Learning curve generation method, apparatus, device, and storage medium

Abstract

The application relates to the technical field of artificial intelligence, and discloses a learning curve generation method, device and equipment and a storage medium, wherein learning history data of students is acquired, a generative model is constructed based on the learning history data, and the generative model is used to generate personalized sub-sample test questions according to a learning theme; the answering data of the students is collected in real time, the answering data is standardized and normalized, key features are extracted from the processed answering data, and answering feature data is obtained; a personalized learning curve of the students is generated according to the answering feature data, the personalized learning curve is analyzed, and the sub-sample test questions of the students are dynamically adjusted according to the personalized learning curve; learning feedback information of the students is collected, and a Harris hawk optimization algorithm is used to adjust a learning strategy based on the learning feedback information; the application improves the pertinence and effectiveness of learning, provides more accurate learning progress prediction, continuously optimizes the algorithm, and improves the accuracy and practicability of the learning curve.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a method, apparatus, device, and storage medium for generating learning curves. Background Technology

[0002] Learning curves are an important tool for assessing changes in the performance of individuals or systems during the learning process. They are widely used in education, training, and psychological research. Traditional learning curves usually describe learning progress through simple mathematical models or rules of thumb, such as exponential curves or linear models. However, these methods often ignore the complexity of the learning process and cannot accurately reflect the learning performance of different individuals in diverse environments and tasks. Summary of the Invention

[0003] The purpose of this invention is to solve the above-mentioned problems by designing a learning curve generation method, apparatus, device, and storage medium.

[0004] The first aspect of this invention provides a method for generating a learning curve, the method comprising the following steps:

[0005] Acquire students' learning history data, construct a generative model based on the learning history data, and use the generative model to generate personalized sample test questions according to the learning theme;

[0006] Students' answer data is collected in real time, the answer data is standardized and normalized, and key features are extracted from the processed answer data to obtain answer feature data;

[0007] Based on the answer feature data, a personalized learning curve for each student is generated, the personalized learning curve is analyzed, and the sample test questions for each student are dynamically adjusted based on the personalized learning curve.

[0008] Collect students' learning feedback information and adjust the learning strategy based on the learning feedback information using the Harris Eagle optimization algorithm.

[0009] Optionally, in a first implementation of the first aspect of the present invention, the step of acquiring students' learning history data, constructing a generative model based on the learning history data, and using the generative model to generate personalized sample test questions according to the learning topic includes:

[0010] Collect students' learning history data and use principal component analysis algorithm to reduce the dimensionality of the learning history data, including at least past test scores, homework completion status, and classroom participation performance;

[0011] A variational autoencoder is trained using dimensionality-reduced learning history data. The variational autoencoder consists of an encoder and a decoder. A policy gradient algorithm is introduced into the variational autoencoder for performance optimization.

[0012] The similarity between the generated test questions and the test questions that students have already answered is calculated. A regularization term related to the learning history data is added to the loss function of the variational autoencoder. Let the generated test questions be y, the set of test questions that students have already answered be S, and the similarity be sim(y,S). Then the regularization term can be expressed as:

[0013]

[0014] In the formula, λ represents the regularization coefficient;

[0015] The model parameters that minimize the loss function are selected as the optimal parameters, and the encoder in the variational autoencoder with the optimal parameters is output to obtain the generative model. Personalized subsample test items are generated based on the learning topic and the generative model.

[0016] Optionally, in a second implementation of the first aspect of the present invention, the variational autoencoder is trained using dimensionality-reduced learning history data. The variational autoencoder consists of an encoder and a decoder. A policy gradient algorithm is introduced into the variational autoencoder for performance optimization, including:

[0017] The dimensionality-reduced learning history data is divided into training set, validation set and test set. The parameters of the encoder and decoder of the variational autoencoder are initialized, and the variational autoencoder is trained using the training set data.

[0018] The gradient of the policy network is calculated using the policy gradient algorithm. The policy network generates personalized sample questions based on the current state. The value of the current state is estimated through the value network, where the current state includes the learning topic and the learning history.

[0019] Let the parameters of the policy network be w, and the state at time step t be s. t Action is a t The reward is s t The policy gradient is:

[0020]

[0021] In the formula, τ=(s0,a0,s1,a1,…) represents a trajectory, and π w (a t |s t ) indicates that the policy network is in state s t Output action a t The probability, R(s) t ,a t ) indicates that it is in state st Take action a t The reward obtained, where γ represents the discount factor and J(w) represents the objective function. Indicates that by strategy π w The expected value of the generated trajectory τ;

[0022] The parameters of the variational autoencoder are updated using the stochastic gradient descent algorithm, and the parameters of the policy network and value network are updated using the Adam optimizer, so as to optimize the performance of the variational autoencoder through the policy gradient algorithm.

[0023] Optionally, in a third implementation of the first aspect of the present invention, the real-time collection of students' answer data, the standardization and normalization of the answer data, and the extraction of key features from the processed answer data to obtain answer feature data include:

[0024] Real-time collection of students' answer data, including at least answer time, answer content, accuracy rate and error type; preliminary cleaning of the collected data to remove erroneous data and outliers; and standardization and normalization processing after preliminary cleaning.

[0025] Convolutional neural networks are constructed, including convolutional layers, pooling layers, and fully connected layers. Convolutional layers are used to extract local features, pooling layers are used to reduce feature dimensionality, and fully connected layers are used to integrate and classify the extracted features.

[0026] The Word2Vec model is used to convert the processed answer data into word vector representations. The answer data represented by word vectors is then input into a convolutional neural network model. After passing through convolutional and pooling layers, feature maps are obtained. The feature maps are then integrated into feature vectors through fully connected layers, and finally, the answer feature data is obtained through forward propagation.

[0027] Optionally, in a fourth implementation of the first aspect of the present invention, generating a student's personalized learning curve based on the answer feature data includes:

[0028] We construct a long short-term memory network, introduce an attention mechanism into the long short-term memory network, increase residual connections, and use LeakyReLU as the activation function;

[0029] At each time step, the attention weights are calculated, and then multiplied by the hidden state to obtain the weighted hidden state.

[0030]

[0031] In the formula, h' t h represents the weighted hidden state. i Let α represent the hidden state at time step i. t,iThis represents the attention weight of the t-th time step to the ith time step;

[0032] At each time step, the input and output are summed to obtain the output after residual connection:

[0033] h t =h t +x t

[0034] In the formula, h t Let x represent the hidden state at time step t. t This represents the input at time step t;

[0035] The answer feature data is input into a Long Short-Term Memory (LSTM) network. After processing through the forget gate, input gate, and output gate of the LTM hidden layer, the output layer predicts the student's learning performance indicators at future time steps. With time as the horizontal axis and the predicted learning performance indicators as the vertical axis, the predicted values ​​at each time point are connected to generate the student's personalized learning curve.

[0036] Optionally, in a fifth implementation of the first aspect of the present invention, analyzing the personalized learning curve and dynamically adjusting the student's sample test questions based on the personalized learning curve includes:

[0037] The slope of the personalized learning curve is calculated using numerical differentiation methods to determine the learning speed, and the inflection point of the curve is identified to determine the turning point of learning, thus obtaining the curve characteristics.

[0038] The student's current learning status is assessed based on the curve characteristics. If the curve slope k > 0.2, the student is in a state of rapid progress. If the curve slope 0 < k ≤ 0.2, the student is in a state of steady progress. If the curve slope -0.05 ≤ k ≤ 0.05, the student is in a state of bottleneck. If the curve slope k < -0.05, the student is in a state of regression.

[0039] By combining personalized learning curves and students' learning status, we can identify students' weaknesses and key learning content, and dynamically adjust the sample test questions for students.

[0040] 8. Optionally, in the sixth implementation of the first aspect of the present invention, the step of collecting students' learning feedback information and adjusting the learning strategy based on the learning feedback information using the Harris Eagle optimization algorithm includes:

[0041] Collect learning feedback information from students using sample test questions adjusted for the learning curve, and determine the optimization goals of learning strategies;

[0042] Initialize the Harris eagle population, with each individual representing a combination of learning strategy parameters. Calculate the fitness value of each individual, and select the best individuals as elite individuals based on their fitness values. Let the learning strategy parameters be p = {p1, p2, ..., p...}. n The fitness function is f(p), and the position of an individual Harris Hawk is represented as P = {P1, P2, ..., P}. m The population size is m.

[0043] Entering the iterative process, in each iteration, the Harris Eagle individual updates its position according to different strategies, including the exploration phase and the development phase. In the exploration phase, the Harris Eagle individual adopts a soft-encirclement exploration strategy, and the individual position update formula is:

[0044] P(t+1)=P rand (t)-r1|P rand (t)-2r2P(t)|

[0045] In the formula, P rand (t) is the position of an individual randomly selected from the current population, r1 and r2 are random numbers in the range [0,1], and t represents the current iteration number;

[0046] During the development phase, based on the escape energy E of the prey and the value of the random number r3, a development strategy is selected for position updating. When |E|≥1, a prey-chasing strategy is adopted, and when |E|<1, a gradual soft-encirclement strategy is adopted for position updating.

[0047] After updating the individual's position, the fitness value is recalculated. If the maximum number of iterations is reached, the iteration stops and the optimal combination of learning strategy parameters is output. Otherwise, the next iteration continues.

[0048] The Harris Eagle optimization algorithm is used to output the optimal combination of learning strategy parameters to adjust students' learning strategies.

[0049] A second aspect of the present invention provides a learning curve generation apparatus, the learning curve generation apparatus comprising:

[0050] The test question generation module is used to acquire students' learning history data, construct a generative model based on the learning history data, and use the generative model to generate personalized sample test questions according to the learning theme.

[0051] The data acquisition module is used to collect students' answer data in real time, standardize and normalize the answer data, and extract key features from the processed answer data to obtain answer feature data.

[0052] The learning curve generation module is used to generate a personalized learning curve for students based on the answer feature data, analyze the personalized learning curve, and dynamically adjust the sample test questions for students based on the personalized learning curve.

[0053] The feedback and optimization module is used to collect students' learning feedback information and adjust the learning strategy based on the learning feedback information using the Harris Eagle optimization algorithm.

[0054] A third aspect of the present invention provides a learning curve generation apparatus, the learning curve generation apparatus including a memory and at least one processor, the memory storing instructions; the at least one processor invokes the instructions in the memory to cause the learning curve generation apparatus to perform the steps of the learning curve generation method as described in any of the preceding claims.

[0055] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the learning curve generation method as described in any of the preceding claims.

[0056] The technical solution provided by this invention involves acquiring students' learning history data, constructing a generative model based on the learning history data, and using the generative model to generate personalized sample test questions according to the learning theme; collecting students' answer data in real time, standardizing and normalizing the answer data, and extracting key features from the processed answer data to obtain answer feature data; generating personalized learning curves for students based on the answer feature data, analyzing the personalized learning curves, and dynamically adjusting the sample test questions for students based on the personalized learning curves; collecting students' learning feedback information, and adjusting the learning strategy based on the learning feedback information using the Harris Eagle optimization algorithm; this invention ensures that each student receives tailor-made questions during the learning process by generating test questions, improving the pertinence and effectiveness of learning; the learning curve can be dynamically updated according to the student's real-time performance, reflecting their actual learning situation and providing more accurate learning progress prediction; through data collection and analysis, the test question generation algorithm can be continuously optimized to improve the accuracy and practicality of the learning curve. Attached Figure Description

[0057] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0058] Figure 1 A schematic diagram of the first embodiment of the learning curve generation method provided by the present invention;

[0059] Figure 2A schematic diagram of a second embodiment of the learning curve generation method provided in this invention;

[0060] Figure 3 A schematic diagram of a third embodiment of the learning curve generation method provided in this invention;

[0061] Figure 4 This is a schematic diagram of the learning curve generation device provided in an embodiment of the present invention;

[0062] Figure 5 This is a schematic diagram of the learning curve generation device provided in an embodiment of the present invention. Detailed Implementation

[0063] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0064] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 A schematic diagram of the first embodiment of the learning curve generation method provided by this invention is shown. The method specifically includes the following steps:

[0065] Step 101: Obtain students' learning history data, construct a generative model based on the learning history data, and use the generative model to generate personalized sample test questions according to the learning theme.

[0066] In this embodiment of the invention, learning history data may include students' past completed assignments, exam scores, classroom performance records, and learning trajectories on online learning platforms. This data can be collected from multiple channels such as the school's teaching management system, online learning platforms, and paper archives. For example, for mathematics, learning history data may include students' assignment scores, unit test scores, and records of incorrect answers in different chapters; for English, it may include students' reading comprehension answers, listening practice scores, and oral expression assessments. Based on the learning theme, a generative model is used to generate personalized sample questions. The learning theme can be a specific knowledge point, chapter, or skill requirement within the subject. For example, for the mathematics theme of "functions," the generative model can generate personalized sample questions related to functions based on students' learning history data, such as problems involving evaluating different types of functions, graphing, and analyzing their properties. For the English theme of "reading comprehension," the generative model can generate short passages that match students' reading levels and interests, and pose corresponding questions.

[0067] Step 102: Collect students' answer data in real time, standardize and normalize the answer data, and extract key features from the processed answer data to obtain answer feature data;

[0068] In this embodiment, answer data can be collected in real time through online testing platforms, paper test paper scanning, classroom interaction systems, etc. Answer data includes information such as students' answers to questions, answering time, and answering order. For example, on an online learning platform, after a student completes a question, the system immediately records the student's answer, answering time, and relevant information about the question, such as the knowledge point and difficulty level. For paper test papers, scanning and recognition technologies can be used to convert students' answers into digital data for collection.

[0069] In this embodiment, students' answer data is collected in real time, including at least the answer time, answer content, accuracy rate, and error type. The collected data undergoes preliminary cleaning to remove erroneous data and outliers, followed by standardization and normalization. A convolutional neural network is constructed, including convolutional layers, pooling layers, and fully connected layers. The convolutional layers are used to extract local features, the pooling layers are used to reduce feature dimensionality, and the fully connected layers are used to integrate and classify the extracted features. The processed answer data is converted into word vector representations using a Word2Vec model. The word vector representations are then input into the convolutional neural network model. After passing through the convolutional and pooling layers, feature maps are obtained. The fully connected layers integrate the feature maps into feature vectors, which are then forward-propagated to obtain the answer feature data.

[0070] In this embodiment, key features can reflect a student's learning status and ability level. These key features may include answer accuracy, answer time, mastery of knowledge points, and answering strategies. For example, a student's mastery of various knowledge points can be assessed by calculating their answer accuracy for different types of questions. By analyzing the distribution of students' answering time, their answering speed and efficiency can be understood. Furthermore, students' answering strategies and thought processes can be inferred based on their answering order and revisions.

[0071] Step 103: Generate personalized learning curves for students based on their answer characteristics data, analyze the personalized learning curves, and dynamically adjust the sample test questions for students based on the personalized learning curves.

[0072] In this embodiment, the slope of the personalized learning curve is calculated using numerical differentiation to determine the learning speed, and the inflection point of the curve is identified to determine the turning point of learning, thus obtaining the curve characteristics. Based on the curve characteristics, the student's current learning status is evaluated: if the curve slope k > 0.2, the student is in a state of rapid progress; if the curve slope 0 < k ≤ 0.2, the student is in a state of stable progress; if the curve slope -0.05 ≤ k ≤ 0.05, the student is in a bottleneck state; and if the curve slope k < -0.05, the student is in a state of regression. Combining the personalized learning curve and the student's learning status, the student's weak points and key learning content are determined, and the sample test questions for the student are dynamically adjusted.

[0073] In this embodiment, personalized learning curves are analyzed to understand students' learning status and needs. This analysis can include trend analysis, fluctuation analysis, and comparative analysis. Trend analysis can observe the upward or downward trend of the learning curve to determine whether the student's learning is progressing or regressing. Fluctuation analysis can check the fluctuation of the learning curve to understand whether the student's learning is stable. Comparative analysis can compare the student's learning curve with other students or a standard learning curve to identify the student's strengths and weaknesses. For example, if the student's learning curve shows an upward trend, it indicates that the student's learning is progressing, and more challenging sample questions can be provided. If the learning curve fluctuates greatly, it may mean that the student's learning is unstable, and further analysis of the reasons and adjustment of learning strategies are needed.

[0074] In this embodiment, based on the analysis results of personalized learning curves, the sample test questions for students are dynamically adjusted. Adjustments may include aspects such as test difficulty, knowledge point coverage, and question types. If a student's learning curve shows a good grasp of a certain knowledge point, the difficulty of that knowledge point can be increased, or related extended knowledge points can be introduced. If a student's learning curve shows difficulty with a certain knowledge point, the difficulty of the test questions can be reduced, or more practice and explanations can be provided. For example, if a student's learning curve for the "function" knowledge point shows a good grasp of the properties of basic functions, sample test questions involving more difficult knowledge points such as composite functions and inverse functions can be generated. If a student's learning curve for the "reading comprehension" knowledge point shows difficulty in understanding long articles, shorter, less difficult articles can be provided for practice.

[0075] Step 104: Collect students' learning feedback information and adjust the learning strategy based on the learning feedback information using the Harris Eagle optimization algorithm.

[0076] Collect student learning feedback information from sample test questions adjusted for learning curves to determine the optimization objective of the learning strategy; initialize the Harris eagle population, with each individual representing a combination of learning strategy parameters, calculate the fitness value of each individual, and select the best individuals as elite individuals based on their fitness values. Let the learning strategy parameters be p = {p1, p2, ..., p...} n The fitness function is f(p), and the position of an individual Harris Hawk is represented as P = {P1, P2, ..., P}. m The population size is m; The process begins with an iteration. In each iteration, individual Harris eagles update their positions according to different strategies, including an exploration phase and a development phase. During the exploration phase, Harris eagles employ a soft-encirclement exploration strategy, and the individual position update formula is:

[0077] P(t+1)=P rand (t)-r1|P rand (t)-2r2P(t)|

[0078] In the formula, P rand (t) represents the position of an individual randomly selected from the current population, r1 and r2 are random numbers in the range [0,1], and t represents the current iteration number. During the development phase, based on the escape energy E of the prey and the value of the random number r3, a development strategy is selected for position updates. When |E|≥1, a prey-chasing strategy is adopted; when |E|<1, a gradual soft-encirclement strategy is adopted for position updates. After updating the individual's position, the fitness value is recalculated. If the maximum number of iterations is reached, the iteration stops, and the optimal learning strategy parameter combination is output. Otherwise, the next iteration continues. The Harris Eagle optimization algorithm is used to output the optimal learning strategy parameter combination to adjust the student's learning strategy.

[0079] In this embodiment of the invention, generative artificial intelligence technology can generate personalized learning questions and learning curves based on students' learning progress, historical data, and mastery of knowledge points. This personalized design effectively improves students' learning outcomes and motivation. The learning curve can be dynamically updated based on real-time answer data, ensuring the flexibility and adaptability of the learning path. Through dynamic adjustment of the learning curve, the system can provide real-time learning suggestions and feedback to help students effectively cope with different learning challenges. By continuously optimizing the question generation algorithm, the difficulty and focus of the questions can be adjusted in real time based on students' answer data and learning curves, ensuring that students receive appropriate learning resources at each learning stage, thereby improving learning efficiency. Furthermore, the system can predict students' learning progress based on the learning curve and provide personalized learning feedback based on the prediction results, helping students adjust their learning strategies and ensuring the successful achievement of their final learning goals.

[0080] Please see Figure 2 A schematic diagram of a second embodiment of the learning curve generation method provided in this invention, the method comprising:

[0081] Step 201: Collect students' learning history data and use principal component analysis algorithm to reduce the dimensionality of the learning history data, including at least past test scores, homework completion status, and classroom participation performance;

[0082] Step 202: Train the variational autoencoder using the dimensionality-reduced learning history data. The variational autoencoder consists of an encoder and a decoder. Introduce the policy gradient algorithm into the variational autoencoder for performance optimization.

[0083] In this embodiment, the dimensionality-reduced learning history data is divided into a training set, a validation set, and a test set. The parameters of the encoder and decoder of the variational autoencoder are initialized, and the variational autoencoder is trained using the training set data. The gradient of the policy network is calculated using the policy gradient algorithm. The policy network generates personalized sample questions based on the current state, and the value of the current state is estimated through the value network, where the current state includes the learning topic and the learning history. Let the parameters of the policy network be w, and the state at time step t be s. t Action is a t The reward is s t The policy gradient is:

[0084]

[0085] In the formula, τ=(s0,a0,s1,a1,…) represents a trajectory, and π w (a t |s t ) indicates that the policy network is in state st Output action a t The probability, R(s) t ,a t ) indicates that it is in state s t Take action a t The reward obtained, where γ represents the discount factor and J(w) represents the objective function. Indicates that by strategy π w The expected value of the generated trajectory τ is obtained; the parameters of the variational autoencoder are updated using the stochastic gradient descent algorithm, and the parameters of the policy network and value network are updated using the Adam optimizer, so as to complete the performance optimization of the variational autoencoder through the policy gradient algorithm.

[0086] Step 203: Calculate the similarity between the generated test questions and the test questions that students have already answered, and add a regularization term related to the learning history data to the loss function of the variational autoencoder;

[0087] In this embodiment, let the generated test question be y, the set of test questions that students have already answered be S, and the similarity be sim(y,S). Then the regularization term can be expressed as:

[0088]

[0089] In the formula, λ represents the regularization coefficient;

[0090] Step 204: Select the model parameters that minimize the loss function as the optimal parameters, and output the encoder in the variational autoencoder of the optimal parameters to obtain the generative model. Generate personalized subsample test questions based on the learning topic and the generative model.

[0091] Please see Figure 3 A schematic diagram of a third embodiment of the learning curve generation method provided in this invention, the method comprising:

[0092] Step 301: Construct a long short-term memory network, introduce an attention mechanism into the long short-term memory network, increase residual connections, and use LeakyReLU as the activation function;

[0093] In this embodiment, the cell state can be viewed as a memory unit used to store long-term information. The forget gate determines which information is discarded from the cell state, the input gate determines which new information is added to the cell state, and the output gate determines which information is output from the cell state. For example, when processing text data, LSTM can read the text word by word, update the cell state based on the current word and previous memory states, and output the prediction result. For time series data, LSTM can predict future values ​​based on historical data. The attention mechanism allows the model to focus more on important parts of the input data, thereby improving the model's performance. Introducing an attention mechanism into LSTM allows the model to focus more on different input features at different time steps. For example, when processing text data, the attention mechanism allows the model to focus more on keywords or key sentences in the text. For image data, attention mechanisms allow models to focus more on important regions of the image; residual connections make models easier to train and improve their performance. Adding residual connections to LSTM allows the model to directly pass information between different layers, avoiding information loss. Residual connections are typically implemented by adding the input data to the output data after processing through several layers. This allows the model to learn the features of the input data more easily during training and avoids gradient vanishing or exploding problems. For example, when processing deep neural networks, residual connections allow the model to directly pass information between different layers, thereby improving model performance and training efficiency. When processing deep neural networks, LeakyReLU allows the model to achieve some activation in negative regions, thus improving model performance and generalization ability.

[0094] Step 302: Calculate the attention weights at each time step, and then multiply the attention weights by the hidden state to obtain the weighted hidden state;

[0095] In this embodiment, the calculation formula is:

[0096]

[0097] In the formula, h' t h represents the weighted hidden state. i Let α represent the hidden state at time step i. t,i This represents the attention weight of the t-th time step to the ith time step;

[0098] Step 303: At each time step, add the input and output to obtain the output after residual connection;

[0099] In this embodiment, the calculation formula is:

[0100] h t =h t +x t

[0101] In the formula, h t Let x represent the hidden state at time step t. t This represents the input at time step t;

[0102] Step 304: Input the answer feature data into the Long Short-Term Memory (LSTM) network. After processing through the forget gate, input gate, and output gate of the LTM network's hidden layer, the output layer predicts the student's learning performance indicators at future time steps. Using time as the horizontal axis and the predicted learning performance indicators as the vertical axis, connect the predicted values ​​at each time point to generate the student's personalized learning curve.

[0103] In this embodiment, the answer feature data can be characteristics such as the student's answer accuracy, answer time, and mastery of knowledge points. The answer feature data is used as input to the Long Short-Term Memory Network. After processing by the network, the predicted value of the student's learning performance index at future time steps can be obtained. For example, the student's answer accuracy at different time points can be used as input data and input into the Long Short-Term Memory Network. The network can predict the student's answer accuracy at future time points based on the historical answer accuracy. After processing through the forget gate, input gate, and output gate, the hidden layer of the Long Short-Term Memory Network consists of the forget gate, input gate, and output gate. These gates control the flow of information, thus enabling the handling of long-term dependency problems. The forget gate determines which information is discarded from the cell state, the input gate determines which new information is added to the cell state, and the output gate determines which information is output from the cell state. For example, when processing answer feature data, the forget gate can determine which historical answer information should be forgotten based on the input of the current time step and the hidden state of the previous time step; the input gate can determine which new answer information should be added to the cell state based on the input of the current time step and the hidden state of the previous time step; and the output gate can determine which information should be output from the cell state based on the input of the current time step and the cell state. After processing by the Long Short-Term Memory network, the output layer can predict students' learning performance indicators at future time steps. These indicators can be answer accuracy, answer time, knowledge mastery, etc. The output layer typically consists of a fully connected layer that maps the hidden state to the predicted values ​​of the learning performance indicators. For example, when processing answer feature data, the output layer can predict students' answer accuracy at future time points. The output of the output layer can be a value between 0 and 1, representing the predicted accuracy of the student's answers at a future point in time.

[0104] Please see Figure 4 A schematic diagram of the learning curve generation device provided in this embodiment of the invention. The device includes:

[0105] The test question generation module 401 is used to acquire students' learning history data, construct a generative model based on the learning history data, and use the generative model to generate personalized sample test questions according to the learning theme.

[0106] The data acquisition module 402 is used to collect students' answer data in real time, standardize and normalize the answer data, and extract key features from the processed answer data to obtain answer feature data.

[0107] The learning curve generation module 403 is used to generate personalized learning curves for students based on answer feature data, analyze the personalized learning curves, and dynamically adjust the sample test questions for students based on the personalized learning curves.

[0108] The feedback and optimization module 404 is used to collect students' learning feedback information and adjust the learning strategy based on the learning feedback information using the Harris Eagle optimization algorithm.

[0109] In this embodiment of the invention, by generating test questions, it is ensured that each student receives tailor-made questions during the learning process, thereby improving the pertinence and effectiveness of learning; the learning curve can be dynamically updated according to the student's real-time performance, reflecting their actual learning situation and providing more accurate prediction of learning progress; through data collection and analysis, the test question generation algorithm can be continuously optimized to improve the accuracy and practicality of the learning curve.

[0110] above Figure 4 The learning curve generation device in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The learning curve generation device in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0111] Figure 5 This is a schematic diagram of a learning curve generation device provided in an embodiment of the present invention. The learning curve generation device 600 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 610 (e.g., one or more processors) and a memory 620, and one or more storage media 630 (e.g., one or more mass storage devices) for storing application programs 633 or data 632. The memory 620 and storage media 630 can be temporary or persistent storage. The program stored in the storage media 630 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the learning curve generation device 600. Furthermore, the processor 610 may be configured to communicate with the storage media 630 and execute the series of instruction operations in the storage media 630 on the learning curve generation device 600 to implement the method provided in the above embodiment.

[0112] The learning curve generation device 600 may also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input / output interfaces 660, and / or one or more operating devices 631, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 5 The learning curve generation device structure shown does not constitute a limitation on the computer device provided by the present invention. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0113] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the various steps of the learning curve generation method provided in the above embodiments.

[0114] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described equipment or apparatus / unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0115] 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 technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0116] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A learning curve generation method characterized by, The learning curve generation method comprises the following steps: Acquire the learning history data of the student, construct a generative model based on the learning history data, and generate personalized sub-sample test questions according to the learning theme by using the generative model; Real-time collection of the student's answer data, standardization and normalization processing of the answer data, and extraction of key features from the processed answer data to obtain answer feature data; Generating a personalized learning curve for the student according to the answer feature data, including: constructing a long short-term memory network, inputting the answer feature data into the long short-term memory network, processing through the forget gate, input gate and output gate of the long short-term memory network hidden layer, and outputting the learning performance index of the student at the future time step by the output layer; taking time as the horizontal axis and the predicted learning performance index as the vertical axis, connecting the predicted values at each time point to generate the personalized learning curve of the student; analyzing the personalized learning curve, and dynamically adjusting the sub-sample test questions of the student according to the personalized learning curve, including: using the numerical differentiation method to calculate the slope of the personalized learning curve to determine the learning speed, finding the inflection point of the curve to determine the turning point of the learning, and obtaining the curve characteristics; evaluating the current learning state of the student according to the curve characteristics, combining the personalized learning curve and the learning state of the student to determine the weak links and key learning content of the student, and dynamically adjusting the sub-sample test questions of the student; Collecting student learning feedback information and adjusting the learning strategy based on this feedback information using the Harris Eagle optimization algorithm includes: collecting student learning feedback information on sample test questions after learning curve adjustment, determining the optimization objective of the learning strategy; initializing the Harris Eagle population, with each individual representing a combination of learning strategy parameters, calculating the fitness value of each individual, and selecting excellent individuals as elite individuals based on their fitness values. Let the learning strategy parameters be p = {p1, p2, ..., p...} n The fitness function is f(p), and the position of an individual Harris Hawk is represented as P = {P1, P2, ..., P}. m The population size is m. The process begins with an iteration where individuals of the Harris Eagle update their position according to different strategies in each iteration, including exploration and development phases. During the exploration phase, Harris Eagles employ a soft-surrounding exploration strategy, and the individual position update formula is: P(g+1) = P rand (g)-r1|P rand (g)-2r2P(g)|, where P rand (g) represents the position of an individual randomly selected from the current population, where r1 and r2 are random numbers in the range [0, 1], and g represents the current iteration number. During the development phase, based on the escape energy E of the prey and the value of the random number r3, a development strategy is selected for position updates. When |E|≥1, a prey-chasing strategy is adopted; when |E|<1, a gradual soft-encirclement strategy is adopted for position updates. After updating the individual's position, the fitness value is recalculated. If the maximum number of iterations is reached, the iteration stops, and the optimal learning strategy parameter combination is output. Otherwise, the next iteration continues. The Harris Eagle optimization algorithm is used to output the optimal learning strategy parameter combination to adjust the student's learning strategy.

2. The learning curve generation method of claim 1, wherein, The acquisition of the learning history data of the student, the construction of the generative model based on the learning history data, and the generation of personalized sub-sample test questions according to the learning theme by using the generative model, comprising: Collecting the learning history data of the student, and performing dimensionality reduction processing on the learning history data by using a principal component analysis algorithm, at least including past examination results, homework completion, and classroom participation performance; Training a variational autoencoder by using the dimensionally reduced learning history data, the variational autoencoder comprising an encoder and a decoder, and introducing a policy gradient algorithm into the variational autoencoder for performance optimization; Calculating the similarity between the generated test questions and the test questions already done by the student, adding a regularization term related to the learning history data in the loss function of the variational autoencoder, setting the generated test question as y, the set of test questions already done by the student as S, and the similarity as sim(y, S), then the regularization term is expressed as: In the formula, λ represents the regularization coefficient; Selecting the model parameters that minimize the loss function as the optimal parameters, and outputting the encoder of the variational autoencoder with the optimal parameters to obtain the generative model, and generating personalized sub-sample test questions according to the learning theme and the generative model.

3. The learning curve generation method of claim 2, wherein, The training of the variational autoencoder by using the dimensionally reduced learning history data, the variational autoencoder comprising an encoder and a decoder, and the introduction of a policy gradient algorithm into the variational autoencoder for performance optimization, comprising: Dividing the dimensionally reduced learning history data into a training set, a validation set and a test set, initializing the parameters of the encoder and the decoder of the variational autoencoder, and training the variational autoencoder using the training set data; The gradient of the policy network is calculated using a policy gradient algorithm, the policy network generates personalized subtest questions according to the current state, and the value of the current state is estimated by the value network, wherein the current state includes learning topics and learning history; Let the parameters of the policy network be w, the state at time step t be s t , the action be a t , the reward be s t , and the policy gradient be: In the formula, τ=(s0,a0,s1,a1,…) represents a trajectory, and π w (a t |s t ) indicates that the policy network is in state s t Output action a t The probability, R(s) t ,a t ) indicates that it is in state s t Take action a t The reward obtained, where γ represents the discount factor and J(w) represents the objective function. Indicates that by strategy π w The expected value of the generated trajectory τ; The parameters of the variational autoencoder are updated using a stochastic gradient descent algorithm, and the parameters of the policy network and the value network are updated using an Adam optimizer to optimize the performance of the variational autoencoder through a policy gradient algorithm.

4. The learning curve generation method of claim 1, wherein, The student's answer data is collected in real time, the answer data is standardized and normalized, and key features are extracted from the processed answer data to obtain answer feature data, including: The student's answer data is collected in real time, including at least answer time, answer content, accuracy and error type, and the collected data is preliminarily cleaned to remove error data and outliers, and then standardized and normalized; A convolutional neural network is constructed, including convolutional layers, pooling layers and fully connected layers, the convolutional layers are used to extract local features, the pooling layers are used to reduce the feature dimension, and the fully connected layers are used to integrate and classify the extracted features; The processed answer data is converted into word vector representation using the Word2Vec model, the answer data represented by the word vector is input into the convolutional neural network model, and after the convolutional layer and the pooling layer, the feature map is obtained, the feature map is integrated into a feature vector through the fully connected layer, and then the answer feature data is obtained through the forward propagation.

5. The learning curve generation method of claim 1, wherein, According to the answer feature data, a personalized learning curve of the student is generated, including: A long short-term memory network is constructed, an attention mechanism is introduced into the long short-term memory network, a residual connection is added, and LeakyReLU is used as an activation function; At each time step, the attention weight is calculated, and then the attention weight is multiplied by the hidden state to obtain the weighted hidden state: where h t denotes the weighted hidden state, h i denotes the hidden state at the i-th time step, α t,i denotes the attention weight from the t-th time step to the i-th time step; At each time step, the input and output are added to obtain the output after the residual connection: h t1 = h t + x t where h t denotes the hidden state at the t-th time step, x t denotes the input at the t-th time step; The answer feature data is input into the long short-term memory network, and after processing by the forget gate, the input gate and the output gate of the long short-term memory network hidden layer, the output layer predicts the learning performance index of the student at the future time step, and the personalized learning curve of the student is generated by connecting the predicted values at each time point.

6. The learning curve generation method of claim 1, wherein, The personalized learning curve is analyzed, and the subtest questions of the student are dynamically adjusted according to the personalized learning curve, and further including: According to the curve characteristics, the current learning state of the student is evaluated; the current learning state includes fast progress state, stable progress state, bottleneck state and regression state.

7. A learning curve generation device characterized by comprising: The learning curve generation device includes a memory and at least one processor, the memory stores instructions; the at least one processor calls the instructions in the memory to make the learning curve generation device execute the steps of the learning curve generation method in any one of claims 1-6.

8. A computer-readable storage medium having stored thereon instructions, the computer-readable storage medium comprising: The instructions are executed by the processor to implement the steps of the learning curve generation method in any one of claims 1-6.

Images