Question group generation method and device, electronic equipment, storage medium and computer program product

Abstract

The application provides a test paper compiling method and device, electronic equipment, a storage medium and a computer program product. The method comprises the following steps: receiving a test paper compiling instruction of a user; performing intention analysis on the test paper compiling instruction to obtain a test paper compiling intention of the user; and customizing a personalized test paper of the user based on the test paper compiling intention and portrait information of the user. The application fully considers the objective portrait information and subjective test paper compiling intention of the user, so that a test paper that meets the personalized needs of the user can be generated, the learning efficiency and learning effect of the user are improved, and problems such as mismatch between the content of the test paper and the needs of the user and weak pertinence in the traditional test paper compiling method are avoided.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, apparatus, electronic device, storage medium, and computer program product for creating computer files. Background Technology

[0002] With the continuous development of educational informatization, the demand for personalized learning is growing. As a crucial part of the teaching process, test paper creation directly impacts students' learning outcomes. How to efficiently and accurately provide personalized test paper creation services for different student groups is a pressing issue in the education field.

[0003] Currently, the most common method for creating test papers is to rely on teachers to create them manually. While this method can incorporate teaching experience, the scope of the test papers is often based on broad, general rules, making it difficult to take into account the individual learning needs of each student. This results in test paper results that lack specificity and cannot effectively meet the needs of individual students to strengthen their weaknesses. Summary of the Invention

[0004] This invention provides a method, apparatus, electronic device, storage medium, and computer program product for compiling documents, in order to overcome the deficiencies existing in the prior art.

[0005] This invention provides a method for creating test papers, comprising the following steps: Receive the user's volume creation command; The intent of the volume creation command is analyzed to obtain the user's volume creation intent; Based on the stated test paper creation intent and the user's profile information, a personalized test paper is customized for the user.

[0006] According to a method for creating test papers provided by the present invention, the step of customizing a personalized test paper for a user based on the test paper creation intention and the user's profile information includes: Based on the profile information, determine multiple group types that match the user; Based on the stated test paper creation intent, the recommendation weight of each question under each test paper type is determined. The test paper creation intent is used to characterize the user's test paper creation needs under different dimensions. Based on the recommendation weights of each question under each test paper type, a personalized test paper for the user is obtained.

[0007] According to a method for creating a series of documents based on the user profile information, determining multiple series of document types matching the user includes: Based on the profile information, multiple test paper types matching the user are determined, as well as the percentage of knowledge point difficulty under each test paper type.

[0008] According to a test paper generation method provided by the present invention, determining the weight of each question under each test paper generation type based on the test paper generation intention includes: Based on the proportion of knowledge point difficulty in each test paper type, the first recommended weight of each question in each test paper type is determined. Based on the stated test paper composition intention, determine the second recommended weight for each question under each test paper composition type; The recommended weight for each question is obtained by combining the first and second recommendation weights of each question.

[0009] According to a test paper generation method provided by the present invention, determining the second recommended weight of each question under each test paper generation type based on the test paper generation intention includes: Based on the knowledge point type of each question under each test paper type, the initial recommendation weight of each question is determined; Based on the stated test paper composition intention, the initial weights of each test item are adjusted to obtain the second recommended weights for each test item.

[0010] According to a method for generating test papers provided by the present invention, the test paper type carries a priority label; the step of generating test papers based on the recommendation weight of each question under each test paper type to obtain the user's personalized test paper includes: Based on the recommended weight of each question under each test paper type, test papers are generated to obtain personalized test papers corresponding to each test paper type. Based on the priority tags of each test paper type, the personalized test papers corresponding to each test paper type are prioritized.

[0011] According to a test paper generation method provided by the present invention, the step of generating test papers based on the recommendation weights of each question under each test paper generation type to obtain the user's personalized test paper further includes: Receive the user's tuning instructions; Based on the optimization information carried by the optimization instructions, the recommended weight of each question under each test paper type is optimized; Based on the optimized recommendation weights of each question, a personalized test paper is generated.

[0012] According to a method for creating a book volume provided by the present invention, the step of performing intent analysis on the book volume creation instruction to obtain the user's book volume creation intent includes: Based on the portrait information, multiple candidate grouping intentions are determined; Based on the volume creation instruction, the user's volume creation intent is determined from the plurality of candidate volume creation intents.

[0013] The present invention also provides a paper-building device, comprising the following modules: The receiving unit is used to receive the user's volume assembly instructions; The analysis unit is used to perform intent analysis on the volume creation command to obtain the user's volume creation intent; The customization unit is used to customize a personalized test paper for the user based on the test paper creation intention and the user's profile information.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the volume assembly method as described above.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the volume assembly method as described above.

[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the volume assembly method as described above.

[0017] The test paper generation method, apparatus, electronic device, storage medium, and computer program product provided by this invention analyze the user's test paper generation instructions to obtain the user's test paper generation intention, and combine this with the user's profile information to ultimately generate a personalized test paper. Because this invention fully considers both the user's objective profile information and subjective test paper generation intention, it can generate test papers that better meet the user's personalized needs, improving the user's learning efficiency and learning outcomes, and avoiding problems such as mismatch between test paper content and user needs, and lack of specificity in traditional test paper generation methods. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is one of the flowcharts of the document assembly method provided by the present invention.

[0020] Figure 2 This is a flowchart illustrating an implementation of step 130 in the document assembly method provided by the present invention.

[0021] Figure 3 This is a flowchart illustrating an implementation of step 132 in the document assembly method provided by the present invention.

[0022] Figure 4 This is the second flowchart of the document assembly method provided by the present invention.

[0023] Figure 5 This is the third flowchart of the document assembly method provided by the present invention.

[0024] Figure 6 This is the fourth flowchart of the document assembly method provided by the present invention.

[0025] Figure 7 This is a schematic diagram of the structure of the roll-building device provided by the present invention.

[0026] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0028] Currently, common test paper generation methods mainly include the following two types: First, relying on teachers to manually generate test papers. While this method can incorporate teaching experience, the scope of test paper generation is often based on broad, general rules, making it difficult to consider the individual learning needs of each student. This results in a lack of targeted test paper results and an inability to effectively address individual student weaknesses. Second, test paper generation products based on fixed rules. Existing products of this type often use fixed dimensions or question banks. After users make selections, only specific results are generated, lacking a personalized test paper generation process and results. It is difficult to dynamically adjust and optimize based on individual student situations. Furthermore, the accuracy of recognizing user intent needs improvement, failing to truly understand students' real needs. Therefore, traditional test paper generation methods have significant shortcomings in terms of personalization and intelligence, failing to meet the growing demand for personalized learning.

[0029] To address this issue, the present invention provides a method for creating test papers. Wherein, Figure 1 This is one of the flowcharts illustrating the document assembly method provided by this invention, such as... Figure 1 As shown, the method includes steps 110, 120 and 130.

[0030] Step 110: Receive the user's volume creation instruction.

[0031] Here, "user" refers to the entity that needs to be assessed through the generated test papers. This entity can be a student, a school, or an educational district; this embodiment of the invention does not impose specific limitations on this. The test paper generation instruction refers to the instruction information used to express the user's test paper generation needs. It characterizes the features and requirements that the user hopes the generated test paper should possess, such as difficulty, question types, knowledge point coverage, and emphasis.

[0032] The test paper creation instructions can include: ① Voice instructions, which are the user's needs expressed through voice input. For example, a user can say, "I want a physics test paper that focuses on the law of conservation of momentum, with moderate difficulty, and includes multiple-choice and calculation questions." ② Text instructions, which are text information entered by the user via keyboard, touchscreen, etc., to describe in detail their specific requirements for the test paper. For example, a user can enter in the dialog box, "Generate a test paper on electromagnetic induction, with moderate difficulty, including multiple-choice, fill-in-the-blank, and calculation questions, focusing on the application of Faraday's law of electromagnetic induction." ③ Preset function button instructions, which are instructions where the user can trigger the generation of a test paper with specific attributes by clicking a pre-set function button on the learning device. For example, the learning device can have function buttons such as "Redo Incorrect Questions," "Reinforce Knowledge Points," and "Mock Exam" for the user to choose from.

[0033] Considering that users' intentions may be ambiguous or incomplete, after receiving the user's test paper creation instruction, the instruction can first be preprocessed, such as by text cleaning and removing stop words. Then, a large language model can be used to semantically understand and complete the instruction, transforming the ambiguous instruction into a clear and explicit description of the intention. Finally, the instruction can be supplemented or adjusted based on the user's historical learning data. For example, if the user only requests "midterm review" in the test paper creation instruction, specific knowledge points, question types, and difficulty information can be automatically supplemented based on the user's midterm exam scope, historical wrong answer records, and knowledge point mastery, generating a test paper that better meets the user's personalized needs. This improves the accuracy and efficiency of subsequent intention analysis and ensures the relevance and effectiveness of the test paper creation results.

[0034] In addition, to further enhance the user experience, preset test paper creation instruction templates or prompts can be provided to guide users to more clearly express their test paper creation needs. For example, a template such as "Please generate a test paper for [subject] of [difficulty], focusing on [knowledge points]" can be provided for users to refer to and modify.

[0035] Step 120: Perform intent analysis on the volume creation command to obtain the user's volume creation intent.

[0036] Specifically, test creation intent refers to the user's expectations and goals for the test paper. It represents the assessment objectives and content scope that the user hopes to achieve through the test paper. Test creation intent is obtained after performing intent analysis on the test creation command. For example, by analyzing the user's input natural language command "Generate a Grade 11 Mechanics final exam review paper," we can know that the user's test creation intent is to generate a test paper suitable for Grade 11 students, with content focusing on mechanics, for final exam review.

[0037] The intent behind test paper creation often encompasses multiple dimensions of user needs, such as the difficulty of knowledge points, knowledge point attributes (including frequently tested points, common mistakes, and weak points), local exam trends, and question types. For example, under the dimension of knowledge point difficulty, users can select difficulty levels such as "easy," "medium," and "difficult" to control the overall difficulty of the test paper. Under the dimension of knowledge point attributes, users can select attributes such as "frequently tested points," "common mistakes," and "weak points" to highlight the key content tested in the test paper. Under the dimension of local exam trends, users can select "high-frequency local test points" to make the test paper more aligned with local exam trends. Under the dimension of question types, users can select question types such as "multiple choice," "fill in the blank," and "problem-solving" to meet different assessment needs.

[0038] As an optional implementation, a large language model can be used to perform deep semantic analysis on the test paper creation instructions, extracting keywords and key information contained within them. These keywords and key information are then matched with a pre-built knowledge graph to determine the user's test paper creation intent, such as the range and difficulty level of knowledge points the user wishes to assess. Furthermore, the user's historical learning data and behavioral habits can be combined to more accurately predict and judge the user's test paper creation intent. For example, if a user frequently makes mistakes on a certain knowledge point, it can be inferred that the user wants a test paper that focuses on assessing that knowledge point.

[0039] Furthermore, to improve the accuracy of intent analysis, multi-turn dialogues can be used to interact with users, guiding them to more clearly express their test paper creation needs. For example, if the user's input instructions cannot be accurately understood, the system can proactively ask, "What knowledge points do you hope the generated test paper will primarily test?", thereby more accurately grasping the user's intent in creating the test paper.

[0040] Step 130: Based on the test paper creation intent and user profile information, customize personalized test papers for users.

[0041] Specifically, user profile information refers to the description of user characteristics, which is used to represent information such as the user's learning level, knowledge mastery, and learning preferences. User profile information may include information such as the user's grade level, subject, history learning situation, weak knowledge points, frequently tested knowledge points, and study time.

[0042] The methods for determining user profile information may include: ① Obtaining basic user information and learning status through user-initiated questionnaires. ② Constructing a user's knowledge graph by analyzing historical learning data, such as homework completion, exam scores, and incorrect answer records. ③ Inferring user learning interests and preferences by combining user social information and behavioral data, such as the activity level of learning communities and search history. It is understood that other methods can also be used to determine user profile information, and this embodiment of the invention does not specifically limit these methods.

[0043] Unlike traditional methods, this invention, when generating personalized test papers, not only considers the user's objective profile information, but more importantly, fully incorporates the user's subjective intention in creating the test paper. This approach gives users greater autonomy and flexibility, enabling the generated test papers to truly reflect the user's will and needs, achieving a personalized learning experience tailored to each individual.

[0044] Specifically, when generating test papers, the first step is to determine the test paper type and the proportion of knowledge points with varying difficulty levels based on the user's profile. For example, if a user's historical learning indicates they are currently in the mid-term review stage and have made many mistakes in the mechanics section, a mid-term review-type test paper can be prioritized, with an appropriate increase in the proportion of medium-difficulty mechanics questions. Next, the test paper can be fine-tuned based on the user's test paper creation intentions. For instance, if a user expresses a desire for a test paper that "focuses on the law of conservation of energy and includes some application problems related to real-life situations," then, while meeting the requirements for mid-term review and the proportion of mechanics questions, questions related to the law of conservation of energy can be prioritized, and application problems connected to real-life situations can be included to enhance the user's learning interest and problem-solving abilities.

[0045] In addition, to further enhance the personalization of the test paper, details such as the layout style, font size, and question presentation method can be adjusted according to the user's learning preferences. For example, if the user is used to studying on a tablet, a test paper suitable for tablet reading can be generated, and pictures and animations can be appropriately added to the questions to improve the user's reading experience.

[0046] In summary, by combining objective user profile information with subjective test paper creation intentions, this invention can generate a highly personalized test paper that meets the different learning needs and preferences of users, thereby improving their learning efficiency and learning outcomes.

[0047] For example, User A's profile indicates that they are a student with weak foundational knowledge, and in their test creation instructions, User A expressed a desire for a test paper that is relatively easy and includes detailed explanations to help them solidify their basic knowledge. Based on this, a test paper with lower difficulty can be generated according to User A's profile, and detailed explanations can be added after each question according to User A's test creation instructions to help User A understand the problem-solving approaches and methods.

[0048] The test paper generation method in this embodiment of the invention is more like an intelligent agent. It can understand the user's needs and dynamically adjust and optimize the content of the test paper according to the user's personalized characteristics, generating a truly personalized test paper that meets the user's expectations. Compared with the traditional semi-customized template test paper generation method, this invention is more flexible and versatile, better reflects the student's will, and can effectively improve the student's learning efficiency and interest.

[0049] It should be noted that steps 120 and 130 above can be implemented using a large language model (hereinafter referred to as "large model"). For example, in step 120, the large model can analyze the user's input test paper creation instructions to extract the user's test paper creation intent; in step 130, the large model can select suitable questions from the question bank based on the user's test paper creation intent and profile information, and then combine and sort them to finally generate a personalized test paper. Alternatively, steps 120 and 130 can be implemented using the large model at the same time, such as inputting the user's profile information and test paper creation instructions into the large model, which can directly generate a complete test paper. This embodiment of the invention does not specifically limit this approach. Here, "large model" refers to a deep learning model trained on massive amounts of data with powerful natural language processing and generation capabilities, such as the Transformer model or the BERT model.

[0050] The test paper generation method provided in this invention analyzes the user's test paper generation instructions to obtain the user's test paper generation intent, and combines this with the user's profile information to ultimately generate a personalized test paper. Because this invention fully considers both the user's objective profile information and subjective test paper generation intent, it can generate test papers that better meet the user's personalized needs, improving the user's learning efficiency and learning outcomes, and avoiding problems such as mismatch between test paper content and user needs, and lack of specificity, found in traditional test paper generation methods.

[0051] Based on the above embodiments, Figure 2 This is a flowchart illustrating an embodiment of step 130 provided by the present invention, as shown below. Figure 2 As shown, step 130 includes: Step 131: Based on the profile information, determine multiple volume types that match the user.

[0052] Specifically, test paper type refers to the overall positioning and format classification of the test paper, which is used to characterize the design direction and focus of the test paper. Test paper types can include standard test papers (such as standard-achieving papers, advanced papers, etc.), specialized practice papers (such as papers focusing on individual weaknesses, local common mistakes, and local high-frequency test points), mock test papers (such as middle school entrance exam mock papers and college entrance exam mock papers), and question package test papers (such as basic consolidation question packages and advanced question packages). For example, standard-achieving papers focus on assessing students' mastery of basic knowledge, while advanced papers focus on assessing students' comprehensive application ability; papers focusing on individual weaknesses provide intensive training on students' weak knowledge points, while papers focusing on local common mistakes provide focused practice on common errors made by local students.

[0053] Considering that the profile information includes information such as the user's grade, subject, historical grades, weak knowledge points, frequently tested knowledge points, and study time, the relevant test paper types can be initially selected based on the user's grade and subject. Then, the test paper types can be finely adjusted based on the user's historical grades and weak knowledge points.

[0054] For example, if a user's profile shows that the user is a second-year high school student with poor mechanics grades and a weak point in the "work-energy theorem" knowledge point, a "Second-Year High School Mechanics - Work-Energy Theorem - Personal Weak Point Practice Paper" can be recommended.

[0055] For example, if a user's profile shows that the user is a ninth-grade student with average math scores, has recently performed poorly in the "quadratic function" section, and has not studied for a long time, then the test paper types that can be recommended for them include: "Ninth Grade Math - Quadratic Function - Basic Consolidation Test Pack" and "Ninth Grade Math - Quadratic Function - Standardized Test".

[0056] As an optional implementation, candidate test paper types can first be selected from a predefined test paper type library based on the user's grade and subject. Then, the matching degree between the user profile information and each candidate test paper type can be calculated, which can be based on factors such as the user's historical grades, weak knowledge points, and study time. Next, the candidate test paper types are sorted according to the matching degree, and the test paper types with the highest matching degree are selected as the test paper types that match the user. Finally, these test paper types can be appropriately adjusted and optimized, for example, the weight of weak knowledge points can be adjusted according to the user's latest learning situation, and finally multiple test paper types that match the user can be obtained.

[0057] Understandably, after obtaining multiple document composition types, these types can be displayed (e.g., in a list or card format) to allow users to select the type they are most interested in or need. If a user selects a document composition type, they can directly proceed to the document composition process for that type. If a user is not satisfied with any of the document composition types, they can be guided to manually enter their document composition requirements, or the option to customize document composition types can be provided.

[0058] Step 132: Based on the test paper creation intent, determine the recommendation weight of each question under each test paper type. The test paper creation intent is used to represent the user's test paper creation needs under different dimensions.

[0059] Specifically, the recommendation weight of each question refers to the probability that the question will be selected for the final exam paper under a specific exam paper type. It characterizes the degree to which the question matches the user's exam paper creation intention and its contribution to improving the quality of the exam paper. A higher recommendation weight indicates that the question better matches the user's exam paper creation intention, and therefore, the higher the priority for the corresponding exam paper to be recommended to the user as part of the exam paper creation results. For example, for an exam paper type like "Grade 11 Mechanics - Kinetic Energy Theorem - Personal Weakness Point Practice Paper," questions that test knowledge points related to the kinetic energy theorem and are of moderate difficulty can be assigned a higher recommendation weight.

[0060] Furthermore, the test paper creation intent encompasses user needs across multiple dimensions. These different dimensions refer to the user's specific requirements for the test paper in terms of knowledge points, difficulty, question types, and exam conditions. For example, a user might desire a test paper of moderate difficulty, focusing on the application of the kinetic energy theorem, and including a certain proportion of multiple-choice and open-ended questions. Based on this, this embodiment of the invention comprehensively considers user needs across various dimensions and combines this with the characteristics of the test questions themselves (such as the knowledge points tested, difficulty, question types, and local exam conditions) to calculate the recommendation weight for each question, ultimately obtaining the recommendation weight for each question under each test paper type.

[0061] For example, if a user specifies test paper requirements such as "medium difficulty", "tests the kinetic energy theorem", and "contains 50% multiple choice questions and 50% problem-solving questions", then for a multiple choice question testing the application of the kinetic energy theorem, if its difficulty is rated as medium and it is consistent with the local test situation, its recommendation weight will be relatively high; conversely, if the question is too difficult or too easy, or if it tests other knowledge points, its recommendation weight will be relatively low.

[0062] As an optional implementation, candidate questions related to the knowledge points specified by the user can be selected first. Then, the scores of each candidate question can be calculated in terms of difficulty, question type, local test conditions, etc. The score calculation can be based on predefined rules or machine learning models. Next, the scores in each dimension are weighted and summed to obtain the comprehensive score of each candidate question, which is the recommendation weight of the question. Finally, the recommendation weights of all questions are normalized so that their values ​​are between 0 and 1, and the recommendation weights of each question in each test paper type are finally obtained.

[0063] Understandably, before determining the recommendation weight of each question, the questions in the question bank can be preprocessed, such as by marking the knowledge points, rating the difficulty, classifying the question types, and analyzing the local test situation, so that the subsequent calculation of recommendation weights can be more efficient and accurate.

[0064] Step 133: Based on the recommendation weight of each question under each test paper type, generate test papers to obtain personalized test papers for users.

[0065] Specifically, the recommendation weight of each question represents its degree of matching with user needs and its contribution to improving the quality of the test paper. Based on these recommendation weights, questions can be sorted and filtered, and those with higher recommendation weights can be combined into a complete test paper for the user. Since this test paper is generated based on the user's personalized profile and test paper creation intentions, it better meets the user's learning needs, more effectively strengthens the user's weak knowledge points, and improves the user's learning efficiency. Therefore, this test paper has a higher degree of personalization and better learning outcomes.

[0066] As an optional implementation, all questions can first be sorted in descending order according to their recommended weights. Then, questions can be selected from the sorted list and added to the test paper in sequence until the preset number of questions or total score requirements of the test paper are met. During the selection of questions, the correlation of knowledge points and the combination of difficulty between questions need to be considered to ensure the overall quality of the test paper and obtain the user's test paper.

[0067] As another optional embodiment, a test paper generation method based on optimization algorithms can also be adopted. For example, the test paper generation problem can be modeled as an integer programming problem, where the objective function is to maximize the sum of the recommended weights of all questions in the test paper. The constraints include the number of questions, total score, knowledge point coverage, difficulty distribution, etc. Then, optimization algorithms (such as linear programming, genetic algorithms, etc.) are used to solve the integer programming problem to obtain the optimal test paper and the user's test paper.

[0068] The test paper generation method provided in this invention determines the user's test paper generation intention and profile information, identifies multiple test paper types matching the user based on the profile information, and determines the recommendation weight of each question under each test paper type based on the test paper generation intention. This allows for targeted generation of personalized test papers to meet diverse learning needs. By fully considering the user's learning level, knowledge mastery, learning preferences, assessment objectives, and content scope, the generated test papers are more closely aligned with the user's actual situation, better helping them identify and fill knowledge gaps, consolidate knowledge, and improve abilities. This improves the efficiency and quality of test paper generation, enhances the user's learning experience, and achieves personalized intelligent test paper generation. It effectively improves students' learning efficiency and outcomes, avoiding the problems of weak test paper targeting and inability to meet personalized user needs in traditional test paper generation methods, as well as the low efficiency and difficulty in guaranteeing quality in manual test paper generation.

[0069] Based on the above embodiments, multiple volume types matching the user are determined based on the profile information, including: Based on user profile information, multiple test paper types matching the user are determined, as well as the percentage of knowledge point difficulty under each test paper type.

[0070] Specifically, the proportion of knowledge point difficulty refers to the percentage of knowledge points of different difficulty levels under a specific test paper type. For example, in the "basic consolidation" test paper type, the proportion of easy-difficulty knowledge points may be higher, while the proportion of difficult-difficulty knowledge points may be lower; in the "ability improvement" test paper type, the proportion of difficult-difficulty knowledge points may be higher, while the proportion of easy-difficulty knowledge points may be lower.

[0071] Considering the differences in learning levels and knowledge mastery among users, and the different assessment objectives and content focuses of different test paper types, it is also necessary to consider the difficulty ratio of knowledge points when creating test papers. This is to ensure that the generated test papers can better meet the personalized needs of users and achieve the corresponding assessment objectives. Furthermore, the difficulty ratio of knowledge points can be dynamically adjusted based on user profile information and test paper type to improve the relevance and effectiveness of the test papers. This avoids generating test papers that are too difficult or too easy, which would prevent users from effectively assessing their learning or achieving the corresponding learning objectives.

[0072] As an optional implementation, a default knowledge point difficulty ratio can first be set for each test paper type according to predefined rules. Then, based on the user's historical scores and weak knowledge points, the default knowledge point difficulty ratio can be adjusted. For example, if the user's weak knowledge points are concentrated at a certain difficulty level, the proportion of knowledge points at that difficulty level can be appropriately increased. Finally, the knowledge point difficulty ratio can be further optimized based on user feedback to ultimately determine the knowledge point difficulty ratio for each test paper type. The predefined rules mentioned above can be based on the characteristics of the test paper type itself to set the default knowledge point difficulty ratio. For example, the default knowledge point difficulty ratio for the "Basic Consolidation" test paper is "Easy: Medium: Difficult = 70%: 20%: 10%", the default for the "Ability Improvement" test paper is "Easy: Medium: Difficult = 30%: 40%: 30%", and the default for the "Pre-exam Intensive" test paper is "Easy: Medium: Difficult = 20%: 50%: 30%".

[0073] As another optional embodiment, a machine learning model, such as a regression model or a classification model, can be used to predict the optimal knowledge point difficulty ratio for a user under a given test group type based on the user's profile information and test group type. The training data for the model can include the user's historical learning data, the learning data of other users, and the knowledge point difficulty ratio set by experts, etc., to ultimately determine the knowledge point difficulty ratio under each test group type.

[0074] Furthermore, considering that users' learning progress may change over time and that user feedback may contain certain biases, after determining the proportion of knowledge point difficulty, the proportion of knowledge point difficulty can be evaluated and adjusted periodically. For example, by analyzing users' problem-solving performance, exam scores, and feedback, it can be determined whether the current proportion of knowledge point difficulty is reasonable, and adjustments can be made accordingly based on the evaluation results.

[0075] Based on any of the above embodiments Figure 3 This is a flowchart illustrating an implementation method for step 132 in the document assembly method provided by the present invention, as shown below. Figure 3 As shown, step 132, based on the test paper composition intention, determines the weight of each question under each test paper type, including: Step 1321: Based on the proportion of knowledge point difficulty under each test paper type, determine the first recommended weight of each question under each test paper type; Step 1322: Based on the test paper composition intention, determine the second recommended weight of each question under each test paper type; Step 1323: Combine the first recommendation weight and the second recommendation weight of each test item to obtain the recommendation weight of each test item.

[0076] Specifically, the first recommendation weight refers to the degree of recommendation for a question within a specific test paper type, determined by the proportion of knowledge point difficulty. It characterizes the degree to which the question's difficulty matches the expected difficulty distribution of that test paper type. Since the proportion of knowledge point difficulty reflects the percentage of knowledge points of different difficulty levels that should be present in that test paper type, a higher first recommendation weight indicates that the question's difficulty better matches the overall difficulty requirements of that test paper type. Therefore, the first recommendation weight determined by the proportion of knowledge point difficulty ensures that the overall difficulty structure of the test paper conforms to the preset design. For example, for a "basic consolidation" test paper type, questions with lower difficulty may have a relatively higher first recommendation weight.

[0077] As an optional implementation, the first recommended weight of a test question can be calculated based on its difficulty level and the expected proportion of that difficulty level within the test paper type. For example, if a test question has a difficulty level of "easy" and the expected proportion of "easy" difficulty level within the test paper type is 70%, then the first recommended weight of that test question can be set to 0.7, thus determining the first recommended weight of each test question.

[0078] As another optional embodiment, fuzzy matching can also be used to perform fuzzy matching between the difficulty level of a test question and the expected difficulty distribution under this test paper type, and determine the first recommended weight of the test question based on the degree of matching. For example, a Gaussian function can be used to describe the expected difficulty distribution under this test paper type, and then the similarity between the difficulty level of the test question and the Gaussian function can be calculated. This similarity is the first recommended weight of the test question, and the first recommended weight of each test question is determined.

[0079] Furthermore, the second recommendation weight refers to the degree of recommendation of test questions based on the user's test-setting intention under a specific test-setting type. It is used to characterize the degree of matching between the test questions and the user's explicitly specified test-setting needs. Since test-setting intentions include multiple dimensions of user needs, such as knowledge points, question types, and local exam conditions, a higher second recommendation weight indicates that the test questions better meet the user's personalized test-setting needs. Therefore, the second recommendation weight determined based on test-setting intentions can ensure that the content structure of the test paper meets the user's specific needs. For example, if the user specifies that they want to focus on testing the "kinetic energy theorem," then the second recommendation weight of test questions testing knowledge points related to the "kinetic energy theorem" will be relatively high.

[0080] As an optional implementation, the second recommendation weight of a test question can be calculated by matching it with the user-specified test paper generation requirements based on the knowledge points, question types, local test conditions, and other attributes tested in the test question. For example, a keyword matching method can be used to calculate the number of user-specified keywords contained in the test question description. The more keywords there are, the higher the second recommendation weight of the test question, thus determining the second recommendation weight of each test question.

[0081] As another optional embodiment, a deep learning model can also be used to perform semantic analysis on the test questions and the user's test paper generation needs, calculate the semantic similarity between them, and the semantic similarity is the second recommendation weight of the test question, thereby determining the second recommendation weight of each test question.

[0082] Considering that simply basing test paper creation on the first recommended weight of each question might overlook users' personalized needs, resulting in generic test papers that fail to meet users' specific learning objectives and thus lead to a poor user experience. For example, even with a "basic consolidation" test paper type, different users may want to focus on different knowledge points.

[0083] Furthermore, considering that compiling test papers solely based on the second recommended weight of each question might overlook the overall difficulty structure of the test paper, resulting in an unreasonable distribution of difficulty and an inability to effectively assess the user's learning level, thus compromising the quality of the test paper. For example, even if the user specifies that a particular knowledge point should be emphasized, it is still necessary to ensure that the test paper contains a certain proportion of questions at different difficulty levels.

[0084] Based on this, the embodiments of the present invention integrate the first recommended weight and the second recommended weight of each test question to obtain the final recommended weight of each test question. This can take into account both the overall difficulty structure of the test paper and the personalized needs of users, ensuring both the quality of the test paper and meeting the specific learning objectives of users, avoiding the problems caused by considering only one aspect while ignoring the other. For example, for a test question testing the "work energy theorem", if its difficulty level is "medium", and the user specifies that the "work energy theorem" should be tested in detail, then its final recommended weight will be calculated by combining its first recommended weight (reflecting the degree of difficulty level "medium") and the second recommended weight (reflecting the degree of testing of the "work energy theorem").

[0085] As an optional implementation, a linear weighting method can be used to sum the first and second recommended weights of each question to obtain the final recommended weight for each question. For example, the final recommended weight = α × first recommended weight + (1-α) × second recommended weight, where α is an adjustable parameter used to control the relative importance of the first and second recommended weights to obtain the recommended weight for each question. For example, if the user has high requirements for the overall difficulty of the test paper, but does not particularly emphasize the examination of specific knowledge points, a higher value for α (e.g., 0.7) can be assigned to give more importance to the first recommended weight, ensuring that the overall difficulty structure of the test paper meets the preset requirements. If the user explicitly specifies that certain knowledge points should be emphasized, but does not have specific requirements for the overall difficulty of the test paper, a lower value for α (e.g., 0.3) can be assigned to give more importance to the second recommended weight, ensuring that the content structure of the test paper meets the user's specific needs.

[0086] Based on any of the above embodiments, and based on the test paper compilation intention, the second recommended weight of each question under each test paper type is determined, including: Based on the knowledge point type of each question under each test paper type, the initial recommendation weight of each question is determined; Based on the test paper compilation intention, the initial weights of each test question were adjusted to obtain the second recommended weights for each test question.

[0087] Specifically, the knowledge point type of each test question refers to the attribute of the knowledge point being tested. For example, the knowledge point type of each test question can include core knowledge points, frequently tested points, common mistakes, weak points, and newly added test points. Furthermore, each test question in the question bank can be pre-labeled with a corresponding type tag, and the corresponding knowledge point type can be obtained based on the type tag of each test question.

[0088] The initial recommendation weight refers to the degree of recommendation determined based on the knowledge point type of the test questions, without considering the user's personalized test paper creation intention. For example, for test paper types of "local high-frequency test point practice", high-frequency test points can be assigned a higher initial recommendation weight.

[0089] The initial recommendation weight for each question can be determined based on preset rules. For example, the initial recommendation weight is set to 0.8 for core knowledge points; 0.7 for frequently tested points; 0.6 for error-prone points; 0.5 for weak points; and 0.4 for newly added test points. Alternatively, an expert system can be used to assign different initial recommendation weights to questions of different knowledge point types based on factors such as the importance of the knowledge point, exam frequency, and student mastery. For instance, the expert system can set different initial recommendation weights for questions of different knowledge point types based on the analysis of past exam papers.

[0090] Furthermore, considering that the initial recommended weights for each question only take into account the knowledge point type of the question itself, without taking into account the user's personalized test paper creation intention, if the test paper is created solely based on the initial recommended weights for each question, the generated test paper may fail to meet the user's specific needs. For example, the user may want to focus on testing a specific knowledge point, or may want to avoid testing a knowledge point that they have already mastered, resulting in a poorly targeted test paper that cannot effectively help the user improve their learning outcomes.

[0091] Based on this, embodiments of the present invention adjust the initial weights of each test question according to the test paper compilation intention, so that the obtained second recommended weights can reflect both the importance of the knowledge point type of the test question itself and meet the user's personalized test paper compilation intention. For example, if the user specifies that the "kinetic energy theorem" should be emphasized, the recommended weight of test questions that test the knowledge point related to the "kinetic energy theorem" can be increased; if the user specifies that a certain knowledge point that has already been mastered should be avoided, the recommended weight of test questions that test the knowledge point related to that knowledge point can be decreased.

[0092] As an optional implementation, keywords can be extracted first according to the user's specified test paper creation intent; then, the similarity between the description text of each test question and these keywords can be calculated; next, the initial recommendation weight of each test question can be adjusted according to the similarity. For example, the second recommendation weight can be determined using the following formula: Second recommendation weight = Initial recommendation weight × (1 + similarity); finally, the second recommendation weight of all test questions is normalized so that its value is between 0 and 1, thus obtaining the second recommendation weight of each test question.

[0093] Based on any of the above embodiments, and based on the test paper compilation intention, the recommended weight of each question under each test paper type is determined, including: Based on the intent to create the document, determine the user's document creation needs across multiple different dimensions; Based on the test paper generation requirements under various dimensions, the recommended weight of each question under each test paper generation type is determined.

[0094] Specifically, test paper requirements across different dimensions refer to users' specific requirements for the test paper in terms of knowledge points, difficulty, question types, and exam content. For example, a user might want the test paper to focus on the "kinetic energy theorem," with a "medium" difficulty level, including both "multiple-choice" and "problem-solving" questions, and containing "highly frequently tested local knowledge points." Because users may have different preferences and needs across various dimensions, test paper creation requires comprehensive consideration of these requirements to generate a test paper that truly meets user expectations. Otherwise, considering only one or a few dimensions may result in a test paper that fails to meet the user's overall requirements.

[0095] To further improve the accuracy and personalization of test paper generation, better meet users' needs in different dimensions, and avoid deviations between the generated test papers and users' expectations, this embodiment of the invention first determines the user's test paper generation needs in multiple different dimensions based on the test paper generation intention. This allows for a more comprehensive understanding of the user's test paper generation goals, a more accurate grasp of the user's test paper generation preferences, and a more effective improvement in the quality and efficiency of test paper generation, thus achieving truly personalized and intelligent test paper generation.

[0096] As an optional implementation, natural language processing (NLP) technology can be used to analyze the user's test-taking intentions, extract keywords and key information, and then map these keywords and key information to different dimensions to determine the user's test-taking needs across multiple dimensions. For example, named entity recognition (NER) technology can be used to identify entities such as knowledge points, difficulty levels, and question types in the user's intentions and categorize them into corresponding dimensions to determine the user's test-taking needs across multiple dimensions.

[0097] After determining the test paper requirements across different dimensions, a comprehensive score can be calculated for each question based on the user's needs in each dimension. This comprehensive score is then used as the recommendation weight for that question, resulting in the recommendation weight for each question across different test paper types. For example, a weighted summation method can be used to sum the scores of each question across different dimensions. The weights can be adjusted based on the user's emphasis on each dimension; if a user places particular emphasis on a certain dimension, its weight can be increased.

[0098] Considering that test questions in the question bank usually need to be pre-labeled, and the labeling information includes the knowledge points tested by the test questions, the difficulty level, the question type, etc., the test questions in the question bank can be pre-processed before determining the recommendation weight of each test question. For example, knowledge graph technology can be used to divide and associate the knowledge points tested by the test questions in a more granular way, so as to more accurately calculate the degree of matching between the test questions and user needs.

[0099] Specifically, a knowledge graph can be used to organize knowledge points into a network, establishing hierarchical relationships (e.g., "force" → "Newton's laws of motion" → "Newton's second law") and associative relationships (e.g., "kinetic energy theorem" and "law of conservation of energy"). This way, when a user specifies a knowledge point, it can automatically expand to related knowledge points, thereby broadening the range of questions and ensuring the knowledge point coverage of the exam. Furthermore, each question in the question bank can be associated with a knowledge point in the knowledge graph; one question can correspond to multiple knowledge points, or one knowledge point can correspond to multiple questions. The association weight can be determined based on the degree to which the question tests the knowledge point. Finally, the knowledge graph can be used for knowledge reasoning. For example, if a user wants to test their understanding of the "kinetic energy theorem," they can use knowledge reasoning to find questions that require the "kinetic energy theorem" to solve. This knowledge graph can support multi-dimensional knowledge point matching; for example, it can match not only the name of the knowledge point but also its concept, attributes, and application scenarios. Knowledge graphs can also be updated dynamically. For example, they can automatically adjust the relationships and weights between knowledge points based on data such as the user's question-answering behavior and exam scores, making the knowledge graph more relevant to the user's actual situation.

[0100] Based on any of the above embodiments, the test paper type carries a priority label; based on the recommendation weight of each question under each test paper type, test papers are generated to obtain the user's personalized test paper, including: Based on the recommended weight of each question under each test paper type, test papers are generated to obtain personalized test papers corresponding to each test paper type. Based on the priority tags of each test paper type, the personalized test papers corresponding to each test paper type are prioritized.

[0101] Specifically, the priority labels for each test set type are identifiers used to distinguish the importance or recommended order of different test set types. They represent which test set type best suits the user's current needs or learning goals among multiple available options. These priority labels can be determined based on user profile information. For example, based on user profile information, users can be divided into three categories: high-achieving, average, and low-achieving. For high-achieving users, the priority labels for test set types are: "Weakness-focused practice tests > Standard practice tests > Improvement tests, Common mistake-focused practice tests > Frequently tested point practice tests"; for average users, the priority labels are: "Weakness-focused practice tests > Common mistake-focused practice tests > Standard practice tests, Frequently tested point practice tests > Improvement tests"; and for low-achieving users, the priority labels are: "Weakness-focused practice tests > Common mistake-focused practice tests > Frequently tested point practice tests, Standard practice tests > Improvement tests".

[0102] In addition, by combining users' historical learning data, test paper creation records, and feedback information, machine learning models can be used to predict users' preference for different test paper types, and this preference can be used as a priority label for test paper types. This embodiment of the invention does not specifically limit this.

[0103] Considering that different test paper types emphasize different assessment objectives and content, and that different users have different needs for different test paper types, if the priority of test paper types is not distinguished and all test paper types are treated equally, users may be at a loss when choosing test papers, or mistakenly choose test papers that do not meet their own needs, resulting in a poor user experience and failing to effectively improve learning outcomes.

[0104] Based on this, this embodiment of the invention first generates test papers based on the recommended weights of each question under each test paper type, obtaining test papers corresponding to each test paper type. Then, based on the priority tags of each test paper type, the test papers corresponding to each test paper type are prioritized, such as sorting the test papers corresponding to each test paper type in descending order of priority. This can more effectively guide users to select the test paper that best suits their needs, improve the utilization rate and learning effect of the test papers, better meet the personalized learning needs of users, and avoid users feeling confused or making incorrect choices when selecting test papers. The test paper corresponding to each test paper type refers to the set of questions generated according to the recommended weights of that test paper type. It can be a complete test paper or a question package containing multiple questions; this embodiment of the invention does not specifically limit this.

[0105] For example, if three types of test papers are generated simultaneously for a high-achieving user, namely "Weaknesses Practice Test", "Standard Test", and "Advanced Test", then when the test papers are displayed to the user, the "Weaknesses Practice Test" will be placed first, because this type better meets the current learning needs of the high-achieving user.

[0106] Understandably, after prioritizing the personalized test papers corresponding to each test paper type, personalized test papers can be recommended to users based on this priority, or displayed in the test paper list according to priority order. This can guide users to prioritize the personalized test papers that best suit their needs, improving the utilization rate of the test papers and the learning effect. For example, the personalized test papers with the highest priority can be prominently displayed on the homepage, or personalized test papers can be arranged in priority order in the test paper list.

[0107] Based on any of the above embodiments, test papers are generated according to the recommendation weights of each question under each test paper type to obtain the user's personalized test paper, and then the process further includes: Receive user tuning instructions; Based on the optimization information carried by the optimization instructions, the recommended weight of each question under each test paper type is optimized; Based on the optimized recommendation weights of each question, a personalized test paper is generated.

[0108] Specifically, a user's optimization command refers to an instruction issued by the user to adjust the test paper composition strategy when the user is dissatisfied with the generated test paper. This is typically generated after the user previews or uses the generated test paper. For example, a user might feel the test paper is too difficult, the knowledge point coverage is insufficient, or the question types do not meet their needs, thus issuing an optimization command. The generation methods for optimization commands can include: ① The user manually adjusting parameters such as the difficulty coefficient, knowledge point weight, and question type ratio of the test paper. ② The user replacing or deleting individual questions in the test paper. ③ The user providing natural language commands, such as "too difficult," "incomplete knowledge point coverage," or "question types do not meet requirements," automatically adjusting the test paper composition strategy based on the user's natural language commands. The above are examples of optimization command generation methods; this embodiment of the invention does not specifically limit the generation methods of optimization commands.

[0109] Upon receiving an optimization instruction, it indicates that the user is dissatisfied with the currently generated test paper and requests adjustments based on their feedback to generate a test paper that better meets their needs. In this case, based on the optimization information carried in the optimization instruction, the recommendation weight of each question under each test paper type is optimized. For example, if the user feels the test paper is too difficult, the recommendation weight of more difficult questions can be reduced, while the recommendation weight of less difficult questions can be increased. The optimization information refers to the information provided by the user in the optimization instruction to guide the adjustment of the test paper generation strategy. This information may include parameter values ​​manually adjusted by the user, question information replaced or deleted by the user, and the user's evaluation of the test paper.

[0110] As an optional implementation, the optimization instructions can be parsed first to extract optimization information; then, based on the optimization information, the recommended weights of each question under each test paper type can be adjusted. For example, if the user manually adjusts the difficulty coefficient, the recommended weights of questions at each difficulty level can be adjusted accordingly based on the adjustment ratio of the difficulty coefficient; finally, the adjusted recommended weights are normalized so that their values ​​are between 0 and 1, thereby optimizing the recommended weights of each question under each test paper type.

[0111] After optimizing the recommendation weights of each question under each test paper type, the test papers are reassembled based on the optimized recommendation weights to generate new test papers, resulting in optimized test papers. These optimized test papers better meet user needs and learning goals, thereby improving the user's learning experience and learning outcomes.

[0112] Understandably, if the optimized test paper still fails to fully meet the user's needs, the user can issue another optimization command, providing more detailed feedback information to generate another optimization command and continue to optimize the recommendation weight of each question under each test paper type, so that the optimized test paper becomes more and more in line with the user's actual situation, and finally generate a personalized test paper that fully meets the user's expectations.

[0113] Based on any of the above embodiments, intent analysis is performed on the volume creation command to obtain the user's volume creation intent, including: Based on the profile information, multiple candidate document grouping intentions were determined; Based on the volume creation command, the user's volume creation intent is determined from multiple candidate volume creation intents.

[0114] Specifically, candidate test paper intentions refer to the goals and content scope of test papers that a user might want to generate, inferred from the user's profile information. These are alternatives to the user's true test paper intentions. Since the profile information may include the user's grade level, subject, historical grades, weak knowledge points, frequently tested knowledge points, study time, current textbook version, and textbook volume, the user's learning stage and learning content can be analyzed based on the profile information to infer the user's learning goals and needs, thereby generating multiple candidate test paper intentions. For example, if the user is a third-grade elementary school student learning the "addition and subtraction" unit, it can be inferred that the user might need candidate test paper intentions such as "basic addition and subtraction practice" or "addition and subtraction word problem practice." If the user is a second-year junior high school student with the midterm exam approaching, it can be inferred that the user might need candidate test paper intentions such as "midterm review paper" or "weak knowledge point reinforcement paper," resulting in multiple candidate test paper intentions.

[0115] After identifying multiple candidate test paper creation intentions, a large-scale model can be used to perform semantic analysis on the test paper creation instructions, extracting keywords and key information contained within them. These keywords and key information are then matched with the candidate test paper creation intentions, calculating the similarity between each candidate intention and the instruction. The candidate intention with the highest similarity is selected as the user's true test paper creation intention. For example, if the user's input instruction is "The midterm exam is coming soon, please generate a review paper for me," the similarity between this instruction and candidate intentions such as "midterm review paper" and "weak knowledge point reinforcement paper" can be calculated, ultimately determining "midterm review paper" as the user's true test paper creation intention.

[0116] In addition, to improve the accuracy of intent recognition, a multi-turn dialogue approach can be used to interact with users, guiding them to choose from multiple candidate test paper intentions. For example, users can be shown candidate test paper intentions such as "Midterm Review Test" and "Weak Knowledge Point Reinforcement Test," and asked, "Which type of test paper would you prefer to generate?" The final test paper intention is determined based on the user's choice.

[0117] Figure 4 This is the second flowchart of the document assembly method provided by the present invention, as shown below. Figure 4As shown, the method includes: First, based on the user's login information in the learning device (such as a learning machine), determining the user's profile information, which includes the user's grade level, the current textbook version, and the textbook volume. Based on the user's profile information, determining multiple candidate test paper creation intentions, each corresponding to at least one test paper, such as questions like "Help me find a moving point on a number line," "Help me find the shortest path on an axis of symmetry," or "Help me find a customized midterm test paper for grade X (first semester) of textbook X." Next, the user selects the final test paper creation intention from the multiple candidate intentions, and generates the corresponding test paper based on that intention.

[0118] Figure 5 This is the third flowchart of the document assembly method provided by the present invention, as shown below. Figure 5 As shown, the method includes: displaying multiple function entry points (such as New Curriculum Standards, Customized Test Papers, Commonly Missed Questions, Weak Questions, and Frequently Tested Questions) on the display interface of the learning device (such as a learning machine). When a user selects any function entry point, it indicates that the user's intention to create a test paper is to select questions within the range represented by that function entry point. For example, if the user selects "New Curriculum Standards," it indicates that the user's intention is to generate a test paper that meets the requirements of the new curriculum standards. After the user selects any function entry point, based on the user's profile information, multiple candidate test paper creation intentions under the corresponding function entry point are displayed, allowing the user to select the final test paper creation intention from among the multiple candidate intentions. After determining the test paper creation intention, the intention is input into a large model, which then outputs the corresponding test paper.

[0119] For example, for "New Curriculum Standards," the corresponding candidate test paper intention could include "Find New Contextual Literacy Questions for Unit N," where Unit N is selected from "Units with New Curriculum Standards." For "Customized Test Papers," the corresponding candidate test paper intention could include "Find Customized Test Papers for Unit N," where Unit N can be displayed based on the user's semester month to match the user's learning progress as much as possible; for example, N would be displayed as 5 in January, 1 in February, 2 in March, 3 in April, and so on. For "Commonly Missed Questions," the corresponding candidate test paper intention could include "Find Practice Questions for Commonly Missed Questions in Unit N," where Unit N can also be displayed based on the user's semester month to match the user's learning progress as much as possible. For "Weak Questions," the corresponding candidate test paper intention could include "Help me find weak points in Unit N for focused practice." For "Frequently Tested Questions," the corresponding candidate test paper intention could include "Help me find frequently tested practice questions in Unit N."

[0120] Figure 6 This is the fourth flowchart of the document assembly method provided by the present invention, as shown below. Figure 6As shown, the method includes: obtaining user profile information and test paper creation intent. The determination of test paper creation intent can include: ① Voice dialogue: The user interacts with the system through voice, expressing their needs for the test paper, such as "I want to do a set of practice questions on the kinetic energy theorem"; the user's voice is converted into text through speech recognition technology, and the user's intent is further analyzed to extract keywords and key information. ② Text: The user directly enters text in the input box, expressing their needs for the test paper, such as "Generate a high school sophomore year final review paper"; the user's intent is analyzed through natural language processing technology to extract keywords and key information. ③ Preset Intents: Pre-set commonly used test paper generation intents, such as "Final Exam Review," "Strengthening Weak Knowledge Points," and "High-Frequency Exam Point Practice." Users can directly select these intents to quickly generate test papers. The test paper generation intents are input into the large model, which analyzes the user's profile, the difficulty level (easy, medium, and difficult) of each knowledge point in the question bank, the type of each knowledge point in the question bank (including frequently tested points, common mistakes, and weak points), and the local exam situation (including the overall exam situation in the user's region, and relevant knowledge points in important local midterm and final exams). The model then outputs the corresponding test paper. If the user is not satisfied with the output test paper, they can re-enter the test paper generation intent using the three methods mentioned above. For example, a multi-turn dialogue method can be used to input the test paper generation intent until a satisfactory test paper is obtained.

[0121] The following describes the paper-building device provided by the present invention. The paper-building device described below and the paper-building method described above can be referred to in correspondence.

[0122] Based on any of the above embodiments Figure 7 This is a schematic diagram of the structure of the roll-building device provided by the present invention, as shown below. Figure 7 As shown, the device includes: The receiving unit 710 is used to receive the user's volume assembly instruction; Analysis unit 720 is used to perform intent analysis on the volume creation command to obtain the user's volume creation intent; Customization Unit 730 is used to customize personalized test papers for users based on test paper creation intent and user profile information.

[0123] Based on any of the above embodiments, and based on the test paper creation intent and user profile information, a personalized test paper for the user is customized, including: Based on the user profile information, determine multiple volume types that match the user; Based on the purpose of test paper compilation, the recommended weight of each question under each test paper type is determined; Personalized test papers for users are generated based on the recommended weights of each question under each test paper type.

[0124] Based on any of the above embodiments, based on the profile information, multiple volume types matching the user are determined, including: Based on user profile information, multiple test paper types matching the user are determined, as well as the percentage of knowledge point difficulty under each test paper type.

[0125] Based on any of the above embodiments, and based on the test paper compilation intention, the weight of each question under each test paper type is determined, including: Based on the proportion of knowledge point difficulty in each test paper type, the first recommended weight of each question in each test paper type is determined. Based on the intent of test paper compilation, determine the second recommended weight for each question under each test paper type; The recommended weight for each question is obtained by combining the first and second recommendation weights of each question.

[0126] Based on any of the above embodiments, and based on the test paper compilation intention, the second recommended weight of each question under each test paper type is determined, including: Based on the knowledge point type of each question under each test paper type, the initial recommendation weight of each question is determined; Based on the test paper compilation intention, the initial weights of each test question were adjusted to obtain the second recommended weights for each test question.

[0127] Based on any of the above embodiments, the test paper type carries a priority label; based on the recommendation weight of each question under each test paper type, test papers are generated to obtain the user's personalized test paper, including: Based on the recommended weight of each question under each test paper type, test papers are generated to obtain personalized test papers corresponding to each test paper type. Based on the priority tags of each test paper type, the personalized test papers corresponding to each test paper type are prioritized.

[0128] Based on any of the above embodiments, test papers are generated according to the recommendation weights of each question under each test paper type to obtain the user's personalized test paper, and then the process further includes: Receive user tuning instructions; Based on the optimization information carried by the optimization instructions, the recommended weight of each question under each test paper type is optimized; Based on the optimized recommendation weights of each question, a personalized test paper is generated.

[0129] Based on any of the above embodiments, intent analysis is performed on the volume creation command to obtain the user's volume creation intent, including: Based on the profile information, multiple candidate document grouping intentions were determined; Based on the volume creation command, the user's volume creation intent is determined from multiple candidate volume creation intents.

[0130] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 8As shown, the electronic device may include a processor 810, a communications interface 820, a memory 830, and a communication bus 840, wherein the processor 810, communications interface 820, and memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a test paper creation method, which includes: receiving a user's test paper creation instruction; performing intent analysis on the test paper creation instruction to obtain the user's test paper creation intent; and customizing a personalized test paper for the user based on the test paper creation intent and the user's profile information.

[0131] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0132] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the test paper generation method provided by the above methods. The method includes: receiving a user's test paper generation instruction; performing intent analysis on the test paper generation instruction to obtain the user's test paper generation intent; and customizing a personalized test paper for the user based on the test paper generation intent and the user's profile information.

[0133] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the test paper generation method provided by the above methods, the method comprising: receiving a user's test paper generation instruction; performing intent analysis on the test paper generation instruction to obtain the user's test paper generation intent; and customizing a personalized test paper for the user based on the test paper generation intent and the user's profile information.

[0134] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0135] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating test papers, characterized in that, include: Receive the user's volume creation command; The semantic analysis of the volume creation command is performed using a large language model to obtain the user's volume creation intent; Based on the stated test paper creation intent and the user's profile information, a personalized test paper is customized for the user, including: Based on the profile information, determine multiple group types that match the user; Using adjustable parameters, the first and second recommendation weights of each question are weighted and summed to obtain the recommendation weight of each question. Based on the recommendation weights of each question, a personalized test paper is generated for the user. The adjustable parameters are determined based on the user's perception of the overall difficulty of the test paper and the examination of specific knowledge points. The first recommendation weight is used to characterize the degree of matching between the difficulty of the corresponding question and the expected difficulty distribution of the test paper type, and the second recommendation weight is used to characterize the degree of matching between the corresponding question and the user's explicitly specified test paper requirements. Before determining the recommendation weight of each test question, a knowledge graph is constructed based on the hierarchical and relational relationships between the knowledge points tested by the test questions in the question bank. Each test question in the question bank is associated with a knowledge point in the knowledge graph. The knowledge graph is used to match knowledge points according to the test paper assembly intention in order to determine the test questions corresponding to the test paper assembly intention from the question bank.

2. The method for assembling test papers according to claim 1, characterized in that, The process of customizing a personalized test paper for a user based on the test paper creation intent and the user's profile information includes: Based on the profile information, determine multiple group types that match the user; Based on the stated test paper composition intention, the recommended weight of each question under each test paper composition type is determined; Based on the recommendation weights of each question under each test paper type, a personalized test paper for the user is obtained.

3. The method for assembling test papers according to claim 2, characterized in that, The step of determining multiple group types matching the user based on the profile information includes: Based on the profile information, multiple test paper types matching the user are determined, as well as the percentage of knowledge point difficulty under each test paper type.

4. The method for assembling test papers according to claim 3, characterized in that, The determination of the recommended weight of each question under each test type based on the test paper composition intention includes: Based on the proportion of knowledge point difficulty in each test paper type, the first recommended weight of each question in each test paper type is determined. Based on the stated test paper composition intention, determine the second recommended weight for each question under each test paper composition type; The recommended weight for each question is obtained by combining the first and second recommendation weights of each question.

5. The method for assembling test papers according to claim 4, characterized in that, The determination of the second recommended weight for each question under each test paper type based on the test paper composition intention includes: Based on the knowledge point type of each question under each test paper type, the initial recommendation weight of each question is determined; Based on the stated test paper composition intention, the initial recommended weights of each test question are adjusted to obtain the second recommended weights of each test question.

6. The method for assembling a test paper according to any one of claims 2 to 5, characterized in that, The test paper type carries a priority tag; the process of generating test papers based on the recommendation weight of each question under each test paper type to obtain the user's personalized test paper includes: Based on the recommended weight of each question under each test paper type, test papers are generated to obtain personalized test papers corresponding to each test paper type. Based on the priority tags of each test paper type, the personalized test papers corresponding to each test paper type are prioritized.

7. The method for compiling test papers according to any one of claims 2 to 5, characterized in that, The process of creating personalized test papers for the user based on the recommendation weights of each question under each test paper type, followed by: Receive the user's tuning instructions; Based on the optimization information carried by the optimization instructions, the recommended weight of each question under each test paper type is optimized; Based on the optimized recommendation weights of each question, a personalized test paper is generated.

8. The method for compiling test papers according to any one of claims 1 to 5, characterized in that, The process of performing intent analysis on the volume creation command to obtain the user's volume creation intent includes: Based on the portrait information, multiple candidate grouping intentions are determined; Based on the volume creation instruction, the user's volume creation intent is determined from the plurality of candidate volume creation intents.

9. A roll-building device, characterized in that, include: The receiving unit is used to receive the user's volume assembly instructions; The analysis unit is used to perform semantic analysis on the volume creation command using a large language model to obtain the user's volume creation intent. A customization unit, used to customize a personalized test paper for the user based on the test paper creation intention and the user's profile information, includes: Based on the profile information, determine multiple group types that match the user; Using adjustable parameters, the first and second recommendation weights of each question are weighted and summed to obtain the recommendation weight of each question. Based on the recommendation weights of each question, a personalized test paper is generated for the user. The adjustable parameters are determined based on the user's perception of the overall difficulty of the test paper and the examination of specific knowledge points. The first recommendation weight is used to characterize the degree of matching between the difficulty of the corresponding question and the expected difficulty distribution of the test paper type, and the second recommendation weight is used to characterize the degree of matching between the corresponding question and the user's explicitly specified test paper requirements. Before determining the recommendation weight of each test question, a knowledge graph is constructed based on the hierarchical and relational relationships between the knowledge points tested by the test questions in the question bank. Each test question in the question bank is associated with a knowledge point in the knowledge graph. The knowledge graph is used to match knowledge points according to the test paper assembly intention in order to determine the test questions corresponding to the test paper assembly intention from the question bank.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the volume assembly method as described in any one of claims 1 to 8.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the volume assembly method as described in any one of claims 1 to 8.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the volume assembly method as described in any one of claims 1 to 8.

Images