Multi-dimensional dynamic question generation method and device for intelligent question bank, equipment and medium

Abstract

The invention provides a multi-dimensional dynamic question generation method for an intelligent question bank, and the method specifically comprises the steps: collecting question generation parameters inputted by a user, determining a target question template according to the question generation parameters, and the question generation parameters comprise at least one of knowledge points, question types and difficulty; a manifold model is constructed based on question configuration parameters, an optimization objective function is defined, optimal question configuration parameters are determined in a multi-dimensional parameter space of the manifold model, and the question configuration parameters comprise question types, knowledge points, variable ranges and mastery degrees; filling the target question template according to the optimal question configuration parameter to generate a question; or, according to the optimal question configuration parameter and the target question template, the question of the converted question type is generated, the question is automatically generated, the adaptation degree is high, and the learning efficiency is effectively improved. The invention further provides a multi-dimensional dynamic question generation device for the intelligent question bank, computer equipment and a readable storage medium.

Description

Technical Field

[0001] The present invention relates to the field of intelligent assisted learning technology, and more specifically, to a method, device, computer equipment and readable storage medium for generating multi-dimensional dynamic questions in an intelligent question bank. Background Art

[0002] With the advancement of science and technology, computers are increasingly used, and intelligent-assisted learning based on computing devices is also gaining popularity. Users can use online or offline intelligent question banks to assist in learning through devices such as computers, tablets, and mobile phones, improving the efficiency of independent learning. However, existing intelligent question banks are not highly intelligent. While they can automatically generate questions, most of them only perform variable substitutions and fail to understand the underlying principles of the questions. This makes them unable to meet the needs of highly personalized learning, resulting in a poor user experience and a need for improved technical solutions. Summary of the Invention

[0003] In light of the above, the present invention aims to address at least one of the technical problems existing in the prior art. To this end, the present invention proposes a multi-dimensional dynamic question generation method, apparatus, computer device, and readable storage medium for an intelligent question bank. By constructing a manifold model, determining optimal question configuration parameters, and generating questions accordingly, the adaptability of the questions is improved, meeting the user's personalized learning assistance needs. Furthermore, this processing method features a simple process and high efficiency.

[0004] To this end, in a first aspect, an embodiment of the present invention provides a method for generating multi-dimensional dynamic questions in an intelligent question bank, the method specifically comprising:

[0005] Collecting question generation parameters input by a user, and determining a target question template according to the question generation parameters, wherein the question generation parameters include at least one of a knowledge point, a question type, and a difficulty level;

[0006] Constructing a manifold model based on question configuration parameters, defining an optimization objective function, and determining optimal question configuration parameters in the multidimensional parameter space of the manifold model, wherein the question configuration parameters include question type, knowledge points, variable range, and mastery level;

[0007] Fill the target question template according to the optimal question configuration parameters to generate a question; or generate a question of a converted question type according to the optimal question configuration parameters and the target question template.

[0008] Preferably, the method further comprises:

[0009] At least two problem-solving paths are generated according to the problem-solving steps of the problem, and at least one of the at least two problem-solving paths is selected for display.

[0010] Preferably, generating at least two problem-solving paths based on the problem-solving steps of the problem specifically includes:

[0011] The answer to the question is selected based on the question and the preset mathematical model, and a detailed step-by-step analysis is generated based on the question structure.

[0012] Preferably, the step analysis includes formula derivation, calculation steps, and solution description; and / or,

[0013] Based on natural language processing (NLP) and machine learning, it detects logical loopholes in the parsing step and generates corresponding supplementary explanations.

[0014] Preferably, the selecting at least one of the at least two problem-solving paths for display specifically includes:

[0015] Construct a problem-solving path model, and determine the problem-solving path with the optimal step combination based on the problem-solving path model, and display the problem-solving path. The problem-solving path model is:

[0016]

[0017] where σ i Is whether to adopt the i-th problem-solving step, 0 means not to choose, 1 means to choose; J ij is the collaborative weight between the i-th problem-solving step and the j-th problem-solving step; Γ is a global control parameter used to balance the breadth of exploration and path length. A large Γ value corresponds to fewer problem-solving steps.

[0018] Preferably, the manifold model is defined as:

[0019]

[0020] g j (p)≥0,j=1,…,m;

[0021] in, is a parameter manifold consisting of a vertex set V, an edge set E, and a weight ω; R(p) is the manifold regularization term, g j (p) is the subject constraint, f(p i ) is the function for generating questions; t i The target value set by the user; g j (p) is a hard constraint in discipline or logic; λ is the coefficient that controls the influence of the regularization term.

[0022] Preferably, the method further comprises:

[0023] Review the questions according to the review rules, including formulas, question structure, and answers; and / or,

[0024] Modifying the topic in response to external topic review information and modification data; and / or,

[0025] Classify and label questions according to classification parameters, including question type, knowledge points, and difficulty; and / or,

[0026] Predict the difficulty of the questions and update the difficulty information.

[0027] Preferably, the method further comprises:

[0028] The difficulty of the questions is predicted based on the Hypergraph Neural Network (HGNN), and the difficulty information is updated to match the corresponding questions. The prediction model is:

[0029] H (l+1) =σ(D -1 H (l) W (l) );

[0030] Among them, H (l) is the feature representation of the llth layer node, and the feature representation includes question type, knowledge point, and error rate; W (l) is the trainable weight matrix of the lllth layer; D is the degree matrix of the node; σ is the activation function.

[0031] Preferably, the method further comprises:

[0032] The question generation strategy is optimized according to the reinforcement learning proximal policy optimization (PPO) algorithm, and the reward function is:

[0033] R=0.6*p+0.3*h+0.1d;

[0034] Among them, p is the difficulty matching degree, h is the user's stay time, and d is the error rate reduction.

[0035] In a second aspect, an embodiment of the present invention further provides a multi-dimensional dynamic question generation device 100 for an intelligent question bank, the device specifically comprising:

[0036] a question template selection module, configured to collect question generation parameters input by a user and determine a target question template based on the question generation parameters, wherein the question generation parameters include at least one of a knowledge point, a question type, and a difficulty level;

[0037] A question configuration parameter determination module is used to construct a manifold model based on question configuration parameters, define an optimization objective function, and determine the optimal question configuration parameters in the multidimensional parameter space of the manifold model. The question configuration parameters include question type, knowledge points, variable range, and mastery level; and

[0038] The question generation module is used to fill the target question template according to the optimal question configuration parameters to generate a question; or to generate a question of a converted question type according to the optimal question configuration parameters and the target question template.

[0039] In a third aspect, an embodiment of the present invention further provides a computer device comprising: a memory and a processor and a computer program stored in the memory. When the computer program is executed on the processor, the multi-dimensional dynamic question generation method of the intelligent question bank as described in the first aspect is implemented.

[0040] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for generating multi-dimensional dynamic questions in the intelligent question bank as described in the first aspect is implemented.

[0041] The multi-dimensional dynamic question generation method of the intelligent question bank provided by the embodiment of the present invention constructs a manifold model to determine the optimal choice of question configuration parameters, and then fills in or converts questions according to the optimal question configuration parameters and the selected question type, and automatically generates questions. The generated questions are highly matched with the user's learning needs, have strong adaptability, can efficiently assist learning, and significantly improve learning efficiency. The multi-dimensional dynamic question generation method of the intelligent question bank has a simple process, good auxiliary learning effect, and provides users with a good user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flowchart of a method for generating multi-dimensional dynamic questions in an intelligent question bank according to an embodiment of the present invention;

[0043] Figure 2 A flowchart of another embodiment of the multi-dimensional dynamic question generation method for the intelligent question bank provided by an embodiment of the present invention;

[0044] Figure 3 A schematic diagram of the framework structure of a multi-dimensional dynamic question generation device for an intelligent question bank provided by an embodiment of the present invention;

[0045] Figure 4 A schematic diagram of the system architecture of a multi-dimensional dynamic question generation system for an intelligent question bank provided by an embodiment of the present invention;

[0046] Figure 5 A schematic diagram of the framework structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0048] The disclosure below provides many different embodiments or examples for realizing different structures of the present invention. In order to simplify the disclosure of the present invention, the components and settings of specific examples are described below. Of course, they are merely examples and are not intended to limit the present invention. In addition, the present invention may repeat reference numbers and / or letters in different examples. This repetition is for the purpose of simplicity and clarity and does not in itself indicate the relationship between the various embodiments and / or settings discussed. In addition, the present invention provides examples of various specific processes and materials, but those skilled in the art will appreciate the applicability of other processes and / or the use of other materials.

[0049] Please refer to Figure 1 The present invention aims to provide a method for generating multi-dimensional dynamic questions in an intelligent question bank, the method specifically comprising the steps of:

[0050] S10, collecting question generation parameters input by the user, and determining a target question template according to the question generation parameters, wherein the question generation parameters include at least one of a knowledge point, a question type, and a difficulty level;

[0051] S20, constructing a manifold model based on question configuration parameters, defining an optimization objective function, and determining optimal question configuration parameters in the multidimensional parameter space of the manifold model, wherein the question configuration parameters include question type, knowledge points, variable range, and mastery level;

[0052] S30. Fill the target question template according to the optimal question configuration parameters to generate a question; or, generate a question of a converted question type according to the optimal question configuration parameters and the target question template.

[0053] Please refer to Figure 4 , further, the method also includes:

[0054] S40: Generate at least two problem-solving paths according to the problem-solving steps of the problem, and select at least one of the at least two problem-solving paths for display.

[0055] Specifically, step S40 generates at least two solution paths based on the problem-solving steps of the problem, specifically including:

[0056] The answer to the question is selected based on the question and the preset mathematical model, and a detailed step-by-step analysis is generated based on the question structure.

[0057] Specifically, the step analysis includes formula derivation, calculation steps, and solution description; and / or,

[0058] Based on natural language processing (NLP) and machine learning, it detects logical loopholes in the parsing step and generates corresponding supplementary explanations.

[0059] Specifically, the step S40 of selecting at least one of the at least two problem-solving paths for display includes:

[0060] Construct a problem-solving path model, and determine the problem-solving path with the optimal step combination based on the problem-solving path model, and display the problem-solving path. The problem-solving path model is:

[0061]

[0062] where σ i Is whether to adopt the i-th problem-solving step, 0 means not to choose, 1 means to choose; J ij is the collaborative weight between the i-th problem-solving step and the j-th problem-solving step; Γ is a global control parameter used to balance the breadth of exploration and path length. A large Γ value corresponds to fewer problem-solving steps.

[0063] Next, we will use an example to illustrate the above problem-solving path model.

[0064] To solve the quadratic equation x 2 For example, -5x+6=0 is used to illustrate. For solving a quadratic equation, the corresponding optional problem-solving steps are shown in Table 1.

[0065] Table 1 Problem-solving steps and descriptions

[0066]

[0067]

[0068] Define the Hamiltonian (energy function), which is the problem-solving path model:

[0069]

[0070] Among them, the state variable State variable σ i ∈{0,1}∈{0,1 represents whether the i-th step is selected.

[0071] Among them, the relationship weight J between steps ijReflects the compatibility or constraints between steps. For example, step 1 must be completed before steps 2, 3, and 4 can occur. Steps 2, 3, and 4 are mutually exclusive (the discriminant determines which path to use). Step 5 is a verification step and must be combined with the root-finding step. Steps 6 and 7 are alternative root-finding methods that can be mutually exclusive with step 2 but cannot appear at the same time.

[0072] Among them, the node bias Γ i Indicates the "cost" or "preference" of a step, for example: simple steps have low costs, complex steps have high costs, and the bias is adjusted based on user preferences.

[0073] Furthermore, σ i σ j represents the interaction when two steps are selected simultaneously; J ij >0 means the two steps are compatible, and choosing two steps can reduce the energy; J ij <0 indicates that the two steps conflict and cannot be selected simultaneously. The step construction relationship and weight are shown in Table 2.

[0074] Table 2 Examples of constructed relationships and weights

[0075] edge (i,j) <![CDATA[J ij ]]> illustrate (1,2) +2 Discriminant>0, steps 1 and 2 are selected at the same time, and the energy decreases (1,3) +2 Discriminant = 0, steps 1 and 3 are selected at the same time, energy decreases (1,4) +2 Discriminant < 0, steps 1 and 4 are selected at the same time, and the energy decreases (2,3) -5 Steps 2 and 3 are mutually exclusive and cannot be selected at the same time. (2,4) -5 Steps 2 and 4 are mutually exclusive and cannot be selected at the same time. (3,4) -5 Steps 3 and 4 are mutually exclusive and cannot be selected at the same time. (2,5) +3 Steps 2 and 5 are compatible, verify the root (3,5) +3 Steps 3 and 5 are compatible, verify the root (4,5) 0 When the discriminant < 0, there is no root and no verification is required. (2,6) -4 Steps 2 and 6 are mutually exclusive and cannot be selected at the same time. (2,7) -4 Steps 2 and 7 are mutually exclusive and cannot be selected at the same time. (6,7) -4 Steps 6 and 7 are mutually exclusive

[0076] Bias Γ i As shown in Table 3, it is assumed that the user prefers the root-finding formula.

[0077] Table 3 Bias Γ i Example

[0078] Step i <![CDATA[Γ i (Offset)]]> illustrate 1 +1 Necessary steps, incentive selection 2 +3 Root-finding formula path, higher preference 3 +1 Single root path, lower preference 4 +0 No root path, default offset 5 +2 Verification step, encourage to select 6 +1 Factoring, Alternative Methods 7 +1 Preparation method, backup method

[0079] Solution example, based on the above parameters, the goal is to find σ i combination, making Minimum. Assume that the calculation result (solved by heuristic search, simulated annealing, quantum annealing, etc.) is:

[0080] σ i =(σ1,σ2,σ3,σ4,σ5,σ6,σ7)=(1,1,0,0,1,0,0);

[0081] This means: step 1 (calculate the discriminant) is selected, step 2 (discriminant > 0, find the root formula) is selected, steps 3 and 4 are not selected (because the discriminant > 0, it is not a single root or no root), step 5 (verify the root) is selected, and steps 6 and 7 are not selected (no alternative method is selected).

[0082] The system automatically generates a solution: calculate the discriminant Δ = 1; since Δ > 0, use the root-finding formula to calculate the two real roots; and verify that the roots hold when substituted into the original equation. This path is logical, cost-effective, and optimal.

[0083] If the user prefers an alternative approach, adjust the bias Γ = +4 (to encourage factorization) and re-solve the result:

[0084] σ i =(1,0,0,0,1,1,0);

[0085] That is, select the factorization step, discard the root-finding formula, and still verify the result.

[0086] This step aims to dynamically generate personalized questions based on the user's input learning objectives, knowledge points, and ability level, breaking through the limitations of traditional template questions, which can only replace numerical values. The algorithm, based on parametric manifold modeling, optimizes the objective function to find the optimal question configuration in a multidimensional parameter space.

[0087] Preferably, the manifold model is defined as:

[0088]

[0089]

[0090] g j (p)≥0,j=1,…,m;

[0091] in, is a parameter manifold consisting of a vertex set V, an edge set E, and a weight ω; R(p) is the manifold regularization term, g j (p) is the subject constraint, f(p i ) is the function for generating questions; t i The target value set by the user; g j (p) is a hard constraint in discipline or logic; λ is the coefficient that controls the influence of the regularization term.

[0092] After comprehensively analyzing parameters such as question types, knowledge points, variable ranges, and student mastery, the most appropriate combination is found in the "Question Space Map" to ensure that the questions are challenging but not beyond the students' level, and that the structure, difficulty, and legality all meet the requirements.

[0093] Next, we will illustrate the above manifold modeling with examples.

[0094] For example, we want to generate a junior high school math function question for a user, where the user expects:

[0095] The target setting is t = (knowledge point: function, medium difficulty, question type: fill in the blank), and the existing parameters in the sample question set V are shown in Table 4.

[0096] Table 4 Question parameters

[0097] Question number x1(knowledge point) x2 (difficulty) x3(question type) A 0.3 (algebra) 0.2 0.2 (select) B 0.5 (function) 0.4 0.5 (fill in the blank) C 0.8 (function + parsing) 0.7 0.8 (answer)

[0098] Among them, pA=(0.3, 0.2, 0.2), pB=(0.5, 0.4, 0.5), pC=(0.8, 0.7, 0.8), V=pA,pB,pC.

[0099] E: Connect by similarity between questions (Euclidean distance), weight

[0100] Disassemble the objective function, the main goal is to minimize the distance from the user's target parameter t. Let the user's target parameter be

[0101] t=(0.6,0.4,0.5);

[0102] If f(p) = p (the parameter directly controls the problem), the objective function becomes:

[0103]

[0104] Among them, the regular term R(p):

[0105] R(p)=∑ (i,j)∈E ω ij ||p i -p j || 2 ;

[0106] g1(p)=1(the knowledge point is legal);

[0107] g2(p)=I{0.2≤x2≤0.6}(medium difficulty);

[0108] g3(p)=I{x3=0.5}(fill in the blank).

[0109] Further optimal point selection is performed and candidates are generated using weighted interpolation:

[0110] p*=αpB+(1-α)pC, where α=0.7;

[0111] p*=0.7*(0.5,0.4,0.5)+0.3*(0.8,0.7,0.8)=(0.59,0.49,0.59);

[0112] The target value is close to t = (0.6, 0.4, 0.5) and all constraints are met.

[0113] To further generate questions, based on the parameter p*=(0.59, 0.49, 0.59), the system selects the appropriate template:

[0114] Question: Given the function y=ax 2The graph of +bx+c passes through the points A(1,2), B(2,3), and C(3,2). Find the analytical expression of the function.

[0115] Note: The knowledge point is function → satisfying x1 = 0.59, the difficulty is medium to high → satisfying x2 = 0.49, the question type is fill in the blanks → x3 = 0.59, which is a fill-in-the-blank question.

[0116] Furthermore, the method further includes step S50 of reviewing the question according to review rules, where the review information includes formulas, question structure, and answers. In this embodiment, after the question is generated, the system automatically reviews it using built-in review rules to ensure that the question complies with regulations and is logically sound. Review content includes, but is not limited to, the correctness of the formula, the completeness of the question structure, and the accuracy of the answer.

[0117] Furthermore, the method further includes step S60 of modifying the question in response to the external question review information and modification data. In this embodiment, even for questions that have passed automatic review, the administrator can still use the manual review function to adjust the question and correct errors or inappropriate parts to ensure the quality of the question. It is understood that questions that have not been reviewed by the system can also be directly reviewed by the administrator.

[0118] Furthermore, the method further includes the step of: S70, categorizing and labeling the questions according to classification parameters, where the classification parameters include question type, knowledge points, and difficulty. In this embodiment, after questions are generated, the system intelligently categorizes them according to question type, knowledge points, difficulty, etc., and generates tags to facilitate administrator management and user query. The question bank supports rapid retrieval and updating.

[0119] Furthermore, the method further includes the step: S80, predicting the difficulty of the question and updating the difficulty information, so as to simulate the real thinking path of solving the question, provide multiple reasonable solutions for the same question, and recommend the best one to the user.

[0120] Furthermore, the method further includes: S90, predicting the difficulty of the question based on a hypergraph neural network (HGNN), and updating the difficulty information to match the corresponding question, the prediction model is:

[0121] H (l+1) =σ(D -1 H (l) W (l) );

[0122] Among them, H (l) is the feature representation of the llth layer node, and the feature representation includes question type, knowledge point, and error rate; W (l) is the trainable weight matrix of the lllth layer; D is the degree matrix of the node; σ is the activation function.

[0123] Furthermore, the method further includes: step S95, optimizing the question generation strategy according to the reinforcement learning proximal policy optimization (PPO) algorithm, and the reward function is:

[0124] R=0.6*p+0.3*h+0.1d;

[0125] Where p is the difficulty match, h is the user dwell time, and d is the error rate reduction. The above model can output a predicted difficulty value for the current question (e.g., 0.65 indicates moderately difficult) and use this result to control the generation strategy for subsequent questions.

[0126] Specifically, within the difficulty prediction model, difficulty match (accounting for the largest percentage, 60%) refers to the degree of match between the difficulty of the question and the student's current level. Questions that are too easy or too difficult are not good; the best questions are those that are "student-solvable but still slightly challenging." For example, if a student has just learned a linear equation and the system gives them a "system of linear equations" question, that's too difficult → the match is low. If they give them a question like "2+3=?", that's too easy → the match is also low. If they give them a question like "3x+2=8," that's just enough for practice → the match is high.

[0127] User dwell time (30%): This refers to the amount of time students spend on the question page, reflecting the question's appeal. Generally, this is understood as: finishing / skipping too quickly means the question is boring; pausing moderately means the question is worth pondering; pausing too long means the question is too difficult or incomprehensible. For example: a student is presented with a multiple-choice question, answers it quickly, and submits the paper in 3 seconds. The system assumes they didn't think carefully enough, resulting in a low score. The student carefully examines the question, ponders it for 15 seconds, and answers correctly. This indicates a good answer, resulting in a high score. Pausing for 3 minutes and still not finishing the question may be too difficult, resulting in a low score.

[0128] Error rate reduction (10%): This refers to whether students' error rates on the same knowledge point have decreased. This reflects whether the questions have helped students "truly learn." For example, a student's error rate in "Fractional Calculation" was 70%. After working on a few system-generated questions, the error rate dropped to 40%. This indicates a positive outcome, resulting in a higher reward. If the error rate remains unchanged or even increases, the reward will be lower.

[0129] The goal of the reward function is to come up with a question → which is just a little challenging for the student → the student thinks carefully → as a result, he or she gradually masters the knowledge point → the system gives extra points.

[0130] The following example illustrates the difficulty prediction.

[0131] Difficulty prediction model definition:

[0132] H (l+1) =σ(D -1 H (l) W (l) );

[0133] H (l) : Feature representation of the lllth layer nodes (such as question type, knowledge points, error rate, etc.);

[0134] W (l) : The trainable weight matrix of the lll layer;

[0135] D: degree matrix of the node (indicates the connection strength of each node with other nodes);

[0136] σ: activation function (used to enhance the expressive power of the model).

[0137] The initial features of each question are composed of three dimensions: question type code (multiple choice = 1, fill-in-the-blank = 2, short-answer = 3); number of associated knowledge points (a simple quantitative representation of the number of associated knowledge points); and historical error rate (a value between 0 and 1). Table 5 shows an example of the initial features.

[0138] Table 5 Initial feature examples

[0139] Question number Question Type Number of related knowledge points Error rate 1 1 3 0.2 2 2 2 0.5 3 1 1 0.8

[0140] Then the initial feature matrix is:

[0141]

[0142] Construct hypergraph connections, adjacency and degree matrices, assuming that there are connections between questions based on the sharing of knowledge points: Question 1 and Question 2 share 2 knowledge points, with a weight of 2; Question 1 and Question 3 share 1 knowledge point, with a weight of 1; Question 2 and Question 3 have no direct connection.

[0143] The adjacency matrix A is:

[0144]

[0145] The degree matrix D is a row and diagonal matrix:

[0146]

[0147] Inverse of the degree matrix:

[0148]

[0149] Training weight matrix W (0) , assuming the model weights are simply:

[0150]

[0151] Here the weight matrix maps the features from 3 dimensions to 2 dimensions, for example, into an implicit difficulty feature vector.

[0152] Calculate updated difficulty data:

[0153]

[0154]

[0155] Difficulty prediction output:

[0156] d i =Sigmoid(αh i1 +βh i2 +b)d i =Sigmoid(0.5×h i1 +0.5×h i2 -2);

[0157]

[0158] d3≈0.36.

[0159] Please refer to Figure 2 In a second aspect, an embodiment of the present invention further provides a multi-dimensional dynamic question generation device for an intelligent question bank, the device specifically comprising:

[0160] The question template selection module 10 is used to collect question generation parameters input by the user and determine a target question template according to the question generation parameters, wherein the question generation parameters include at least one of a knowledge point, a question type, and a difficulty level;

[0161] The question configuration parameter determination module 20 is used to construct a manifold model based on the question configuration parameters, define an optimization objective function, and determine the optimal question configuration parameters in the multidimensional parameter space of the manifold model. The question configuration parameters include question type, knowledge points, variable range, and mastery level; and

[0162] The question generating module 30 is configured to fill the target question template according to the optimal question configuration parameters to generate a question; or to generate a question of a converted question type according to the optimal question configuration parameters and the target question template.

[0163] Please refer to Figure 3 The embodiment of the present invention also provides a multi-dimensional dynamic question generation system for an intelligent question bank, the specific architecture of which includes:

[0164] User interaction layer: provides a visual parameter configuration panel, real-time question preview, multi-path analysis display, and difficulty feedback interface.

[0165] Business logic layer: Implements a dynamic generation engine that supports multi-dimensional parameter input and real-time adjustment. It also includes an autonomous problem-solving core, a CSP solver, a symbolic computation engine, and a quantum-inspired optimizer, enabling an intelligent audit module.

[0166] Question analysis engine layer: Uses self-developed mathematical models for high-precision calculations to generate detailed problem-solving steps and explanations.

[0167] Database management layer: stores question bank data, supports question classification, storage and retrieval, administrator review and manual modification, dynamic classification and version control.

[0168] AI Optimization Center: Difficulty prediction model: Hypergraph Neural Network (HGNN), input dimension 128, 3 convolution layers.

[0169] The data structure design of the multi-dimensional dynamic question generation system for the intelligent question bank includes: a question bank template database: which stores question templates, question types, knowledge points, difficulty, and other information; a parameter range database: which stores the legal range of each parameter to facilitate random question generation; and a user answer record: which stores user answer history and feedback data for intelligent adjustment of question generation and difficulty.

[0170] The operating process of the multi-dimensional dynamic question generation system for the intelligent question bank is as follows: users log in and select a learning objective or knowledge point. The system dynamically generates appropriate questions based on the user's input parameters and displays a preview of the questions. After the user completes the question, the system automatically generates the answer and explanation, providing detailed step-by-step explanations. Administrators can review, categorize, and update the question bank.

[0171] The multi-dimensional dynamic question generation system of the intelligent question bank has the following characteristics:

[0172] Modular structure: parameter input module, users input question parameters (variable range, question type, etc.) through the interface.

[0173] Template matching module: users search for question templates in the question bank and match them with the required question templates. Generation module: generates questions based on templates and parameters. Analysis module: calls algorithms to generate answers and detailed analysis.

[0174] Dynamic logic design: Uses branching structures to implement processing logic for different question types. Integrates AI technology to improve the intelligence of parsing generation.

[0175] Visual Results Module: Provides question preview, answer verification, and export functions. Supports mixed text and graphics for questions and solutions.

[0176] Please refer to Figure 4 In a third aspect, an embodiment of the present invention further provides a computer device 200, comprising: a memory 201, a processor 202, and a computer program 203 stored on the memory 201. When the computer program 203 is executed on the processor 202, the multi-dimensional dynamic question generation method of the intelligent question bank as described in the first aspect is implemented.

[0177] The computer device 200 may be a desktop computer, a notebook computer, a PDA, a cloud server, or other computing devices. The computer device 200 may include, but is not limited to, a processor 202 and a memory 201. Those skilled in the art will appreciate that Figure 3 This is merely an example of the computer device 200 and does not constitute a limitation on the computer device 200 . The computer device 200 may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the computer device 200 may also include input and output devices, network access devices, etc.

[0178] In some embodiments, the memory 201 may be an internal storage unit of the computer device 200, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 201 may also be an external storage device of the computer device 200, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card, etc. equipped on the computer device 200. Furthermore, the memory 201 may also include both an internal storage unit of the computer device 200 and an external storage device. The memory 201 is used to store operating devices, application programs, boot loaders (Boot Loader), data and other programs, such as the program code of the computer program. The memory 201 may also be used to temporarily store data that has been output or is to be output.

[0179] The processor 202 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0180] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for generating multi-dimensional dynamic questions in the intelligent question bank as described in the first aspect is implemented.

[0181] In this embodiment, if the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the process of the above-mentioned method embodiment by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can at least include: any entity or device capable of carrying computer program code to the camera / terminal device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium. For example, a USB flash drive, mobile hard drive, magnetic disk, or optical disk. In some jurisdictions, based on legislation and patent practice, computer-readable media cannot be electric carrier signals or telecommunication signals.

[0182] The multi-dimensional dynamic question generation method of the intelligent question bank provided by the embodiment of the present invention determines the optimal choice of question configuration parameters by constructing a manifold model, and then fills in or converts questions according to the optimal question configuration parameters and the selected question type, and automatically generates questions. The generated questions are highly matched with the user's learning needs, have strong adaptability, can efficiently assist learning, and significantly improve learning efficiency. The multi-dimensional dynamic question generation method of the intelligent question bank has a simple process, good auxiliary learning effect, and provides users with a good user experience.

[0183] In the description of this specification, the reference terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0184] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.

Claims

1. A multi-dimensional dynamic question generation method for an intelligent question bank, characterized in that: The method specifically includes: Collecting question generation parameters input by a user, and determining a target question template according to the question generation parameters, wherein the question generation parameters include at least one of a knowledge point, a question type, and a difficulty level; Constructing a manifold model based on question configuration parameters, defining an optimization objective function, and determining optimal question configuration parameters in the multidimensional parameter space of the manifold model, wherein the question configuration parameters include question type, knowledge points, variable range, and mastery level; Fill the target question template according to the optimal question configuration parameters to generate a question; or generate a question of a converted question type according to the optimal question configuration parameters and the target question template.

2. The multi-dimensional dynamic question generation method of the intelligent question bank according to claim 1 is characterized in that: The method further comprises: At least two problem-solving paths are generated according to the problem-solving steps of the problem, and at least one of the at least two problem-solving paths is selected for display.

3. The multi-dimensional dynamic question generation method of the intelligent question bank according to claim 2 is characterized in that: Generating at least two problem-solving paths based on the problem-solving steps specifically includes: The answer to the question is selected based on the question and the preset mathematical model, and a detailed step-by-step analysis is generated based on the question structure.

4. The multi-dimensional dynamic question generation method of the intelligent question bank according to claim 3 is characterized in that: The step analysis includes formula derivation, calculation steps, and solution description; and / or, Based on natural language processing (NLP) and machine learning, it detects logical loopholes in the parsing step and generates corresponding supplementary explanations.

5. The multi-dimensional dynamic question generation method of the intelligent question bank according to claim 2 is characterized in that: The selecting at least one of the at least two problem-solving paths for display specifically includes: Construct a problem-solving path model, and determine the problem-solving path with the optimal step combination based on the problem-solving path model, and display the problem-solving path. The problem-solving path model is: where σ i Is whether to adopt the i-th problem-solving step, 0 means not to choose, 1 means to choose; J ij is the collaborative weight between the i-th problem-solving step and the j-th problem-solving step; Γ is a global control parameter used to balance the breadth of exploration and path length. A large Γ value corresponds to fewer problem-solving steps.

6. The multi-dimensional dynamic question generation method of the intelligent question bank according to claim 1 is characterized in that: The manifold model is defined as: g j (p)≥0,j=1,…,m; in, is a parameter manifold consisting of a vertex set V, an edge set E, and a weight ω; R(p) is the manifold regularization term, g j (p) is the subject constraint, f(p i ) is the function for generating questions; t i The target value set by the user; g j (p) is a hard constraint in discipline or logic; λ is the coefficient that controls the influence of the regularization term.

7. The multi-dimensional dynamic question generation method of the intelligent question bank according to claim 1 is characterized in that: The method further comprises: Review the questions according to the review rules, including formulas, question structure, and answers; and / or, Modifying the topic in response to external topic review information and modification data; and / or, Classify and label questions according to classification parameters, including question type, knowledge points, and difficulty; and / or predict the difficulty of questions and update difficulty information.

8. The multi-dimensional dynamic question generation method of the intelligent question bank according to claim 1 is characterized in that: The method further comprises: The difficulty of the questions is predicted based on the Hypergraph Neural Network (HGNN), and the difficulty information is updated to match the corresponding questions. The prediction model is: H (l+1) =σ(D -1 H (l) W (l) ); Among them, H (l) is the feature representation of the llth layer node, and the feature representation includes question type, knowledge point, and error rate; W (l) is the trainable weight matrix of the lllth layer; D is the degree matrix of the node; σ is the activation function.

9. The multi-dimensional dynamic question generation method of the intelligent question bank according to claim 1 is characterized in that: The method further comprises: The question generation strategy is optimized according to the reinforcement learning proximal policy optimization (PPO) algorithm, and the reward function is: R=0.6*p+0.3*h+0.1d; Among them, p is the difficulty matching degree, h is the user's stay time, and d is the error rate reduction.

10. A multi-dimensional dynamic question generation device for an intelligent question bank, characterized in that: The device specifically includes: a question template selection module, configured to collect question generation parameters input by a user and determine a target question template based on the question generation parameters, wherein the question generation parameters include at least one of a knowledge point, a question type, and a difficulty level; A question configuration parameter determination module is used to construct a manifold model based on question configuration parameters, define an optimization objective function, and determine the optimal question configuration parameters in the multidimensional parameter space of the manifold model. The question configuration parameters include question type, knowledge points, variable range, and mastery level; and The question generation module is used to fill the target question template according to the optimal question configuration parameters to generate a question; or to generate a question of a converted question type according to the optimal question configuration parameters and the target question template.

11. A computer device, characterized in that: include: A memory, a processor, and a computer program stored in the memory, when the computer program is executed on the processor, implements the multi-dimensional dynamic question generation method of the intelligent question bank as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor, the method for generating multi-dimensional dynamic questions in an intelligent question bank as claimed in any one of claims 1 to 9 is implemented.

Images