AI-based automatic question setting experience method and system
Through the automatic question-setting experience method based on AI, the problem that self-students find it difficult to evaluate their own mastery is solved, effective training for weak parts is achieved, and learning effect and experience are improved.
Patent Information
- Application Number
- CN202510125890.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-27
- Publication Date
- 2025-06-13
AI Technical Summary
In the prior art, self-students find it difficult to accurately evaluate their mastery after learning, resulting in weak parts that are not trained effectively, affecting the learning effect.
Using an automatic question-making experience method based on AI, we automatically match the exercises by obtaining learning courses and learning progress, and updating the exercises based on the training results, and outputting better exercises by updating multiple times to ensure that the weak parts are fully trained.
Effectively evaluate students' learning situation, ensure that weak parts are trained, improve learning effectiveness and skill development, and optimize students' learning experience.
Smart Images
Figure CN120144696A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of AI artificial intelligence technology, and particularly to an AI-based automatic question generation experience method and system. Background Art
[0002] Nowadays, people have a variety of skill and knowledge certificates after leaving school. Most people choose to look for courses on the Internet to study. However, due to leaving the systematic education of school, some students with poor self-study ability are not sure about their mastery after choosing or finishing exercises, and can only roughly judge their mastery of knowledge or skills through past examination questions. However, they cannot fully master the knowledge points and examination points. Therefore, randomly training students or completely training students according to after-class exercises cannot enable students to fully master all the knowledge points or skill training of the course. Therefore, there is an urgent need for an AI-based automatic question generation experience method and system to solve the above problems. Summary of the Invention
[0003] The purpose of the present invention is to propose an AI-based automatic question generation experience method and system to solve one or more technical problems in the prior art, and at least provide a beneficial choice or creation condition.
[0004] An AI-based automatic question generation experience method, the method includes the following steps:
[0005] S100: Obtain the system learning courses and the learning progress of the courses;
[0006] S200: Automatically match exercises according to the learning courses and the learning progress of the courses;
[0007] S300: Statistically analyze the course learning situation of the trainees according to the exercise training results;
[0008] S400: Update the automatically matched exercises according to the course learning situation and conduct a second exercise test;
[0009] S500: Output more optimized automatically matched exercises through multiple updates of exercise practice.
[0010] In step S100, according to the feedback of the system background record on the system learning courses and the learning progress of the courses, classify the system learning courses into skill improvement and knowledge education, match different types of exercises for the trainees through different classifications, arrange the different types of exercises, and perform multi-dimensional algorithm matching on the new exercise types.
[0011] In step S200 of outputting automatically matched better exercises through multiple updates of exercise practice, classify and define the learning courses with tags, where the skill improvement is marked as S and the knowledge education is marked as K. Define the learning courses obtained by the system as courseX, where X is the classification and definition tag of the learning course, and calculate the classified courses using different matching algorithms.
[0012] In the learning courses marked as S, establish a weight sequence BL for automatically matching exercises for the course course. The method for optimizing the weight sequence X and establishing the weight sequence BL for automatically matching exercises for the course course is as follows:
[0013] Analyze the content learned by the user through the course progress P, match the after-class exercises through the content, define the first weight score for the completion of the after-class system, define the score of the user's completion of the second-matched exercises as the second weight score, integrate the first weight score and the second weight score to obtain the third weight score value, and integrate the three weight score values to obtain the first matching degree;
[0014] Optimize the first matching degree through the forward weight direction and the reverse weight direction, input the first matching degree for comparison into the neural network, and the network output result is jointly determined by the forward output and the reverse output. The calculation formula is as follows:
[0015]
[0016] Among them, x represents the input value, and the input value is the first matching degree. are respectively the forward weight direction and the reverse weight direction, and the forward weight and the reverse weight are respectively the first weight score and the second weight score. is the network forward output value, W 1 、U 1 、b 1 are respectively the input layer weight, hidden layer weight, and bias vector of the forward output. The input layer weight matrix is constructed by the weight ratios H1 and H2 of the file vector difference t1 and the secondary roof area difference t2. is the network reverse output value, W 2 、U 2 、b 2 are respectively the input layer weight, hidden layer weight, and bias vector of the reverse output. The input layer weight is the result obtained by comparing the top layer when judging the forward output and the reverse output. Then L1 is the input layer weight, L2 is the hidden layer weight, and the bias vector is and The sum of. If the result obtained by comparing the sub-top layer, then L2 is the input layer weight, L1 is the hidden layer weight, and the bias vector is and The sum of;
[0017] O t is the final output value, V and b 3 are the weight matrix and bias vector of the output layer respectively. The symbol is the concatenation operation;
[0018] Put the final output value O t into the neural network model for training. After the training is completed, the fast index X of the input data is output. The index X is the standard for judging whether the comparison data meets the qualification here. Each index X needs to be trained through a separate neural network model for the mapping relationship between the input data and this index. The result is output through the neural network model and the results of multiple output trainings are obtained to get the second matching degree.
[0019] Put the final output value O t into the neural network model for training. After the training is completed, the fast index X of the input data is output. The index X is the standard for judging whether the comparison data meets the qualification here. Each index X needs to be trained through a separate neural network model for the mapping relationship between the input data and this index. The result is output through the neural network model and the results of multiple output trainings;
[0020] In the neural network model, the number of neurons in the network input layer is equal to the number of input data types, and the number of neurons in the output layer is equal to the number of output data types. The number of intermediate layers and the number of neurons in each layer are freely set. The setting of the intermediate layer will directly affect the accuracy of the approximate data of the neural network. It is judged whether to increase or decrease the number of intermediate layers or neurons according to the training results. The calculation method for judging the number of intermediate layers and neurons is as follows:
[0021] Increment the intermediate layer in the neural network model layer by layer, and name the index X corresponding to the number of intermediate layers output. When the number of intermediate layers is 1, output the index X 1 , and so on. And according to the index X, construct a non-linear regression function, a is a constant value, calculate the variance D of the independent variable of the function, where L is the number of intermediate layers at the time of output, X i is the i-th index X of the output,
[0022] The X i is the input data corresponding to the i-th index X of the output. When , A takes the minimum value, where is to find the partial derivative of the function, A is the parameter for determining the number of intermediate layers. Compare the value of A with the variance D. If D > A, put the output index X into the input neural network model and increase the number of intermediate layers by 1. If D ≤ A, output the index X;
[0023] Update the weighting coefficients of the neurons by the gradient descent method:
[0024]
[0025] Among them, To calculate the partial derivative of the function, P and Q are respectively the system output error and the neuron weight increment, both of which are constant values, ω(l) is the neuron weighting coefficient, Δω(l) is the updated neuron weighting coefficient, δ represents the neuron learning rate, the number of neurons set in the intermediate layer is determined by the Δω(l), and the output value O is combined according to the superposition analysis algorithm t And perform deep learning on the algorithm.
[0026] Step 1: Calculate the first matching degree between the Q&A content of the exam questions in the question bank and the Q&A content of the current exam question, obtain the next question of the exam question with the highest first matching degree in the question bank as the candidate exam question, obtain the second matching degree through training by the convolutional neural network, calculate the second matching degree between the candidate exam question and the next question in the question bank, and if the second matching degree is greater than the first matching degree, output the candidate exam question;
[0027] Step 2: If the second matching degree does not satisfy that the second matching degree is greater than the first matching degree, re-match the exam question with the second highest first matching degree, and repeat Step 1;
[0028] Step 3: Encode the Q&A content of the current exam question according to the preset Q&A model to obtain the question encoding and answer encoding of the current exam question;
[0029] Encode the Q&A content of each exam question according to the preset Q&A model to obtain the question encoding and answer encoding of each exam question;
[0030] Calculate the question matching degree between the question encoding of each exam question and the question encoding of the current exam question;
[0031] Calculate the answer matching degree between the answer encoding of each exam question and the answer encoding of the current exam question;
[0032] Calculate the first matching degree according to the question matching degree and the answer matching degree;
[0033] The selection of the next question of the exam question with the highest first matching degree as the candidate exam question includes:
[0034] Select the exam questions in the Q&A content of all exam questions whose question matching degree is greater than the first preset threshold and whose answer matching degree is greater than the second preset threshold as the candidate exam questions;
[0035] Select the next question of the exam question with the highest first matching degree among the candidate exam questions as the candidate exam question.
[0036] Preferably, in step S400, the matching degree is updated by the results of the second exercise:
[0037] Obtain exercise data and assessment data. The keywords are the keywords in the exercise that increase scores or can be used for bonus statements. Define the assessment data as te and the exercise data as ft. Screen the keywords of the te and the exercise data and construct sequences tet and ftt. The tet is the keyword sequence of the assessment data, and the ftt is the keyword sequence of the exercise data. Count the total number of keywords T and L in the sequences tet and ftt, and calculate the keyword weight ratio. The tetK and tetK+1 are the k-th and (k + 1)-th elements in the sequence tet respectively, and the fttK and fttK+1 are the k-th and (k + 1)-th elements in the sequence ftt respectively. Calculate the index weights. According to the given rk and lk assignments, the weight values Q of the keywords in te and ft can be obtained. The Q1 is the weight value of the keywords in the assessment data, and Q2 is the weight value of the keywords in the assessment data. Input the weight values Q1 and Q2 to calculate the Word2Vec vector.
[0038] Calculating the weight values Q1 and Q2 includes calculating the ratio of the exercise data and the assessment data in the matching degree calculation through the ratio of adjacent keyword data, and then obtaining the weight values Q1 and Q2.
[0039] The method for calculating the Word2Vec vector is as follows:
[0040] Construct a weight matrix through the weight values of the keywords in the te and ft. Based on the TextRank algorithm, obtain the word vectors of the keywords in the te and ft through the weight matrix, and define the word vectors of the keywords as p(x, y). The x represents the weight value of the te keyword, and the y represents the weight value of the ft keyword. Calculate the objective weight of the core keyword according to the word vector p(x, y) of the keyword, and calculate the information entropy of the keyword. The calculation formula is:
[0041]
[0042] Among them, ei represents the information entropy of the i-th keyword, and p represents the number of keywords; calculate the entropy weight of the evaluation index. The calculation formula is:
[0043] S6: Furthermore, the objective weight vector Y=(y1, y2,..., yp) of the core keyword can be obtained. Optimally fit the subjective weight and the objective weight vector to obtain the comprehensive weight matrix zi. The calculation formula is:
[0044]
[0045] S7: Calculate the comprehensive weight matrix Z = (z1, z2,..., zp), calculate the Word2Vec vector through the weight matrix Z = (z1, z2,..., zp), and update the first matching degree through the Word2Vec vector.
[0046] The beneficial effects of the present invention are as follows: judge the first weight score through the after-class exercises, obtain the second weight score through the second after-class exercises, combine the two weight scores to obtain the first matching degree, automatically match the exercises through the first matching degree, and give positive and negative weights to the results of the current exercises to optimize the current matching degree, avoiding the situation that the exercises in the weak parts of the students are not trained during learning, enhancing the all-round development of the students in this course, and at the same time optimizing the interactive experience of the students in the skill part. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] By describing the embodiments shown in the accompanying drawings in detail, the above and other features of the present invention will become more obvious. The same reference numerals in the drawings of the present invention represent the same or similar elements. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. In the attached
[0048] In the figure:
[0049] Figure 1 is a flowchart of a performance evaluation method for integration of production and education based on flexible configuration of indicators. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0050] The following will clearly and completely describe the concept, specific structure and technical effects generated by the present invention in combination with the embodiments and the drawings, so as to fully understand the purpose, solution and effects of the present invention. It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.
[0051] An AI-based automatic question generation experience method, the method comprising the following steps:
[0052] S100: Obtain the system learning courses and the learning progress of the courses;
[0053] S200: Automatically match exercises according to the learning courses and the learning progress of the courses;
[0054] S300: Statistically analyze the students' course learning situation according to the exercise training results;
[0055] S400: Update the automatically matched exercises according to the course learning situation and conduct a second exercise test;
[0056] S500: Output automatically matched better exercises through multiple updates of exercise practice.
[0057] In step S100, according to the feedback of the system background record on the system learning courses and the course learning progress, classify the system learning courses into skill improvement and knowledge education, match different types of exercises for the students through different classifications, arrange the different types of exercises, and perform multi-dimensional algorithm matching on the new exercise types.
[0058] In step S200 of outputting automatically matched better exercises through multiple updates of exercise practice, define the classification marks for the learning courses, where the skill improvement is marked as S and the knowledge education is marked as K. Define the learning course obtained from the system as courseX, where X is the classification mark for the learning course, and perform calculations using different matching algorithms on the classified courses.
[0059] In the learning courses marked as S, establish a weight sequence BL for automatically matching exercises for the course course, optimize the weight sequence X, and the method for establishing the weight sequence BL for automatically matching exercises for the course course is as follows:
[0060] Analyze the content learned by the user through the course progress P, match the after-class exercises through the content, define the first weight score for the completion of the after-class system, define the score for the completion of the exercises matched for the second time by the user as the second weight score, integrate the first weight score and the second weight score to obtain the third weight score value, and integrate the three weight score values to obtain the first matching degree;
[0061] Optimize the first matching degree through the positive weight direction and the negative weight direction, input the first matching degree for comparison into the neural network, and the network output result is jointly determined by the positive output and the negative output. The calculation formula is as follows:
[0062]
[0063] O t = g(Vh t + b 3 );
[0064] Among them, x represents the input value, and the input value is the first matching degree. are respectively the positive weight direction and the negative weight direction, and the positive weight and the negative weight are respectively the first weight score and the second weight score. is the network positive output value, W 1 、U 1 、b 1They are the input layer weights, hidden layer weights, and bias vectors for the forward output respectively. The input layer weight matrix is constructed by the weight ratios H1 and H2 of the dossier vector difference t1 and the sub-roof area difference t2. is the network reverse output value, W 2 , U 2 , b 2 They are the input layer weights, hidden layer weights, and bias vectors for the reverse output respectively. The input layer weights are the results obtained through top-level comparison when judging the forward output and the reverse output. Then L1 is the input layer weight, L2 is the hidden layer weight, and the bias vector is the sum of and . If the result obtained through sub-top-level comparison, then L2 is the input layer weight, L1 is the hidden layer weight, and the bias vector is the sum of
[0065] O t is the final output value, V and b 3 are the output layer weight matrix and the bias vector respectively. The symbol is the concatenation operation;
[0066] Put the final output value O t into the neural network model for training. After the training is completed, the fast index X is output for the input data. The index X is the standard for judging whether the comparison data meets the qualification here. Each index X needs to be trained through a separate neural network model for the mapping relationship between the input data and this index. The result is output through the neural network model and the results of multiple output trainings are obtained to get the second matching degree.
[0067] Put the final output value O t into the neural network model for training. After the training is completed, the fast index X is output for the input data. The index X is the standard for judging whether the comparison data meets the qualification here. Each index X needs to be trained through a separate neural network model for the mapping relationship between the input data and this index. The result is output through the neural network model and the results of multiple output trainings;
[0068] In the neural network model, the number of neurons in the network input layer is equal to the number of input data types, the number of neurons in the output layer is equal to the number of output data types, while the number of intermediate layers and the number of neurons in each layer are freely set. The setting of the intermediate layer will directly affect the accuracy of the approximate data of the neural network. It is judged whether to increase or decrease the number of intermediate layers or neurons according to the training results. The calculation methods for judging the number of intermediate layers and neurons are as follows:
[0069] Increment the intermediate layers of the neural network model layer by layer, and name the metric X corresponding to the number of intermediate layers. When the number of intermediate layers is 1, output the metric X 1 , and so on, and construct a non-linear regression function based on the metric X a is a constant value, calculate the variance D of the independent variable of the function where L is the number of intermediate layers at the time of output, X i is the i-th metric X of the output
[0070] The X i is the input data corresponding to the i-th metric X of the output. When , A takes the minimum value, where is to find the partial derivative of the function, A is a parameter for determining the number of intermediate layers. Compare the value of A with the variance D. If D > A, input the output metric X into the neural network model and increment the number of intermediate layers by 1. If D ≤ A, output the metric X
[0071] Update the weighted coefficients of the neurons by the gradient descent method
[0072]
[0073] where is to find the partial derivative of the function, P and Q are the system output error and the neuron weight increment respectively, both are constant values, ω(l) is the neuron weighted coefficient, Δω(l) is the updated neuron weighted coefficient, δ represents the neuron learning rate. Determine the number of neurons set in the intermediate layer through the Δω(l), and combine the output value O according to the superposition analysis algorithm t and perform deep learning on the algorithm
[0074] Step 1: Calculate the first matching degree between the Q&A content of the questions in the question bank and the Q&A content of the current question. Obtain the next question of the question with the highest first matching degree in the question bank as the candidate question. Obtain the second matching degree through the training of the convolutional neural network. Calculate the second matching degree between the candidate question and the next question in the question bank. If the second matching degree is greater than the first matching degree, output the candidate question
[0075] Step 2: If the second matching degree does not satisfy that the second matching degree is greater than the first matching degree, re-match the question with the second highest first matching degree and repeat Step 1
[0076] Step 3: Encode the Q&A content of the current question according to the preset Q&A model to obtain the question encoding and answer encoding of the current question
[0077] Encode the Q&A content of each question according to the preset Q&A model to obtain the question encoding and answer encoding of each question
[0078] Calculate the question matching degree between the question code of each test question and the question code of the current test question;
[0079] Calculate the answer matching degree between the answer code of each test question and the answer code of the current test question;
[0080] Calculate the first matching degree according to the question matching degree and the answer matching degree;
[0081] The selection of the next question of the test question with the highest first matching degree as the candidate test question includes:
[0082] Select the test questions in the question-and-answer content of all test questions whose question matching degree is greater than the first preset threshold and whose answer matching degree is greater than the second preset threshold as the candidate test questions;
[0083] Select the next question of the test question with the highest first matching degree among the candidate test questions as the candidate test question.
[0084] Preferably, in step S400, update the matching degree through the second exercise result:
[0085] Obtain exercise data and assessment data. The keyword is the keyword of the sentence that increases the score or can be bonus-scored in the exercise. Define the assessment data as te, and the assessment data as ft. Screen the te and the keywords of the exercise data and construct the sequences tet and ftt. The tet is the keyword sequence of the assessment data, and the ftt is the keyword sequence of the exercise data. Count the total number of keywords T and L in the sequences tet and ftt, and calculate the keyword weight ratio. The tetK and tetK+1 are the k-th and (k + 1)-th elements in the sequence tet respectively, and the fttK and fttK+1 are the k-th and (k + 1)-th elements in the sequence ftt respectively. Calculate the index weight. According to the given rk and lk assignments, the weight value Q of the keywords in te and ft can be obtained. The Q1 is the weight value of the assessment data keyword, and Q2 is the weight value of the assessment data keyword. Input the weight values Q1 and Q2 to calculate the Word2Vec vector;
[0086] Calculating the weight values Q1 and Q2 includes calculating the ratio of the exercise data and the assessment data in the matching degree calculation through the ratio of adjacent keyword data, and then obtaining the weight values Q1 and Q2;
[0087] The method for calculating the Word2Vec vector is:
[0088] Construct a weight matrix based on the weight values of the keywords in the te and ft. Based on the TextRank algorithm, obtain the word vectors of the keywords in the te and ft through the weight matrix, and define the word vectors of the keywords as p(x, y), where x represents the weight value of the te keyword and y represents the weight value of the ft keyword. Then calculate the objective weight of the core keyword according to the word vector p(x, y) of the keyword, and calculate the information entropy of the keyword. The calculation formula is as follows:
[0089]
[0090] Among them, ei represents the information entropy of the i-th keyword, and p represents the number of keywords; calculate the evaluation index entropy weight, and the calculation formula is as follows:
[0091] S6: Furthermore, the objective weight vector Y=(y1, y2,..., yp) of the core keyword can be obtained. Optimize and fit the subjective weight and the objective weight vector to obtain the comprehensive weight matrix zi. The calculation formula is as follows:
[0092]
[0093] S7: Calculate the comprehensive weight matrix Z=(z1, z2,..., zp). Calculate the Word2Vec vector through the weight matrix Z=(z1, z2,..., zp), and update the first matching degree through the Word2Vec vector.
[0094] Preferably, the method for the AI automatic question generation experience is based on a system. The system includes a processor, a memory, an exercise collection module, an exercise analysis module, an exercise weight analysis module, and a voice interaction module. The exercise collection module, the exercise analysis module, the exercise weight analysis module, and the voice interaction module are computer programs stored in the memory and can run on the processor. When the processor executes the computer program, it realizes the steps in the method for the AI automatic question generation experience described above; the functions of each module are as follows:
[0095] Processor: Communicate the information in multiple modules and process the data generated in the exercise collection module, the exercise analysis module, the exercise weight analysis module, and the voice interaction module;
[0096] Memory: Store the data generated in the processor, the exercise collection module, the exercise analysis module, the exercise weight analysis module, and the voice interaction module, and can read the data from the memory;
[0097] Exercise collection module: Collect the after-class exercises of the course;
[0098] Exercise Analysis Module: Analyze the matching degree between the course and the after-class exercises;
[0099] Exercise Weight Analysis Module: Match the exercise with the first matching degree before automatically matching the exercises;
[0100] Voice Interaction Module: Interact with the user through voice Q&A by outputting exercises.
[0101] The described AI-based automatic question generation experience system can run on computing devices such as desktop computers, laptops, palmtop computers, and cloud data centers. The described AI-based automatic question generation experience system includes, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above examples are only examples of an AI-based automatic question generation experience system and do not constitute a limitation on the AI-based automatic question generation experience system. It may include more or fewer components than the examples, or combine certain components, or different components. For example, the AI-based automatic question generation experience system may also include input / output devices, network access devices, buses, etc.
[0102] The so-called processor can be a Central Processing Unit (CPU), or other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete component gate circuits, or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The processor is the control center of the AI-based automatic question generation experience system, connecting all sub-regions of the entire AI-based automatic question generation experience system through various interfaces and lines.
[0103] The memory can be used to store the computer programs and / or modules. By running or executing the computer programs and / or modules stored in the memory, and by invoking the data stored in the memory, the processor can implement various functions of the AI-based automatic question generation experience system. The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as a hard disk, memory, plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage devices.
[0104] Although the description of the present disclosure has been quite detailed and has particularly described several of the described embodiments, it is not intended to be limited to any of these details or embodiments or any particular embodiment, so as to effectively cover the intended scope of the present disclosure. In addition, the present disclosure is described above in terms of embodiments foreseeable by the inventors for the purpose of providing a useful description, and non-substantive changes to the present disclosure that are not currently foreseeable may still represent equivalent changes to the present disclosure.
Claims
1. A method for automatically setting questions based on AI, characterized in that: The method comprises the following steps: S100: Obtain system learning courses and course learning progress; S200: Automatically match exercises according to the learning course and course learning progress; S300: Calculate the course learning status of the staff according to the results of the exercise training; S400: updating the automatically matched exercises according to the course learning situation and conducting a secondary exercise test; S500: By updating exercises multiple times, the output automatically matches better exercises.
2. According to claim 1, the method for automatically setting questions based on AI is characterized in that: In step S100, the system learning courses and course learning progress are recorded and fed back in the system background, and the system learning courses are classified into skill improvement and knowledge education. Different types of exercises are matched to students through different classifications, and the different types of exercises are arranged, and multi-dimensional algorithm matching is performed on new exercise types.
3. The AI-based automatic question-setting experience method according to claim 1, characterized in that: In step S200 of outputting automatically matched better exercises by updating exercises multiple times, the learning course classification definition marks are marked, where skill improvement is marked as S and knowledge education is marked as K. The learning course acquired from the system is defined as courseX, where X is the learning course classification definition mark, and different matching algorithms are used to calculate the classified courses.
4. The AI-based automatic question-setting experience method according to claim 3 is characterized in that: In a learning course marked as S, a weight sequence BL of automatically matching exercises is established for the course, and the weight sequence X is optimized. The method for establishing the weight sequence BL of automatically matching exercises for the course is: Analyze the user's learning content through the course progress P, match the after-class exercises through the content, define the first weight score for the after-class system completion, define the score for the user's completion of the second matching exercises as the second weight score, and obtain the third weight score value by integrating the first weight score and the second weight score. The first matching degree is obtained by integrating the three weight score values; The first matching degree is optimized by the forward weight direction and the reverse weight direction, and the neural network input is compared with the first matching degree. The network output result is determined by the forward output and the reverse output. The calculation formula is as follows: O t =g(Vh t +b3); Among them, x represents the input value, and the input value is the first matching degree. are the forward weight and reverse weight directions respectively, the forward weight and reverse weight are the first weight score and the second weight score respectively, is the network forward output value, W1, U1, b1 are the input layer weight, hidden layer weight and bias vector of the forward output respectively, and the input layer weight matrix is constructed by the weight ratio H1 and H2 of the volume vector difference t1 and the secondary antenna area difference t2, is the reverse output value of the network, W2, U2, and b2 are the input layer weight, hidden layer weight, and bias vector of the reverse output, respectively. The input layer weight is the result obtained by comparing the top layer when judging the forward output and the reverse output. L1 is the input layer weight, L2 is the hidden layer weight, and the bias vector is and If the result is obtained by comparing the second top layer, L2 is the input layer weight, L1 is the hidden layer weight, and the bias vector is and of and; O t is the final output value, V and b3 are the output layer weight matrix and bias vector respectively, and the symbol For splicing operation; The final output value O t The neural network model is put into training, and after the training is completed, a fast indicator X is output for the input data, and the indicator X is used to judge whether the comparison data meets the qualified standard here. Each indicator X needs to be trained through a separate neural network model to obtain the mapping relationship between the input data and the indicator, and the second matching degree is obtained by outputting the neural network model and outputting the training output results multiple times.
5. The AI-based automatic question-setting experience method according to claim 4 is characterized in that: The final output value O t The neural network model is put into training, and after the training is completed, the input data is quickly outputted as an index X, where the index X is used to judge whether the comparison data meets the qualified standard. Each index X needs to be trained through a separate neural network model to obtain the mapping relationship between the input data and the index, and the output is outputted through the neural network model and the training output is outputted multiple times; In the neural network model, the number of neurons in the network input layer is equal to the number of input data types, the number of neurons in the output layer is equal to the number of output data types, and the number of intermediate layers and the number of neurons in each layer are freely set. The setting of the intermediate layer will directly affect the accuracy of the approximate data of the neural network. Whether the number of intermediate layers or neurons needs to be increased or decreased is determined based on the training results. The calculation method for determining the number of intermediate layers and neurons is: The middle layers in the neural network model are added layer by layer, and the output index X corresponding to the number of middle layers is named. When the number of middle layers is 1, the index X1 is output, and so on. Based on the index X, a nonlinear regression function is constructed. a is a constant value, and the variance D of the independent variable of the function is calculated. Where L is the number of intermediate layers at output, X i is the i-th index of the output, X i ` is the input data corresponding to the output index X at the i-th position. When A takes the minimum value, To obtain the partial derivative of the function, A is a parameter for determining the number of intermediate layers. The value of A is compared with the variance D. If D>A, the output index X is input into the neural network model, and the number of intermediate layers is increased by 1. If D≤A, the output index X is output; Update the weight coefficients of neurons by gradient descent: in, To obtain the partial derivative of the function, P and Q are the system output error and the neuron weight increment, respectively, both of which are constant values, ω(l) is the neuron weight coefficient, Δω(l) is the updated neuron weight coefficient, δ represents the neuron learning rate, and the number of neurons set in the middle layer is determined by the Δω(l). According to the superposition analysis algorithm combined with the output value O t And perform deep learning on the algorithm.
6. The AI-based automatic question-setting experience method according to claim 3 is characterized in that: In the learning course marked as K, obtain the question and answer content of the current test question of the learning course; Step 1: Calculate the first matching degree between the question and answer content of the test questions in the question bank and the question and answer content of the current test question, obtain the next question of the test question with the highest first matching degree in the test bank as a candidate test question, obtain the second matching degree through convolutional neural network training, calculate the second matching degree between the candidate test question and the next question in the test bank, and output the candidate test question if the second matching degree is greater than the first matching degree; Step 2: If the second matching degree does not satisfy the second matching degree greater than the first matching degree, re-match the test questions with the second first matching degree, and repeat step 1; Step 3: Encode the question and answer content of the current test question according to a preset question and answer model to obtain the question code and answer code of the current test question; Encoding the question and answer content of each test question according to a preset question and answer model to obtain a question code and an answer code for each test question; Calculate the problem matching degree between the problem code of each test question and the problem code of the current test question; Calculate the answer matching degree between the answer code of each test question and the answer code of the current test question; Calculating the first matching degree according to the question matching degree and the answer matching degree; The step of selecting the next question of the first question with the highest matching degree as a candidate question comprises: Selecting, from among all the questions and answers, questions whose question matching degree is greater than a first preset threshold and whose answer matching degree is greater than a second preset threshold as test questions to be selected; The next question of the question with the highest first matching degree among the test questions to be selected is selected as a candidate test question.