Individualized clinical medicine online learning test paper generation method and system

Abstract

The application provides an individualized clinical medicine online learning test paper generation method, comprising the following steps: S1, using a first neural network to evaluate the test results of each student, and extracting a first version of the test paper from a candidate question bank according to the evaluation results; S2, adjusting the difficulty coefficients of each question of the first version of the test paper by using a second network according to the answering results of a plurality of students on the first version of the test paper, and extracting the questions from the candidate question bank to generate a second version of the test paper; S3, repeating step S2 to generate N versions of the test paper; and S4, adjusting the difficulty coefficients of the individualized test paper questions of each student by using a third neural network according to the answering results of each student on the first version to the Nth version of the test paper, and extracting the questions from the candidate question bank to generate the individualized test paper. The application can solve the technical problem that there is no individualized test paper generation technology in the prior art which can be customized according to the learning progress and different degrees of knowledge mastery of students.

Description

Technical Field

[0001] This invention relates to the field of clinical medical education technology, specifically to a method and system for generating personalized online clinical medical learning exam papers. Background Technology

[0002] In the prior art, the earlier patent application with publication number CN107590247A provides an intelligent test paper generation method based on group knowledge diagnosis. This method sets constraints for test paper generation according to exam requirements, performs group knowledge diagnosis to determine the knowledge points of the test paper, and extracts questions related to these knowledge points from a question bank to form a candidate question bank. Subsequently, a demand matrix is ​​generated based on the difficulty of the test paper, question types, and the calculated total number of questions in the constraints. A supply matrix is ​​generated based on the candidate question bank. The demand matrix is ​​then subtracted from the supply matrix to obtain a difference matrix, which is used to correct the demand matrix. Based on the generated final demand matrix, questions are extracted from the candidate question bank to generate an initial population. Finally, the generated initial population is used to complete the test paper generation based on a genetic algorithm. However, the intelligent test paper generation method provided by the above technical solution does not fully consider the actual ability level of students when generating the test papers.

[0003] The prior patent application with publication number CN111737450A provides another method for test paper assembly. This method includes: receiving test paper assembly input data, which at least includes test center range information and exam class information; obtaining the corresponding class ability value based on the exam class information; recalling a set of candidate questions from a question bank that correspond to the test center range information and match the class ability value; and generating a preset number of candidate test papers from the set of candidate questions according to the test paper question sorting rules. Using this technical solution, high-quality test papers with evenly distributed test centers, reasonable difficulty, and reasonable arrangement can be automatically generated, demonstrating a higher level of intelligence.

[0004] However, the intelligent test paper generation methods and systems provided by the aforementioned technical solutions are all designed for formal or mock exams. The difficulty level of these tests is tailored to students who have already completed systematic learning of a subject. For certain disciplines, such as clinical medicine, the duration of study is long, the content is extensive, and the knowledge points required for assessment are complex, making it difficult to complete systematic learning in a short period. It is well known that practicing more questions and taking more tests can strengthen understanding and mastery of knowledge points, and is one of the effective methods for quickly mastering a new subject. However, current intelligent test paper generation methods and systems often produce test papers that are ineffective for beginners or students who have not yet completed their studies, lacking specificity and providing poor practice results. This can even lead to students being unable to answer most questions, which can damage their confidence. Therefore, there is an urgent need for a method that can customize and generate individualized test papers based on students' learning progress and varying levels of knowledge mastery. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a method and system for generating personalized online learning test papers in clinical medicine, thereby improving the existing technology's lack of a way to customize and generate personalized test papers according to students' learning progress and varying levels of knowledge mastery.

[0006] The technical solution adopted in this invention is as follows: Firstly, a method for generating personalized online clinical medical learning exam papers is provided, including: S1. The first neural network is used to evaluate the exam scores of each student. Based on the evaluation results, the first version of questions is extracted from the candidate question bank to generate the first version of the test paper. The first version of the test paper is used by all students. S2. Based on the answers of multiple students to the first version of the test paper, the difficulty coefficient of each question in the first version of the test paper is adjusted using a second neural network. Questions are then extracted from the candidate question bank based on the adjusted difficulty coefficients of each question to generate the second version of the test paper. The second version of the test paper is available to all students. S3. Repeat step S2 to generate N versions of the test paper; S4. Based on each student's answers to the first to Nth versions of the test paper, the difficulty coefficient of each student's individualized test paper is adjusted using a third neural network. Based on the adjusted difficulty coefficient of each question, questions are extracted from the candidate question bank to generate individualized test papers.

[0007] Furthermore, the first neural network includes the DHNN neural network.

[0008] Furthermore, when training the DHNN neural network, each student's exam score is used as the input to the neurons, and the difficulty of each question in the exam paper is used as the output of the neurons and fed back to all neurons through connection weights.

[0009] Furthermore, the second neural network includes graph neural networks.

[0010] Furthermore, a graph neural network was used to adjust the difficulty level of each question in the first version of the exam, including: Based on the answers of multiple students to the first version of the test paper, a correlation graph was constructed between all questions on the test paper and the score of each student on each question. Obtain the question embedding vector and the score embedding vector from the question score association graph; Construct a knowledge point correlation graph among the questions, and obtain the knowledge point correlation embedding vector based on the knowledge point correlation graph; Update the knowledge point correlation graph to obtain node weights; The question embedding vector, score embedding vector, and knowledge point relevance embedding vector are used as inputs to a graph neural network. The network is trained based on node weights to obtain the score embedding vector of the hidden layer and the final question embedding vector. Based on the final question embedding vector and the score embedding vector of the hidden layer, the difficulty coefficient of a question for multiple students is predicted. The difficulty level of each question is adjusted based on the predicted difficulty level.

[0011] Furthermore, the third neural network is an LSTM neural network.

[0012] Furthermore, the LSTM neural network has a total of N+1 layers, where N is the number of test paper versions generated in the previous steps; each layer has 128 neurons, and the fully connected layer has 128 neurons; the activation function of the hidden layer is Softmax, the activation function of the fully connected layer is tanH, and the loss function is classification cross-entropy; the gradient descent algorithm is used to find the optimal solution during backpropagation.

[0013] Secondly, a personalized online clinical medicine learning exam paper generation system is provided to implement the personalized online clinical medicine learning exam paper generation method provided in the first aspect, including: The question extraction module is used to extract questions from the candidate question bank and generate test papers; The first version of the test paper difficulty coefficient analysis module and the first version of the test paper generation module include a DHNN neural network, which is used to evaluate the test scores of each student and transmit the evaluation results to the question extraction module. The question extraction module extracts the first version of questions from the candidate question bank based on the evaluation results to generate the first version of the test paper.

[0014] The Nth edition test paper difficulty coefficient analysis module includes a graph neural network, which is used to adjust the difficulty coefficient of each question in the first edition test paper based on the answers of multiple students. The adjusted difficulty coefficient of each question is then transmitted to the question extraction module. The question extraction module extracts questions from the candidate question bank based on the adjusted difficulty coefficient of each question to generate the second edition test paper. The individualized test paper difficulty coefficient analysis module includes an LSTM neural network, which adjusts the difficulty coefficient of each student's personalized test paper questions based on the answer results of each student's first to Nth versions of the test paper; and transmits the adjusted difficulty coefficient of each question to the question extraction module; the question extraction module extracts questions from the candidate question bank according to the adjusted difficulty coefficient of each question to generate the individualized test paper.

[0015] Thirdly, an electronic device is provided, comprising: one or more processors; a storage device for storing one or more programs; and when the one or more programs are executed by the one or more processors, causing the one or more processors to implement the personalized clinical medical online learning test paper generation method provided in the first aspect.

[0016] Fourthly, a computer-readable storage medium storing a computer program is provided, which, when executed by a processor, implements the personalized clinical medical online learning test paper generation method provided in the first aspect.

[0017] As can be seen from the above technical solution, the beneficial technical effects of the present invention are as follows: Individualized test papers can be generated based on students' learning progress and different levels of knowledge mastery. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0019] Figure 1 This is a schematic diagram of the test paper generation method according to Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the test paper generation system architecture of Embodiment 2 of the present invention. Detailed Implementation

[0020] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.

[0021] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application should have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0022] Example 1

[0023] This embodiment provides a method for generating personalized online clinical medical learning exam papers, such as... Figure 1 As shown, it includes: S1. Use a DHNN neural network to evaluate the exam scores of each student. Based on the evaluation results, select the first version of questions from the candidate question bank to generate the first version of the test paper. The first version of the test paper is used by all students.

[0024] DHNN (Deep Path Neural Network) is a discrete recurrent neural network characterized by the fact that the output of any neuron is fed back to all neurons as input through connection weights. The aim is to ensure that the output of any neuron is controlled by the outputs of all neurons, thus creating a mutual constraint between their outputs. In a specific implementation, each student's entrance exam score is used as the input to a neuron, and the difficulty of each question in the exam paper is used as the neuron's output, fed back to all neurons as input through connection weights. The difficulty of the questions in the first version of the exam paper can fully account for the learning gaps caused by individual differences among students.

[0025] DHNN neural network updates network weight matrix The formula is as follows:

[0026] In the above formula, k represents the number of questions in the first edition of the exam. This indicates the question with serial number s in the exam paper. , Let i and j represent the exam scores of the i-th and j-th students respectively, which are connected to the s-th question by weights; in the calculation, let i... j.

[0027] Based on the evaluation results, the first version of questions is drawn from the candidate question bank to generate the first version of the test paper.

[0028] In some embodiments, the difficulty levels of each question in the generated test paper can be graded. The method of difficulty grading is not limited and can be implemented in any feasible way in the prior art. For example, the difficulty coefficient can be set as high, medium and low, or it can be divided into 1 to 5 levels according to Arabic numerals, with level 1 being the lowest and level 5 being the highest.

[0029] In some embodiments, the question types in the first version of the test paper can be selected based on the units that students have already learned, so as to avoid including too many unlearned knowledge points in the test paper.

[0030] S2. Based on the answers given by multiple students to the first version of the exam, a graph neural network is used to adjust the difficulty level of each question in the first version of the exam. Questions are then extracted from the candidate question bank based on the adjusted difficulty levels to generate the second version of the exam. The second version of the exam is available to all students. In typical cases, when the first version of the exam is generated, there is no historical data to support the correlation between the difficulty level of each question and the learning level of each student, making it impossible to establish a relationship. This results in the difficulty levels of individual questions not necessarily being appropriate. Therefore, in this implementation, based on the feedback from multiple students who answered the first version of the exam, the difficulty levels of several questions were adjusted to generate a second version of the exam for all students. Details are as follows: S21. Based on the answers given by multiple students to the first version of the exam, construct a graph showing the relationship between all questions on the exam and each student's score on each question. Based on the answers given by multiple students to the first edition of the test paper, a question-score relationship graph can be constructed by treating each question on the test paper and each student's score on each question as nodes and the relationships between each question and these scores as edges.

[0031] S22. Obtain the question embedding vector and score embedding vector based on the question score association graph; S23. Construct a knowledge point correlation diagram among the questions, and obtain the knowledge point correlation embedding vector based on the knowledge point correlation diagram; S24. Update the knowledge point correlation graph to obtain node weights; in a specific implementation, the node weights are obtained using the following formula. :

[0032] In the above formula, yes Activation function This includes the relevance between the questions, scores, and knowledge points. yes The elements in the table. In subsequent calculations, the question weight, score weight, and knowledge point relevance weight can be expressed as follows: , and .

[0033] S25. Using the question embedding vector, score embedding vector, and knowledge point relevance embedding vector as input to the graph neural network, train the network according to the node weights to obtain the score embedding vector of the hidden layer and the final question embedding vector. Based on the final question embedding vector and the score embedding vector of the hidden layer, predict the difficulty coefficient of a question for multiple students.

[0034] S26. Based on the predicted difficulty coefficients, comprehensively adjust the difficulty coefficients of each question, and draw a second set of questions from the candidate question bank based on the adjusted difficulty coefficients of each question. In specific implementation methods, to avoid a large number of questions with excessively high difficulty levels in the automatically generated test papers, an evaluation standard can be set. For example, if 60% of students do not score on a certain question, it indicates that the question has an excessively high difficulty level. In this case, a similar question that tests the same knowledge point but has a lower difficulty level should be found from the test papers.

[0035] S27. Generate a second version of the test paper based on the second version of the questions; the second version of the test paper is available to all students.

[0036] By implementing the technical solution in this step, a second version of the test paper can be obtained, with the difficulty level of each question adjusted to suit the students' actual learning situation.

[0037] S3. Repeat step S2 to generate N versions of the test paper; In a specific implementation, based on the answers given by multiple students to the second version of the test paper, a graph neural network is used to comprehensively adjust the difficulty coefficient of each question in the second version of the test paper, resulting in multiple questions with the difficulty coefficient adjusted for the second time. The third version of the test paper is then generated based on these multiple questions. The third version of the test paper is used by all students. This process is repeated to generate N versions of the test paper as needed, where N is greater than 2.

[0038] By implementing the technical solution in this step, we can obtain the answer results data of multiple students, and use this answer results data as historical data for subsequent generation of personalized test papers.

[0039] S4. Based on each student's answers to the first to Nth versions of the test paper, use an LSTM neural network to adjust the difficulty coefficient of each student's individualized test paper questions. Based on the adjusted difficulty coefficient of each question, extract questions from the candidate question bank to generate individualized test papers.

[0040] LSTM (Laser-Swept Memory) neural networks are a type of recurrent neural network. When generating personalized exams for each student using an LSTM network, the difficulty coefficients of each question in the personalized exam can be linked to previous or subsequent versions of the exam, memorizing the difficulty coefficients of each question and applying them to the generation of the current personalized exam. Specifically, when generating personalized exams, the LSTM neural network takes the answers and difficulty coefficients of each question within a certain time period as input, correlates the answers and difficulty coefficients of each question at different points in time, and combines them according to the student's learning progress and knowledge level, forming different difficulty coefficients for each question type tailored to each student. Then, based on the adjusted difficulty coefficients of each question, questions are drawn from a candidate question bank to generate the personalized exam.

[0041] In this specific implementation, the LSTM neural network has N+1 layers, where N is the number of test paper versions generated in the previous steps. Considering that a test paper generally has no more than 100 questions, each layer has 128 neurons, and the fully connected layer has 128 neurons. The activation function for the hidden layers is Softmax, the activation function for the fully connected layers is tanH, and the loss function is classification cross-entropy. Gradient descent is used to find the optimal solution during backpropagation.

[0042] Example 2

[0043] This embodiment provides a personalized clinical medicine online learning exam paper generation system, used to implement the personalized clinical medicine online learning exam paper generation method provided in Example 1, including: The question extraction module is used to extract questions from the candidate question bank and generate test papers; the implementation method of the question extraction module is not limited and can be implemented in any feasible way in the existing technology.

[0044] The first version of the test paper difficulty coefficient analysis module and the first version of the test paper generation module include a DHNN neural network. The DHNN neural network is a trained model used to evaluate the test scores of each student and transmit the evaluation results to the question extraction module. The question extraction module extracts the first version of questions from the candidate question bank based on the evaluation results to generate the first version of the test paper.

[0045] The Nth edition exam paper difficulty coefficient analysis module includes a graph neural network, which is a trained model used to adjust the difficulty coefficient of each question in the first edition exam paper based on the answers of multiple students. The adjusted difficulty coefficients of each question are then transmitted to the question extraction module. The question extraction module extracts questions from the candidate question bank based on the adjusted difficulty coefficients of each question to generate the second edition exam paper. The individualized test paper difficulty coefficient analysis module includes an LSTM neural network, which is a pre-trained model used to adjust the difficulty coefficient of each student's personalized test paper questions based on the answer results of each student's first to Nth versions of the test paper. The adjusted difficulty coefficients of each question are then transmitted to the question extraction module. The question extraction module extracts questions from the candidate question bank based on the adjusted difficulty coefficients of each question to generate the individualized test paper.

[0046] The personalized clinical medicine online learning test paper generation method and system provided in the above embodiments can customize and generate personalized test papers according to students' learning progress and different levels of knowledge mastery.

[0047] Example 3

[0048] This embodiment provides an electronic device, including: One or more processors; Storage device for storing one or more programs; When one or more programs are executed by one or more processors, the one or more processors implement the personalized clinical medical online learning test paper generation method provided in Example 1.

[0049] Example 4

[0050] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the personalized clinical medical online learning test paper generation method provided in Example 1.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for generating personalized online clinical medical learning exam papers, characterized in that, include: S1. The first neural network is used to evaluate the exam scores of each student. Based on the evaluation results, the first version of questions is extracted from the candidate question bank to generate the first version of the test paper. The first version of the test paper is used by all students. S2. Based on the answers given by multiple students to the first version of the exam, a second neural network is used to adjust the difficulty level of each question in the first version of the exam. This includes: constructing a question score correlation graph for all questions in the exam and the score of each student on each question, based on the answers given by multiple students to the first version of the exam; obtaining question embedding vectors and score embedding vectors from the question score correlation graph; constructing a knowledge point correlation graph between questions and obtaining knowledge point correlation embedding vectors from the knowledge point correlation graph; updating the knowledge point correlation graph and obtaining the node weights using the following formula. : In the above formula, yes Activation function This includes the relevance between the questions, scores, and knowledge points. yes Elements in; The graph neural network is trained using question embedding vectors, score embedding vectors, and knowledge point relevance embedding vectors as inputs, based on node weights, to obtain the score embedding vectors of the hidden layers and the final question embedding vectors. Based on the final question embedding vectors and the score embedding vectors of the hidden layers, the difficulty coefficient of a question for multiple students is predicted. Based on the predicted difficulty coefficients, the difficulty coefficients of each question are adjusted. The second neural network is a graph neural network; Based on the adjusted difficulty level of each question, questions were drawn from the question bank to generate a second version of the test paper; the second version of the test paper is available to all students. S3. Repeat step S2 to generate N versions of the test paper; including: based on the answers of multiple students to the second version of the test paper, use a second neural network to comprehensively adjust the difficulty coefficient of each question in the second version of the test paper to obtain multiple questions with the difficulty coefficient adjusted for the second time, and generate a third version of the test paper based on the multiple questions; the third version of the test paper is used by all students; and so on, to generate the Nth version of the test paper, where N is greater than 2. S4. Based on each student's answers to the first to Nth versions of the test paper, the difficulty coefficient of each student's individualized test paper is adjusted using a third neural network. Based on the adjusted difficulty coefficient of each question, questions are extracted from the candidate question bank to generate individualized test papers.

2. The test paper generation method according to claim 1, characterized in that, The first neural network includes a DHNN neural network.

3. The test paper generation method according to claim 2, characterized in that, When training the DHNN neural network, each student's exam score is used as the input to the neurons, and the difficulty of each question in the exam paper is used as the output of the neurons and fed back to all neurons through connection weights.

4. The test paper generation method according to claim 1, characterized in that, The third neural network is an LSTM neural network.

5. The test paper generation method according to claim 4, characterized in that, The LSTM neural network has a total of N+1 layers, where N is the number of test paper versions generated in the previous steps; each layer has 128 neurons, and the fully connected layer has 128 neurons; the activation function of the hidden layer is Softmax, the activation function of the fully connected layer is tanH, and the loss function is classification cross-entropy; the gradient descent algorithm is used to find the optimal solution during backpropagation.

6. A personalized clinical medicine online learning test paper generation system, characterized in that, The method for generating personalized clinical medical online learning test papers according to any one of claims 1-5 includes: The question extraction module is used to extract questions from the candidate question bank and generate test papers; The first version of the test paper difficulty coefficient analysis module and the first version of the test paper generation module include a DHNN neural network, which is used to evaluate the test scores of each student and transmit the evaluation results to the question extraction module; the question extraction module extracts the first version of questions from the candidate question bank based on the evaluation results to generate the first version of the test paper. The second to Nth version exam difficulty coefficient analysis module includes a graph neural network. This network adjusts the difficulty coefficients of each question in the first version exam based on the answers from multiple students, and then transmits the adjusted difficulty coefficients to the question extraction module. The question extraction module then extracts questions from a candidate question bank based on the adjusted difficulty coefficients, generating the second version exam. It further uses a second neural network to comprehensively adjust the difficulty coefficients of each question in the second version exam based on the answers from multiple students, resulting in multiple questions with the second adjusted difficulty coefficients. These questions are then used to generate the third version exam, which is used by all students. This process continues until the Nth version exam is generated, where N is greater than 2. The individualized test paper difficulty coefficient analysis module includes an LSTM neural network, which adjusts the difficulty coefficient of each student's personalized test paper questions based on the answer results of each student's first to Nth versions of the test paper; and transmits the adjusted difficulty coefficient of each question to the question extraction module; the question extraction module extracts questions from the candidate question bank according to the adjusted difficulty coefficient of each question to generate the individualized test paper.

7. An electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the personalized clinical medical online learning test paper generation method according to any one of claims 1-5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for generating personalized clinical medical online learning test papers as described in any one of claims 1-5.

Images