Test question construction method and device, electronic equipment and storage medium

Abstract

The invention provides a test question construction method and device, electronic equipment and a storage medium, and belongs to the technical field of artificial intelligence, and the method comprises the steps: obtaining target knowledge points for performing test question construction for a target examinee and a recommendation weight of each target knowledge point; according to the recommendation weight of the target knowledge point, screening a first recommended test question set from a test question bank, and pushing related test questions to the target examinee; the target knowledge point and the recommendation weight of the target knowledge point are determined based on the learning ability portrait of the target examinee. According to the test question construction method and device, the electronic equipment and the storage medium provided by the invention, while the related target knowledge points for providing the test questions are determined through the learning ability portraits of the examinees, the recommendation weight of each target knowledge point is defined, so that the problems that the types and contents are relatively fixed during traditional test question construction and the construction efficiency is poor are solved. According to the method and the system, the defect that personalized requirements of medical examinees at different levels are difficult to meet is overcome, and personalized simulation test questions can be intelligently generated for each examinee accurately, efficiently and flexibly.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a test question construction method, device, electronic equipment and storage medium. Background Art

[0002] Educational examinations (such as medical education examinations) have extremely high requirements on the examinees' knowledge mastery and application capabilities.

[0003] The traditional test question generation method is often based on a fixed test question bank. It manually marks knowledge points, difficulty coefficients and other attributes, and constructs the test paper according to fixed rules (such as random question selection, chapter ratio, etc.). It lacks precise matching for individual differences, resulting in the difficulty of the test questions and content distribution not meeting the actual needs of the candidates, affecting the review efficiency and test scores.

[0004] Therefore, the question types and contents of traditional question banks are relatively fixed, making it difficult to meet the personalized needs of medical examinees of different levels. Summary of the invention

[0005] The present invention provides a test question construction method, device, electronic device and storage medium, which can use big data analysis technology to accurately evaluate the examinees' mastery of various knowledge points, and then can construct personalized test questions for different examinees based on their learning ability portraits, which can effectively improve the examinees' review efficiency.

[0006] The present invention provides a test question construction method, comprising the following steps: Obtain all target knowledge points for constructing test questions for target examinees and the recommended weight of each target knowledge point; Filtering a first recommended test question set from a test question bank according to the recommendation weight of each target knowledge point; Pushing the test questions in the first recommended test question set to the target examinee; The target knowledge points and the recommendation weight of each target knowledge point are determined based on the learning ability portrait of the target examinee, and the portrait label of the learning ability portrait records the dynamic mastery index of the target examinee on the knowledge points within the subject.

[0007] According to a test question construction method provided by the present invention, based on the learning ability portrait, the target knowledge points and the recommendation weight of each target knowledge point are screened out from the knowledge points in the subject, specifically including: Inputting the dynamic mastery index of the learning ability portrait into a knowledge point mining model to obtain a first knowledge point recommendation result output by the knowledge point mining model; Using the learning ability profile to perform collaborative filtering on all other candidates, and determining the second knowledge point recommendation result; Fusion the first knowledge point recommendation result and the second knowledge point recommendation result to determine all the target knowledge points and the recommendation weight of each target knowledge point; The knowledge point mining model is obtained by training a first initial network model based on a first training sample set, wherein the first training sample set is composed of a plurality of mastery index samples, and each of the mastery index samples carries a knowledge point recommendation result label.

[0008] According to a test question construction method provided by the present invention, the fusing of the first knowledge point recommendation result and the second knowledge point recommendation result to determine all the target knowledge points and the recommendation weight of each target knowledge point includes: Determining weight distribution of the first knowledge point recommendation result and the second knowledge point recommendation result; The weight distribution is used to perform weighted averaging processing on the first knowledge point recommendation result and the second knowledge point recommendation result to obtain the target knowledge point and the recommendation weight of each target knowledge point.

[0009] According to a test question construction method provided by the present invention, each test question in the test question bank is marked with a knowledge point identifier; The step of selecting a first recommended test question set from a test question bank according to the recommendation weight of each target knowledge point comprises: Sorting all the target knowledge points according to the recommendation weights from large to small; Keep multiple target knowledge points that are at the top of the ranking as valid knowledge points; Filter out valid test questions whose knowledge point identifier is any of the valid knowledge points from the test question bank; Some of the valid test questions are screened out to construct the first recommended test question set.

[0010] According to a test question construction method provided by the present invention, the step of screening out some of the valid test questions to construct the first recommended test question set includes: According to the question type category identifiers and difficulty coefficient identifiers of the valid test questions, a first number of valid test questions whose question type category is a target question type category are selected from all the valid test questions, and a second number of valid test questions whose difficulty coefficient is a target difficulty coefficient are selected; The first recommended question set is constructed based on the first number of valid questions and the second number of valid questions that are screened out.

[0011] According to a test question construction method provided by the present invention, the dynamic mastery index includes: the learning ability profile of the target examinee is determined based on the following method: Determining the dynamic mastery index of the target examinee, and determining a portrait label for constructing the learning ability portrait of the target examinee based on the dynamic mastery index; Constructing the learning ability portrait according to the portrait label; All knowledge points mastered, as well as at least one indicator of the initial mastery of each knowledge point, the last learning time, the learning interval, the current memory retention and the current basic retention rate.

[0012] According to a test question construction method provided by the present invention, determining the dynamic mastery index of the target examinee includes: Acquiring historical learning data of the target examinee; Performing statistical analysis on the historical learning data to determine key evaluation indicators related to the target examinee's mastery of each knowledge point in the subject, as well as the last learning time and learning interval of each knowledge point; Inputting the key evaluation indicators into the knowledge point mastery evaluation model, obtaining all the knowledge points mastered by the target examinee and the initial mastery of each knowledge point output by the knowledge point mastery evaluation model; Inputting the learning interval time into a forgetting curve model to obtain the current memory retention; determining the current base retention rate based on the initial mastery and the current memory retention; The knowledge point mastery evaluation model is obtained by training the second initial network model based on the second training sample set, wherein the second training sample set is composed of a plurality of evaluation index samples, and each of the evaluation index samples carries a mastery label.

[0013] According to a test question construction method provided by the present invention, when a portrait update triggering condition is met, the learning ability portrait is updated; The portrait update triggering conditions include one or more of the following conditions: Reaching a preset time interval, determining that the target candidate has completed a preset amount of learning content, collecting the target candidate's answer feedback, and updating the examination syllabus.

[0014] According to a test question construction method provided by the present invention, the test question bank is continuously updated based on the following methods: Collecting updated information within the subject at preset time intervals, wherein the updated information includes one or more of newly published research papers, subject hot spots, and syllabus change information; Generate updated test questions according to the updated information; Updating the test question bank using the updated test questions; Among them, newly published research papers are collected based on the dynamic expansion mechanism of the knowledge graph.

[0015] According to a test question construction method provided by the present invention, generating updated test questions according to the updated information includes: Extract key information of the update information and determine the knowledge point identifier of the update test question to be generated; Determining the question type category identifier of the updated test question; Setting a difficulty coefficient identifier of the updated test question, wherein the difficulty coefficient identifier is determined based on the knowledge point identifier and the test syllabus requirements; Input the key information, the knowledge point identifier, the question type identifier and the difficulty coefficient identifier as input data into a test question generation model, and obtain the updated test question output by the test question generation model; The test question generation model is obtained by training the third initial network model based on the third training sample set, and the third training sample set is composed of multiple input samples, each of which carries a test question label and a test question scoring label.

[0016] According to a test question construction method provided by the present invention, after pushing the test questions in the first recommended test question set to the target examinee, the method further includes: Collecting the target examinee's answer feedback, wherein the answer feedback includes at least one of the following: test question difficulty level feedback, learning performance feedback, and satisfaction level feedback; Based on the answer feedback, the recommendation weight of each target knowledge point is adjusted when constructing the next test question.

[0017] The present invention also provides a test question construction device, comprising the following modules: A knowledge point determination module is used to obtain all target knowledge points for constructing test questions for target examinees and a recommended weight for each target knowledge point; A test question set construction module, used for selecting a first recommended test question set from a test question bank according to the recommendation weight of each target knowledge point; A test question pushing module, used for pushing the test questions in the first recommended test question set to the target examinee; The target knowledge points and the recommendation weight of each target knowledge point are determined based on the learning ability portrait of the target examinee, and the portrait label of the learning ability portrait records the dynamic mastery index of the target examinee on the knowledge points within the subject.

[0018] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any one of the test question construction methods described above is implemented.

[0019] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the test question construction method as described in any one of the above is implemented.

[0020] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned test question construction methods.

[0021] The test question construction method, device, electronic device and storage medium provided by the present invention determine the relevant target knowledge points for providing test questions to the examinee through the examinee's learning ability portrait, and at the same time clarify the recommended weight of each target knowledge point, thereby solving the problem that the types and contents of traditional test questions are relatively fixed during construction, making it difficult to meet the personalized needs of medical examinees of different levels. The method, device, electronic device and storage medium provided by the present invention can accurately, efficiently and flexibly intelligently generate personalized simulation test questions for each examinee. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 It is a flow chart of the test question construction method provided by the present invention.

[0024] Figure 2 It is a flowchart of screening target knowledge points according to learning ability portraits and determining the recommendation weights of each target knowledge point provided by the present invention.

[0025] Figure 3 It is a flow chart of selecting a first recommended test question set from a test question bank according to the recommendation weight of each target knowledge point provided by the present invention.

[0026] Figure 4 It is a schematic diagram of the process of constructing a learning ability portrait provided by the present invention.

[0027] Figure 5 It is a schematic diagram of a flow chart for determining a dynamic mastering index provided by the present invention.

[0028] Figure 6 It is a curve schematic diagram of the forgetting curve model provided by the present invention.

[0029] Figure 7 It is a structural schematic diagram of the test question construction device provided by the present invention.

[0030] Figure 8 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] It should be noted that, in the description of the present invention, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0033] The terms "first", "second", etc. in the present invention are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of the same type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" means at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0034] The present invention proposes a test question construction method, device, electronic device and storage medium, which mainly aims to provide a solution that can generate personalized simulation test questions based on the candidate's learning ability profile by accurately evaluating the candidate's mastery of knowledge points, combining a dynamically updated question bank and an intelligent recommendation algorithm to tailor simulation test questions for the candidate, thereby improving review efficiency and test scores.

[0035] It should be noted that the test question construction method, device, electronic device and storage medium proposed in the present invention have wide applicability and flexibility, and can be applied to various types of candidate groups and different educational examination fields. Specifically, candidates can be divided according to different learning stages, such as elementary school students, junior high school students, high school students, college students, etc., and can be further refined into specific semesters of each grade, such as the first semester of the first grade of elementary school, the second semester of the second grade of junior high school, etc. In addition, candidates can also be divided according to different professional needs, such as medical candidates, law candidates, engineering candidates, normal school candidates, etc. Whether it is a subject examination in the basic education stage or a professional qualification examination in the field of vocational education, the present invention can provide personalized test question construction services.

[0036] The core of this invention is to determine personalized test question construction strategies based on the learning ability portrait of the examinees. This innovative method is also applicable to other education stages and examination fields. It aims to provide accurate, efficient and flexible test question construction services for the majority of examinees, and meet the personalized needs of different examinee groups in different examination scenarios.

[0037] For ease of expression, in the subsequent embodiments, the field of medical education examinations will be taken as an example to illustrate that the present invention can accurately construct personalized simulation test questions for medical examinees based on their learning ability and knowledge mastery. For example, for examinees who are preparing for the practicing physician examination, targeted test questions can be generated based on their mastery of medical knowledge points, such as knowledge points in different disciplines such as physiology, pathology, and pharmacology, to help them review and prepare for the exam efficiently. However, it should be noted that this implementation method using medical examinees as an example is only one embodiment of the wide application scenario of the present invention, and does not mean that the scope of protection of the present invention is limited to this.

[0038] The execution subject of the test question construction method provided by the present invention can be a mobile terminal (such as a phone watch, a learning machine, etc.), a laptop computer, a desktop computer, or other electronic devices with functions such as data storage, data processing, and data transmission (such as a television, etc.), and the present invention does not make specific limitations on this. In the subsequent embodiments, for steps that are not specifically mentioned or are described in an omitted manner, the execution subject can be regarded as a test question system (referred to as the system) as an example.

[0039] Figure 1 It is a flow chart of the test question construction method provided by the present invention, such as Figure 1 As shown, it mainly includes but is not limited to the following steps: Step 11, obtaining all target knowledge points for constructing test questions for target candidates and the recommended weight of each target knowledge point.

[0040] Among them, the target knowledge points and the recommended weight of each target knowledge point are determined based on the learning ability portrait of the target candidates. The portrait label of the learning ability portrait records the dynamic mastery indicators of the target candidates on the knowledge points within the subject.

[0041] Target candidates refer to any candidate who needs to review test questions. The target candidates can be determined by active request or automatic push.

[0042] For example, the automatic request method can be used to determine the target candidates: when any candidate has the need to practice simulated test questions, he or she can actively initiate a question request to the test system through the online learning platform or mobile learning application. After the candidate logs in to the platform, he or she finds the simulated test question generation entrance in the personal center or practice module, clicks in and selects the corresponding test subject, question type and other preference settings, and the test system will receive the candidate's question request and determine him or her as the target candidate.

[0043] For another example, the automatic push method is used to determine the target examinees. The test system can automatically push simulation test questions to the target examinees regularly or irregularly based on the examinees' historical learning records and learning plans. For example, for an examinee who is reviewing according to a set learning plan, the test system automatically triggers the simulation test question generation process after the examinee completes a certain stage of learning tasks, and determines the examinee as a target examinee.

[0044] Knowledge points refer to knowledge points with relative independence and integrity in the subject content, and are the basic units of test questions. For example, in the field of medical education, knowledge points can be specific disease names, treatment methods, drug action mechanisms, etc., such as "diagnostic criteria for hypertension", "treatment methods for diabetes", "mechanism of action of penicillin", etc. In the field of law, knowledge points can be legal provisions, legal principles, legal systems, etc., such as "offer and promise in contract law", "legitimate defense system in criminal law", etc.

[0045] The test system can build a learning ability profile based on the historical learning data of each candidate. A distributed database system can be established to store the historical learning data of each candidate. These historical learning data mainly include: the candidate's answer record, correct rate, answer time, learning time, wrong question record, etc. on the online platform. It can be collected in real time through the online learning platform or mobile learning application.

[0046] By analyzing these historical learning data, we can dig out the candidates’ mastery of various knowledge points in the subject and form a profile label. For example, by analyzing the candidates’ performance in questions related to cardiovascular diseases, we can find out their mastery of knowledge points such as “diagnosis of coronary heart disease” and “treatment of heart failure”.

[0047] Specifically, the test system will count the dynamic mastery indicators such as the correct answer rate, number of answers, and study time distribution of each knowledge point. For example, it is counted in real time that a candidate's correct answer rate for the knowledge point of "diagnosis of hypertension" is 80%, the number of answers is 20, and the average study time is 10 minutes; the correct answer rate for the knowledge point of "treatment methods of diabetes" is 50%, the number of answers is 15, and the average study time is 15 minutes.

[0048] The recommended weight reflects the importance of the knowledge point to the target candidate's learning and is used to guide the subsequent test question screening. When calculating the recommended weight, factors such as the candidate's mastery of the knowledge point, the importance of the knowledge point in the test syllabus, and the candidate's learning preference can be comprehensively considered. For example, for knowledge points that the candidate has a low mastery and account for a large proportion in the test, a higher recommended weight will be given so that the candidate can focus on practicing.

[0049] The specific calculation method of the recommended weight can be: first determine the basic importance weight of each knowledge point according to the examination syllabus; then adjust the basic importance weight based on the candidate's mastery of the knowledge point (such as accuracy). The lower the mastery, the higher the final recommended weight.

[0050] The learning preferences of candidates can also be taken into consideration, such as combining the candidate's recent preference for a certain subject or knowledge point to appropriately adjust the recommended weight of the relevant knowledge point. For example, if the basic importance weight of the knowledge point "treatment methods for diabetes" in the examination syllabus is 0.6, and the candidate's mastery of this knowledge point is low (such as a correct rate of 50%), then according to the preset weight calculation rules, the recommended weight of this knowledge point will be adjusted to 0.7; if the candidate has a high preference for internal medicine subjects recently (such as a long study time, sufficient answer volume, etc.), then the recommended weight of the internal medicine related knowledge points will be reduced by 0.2.

[0051] Step 12: Filter out a first recommended test question set from the test question bank according to the recommendation weight of each target knowledge point.

[0052] The question bank is a database that stores a large number of medical test questions, covering all subject knowledge points, test questions with different difficulty coefficients and different question types. The sources of test questions can include real test questions from previous years, test questions written by medical education experts, and test questions transformed from the latest medical research results. For example, for medical education examinations, the test questions in the question bank can involve various disciplines such as basic medicine, clinical medicine, and preventive medicine, covering human anatomy, physiology, pathology, pharmacology, internal medicine, surgery, obstetrics and gynecology, pediatrics and many other course contents. For legal education examinations, the question bank covers test questions from multiple disciplines such as basic legal theory, constitutional law, civil law, criminal law, commercial law, and administrative law.

[0053] Furthermore, each test question has a corresponding record in the test question bank, which includes multiple fields such as test question content, knowledge point identification, question type identification, and difficulty coefficient identification. The knowledge point identification is used to clarify the knowledge points involved in the test question; the question type identification distinguishes different types of questions such as multiple-choice questions, fill-in-the-blank questions, short-answer questions, case analysis questions, and discussion questions; the difficulty coefficient identification is set according to factors such as the complexity of the test question and the depth of the knowledge points involved, and is generally divided into three levels: easy, medium, and difficult.

[0054] For example, the content of a medical multiple-choice question is "Which of the following drugs is the first-line drug for the treatment of hypertension, A. Nifedipine B. Nitroglycerin C. Propranolol D. Lidocaine", and its knowledge point is marked as "Drugs for the treatment of hypertension", the question type is marked as "Single-choice question", and the difficulty level is marked as "Medium". The content of a legal case analysis question is "A accidentally hit B, and B sued A in court for compensation. Please analyze whether A's behavior constitutes infringement, and what kind of civil liability A should bear", and its knowledge point is marked as "Tort Liability", the question type is marked as "Case Analysis Question", and the difficulty level is marked as "Difficult".

[0055] After obtaining the target knowledge points and their recommended weights, all target knowledge points can be sorted from high to low according to the recommended weights. This ensures that when screening test questions, priority is given to knowledge points that are more important to the candidates and urgently need to be strengthened. For example, if the candidate's learning ability portrait shows that the recommended weight of the target knowledge point "treatment of diabetes" is 0.7, the recommended weight of the target knowledge point "treatment of heart failure" is 0.6, and the recommended weight of the target knowledge point "diagnosis of hypertension" is 0.5, then these three target knowledge points will be processed in sequence according to the sorted order, and test questions will be screened in a targeted manner, that is, test questions that match the target knowledge points will be screened from the test question bank to construct the first recommended test question set.

[0056] Optionally, in order to meet the diverse examination requirements and the comprehensive review needs of candidates, the selected test questions will be further screened and combined in combination with the question type and difficulty coefficient to construct the first recommended test question set. For example, if the examination syllabus stipulates that multiple-choice questions account for a large proportion, and candidates need to strengthen their practice of medium-difficulty test questions, then during the screening process, the test questions that match the target knowledge points and have a multiple-choice question type and a medium difficulty coefficient will be given priority until the predetermined number of test questions is reached.

[0057] Specifically, the test system will traverse the test records in the test database, search for matching test questions for each target knowledge point, and filter them according to question type and difficulty requirements. For example, for the target knowledge point "treatment methods for diabetes", select from the test database several test questions with a medium difficulty coefficient and several test questions with a short answer type and a difficult difficulty coefficient, and then summarize these test questions to construct the first recommended test question set according to the predetermined number and proportion requirements of test questions.

[0058] Step 13: Push the test questions in the first recommended test question set to the target examinees. The selected push method is also set according to actual needs, such as one-time push, theme push, question-by-question push, etc.

[0059] One-time push means that the constructed first recommended test question set is pushed to the target candidates at one time for them to practice together. This push method is suitable for candidates who have a clear review plan and time arrangement, and can focus on completing a complete set of simulation test questions within a period of time. For example, the candidate plans to conduct a systematic simulation test practice on the weekend. After receiving the candidate's question request, the test system will push all the test questions in the first recommended test question set to the candidate at one time. One-time push can present a complete set of simulation test questions, including the order of test questions, question type distribution and time limit, etc., allowing candidates to practice in a simulated real test environment, which helps to improve candidates' test-taking ability and time management ability.

[0060] The theme push method is to push test questions to candidates in batches according to themes such as target knowledge points or question types. For example, test questions related to a certain systemic disease are pushed first, so that candidates can concentrate on consolidating and practicing the knowledge points of the disease system; or multiple-choice questions are pushed first, and then case analysis questions are pushed. This method helps candidates to review and master knowledge gradually and in depth, and avoid confusion and pressure caused by receiving a large number of test questions of different question types at one time. Theme push can be flexibly arranged according to the learning progress and needs of candidates. For example, for medical candidates, when reviewing internal medicine courses, multiple-choice questions and short-answer questions related to digestive system diseases are pushed first. After the candidates have completed the practice and review of this knowledge point, relevant test questions about cardiovascular system diseases are pushed.

[0061] The push channels can be online learning platforms, mobile learning application push, etc. Taking the push of online learning platforms as an example, the test questions are directly displayed to the candidates through the message center, practice module and other channels of the online learning platform used by the candidates. After the candidates log in to the online learning platform, they can view and answer the pushed test questions in the corresponding module. At the same time, the online learning platform will provide auxiliary tools such as answering interface, timing function, automatic saving of answers, etc., to facilitate candidates to take mock exams. For example, the medical education online platform will provide candidates with a special simulation practice module. After entering the module, the candidates can see the list of pushed test questions. Click on each test question to answer it on the special answering interface. The timer of the online learning platform will record the candidate's answering time in real time. When the candidate finishes answering the question, the answer will be automatically saved. The candidate can check his answer record and score evaluation results at any time.

[0062] Suppose a medical student named Xiao Li is preparing for the medical qualification examination. He submits a question request on the medical education online platform, selects the subject of "Internal Medicine" for simulation practice, and specifies the question types including multiple-choice questions, short-answer questions and case analysis questions.

[0063] After receiving Xiao Li's question request, the test system determines that Xiao Li is the target candidate, retrieves Xiao Li's learning ability portrait, and determines all target knowledge points for test question construction and the recommended weight of each target knowledge point based on the learning ability portrait.

[0064] Then, select the test questions corresponding to these target knowledge points from the test question bank. For example, the test questions matching the target knowledge point of "Complications of Diabetes" in the test question bank have various question types and difficulty levels. At this time, according to Xiao Li's learning preferences (selected multiple-choice questions, short-answer questions and case analysis questions) and the requirements of the exam syllabus, select several test questions with a medium difficulty coefficient of multiple-choice questions and several test questions with a difficult difficulty coefficient of case analysis questions.

[0065] Finally, these screened test questions are combined into a first recommended test question set, for example, including 20 multiple-choice questions, 3 short-answer questions and 2 case analysis questions.

[0066] You can choose a one-time push method and push the test questions in the first recommended test set to Xiao Li through the practice module of the online learning platform. After Xiao Li logs in to the online learning platform, he sees the pushed test questions in the practice module and clicks to enter the answering interface to start the simulation practice. During the answering process, the timing function of the online learning platform starts to run and records Xiao Li's answering time. After Xiao Li finishes answering the questions, the online learning platform automatically saves the answers and scores Xiao Li's answers according to the preset scoring standards. At the same time, it provides detailed answer analysis and wrong question analysis reports to help Xiao Li understand his answering situation and knowledge mastery, and provide guidance for subsequent review.

[0067] The test question construction method provided by the present invention determines the relevant target knowledge points for providing test questions to the examinee through the examinee's learning ability portrait, and at the same time clarifies the recommended weight of each target knowledge point, which solves the problem that the types and contents of traditional test questions are relatively fixed when constructing, and it is difficult to meet the personalized needs of medical examinees of different levels. It can accurately, efficiently and flexibly intelligently generate personalized simulation test questions for each examinee.

[0068] On the basis of the above embodiments, the present invention further optimizes the method for determining the target knowledge points and their recommendation weights. Specifically, by combining the knowledge point mining model and collaborative filtering processing, the target knowledge points can be screened out more accurately and their recommendation weights can be calculated.

[0069] Figure 2 is a flow chart of selecting target knowledge points according to learning ability profiles and determining the recommendation weights of each target knowledge point provided by the present invention, such as Figure 2 As shown, it mainly includes but is not limited to the following steps: Step 201 : input the dynamic mastery index of the learning ability portrait into a knowledge point mining model, and obtain the first knowledge point recommendation result output by the knowledge point mining model.

[0070] Among them, the knowledge point mining model can be a network model based on machine learning or deep learning technology, which is mainly used to extract target knowledge points and related recommendation weights from the candidate's learning ability portrait, and specifically predict the candidate's mastery of each target knowledge point and learning needs by analyzing the candidate's dynamic mastery indicators (such as knowledge point mastery, learning time, answer accuracy, etc.).

[0071] For example, in a medical education exam, dynamic mastery indicators can include the correct rate of answers, number of answers, and study time of candidates for different medical knowledge points. For example, the correct rate of answers for the target knowledge point "Hypertension Treatment" is 60%, the number of answers is 15, and the study time is 10 minutes. These data can be input into the knowledge point mining model as dynamic mastery indicators.

[0072] The knowledge point mining model is obtained by training the first initial network model based on the first training sample set. The first initial network model is a predefined machine learning or deep learning model used to process and analyze the candidate's learning data, which can be a multilayer perceptron (MLP), convolutional neural network (CNN) or recurrent neural network (RNN).

[0073] The first training sample set is a set of multiple mastery index samples, each of which carries a knowledge point recommendation result label. These mastery index samples include indicators such as the examinee's correct answer rate, number of answers, and study time on each knowledge point, as well as the corresponding knowledge point recommendation result labels, such as specific knowledge points and corresponding recommendation weights.

[0074] The training process of training the first initial network model using the first training sample set includes: (1) Clean and normalize the mastery index samples in the first training sample set to ensure the accuracy and consistency of the data. For example, normalize the correct answer rate to the interval [0,1] and convert the learning time into standardized units (such as minutes).

[0075] (2) Select a suitable first initial network model and initialize its parameters. For example, select a multilayer perceptron model and initialize its weights and bias parameters.

[0076] (3) Use the first training sample set to train the first initial network model. Adjust the model parameters through the back propagation algorithm so that the first initial network model can learn the mapping relationship between the input dynamic mastery index and the output knowledge point recommendation results. During the training process, the performance of the model can be evaluated by cross-validation and other methods, and the hyperparameters can be adjusted as needed.

[0077] (4) The validation set can be used to evaluate the trained knowledge point mining model, including calculating evaluation indicators such as accuracy, recall, and F1 value. Furthermore, the model can be optimized based on the evaluation results, such as adjusting the network structure, adding regularization terms, etc., to improve the prediction accuracy and generalization ability of the obtained knowledge point mining model.

[0078] Step 202, using the learning ability profile to perform collaborative filtering on all other candidates to determine the second knowledge point recommendation result.

[0079] Collaborative filtering is to find a group of candidates with similar learning patterns to the target candidates by analyzing the learning behaviors and knowledge points mastered by other candidates, so as to recommend target knowledge points to the target candidates. In medical education examinations, collaborative filtering can use data such as other candidates' answer records, learning time and accuracy on the same knowledge points to find other candidates with similar learning patterns to the target candidates. For example, if other candidates have a low accuracy rate in answering questions on the knowledge point of "hypertension treatment", but their accuracy rate improves after practicing certain questions, then these questions that these candidates have done will be recommended to the target candidates.

[0080] Specifically, the learning ability profile of the target examinee is used to perform collaborative filtering on all other examinees to determine the second knowledge point recommendation result, and the implementation steps may include: (1) Calculate the similarity between the target candidate and other candidates on each knowledge point. The similarity can be calculated using methods such as cosine similarity and Pearson correlation coefficient. For example, calculate the similarity between the target candidate and each other candidate in terms of correct answer rate and study time on the knowledge point of "Hypertension Treatment".

[0081] (2) Find a group of candidates with similar learning patterns to the target candidate based on similarity. For example, select the top 10% of candidates as the similar candidate group.

[0082] (3) Analyze the learning effects of similar candidate groups on each knowledge point and determine the recommended knowledge points that are helpful to the target candidates. Specifically, the average mastery and learning effects of similar candidate groups on each knowledge point are counted to generate the second knowledge point recommendation results. For example, if the recommendation weight set by the similar candidate group when practicing the test questions of the knowledge point "Complications of Diabetes" is 0.7, and the correct rate is significantly improved after practicing the test questions, then the knowledge point "Complications of Diabetes" and its recommendation weight of 0.7 will be used as part of the second knowledge point recommendation results.

[0083] Step 203, the first knowledge point recommendation result and the second knowledge point recommendation result are integrated to determine the recommendation weights of all target knowledge points and each target knowledge point. The fusion method used may include weighted average, ensemble learning, etc. For example, a weight of 0.6 may be assigned to the first knowledge point recommendation result, and a weight of 0.4 may be assigned to the second knowledge point recommendation result, and the final target knowledge point recommendation weight may be calculated by weighted average.

[0084] The fused results include all target knowledge points and their recommendation weights. These recommendation weights combine the results of model prediction and collaborative filtering, and can more comprehensively reflect the importance of the knowledge points to the target candidates.

[0085] The test question construction method provided by the present invention combines the knowledge point mining model and the collaborative filtering processing. The knowledge point mining model is used to extract key information from the learning ability portrait of the examinee and predict the examinee's learning needs; the collaborative filtering utilizes the learning experience of other examinees to provide more targeted recommendations for the current examinee. This fusion method not only improves the accuracy of the recommendation, but also enhances the adaptability and flexibility of the system, and can better meet the personalized needs of different examinees.

[0086] As an optional embodiment, the present invention further optimizes the process of fusing the first knowledge point recommendation results and the second knowledge point recommendation results so as to more accurately determine the target knowledge points and their recommendation weights, specifically including: determining the weight distribution of the first knowledge point recommendation results and the second knowledge point recommendation results; using the weight distribution to perform weighted averaging on the first knowledge point recommendation results and the second knowledge point recommendation results to obtain all target knowledge points and the recommendation weights of each target knowledge point.

[0087] Among them, weight allocation refers to assigning different weights to the first knowledge point recommendation result and the second knowledge point recommendation result according to their reliability and importance. The purpose of weight allocation is to balance the contribution of the two recommendation results, so as to obtain more accurate target knowledge point recommendation weights. The first knowledge point recommendation result obtained based on the knowledge point mining model can reflect the personalized needs of candidates; the second knowledge point recommendation result obtained based on collaborative filtering processing can provide more extensive reference information. Therefore, reasonable weight allocation is crucial to comprehensively consider the advantages of both.

[0088] The present invention can determine the weight distribution between the first knowledge point recommendation result and the second knowledge point recommendation result by combining various means such as historical data evaluation, expert experience evaluation or experimental verification. Taking historical data evaluation as an example, the accuracy and reliability of the first knowledge point recommendation result and the second knowledge point recommendation result are evaluated by analyzing historical data. For example, the success rate of the two recommendation results in the past recommendations and the test takers' feedback are counted to define the weight distribution.

[0089] Furthermore, by using weight allocation (assuming that the weight of the first knowledge point recommendation result is 0.6 and the weight of the second knowledge point recommendation result is 0.4), the specific processing flow of weighted averaging the first knowledge point recommendation result and the second knowledge point recommendation result includes: For each target knowledge point, the first knowledge point recommendation result and the second knowledge point recommendation result are multiplied by their corresponding weights respectively, and then summed, specifically including: First, determine the recommendation weight of any target knowledge point in the first knowledge point recommendation result (assuming it is 0.5), and multiply it by the weight of the first knowledge point recommendation result itself, 0.6, to get 0.3. At the same time, determine the recommendation weight of any target knowledge point in the second knowledge point recommendation result (assuming it is 0.1), and multiply it by the weight of the second knowledge point recommendation result itself, 0.4, to get 0.04. Thus, it can be determined that the final recommendation weight of any target knowledge point after fusion is 0.34.

[0090] The test question construction method provided by the present invention considers personalized recommendations based on a knowledge point mining model and group experience based on collaborative filtering by determining weight distribution and performing weighted averaging processing, thereby avoiding the limitations of a single recommendation method and improving the accuracy and reliability of recommendations.

[0091] As an optional embodiment, how to select the first recommended test question set from the test question bank in the test question construction method provided by the present invention will be described in detail.

[0092] Each question in the question bank is pre-marked with a knowledge point identifier to clarify the specific knowledge points involved in the question. The knowledge point identifier is the key basis for question screening, ensuring that the screened questions accurately match the target knowledge points. For example, a multiple-choice question about "hypertension treatment" has a knowledge point identifier of "hypertension treatment"; a case analysis question about "diabetes complications" has a knowledge point identifier of "diabetes complications".

[0093] Figure 3 The flowchart of the present invention for selecting the first recommended test question set from the test question bank according to the recommendation weight of each target knowledge point is shown in 3, which specifically includes but is not limited to the following steps: Step 301, sorting all target knowledge points from large to small according to the recommended weight. The recommended weight reflects the importance of the target knowledge point to the target examinee. The higher the recommended weight, the more important the target knowledge point is to the target examinee's review work.

[0094] Assuming that the target knowledge points and their recommended weights are: the recommended weight of the target knowledge point "Complications of diabetes" is 0.76, the recommended weight of the target knowledge point "Treatment of heart failure" is 0.56, and the recommended weight of the target knowledge point "Diagnosis of hypertension" is 0.45, then after sorting them from high to low according to the recommended weights, the target knowledge points for the target candidates are ranked as follows: "Complications of diabetes" > "Treatment of heart failure" > "Diagnosis of hypertension".

[0095] Step 302, retaining multiple target knowledge points at the top of the ranking as valid knowledge points. According to the ranking result in step 301, multiple target knowledge points at the top of the ranking can be selected as valid knowledge points. The number of valid knowledge points can be flexibly set according to factors such as the candidate's study time, study plan, and examination requirements.

[0096] For example, if a candidate is about to take a comprehensive exam and is in a tight time schedule, the top five knowledge points with the highest recommended weights can be selected as valid knowledge points. In the above example, "Complications of diabetes" and "Treatment of heart failure" are selected as valid knowledge points.

[0097] Step 303, filter out valid test questions with knowledge point identification as any of the valid knowledge points from the test question bank. According to the sorting result, select multiple target knowledge points with the highest sorting as valid knowledge points. The number of valid knowledge points can be flexibly set according to factors such as the examinee's study time, study plan, and test requirements.

[0098] Step 304: Screen out some valid test questions and construct a first recommended test question set. It is necessary to ensure that the screened test questions accurately correspond to the target knowledge points to meet the needs of personalized recommendation.

[0099] For example, for the valid knowledge point "Complications of diabetes", all questions with the knowledge point marked as "Complications of diabetes" are filtered out from the question bank; for the valid knowledge point "Treatment of heart failure", all questions with the knowledge point marked as "Treatment of heart failure" are filtered out.

[0100] Furthermore, from the screened valid test questions, some test questions are further screened as needed to construct a first recommended test question set. When screening, factors such as the type and difficulty of the test questions can be considered to ensure the diversity and rationality of the test question set.

[0101] Optionally, while screening effective test questions, some other test questions can be added according to the requirements of the test syllabus and the learning needs of candidates. For example, if the test syllabus has specific requirements for the distribution of question types for certain knowledge points, some test questions for other knowledge points can be added on the basis of the screened effective test questions to ensure the comprehensiveness and balance of the test question set.

[0102] Assume that 10 questions matching the target knowledge point of "Complications of diabetes" and 8 questions matching the target knowledge point of "Treatment of heart failure" are selected from the question bank. According to the examination requirements and the candidate's study plan, 5 questions of "Complications of diabetes" and 3 questions of "Treatment of heart failure" are selected. At the same time, in order to ensure the comprehensiveness of the question set, 2 questions matching the knowledge point of "Diagnosis of hypertension" are added to finally construct the first recommended question set.

[0103] The test question construction method provided by the present invention can accurately screen out test questions matching the target knowledge points from the test question bank according to the recommended weights of the target knowledge points, and construct a first recommended test question set, which not only improves the accuracy of test question screening, but also ensures the diversity and rationality of the test question set, and can better meet the personalized learning needs of examinees and improve review efficiency.

[0104] The present invention further optimizes the process of screening valid test questions and constructing the first recommended test question set mentioned in step 304 of the above embodiment. The following is a detailed specific implementation method: based on the question type category identifier and difficulty coefficient identifier of the valid test questions, a first number of valid test questions with a target question type category are screened out from all valid test questions, and a second number of valid test questions with a target difficulty coefficient are screened out. Based on the screened first number of valid test questions and the second number of valid test questions, a first recommended test question set is constructed. Among them, the question type category identifier and the difficulty coefficient identifier are the labeling attributes of each test question in the test question library.

[0105] Question type identification is used to identify the question type. Common question types include multiple-choice questions, fill-in-the-blank questions, short-answer questions, case analysis questions, etc. Question type identification can help the test system filter out questions that meet specific question types based on the requirements of the test syllabus and the learning needs of candidates. For example, if multiple-choice questions account for a large proportion of the test syllabus, multiple-choice questions can be filtered out first.

[0106] The difficulty coefficient indicator is used to indicate the difficulty level of the test questions, for example, it is divided into three levels: easy, medium, and difficult. It can also be expressed as a numerical value (such as 0.1 to 1.0). The difficulty coefficient indicator can help the test system to filter out test questions that meet a specific difficulty level based on the level and learning stage of the examinees. For example, for examinees at the basic stage, test questions with a difficulty coefficient of "easy" can be filtered out first; for examinees at the advanced stage, test questions with a difficulty coefficient of "medium" or "difficult" can be filtered out.

[0107] Optionally, the first number of valid test questions with the question type category of the target question type category can be screened out, and the question type category to be screened can be determined based on the requirements of the examination syllabus and the learning needs of the examinees. For example, the examination syllabus has a large proportion of multiple-choice questions, and the test system can use multiple-choice questions as the target question type category. Assuming that the test system needs to screen out 5 multiple-choice questions, 5 test questions with the question type category of multiple-choice questions are further selected from the screened valid test questions.

[0108] Screening out a second number of valid test questions with a difficulty coefficient equal to the target difficulty coefficient can be performed by determining the difficulty coefficient to be screened based on the level and learning stage of the examinee (which can be determined based on the learning ability profile). For example, for examinees at the basic stage, the target difficulty coefficient can be set to "easy"; for examinees at the advanced stage, the target difficulty coefficient can be set to "medium" or "difficult". From all valid test questions, further screen out test questions with a difficulty coefficient equal to the target difficulty coefficient. For example, assuming that the system needs to screen out 3 test questions with a difficulty coefficient of "medium", 3 test questions with a difficulty coefficient of "medium" are further selected from the screened valid test questions.

[0109] Furthermore, the first number of valid test questions with the target test question type category and the second number of valid test questions with the target difficulty coefficient are integrated to form a first recommended test question set. During the integration process, it is also necessary to ensure the diversity of the first recommended test question set to avoid excessive concentration of test questions on a certain test question type or difficulty level. For example, if there are too many test questions of a certain test question type or difficulty level in the screened test questions, the number of screened questions can be appropriately adjusted to ensure the balance of the first recommended test question set constructed subsequently.

[0110] Finally, the integrated test questions are arranged in a certain order to generate a first recommended test question set. The order of the test questions in the first recommended test question set can be adjusted according to the requirements of the examination syllabus or the study plan of the examinee.

[0111] The test question construction method provided by the present invention can screen out test questions that meet specific requirements from valid test questions according to question type categories and difficulty coefficients, and use them to construct a first recommended test question set. This method not only improves the accuracy of test question screening, but also ensures the diversity and balance of the test question set, and can better meet the personalized learning needs of examinees and improve review efficiency.

[0112] Figure 4 is a flow chart of constructing a learning ability profile provided by the present invention, as an optional embodiment, such as Figure 4 As shown, the present invention further describes in detail how to determine the learning ability profile of the target examinee, which mainly includes the following implementation steps: Step 41, determining the dynamic mastery index of the target examinee, and determining a portrait label for constructing a learning ability portrait of the target examinee based on the dynamic mastery index.

[0113] Dynamic mastery indicators are a set of data used to describe the target candidates' mastery of each knowledge point. They are the basic data for constructing a learning ability portrait, including but not limited to the following: all knowledge points mastered, as well as the initial mastery of each knowledge point, the last learning moment, the learning interval, the current memory retention and the current basic retention rate.

[0114] Among them, all knowledge points mastered refer to the list of knowledge points that the candidates have learned or come into contact with; the initial mastery of each knowledge point refers to the candidate's initial mastery of each knowledge point, which can usually be measured by indicators such as the accuracy rate of answering questions and the number of answers; the last learning moment of each knowledge point is the specific moment when the candidate last learned a certain knowledge point; the learning interval time refers to the time interval between two adjacent studies of each knowledge point; the current memory retention refers to the candidate's memory retention degree of each knowledge point at the current moment; the current basic retention rate refers to the candidate's basic knowledge retention rate of each knowledge point at the current moment.

[0115] After obtaining the dynamic mastery indicators of the target candidates, we can build portrait labels for the learning ability portrait based on these indicators. Each portrait label corresponds to a knowledge point and its related dynamic mastery indicators, for example: Knowledge point name: Hypertension treatment; Initial mastery: 60%; Last learning moment: March 14, 2025; Study interval: 31 days; Current memory retention: 21.1%; Current base retention rate: 12.66%.

[0116] Step 42: construct a learning ability portrait based on the portrait tags. Specifically, all knowledge points and their corresponding dynamic mastery indicators can be combined to form a complete portrait tag set, that is, the user's learning ability portrait. For example, Xiao Zhang's learning ability portrait includes the following portrait tag set: Knowledge point name: Hypertension treatment; Initial mastery: 60%; Last learning moment: March 14, 2025; Study interval: 31 days; Current memory retention: 21.1%; Current base retention rate: 12.66%.

[0117] Knowledge point name: Complications of diabetes; Initial mastery: 50%; Last learning moment: March 10, 2025; Study interval: 35 days; Current memory retention: 18.5%; Current base retention rate: 9.25%.

[0118] The test question construction method provided by the present invention can comprehensively and accurately determine the learning ability portrait of the target examinee, and the learning ability portrait is constructed based on the dynamic mastery index, which can reflect the examinee's mastery of each knowledge point in real time, and provide accurate data support for personalized test question recommendation. This method not only improves the accuracy of test question recommendation, but also enhances the adaptability and flexibility of the system, can better meet the examinee's personalized learning needs, and improves review efficiency.

[0119] Figure 5 is a flow chart of determining a dynamic mastering index provided by the present invention, as an optional embodiment, such as Figure 5 As shown, the present invention further describes in detail how to determine the dynamic mastery index of the target examinee, and the following are detailed specific implementation steps: Step 401, obtaining the historical learning data of the target examinee, such as collecting the historical learning data of the target examinee through an online learning platform, including answer records, accuracy, answer time, learning time, wrong question records, etc.; collecting the examinee's learning data on mobile devices through mobile learning applications to ensure the comprehensiveness and real-time nature of the data. Other learning resources, such as the Learning Management System (LMS), online course platforms, etc., can also be used to collect the examinee's learning behavior data in different learning resources.

[0120] Optionally, the collected historical learning data of candidates may contain noise or incomplete records, so these historical learning data can be cleaned first, the purpose of which is to remove these noise and incomplete records to ensure the quality of the data. For example, remove records with abnormally short answering time (such as less than 1 second), because these records may be erroneous operations.

[0121] At the same time, some of the historical learning data can be normalized to keep the data ranges of different indicators consistent, which is convenient for subsequent analysis.

[0122] In addition, you can convert some non-numeric data into numeric data for mathematical calculations, for example, convert the last learning time into the number of days from the current time (learning interval).

[0123] Step 402, statistically analyze the historical learning data to determine key evaluation indicators related to the target examinees' mastery of each knowledge point in the subject, as well as the last learning time and learning interval of each knowledge point.

[0124] Among them, the key evaluation indicators mainly include the number of answers, correctness, and distribution of study time. Among them, the correctness of any knowledge point = (number of correct answers for any knowledge point / total number of answers) × 100%. The number of answers refers to the total number of answers for each knowledge point, which is used to indirectly understand the candidates' familiarity with each knowledge point; study time refers to the total study time of the candidates for each knowledge point, including the time spent reading the questions, giving answers, and reading the analysis.

[0125] The last learning time of each knowledge point records the specific time when the candidate last learned each knowledge point.

[0126] The learning interval is the time interval from the last learning moment to the current moment, which is used for subsequent memory retention calculations.

[0127] Step 403, inputting the key evaluation indicators into the knowledge point mastery evaluation model, obtaining all the knowledge points mastered by the target examinee and the initial mastery of each knowledge point output by the knowledge point mastery evaluation model.

[0128] The knowledge point mastery evaluation model is obtained by training the second initial network model based on the second training sample set. The second training sample set is composed of multiple evaluation index samples, each of which carries a mastery label. The specific training process includes: (1) First, the data in the second training sample set is cleaned and normalized to ensure the accuracy and consistency of the data.

[0129] (2) Select a suitable second initial network model and initialize its parameters. For example, select a multi-layer perceptron (MLP) model and initialize its weight and bias parameters.

[0130] (3) Using the second training sample set to train the second initial network model, adjusting the model parameters through the back propagation algorithm, so that the model can learn the mapping relationship between the key evaluation indicators of the input and the mastery degree of the output.

[0131] (4) Use the validation set to evaluate the trained model and calculate evaluation indicators such as accuracy, recall, and F1 value. Based on the evaluation results, optimize the model, such as adjusting the network structure and adding regularization terms, to improve the model's prediction accuracy and generalization ability.

[0132] (5) The trained model is saved as a knowledge point mastery evaluation model for subsequent mastery evaluation, including inputting the key evaluation indicators obtained from statistical analysis into the knowledge point mastery evaluation model. The knowledge point mastery evaluation model outputs the candidate's initial mastery of each knowledge point. For example, the candidate's initial mastery of the target knowledge point "Hypertension Treatment" is 60%.

[0133] Step 404, input the learning interval into the forgetting curve model to obtain the current memory retention. The forgetting curve model can be a mathematical model based on the Ebbinghaus forgetting curve, which is mainly used to calculate the current memory retention of each knowledge point of the examinee. The model can estimate the memory retention degree (generally expressed as a memory retention percentage) through the learning interval.

[0134] Figure 6 : is a curve diagram of the forgetting curve model provided by the present invention, wherein the horizontal axis represents time, specifically the learning interval time, and the vertical axis represents the memory retention percentage. Figure 6 Taking the example shown in the figure, assuming that the target candidate's learning interval on the target knowledge point of "Hypertension Treatment" is 31 days, then according to the forgetting curve model, it can be directly read that the current target candidate's memory retention of this target knowledge point is only 21.1%.

[0135] Step 405, determine the current basic retention rate based on the initial mastery and the current memory retention. The current basic retention rate can be calculated based on the initial mastery and the current memory retention. For example, if the candidate's initial mastery of the target knowledge point "Hypertension Treatment" is 60%, and the current memory retention is 21.1%, then the current basic retention rate is: 60% 21.1%=12.66%.

[0136] The test question construction method provided by the present invention can comprehensively and accurately determine the dynamic mastery indicators of the target examinees, including the initial mastery degree, current memory retention and current basic retention rate. These dynamic mastery indicators provide accurate data support for constructing learning ability portraits, ensuring the accuracy and real-time nature of learning ability portraits. The learning ability portraits constructed based on these dynamic mastery indicators can better reflect the examinees' mastery of various knowledge points, provide a solid foundation for personalized test question recommendations, and improve review efficiency.

[0137] In the traditional method of building learning ability portraits, the portraits are usually generated based on the learning data of the candidates at a certain stage, and once generated, they are no longer updated. This method has the following defects: On the one hand, the learning ability and knowledge mastery of candidates are dynamic. Over time, the candidate's mastery of certain knowledge points may increase or decrease, and the static learning ability portrait cannot reflect these changes, resulting in subsequent test questions that may no longer be in line with the candidate's actual learning status.

[0138] On the other hand, the traditional method of constructing learning ability portraits lacks real-time feedback on the candidates' learning behavior and cannot be adjusted in time according to the candidates' latest learning data, thus affecting the accuracy of personalized learning plans.

[0139] In order to solve the above problems, the present invention proposes a solution for real-time updating of learning ability portraits, mainly to update the learning ability portraits of candidates when it is determined that the portrait update trigger conditions are met.

[0140] Profile update trigger conditions refer to specific events or conditions that trigger the update of learning ability profiles, which may include but are not limited to the following: reaching a preset time interval, determining that the target candidates have completed a preset amount of learning content, collecting target candidates' answer feedback, and updating the examination syllabus.

[0141] Among them, reaching a preset time interval refers to automatically updating the learning ability portraits of all candidates every certain period of time (for example, every week) to regularly update the learning changes of each candidate during this time period. Determining that the target candidates have completed a preset amount of learning content is a way to update the learning ability portrait of each individual candidate. When the candidate completes a certain number of learning tasks (such as completing the study of a chapter, completing a certain number of test questions), the update of his learning ability portrait is triggered. The feedback on the answers of the target candidates is collected, which refers to the feedback on the answers uploaded to the test system by the candidates after completing the simulation test questions (such as test question difficulty, learning effect) as the basis for triggering the update of the learning ability portrait. In addition, when the test outline changes, the present invention will also promptly update the learning ability portraits of all candidates to ensure that the test question recommendations meet the latest test requirements.

[0142] After determining that the profile update triggering conditions for updating the learning ability profile of any candidate (such as the target candidate) are met, the update operation of the learning ability profile will be immediately executed, mainly including: Collect the target candidates' learning behavior data in real time, including answer records for each knowledge point, study time, error rate, etc., so as to update the portrait labels of each dimension in the learning ability portrait based on these learning behavior data, such as the mastery of each knowledge point, memory retention, basic retention rate, etc.

[0143] The present invention can ensure that the learning ability portrait always reflects the latest learning status of the examinee by updating the learning ability portrait in real time. This dynamic update mechanism not only improves the timeliness and accuracy of the learning ability portrait, but also enhances the adaptability and flexibility of the system, which can better meet the examinee's personalized learning needs and improve review efficiency.

[0144] Assume that a medical candidate named Xiao Zhang is preparing for the Medical Practitioner Qualification Examination. The test system collects Xiao Zhang’s historical learning data through the online learning platform and constructs an initial learning ability profile. The specific data is as follows: Knowledge point name: Hypertension treatment; Initial mastery: 60%; Last study date: March 14, 2025; Study interval: 31 days; Current memory retention: 21.1%; Current base retention rate: 12.66% After determining that the trigger conditions for the portrait update are met, such as Xiao Zhang completing a preset amount of learning content (completing the learning task of a chapter), retrieve Xiao Zhang's real-time learning behavior data, such as answering records (answering 10 questions on the target knowledge point of "Hypertension Treatment", and the accuracy of answering questions increased to 70%), and learning time (the total learning time on the knowledge point of "Hypertension Treatment" increased by 10 minutes, and the last learning time was April 15, 2025).

[0145] Based on the collected learning behavior data, update Xiao Zhang's learning ability profile, such as updating the knowledge point mastery, that is, based on the new answer record, update the mastery of the "hypertension treatment" knowledge point to 70%. Also, update the memory retention and basic retention rate, including: based on the new learning interval (20 days), use the forgetting curve model to recalculate the current memory retention to 25%, and the basic retention rate to 17.5%, then you can get the updated learning ability profile as follows: Knowledge point name: Hypertension treatment; Initial mastery: 70%; Last study date: April 15, 2025; Study interval: 20 days; Current memory retention: 25%; Current base retention rate: 17.5%.

[0146] Through the above-mentioned real-time update mechanism, the test question system can promptly reflect Xiao Zhang's learning progress on the knowledge point of "Hypertension Treatment", provide more accurate data support for subsequent personalized test question recommendations, and help Xiao Zhang review and prepare for the exam more efficiently.

[0147] In traditional question bank management systems, the update of the question bank often relies on manual regular sorting and updating. This approach has the following defects: (1) The manual update cycle is long, which makes it difficult to reflect the latest research results and changes in the teaching syllabus in a timely manner, often resulting in untimely updates.

[0148] (2) Manual updating may miss some important knowledge points or hot issues, resulting in the lag of the content of the question bank.

[0149] (3) Manual updating requires a lot of manpower and time, is inefficient, and cannot meet the rapidly changing educational needs.

[0150] As an optional embodiment, the present invention not only continuously updates the learning ability portrait of each examinee, but also continuously updates the question bank. The method mainly adopts a question bank continuous updating method based on the knowledge graph dynamic expansion mechanism. The specific implementation steps include: First, update information within the subject is collected at preset time intervals, wherein the update information includes one or more of newly published research papers, subject hot spots, and syllabus change information.

[0151] Reasonable preset time intervals can be set according to the characteristics and update frequency of each subject. For example, in the field of medicine, the question bank can be updated daily, weekly or monthly. The preset time interval can also be adjusted according to actual needs to ensure the timeliness and accuracy of the question bank.

[0152] For the collection of updated information, the latest medical research papers can be obtained in real time by connecting to the application programming interface (API) of academic databases such as PubMed. These medical research papers can provide the latest research results and knowledge updates.

[0153] In addition, by monitoring academic forums, professional websites and social media, you can collect hot issues and discussion topics within the current discipline, such as focusing on keywords that suddenly become popular (the number of times they are mentioned has increased fivefold within a week).

[0154] In addition, regularly obtain the latest syllabus revision information from the education authorities or authoritative institutions to ensure that the content of the question bank is consistent with the syllabus. These syllabus revision information usually includes the addition of new knowledge points, adjustments to the depth and breadth of knowledge points, etc.

[0155] After obtaining the updated information, key knowledge points and information are extracted by performing text mining and natural language processing on the collected updated information, such as extracting new treatment methods and drug action mechanisms from research papers.

[0156] Using named entity recognition (NER) technology, key entities in the medical field, such as disease names, drug names, gene names, symptom descriptions, etc., are extracted from research papers. For example, "lung cancer" (disease entity) and "pembrolizumab" (drug entity) mentioned in research papers are identified.

[0157] Relationship extraction techniques, such as dependency parsing and pattern matching, are used to determine the relationships between entities. For example, from the sentence "The study found that pembrolizumab has a significant therapeutic effect on lung cancer" in a research paper, the "treatment" relationship between "pembrolizumab" and "lung cancer" is extracted.

[0158] Furthermore, the extracted entities and relationships can be matched and integrated with the existing medical knowledge graph. If the corresponding entities and relationships already exist in the existing medical knowledge graph, their attribute information is updated, such as adding new research evidence, updating efficacy evaluation, etc.; if they do not exist, new nodes and relationships are created.

[0159] Furthermore, new test questions can be designed based on the extracted key information, including but not limited to the following steps: (1) Determine the test question type: Multiple-choice questions: single-choice questions or multiple-choice questions can be designed according to the characteristics and importance of each knowledge point. For example, for the knowledge point of "the mechanism of action of new anticancer drugs", a multiple-choice question can be designed: "Which of the following mechanisms is the main mechanism of action of the new anticancer drug pembrolizumab: A. Inhibit tumor cell DNA replication, B. Promote tumor cell apoptosis, C. Regulate the immune microenvironment of tumor cells, D. Block the energy metabolism pathway of tumor cells".

[0160] Short-answer questions are designed for knowledge points that require candidates to deeply understand and explain. For example, for the newly added knowledge point "Rare Disease Diagnosis Process", a short-answer question can be designed: "Please briefly describe the diagnostic process for patients suspected of rare diseases, including preliminary screening, laboratory tests, imaging tests, and genetic counseling."

[0161] Case analysis questions, combined with actual clinical cases, incorporate research results or new knowledge points. For example, for the knowledge point of "Clinical application of new anti-cancer drugs", you can design a case analysis question: "The patient, a 56-year-old male, was diagnosed with advanced non-small cell lung cancer and has received 2 cycles of pembrolizumab treatment. Please analyze the patient's treatment response and explain the advantages and precautions of pembrolizumab in the treatment of advanced non-small cell lung cancer."

[0162] (2) Question writing: Write the content of the test questions according to the determined test question types and the extracted key information. Ensure the accuracy, scientificity and standardization of the test questions. For example, when writing multiple-choice questions, ensure the interference and rationality between the options; when writing short-answer questions and case analysis questions, ensure the clarity and pertinence of the questions.

[0163] Set the difficulty coefficient for each question based on the importance, difficulty level and syllabus requirements of the knowledge points. For example, set a lower difficulty coefficient for questions on basic knowledge points; set a higher difficulty coefficient for questions on cutting-edge research results or complex knowledge points.

[0164] (3) Examination question review and verification: Expert review, such as organizing an expert team (invite medical education experts, clinicians, medical researchers, etc. to form an expert review team). These experts have rich teaching and clinical experience and are familiar with the latest medical research results and teaching syllabus requirements.

[0165] The review content is that experts will conduct a comprehensive review of the converted test questions, including the accuracy, scientificity, difficulty setting, and matching degree with knowledge points. For example, experts will evaluate whether the options of multiple-choice questions are reasonable and whether the cases of case analysis questions are representative.

[0166] Feedback and modification, that is, modifying and improving the test questions based on the expert's review opinions. For example, if an expert points out that there is confusion between the options of a multiple-choice question, the options will be modified in a timely manner.

[0167] Machine-assisted review, including review and consistency verification using automatic review tools. Among them, review by automatic review tools refers to the use of intelligent review tools to review the format, language specifications, etc. of test questions. These tools can check whether the test questions conform to the standardized format, whether there are grammatical errors or inappropriate wording, etc. Consistency verification refers to verifying the consistency and differentiation of new test questions with existing test questions in the question bank through comparison and analysis with the existing question bank. Ensure that the new test questions are consistent with the overall question bank in terms of knowledge point coverage, difficulty distribution, etc., while avoiding duplication of test question content.

[0168] Finally, the updated question bank is published using the online learning platform, including: publishing the updated question bank to the online learning platform or mobile learning application for candidates to use. Candidates can obtain the latest test questions through the platform, conduct simulation exercises and self-assessment.

[0169] At the same time, user feedback can be continuously collected, that is, when candidates use new test questions, their feedback can be collected, such as whether the test questions are of appropriate difficulty, whether they are helpful for knowledge understanding and application, etc. Based on the feedback, the test questions can be further optimized and adjusted.

[0170] The test question construction method provided by the present invention can realize the continuous updating of the test question bank and ensure the timeliness and accuracy of the content of the test question bank. The acquisition method based on the dynamic expansion mechanism of the knowledge graph can timely capture the latest research results and hot issues in the discipline and enrich the content of the test question bank. At the same time, through expert review and machine-assisted review, the quality and scientificity of the updated test questions are ensured. This method not only improves the updating efficiency of the test question bank, but also enhances the adaptability and flexibility of the test question system, and can better meet the personalized learning needs of candidates.

[0171] As an optional embodiment, the generating of updated test questions according to the updated information mainly includes but is not limited to: Extract key information of the update information, determine the knowledge point identifier of the updated test question to be generated; determine the question type category identifier of the updated test question; set the difficulty coefficient identifier of the updated test question, the difficulty coefficient identifier is determined based on the knowledge point identifier and the examination syllabus requirements; input the key information, knowledge point identifier, question type category identifier and difficulty coefficient identifier as input data into the test question generation model to obtain the updated test question output by the test question generation model.

[0172] Key information refers to the core content related to knowledge points extracted from updated information (such as newly published research papers, subject hot spots, changes in the syllabus, etc.), which is the basis for generating updated test questions. For example, the key information extracted from a research paper on a new anti-cancer drug may include the drug name, mechanism of action, clinical application, etc.

[0173] Knowledge point identification is a label used to identify the specific knowledge points involved in the test questions, ensuring that the generated test questions are closely related to the syllabus and examination requirements. For example, the knowledge point identification is determined to be "the mechanism of action of new anticancer drugs" from the extracted key information.

[0174] The method for determining the knowledge point identification is mainly to map the extracted key information to the knowledge point system in the syllabus. For example, the "mechanism of action of new anticancer drugs" is mapped to the knowledge point "Oncology-Treatment Methods". Medical education experts can review the extracted key information to ensure the accuracy and rationality of the knowledge point identification.

[0175] The question type category identifier is a label used to identify the type of test question. Common question types include multiple-choice questions, fill-in-the-blank questions, short-answer questions, case analysis questions, etc. For example, based on the characteristics of the knowledge points and the test requirements, the question type category is determined to be "multiple-choice questions" or "case analysis questions". You can choose the appropriate question type based on the complexity of the knowledge points and the teaching requirements. For example, for complex knowledge points (such as the mechanism of action of drugs), you can choose case analysis questions; for basic knowledge points (such as drug names), you can choose multiple-choice questions. In addition, you can also refer to the distribution requirements for question types in the test outline to ensure that the question type category distribution of the generated test questions meets the test standards.

[0176] The difficulty coefficient is a label used to indicate the difficulty of the test questions, which are usually divided into three levels: easy, medium, and difficult. For example, for test questions related to basic knowledge points, the difficulty coefficient can be set to "easy"; for test questions related to cutting-edge research results or complex knowledge points, the difficulty coefficient can be set to "difficult".

[0177] The present invention sets the difficulty coefficient mainly according to the importance and complexity of the knowledge point, but also combines the requirements of the examination syllabus. For example, for the knowledge point of "mechanism of action of new anticancer drugs", the difficulty coefficient can be set to "medium", and the distribution requirements of the difficulty coefficient in the examination syllabus are referred to to ensure that the generated test questions meet the requirements of the examination syllabus.

[0178] The test question generation model is a model based on machine learning or deep learning technology, which is used to generate test questions that meet the teaching syllabus and examination requirements. It generates corresponding test questions by analyzing the input key information, knowledge point identification, question type identification and difficulty coefficient identification.

[0179] The third training sample set used to perform pre-training on the test question generation model consists of multiple input samples, each of which carries a test question label and a test question score label. These input samples are used to train the test question generation model so that it can learn the mapping relationship between input data and output test questions. The specific training process includes: Data preprocessing: Clean and normalize the data in the third training sample set to ensure the accuracy and consistency of the data.

[0180] Model initialization: Select a suitable third initial network model and initialize its parameters. For example, select a multi-layer perceptron model and initialize its weight and bias parameters.

[0181] Training process: Use the third training sample set to train the third initial network model. Adjust the model parameters through the back propagation algorithm so that the model can learn the mapping relationship between input data and output test questions.

[0182] Model evaluation and optimization: Use the validation set to evaluate the trained model and calculate evaluation indicators such as accuracy, recall, and F1 value. Based on the evaluation results, optimize the model, such as adjusting the network structure and adding regularization terms, to improve the model's prediction accuracy and generalization ability.

[0183] Model saving: Save the trained model as a test question generation model for subsequent test question generation.

[0184] After completing the training of the test question generation model, the extracted key information, knowledge point identifier, question type identifier and difficulty coefficient identifier are input into the test question generation model as input data, and the model outputs the corresponding updated test questions.

[0185] For example, if the input data is "The main mechanism of action of the new anticancer drug pembrolizumab is to regulate the immune microenvironment of tumor cells" + "The mechanism of action of the new anticancer drug" + "Multiple-choice question" + "Difficulty coefficient identifier", then the question generation model will output the following question: "Which of the following mechanisms is the main mechanism of action of the new anticancer drug pembrolizumab? A. Inhibit tumor cell DNA replication, B. Promote tumor cell apoptosis, C. Regulate the immune microenvironment of tumor cells, D. Block the energy metabolism pathway of tumor cells".

[0186] The present invention provides a test question generation model based on machine learning that can automatically learn the mapping relationship between input data and output test questions and generate high-quality test questions. This method not only improves the efficiency of test question generation, but also enhances the adaptability and flexibility of the system, and can better meet the personalized learning needs of candidates.

[0187] As an optional embodiment, the present invention further describes in detail that after the test questions in the first recommended test question set are pushed to the target candidates, various methods are used to collect the answer feedback of each candidate, and the recommendation weight of the target knowledge point is adjusted in real time based on the answer feedback.

[0188] Specifically, after pushing the test questions in the first recommended test question set to the target examinees, it also includes: collecting the target examinees' answer feedback, so as to adjust the recommendation weight of each target knowledge point when constructing the next test questions based on the answer feedback.

[0189] Feedback on answers mainly includes but is not limited to feedback on the difficulty of test questions, feedback on learning performance and feedback on satisfaction. Feedback on the difficulty of test questions refers to the candidate's evaluation of the difficulty of the test questions, such as "the test questions are too difficult", "the test questions are too easy" or "the difficulty is moderate". Feedback on learning performance refers to the candidate's answer score after answering the test questions, such as the correct answer rate, answering time, etc. Satisfaction feedback refers to the candidate's overall satisfaction evaluation of the test questions, such as "very satisfied", "satisfied", "unsatisfied", etc. The candidate's answer feedback can be collected through the feedback function of the online learning platform, such as the online learning platform can provide questionnaires, scoring systems or comment areas for candidates to submit feedback. The candidate's answer feedback can also be collected through the feedback module of the mobile learning application, such as the mobile learning application can push feedback questionnaires or provide instant feedback functions. Of course, some traditional methods can also be used to collect answer feedback, such as sending emails or text messages to candidates to collect their feedback.

[0190] The collected feedback is classified and counted according to the difficulty, learning performance, satisfaction and other categories. For example, the proportion of candidates who think that "the test questions are too difficult" and the proportion of test questions that are obviously helpful in improving learning results can be counted. The relationship between the difficulty feedback and the candidate's ability level and the mastery of knowledge points can also be further analyzed. For example, if most candidates think that the test questions of a certain knowledge point are too difficult, analyze whether these candidates generally have a low mastery of this knowledge point.

[0191] Finally, the recommendation weight of each relevant target knowledge point can be adjusted based on the answer feedback, for example: If a large number of candidates report that the test questions on a certain knowledge point are too difficult or too easy, the recommended weight of the test questions on that knowledge point will be lowered in order to provide test questions of appropriate difficulty for candidates of different levels.

[0192] For those knowledge point questions that have been fed back to help improve learning outcomes, their recommendation weight will be increased; while for questions that have received poor feedback on learning outcomes, their weight will be reduced, prompting the system to recommend more useful questions.

[0193] Taking satisfaction feedback into comprehensive consideration, the corresponding recommendation weight will be increased for knowledge points or question types with high candidate satisfaction; the weight will be reduced for those with low satisfaction, and the direction of question recommendation will be optimized.

[0194] The present invention can realize dynamic optimization of test question recommendation by collecting the target examinees' answer feedback and adjusting the recommendation weight based on the feedback. This method not only improves the accuracy and adaptability of test question recommendation, but also enhances the feedback mechanism of the system to ensure that the recommended test questions always meet the examinees' actual learning needs.

[0195] Figure 7 : is a schematic diagram of the structure of the test question construction device provided by the present invention, such as Figure 7 As shown, the present invention also provides a test question construction device, which mainly includes but is not limited to: The knowledge point determination module 1 is used to obtain all target knowledge points for constructing test questions for target examinees and the recommended weight of each target knowledge point.

[0196] The test question set construction module 2 is used to select a first recommended test question set from the test question bank according to the recommendation weight of each target knowledge point.

[0197] The test question pushing module 3 is used to push the test questions in the first recommended test question set to the target examinee.

[0198] The target knowledge points and the recommendation weight of each target knowledge point are determined based on the learning ability portrait of the target examinee, and the portrait label of the learning ability portrait records the dynamic mastery index of the target examinee on the knowledge points within the subject.

[0199] It should be noted that the test question construction device provided by the present invention can execute the test question construction method described in any of the above embodiments during specific operation, which will not be elaborated in this embodiment.

[0200] The test question construction device provided by the present invention determines the relevant target knowledge points of the test questions to be provided to the examinee through the examinee's learning ability portrait, and clarifies the recommended weight of each target knowledge point, which solves the problem that the types and contents of traditional test questions are relatively fixed when constructing, and it is difficult to meet the personalized needs of medical examinees of different levels. It can accurately, efficiently and flexibly intelligently generate personalized simulation test questions for each examinee.

[0201] Figure 8 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 8 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the test question construction method, which includes: obtaining all target knowledge points for test question construction for the target examinee and the recommended weight of each target knowledge point; selecting a first recommended test question set from the test question bank according to the recommended weight of each target knowledge point; pushing the test questions in the first recommended test question set to the target examinee; wherein the target knowledge point and the recommended weight of each target knowledge point are determined based on the learning ability portrait of the target examinee, and the portrait label of the learning ability portrait records the dynamic mastery index of the target examinee on the knowledge points in the subject.

[0202] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0203] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the test question construction method provided by the above-mentioned embodiments, the method including: obtaining all target knowledge points for test question construction for target candidates and the recommended weight of each target knowledge point; according to the recommended weight of each target knowledge point, screening out a first recommended test question set from a test question bank; pushing the test questions in the first recommended test question set to the target candidate; wherein the target knowledge point and the recommended weight of each target knowledge point are determined based on the learning ability portrait of the target candidate, and the portrait label of the learning ability portrait records the dynamic mastery index of the target candidate on the knowledge points within the subject.

[0204] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the test question construction method provided by the above-mentioned embodiments, the method comprising: obtaining all target knowledge points for test question construction for target candidates and the recommended weight of each of the target knowledge points; screening out a first recommended test question set from a test question bank according to the recommended weight of each of the target knowledge points; pushing the test questions in the first recommended test question set to the target candidate; wherein the target knowledge points and the recommended weight of each of the target knowledge points are determined based on the learning ability portrait of the target candidate, and the portrait label of the learning ability portrait records the target candidate's dynamic mastery index of the knowledge points within the subject.

[0205] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0206] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0207] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing test questions, characterized in that: include: Obtain all target knowledge points for constructing test questions for target examinees and the recommended weight of each target knowledge point; Filtering a first recommended test question set from a test question bank according to the recommendation weight of each target knowledge point; Pushing the test questions in the first recommended test question set to the target examinee; The target knowledge points and the recommendation weight of each target knowledge point are determined based on the learning ability portrait of the target examinee, and the portrait label of the learning ability portrait records the dynamic mastery index of the target examinee on the knowledge points within the subject.

2. The test question construction method according to claim 1, characterized in that: According to the learning ability portrait, the target knowledge points and the recommendation weight of each target knowledge point are selected from the knowledge points in the subject, specifically including: Inputting the dynamic mastery index of the learning ability portrait into a knowledge point mining model to obtain a first knowledge point recommendation result output by the knowledge point mining model; Using the learning ability profile to perform collaborative filtering on all other candidates, and determining the second knowledge point recommendation result; Fusion the first knowledge point recommendation result and the second knowledge point recommendation result to determine all the target knowledge points and the recommendation weight of each target knowledge point; The knowledge point mining model is obtained by training a first initial network model based on a first training sample set, wherein the first training sample set is composed of a plurality of mastery index samples, and each of the mastery index samples carries a knowledge point recommendation result label.

3. The test question construction method according to claim 2, characterized in that: The fusing the first knowledge point recommendation result and the second knowledge point recommendation result to determine all the target knowledge points and the recommendation weight of each target knowledge point includes: Determining weight distribution of the first knowledge point recommendation result and the second knowledge point recommendation result; The weight distribution is used to perform weighted averaging processing on the first knowledge point recommendation result and the second knowledge point recommendation result to obtain the target knowledge point and the recommendation weight of each target knowledge point.

4. The test question construction method according to claim 1, characterized in that: Each test question in the test question bank is marked with a knowledge point identifier; The step of selecting a first recommended test question set from a test question bank according to the recommendation weight of each target knowledge point comprises: Sorting all the target knowledge points according to the recommendation weights from large to small; Keep multiple target knowledge points that are at the top of the ranking as valid knowledge points; Filter out valid test questions whose knowledge point identifier is any of the valid knowledge points from the test question bank; Some of the valid test questions are screened out to construct the first recommended test question set.

5. The test question construction method according to claim 4, characterized in that: The step of screening out some of the valid test questions and constructing the first recommended test question set includes: According to the question type category identifiers and difficulty coefficient identifiers of the valid test questions, a first number of valid test questions whose question type category is a target question type category are selected from all the valid test questions, and a second number of valid test questions whose difficulty coefficient is a target difficulty coefficient are selected; The first recommended question set is constructed based on the first number of valid questions and the second number of valid questions that are screened out.

6. The test question construction method according to any one of claims 1 to 5, characterized in that: The learning ability profile of the target candidate is determined based on the following method: Determining the dynamic mastery index of the target examinee, and determining a portrait label for constructing the learning ability portrait of the target examinee based on the dynamic mastery index; Constructing the learning ability portrait according to the portrait label; The dynamic mastery index includes: all the knowledge points mastered, and at least one of the initial mastery degree of each knowledge point, the last learning time, the learning interval time, the current memory retention and the current basic retention rate.

7. The test question construction method according to claim 6, characterized in that: Determining the dynamic mastery index of the target examinee includes: Acquiring historical learning data of the target examinee; Performing statistical analysis on the historical learning data to determine key evaluation indicators related to the target examinee's mastery of each knowledge point in the subject, as well as the last learning time and learning interval of each knowledge point; Inputting the key evaluation indicators into the knowledge point mastery evaluation model, obtaining all the knowledge points mastered by the target examinee and the initial mastery of each knowledge point output by the knowledge point mastery evaluation model; Inputting the learning interval time into a forgetting curve model to obtain the current memory retention; determining the current base retention rate based on the initial mastery and the current memory retention; The knowledge point mastery evaluation model is obtained by training the second initial network model based on the second training sample set, wherein the second training sample set is composed of a plurality of evaluation index samples, and each of the evaluation index samples carries a mastery label.

8. The test question construction method according to claim 7, characterized in that: Also includes: When a portrait update triggering condition is met, updating the learning ability portrait; The portrait update triggering conditions include one or more of the following conditions: Reaching a preset time interval, determining that the target candidate has completed a preset amount of learning content, collecting the target candidate's answer feedback, and updating the examination syllabus.

9. The test question construction method according to claim 1, characterized in that: The question bank is continuously updated based on the following methods: Collecting updated information within the subject at preset time intervals, wherein the updated information includes one or more of newly published research papers, subject hot spots, and syllabus change information; Generate updated test questions according to the updated information; Updating the test question bank using the updated test questions; Among them, newly published research papers are collected based on the dynamic expansion mechanism of the knowledge graph.

10. The test question construction method according to claim 9, characterized in that: The step of generating updated test questions according to the updated information comprises: Extract key information of the update information and determine the knowledge point identifier of the update test question to be generated; Determining the question type category identifier of the updated test question; Setting a difficulty coefficient identifier of the updated test question, wherein the difficulty coefficient identifier is determined based on the knowledge point identifier and the test syllabus requirements; Input the key information, the knowledge point identifier, the question type identifier and the difficulty coefficient identifier as input data into a test question generation model, and obtain the updated test question output by the test question generation model; The test question generation model is obtained by training the third initial network model based on the third training sample set, and the third training sample set is composed of multiple input samples, each of which carries a test question label and a test question scoring label.

11. The test question construction method according to claim 1, characterized in that: After pushing the test questions in the first recommended test question set to the target examinee, the method further includes: Collecting the target examinee's answer feedback, wherein the answer feedback includes at least one of the following: test question difficulty level feedback, learning performance feedback, and satisfaction level feedback; Based on the answer feedback, the recommendation weight of each target knowledge point is adjusted when constructing the next test question.

12. A test question construction device, characterized in that: include: A knowledge point determination module is used to obtain all target knowledge points for constructing test questions for target examinees and a recommended weight for each target knowledge point; A test question set construction module, used for selecting a first recommended test question set from a test question bank according to the recommendation weight of each target knowledge point; A test question pushing module, used for pushing the test questions in the first recommended test question set to the target examinee; The target knowledge points and the recommendation weight of each target knowledge point are determined based on the learning ability portrait of the target examinee, and the portrait label of the learning ability portrait records the dynamic mastery index of the target examinee on the knowledge points within the subject.

13. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the test question construction method as described in any one of claims 1 to 11 is implemented.

14. 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, the test question construction method as described in any one of claims 1 to 11 is implemented.

15. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the test question construction method as described in any one of claims 1 to 11 is implemented.

Images