions corresponding to the removed question, a replacement question is drawn from the remaining candidate questions in that level.

8. The test paper generation method based on multi-round response and spatial constraints according to claim 1, characterized in that, The method further includes: Retrieve questions that have been generated but not yet answered in the current round; The generated but unanswered questions are merged with the user's historical answered questions to form the global exclusion constraint.

9. A test paper generation device based on multi-round response and spatial constraints, characterized in that, The device includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to: Perform the steps of the test paper generation method based on multi-round response and spatial constraints as described in any one of claims 1-8.

10. A non-volatile computer storage medium for generating test papers based on multi-round responses and spatial constraints, storing computer-executable instructions, characterized in that, The computer-executable instructions are set as follows: Perform the steps of the test paper generation method based on multi-round response and spatial constraints as described in any one of claims 1-8.

Images