Construction method of adaptive learning model

Abstract

The invention provides a construction method of an adaptive learning model, and belongs to the technical field of education data processing. The method comprises the steps that a central server constructs a meta-learning model through online learning data, and initializes a global prediction model through the meta-learning model; the client obtains a meta-learning model and a global prediction model parameter, and initializes a local prediction model according to the global prediction model parameter; the client performs personalized updating on the local prediction model through the meta-learning model and the local learning data; integrating a local prediction model and a local recommendation service model; the local recommendation service model outputs recommendation content according to a prediction result of the local prediction model; and updating the local recommendation service model according to the historical recommendation content, the local learning data, the mastering probability of each knowledge point and the knowledge path perfection degree. According to the method, accurate personalized recommendation and global teaching strategy optimization are realized by constructing the adaptive learning model, and the learning experience of the user and the utilization efficiency of teaching resources can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of educational data processing, and in particular to a method for constructing an adaptive learning model. Background Art

[0002] In the current application of educational technology, especially in the field of online learning and intelligent teaching, there are many challenges. On the one hand, the system needs to provide personalized teaching resources and learning paths based on each user's learning behavior and knowledge mastery, which requires the system to have an accurate localization model to adapt to the specific needs of different users. On the other hand, the system also needs to understand and optimize the teaching content and methods from a macro perspective to adapt to a wide range of user groups, which requires the ability to process and analyze massive amounts of online learning data and extract global teaching strategies from it.

[0003] When faced with the needs of personalized teaching and global teaching strategy optimization, the federated learning model demonstrates its unique applicability and advantages. Federated learning allows local learning models to be deployed on each user's device. These models can be trained based on the user's personal data and learning behavior without uploading personal data to a central server, thereby protecting user privacy. While maintaining personal data privacy, federated learning achieves global knowledge extraction and sharing by aggregating local model parameters or updates on each device rather than directly aggregating the data itself.

[0004] However, students exhibit huge differences in their behavior patterns during the learning process, including learning time, learning methods, and concentration. These individual differences require the system to be highly flexible and able to capture and analyze each student's unique behavior patterns in real time. For users who have just joined the platform (especially new users without historical learning data), it is difficult for the system to accurately predict their learning preferences and knowledge mastery, making it difficult to immediately provide effective personalized learning resources, which is the "cold start" problem. The cold start problem may cause the resources obtained by new users in the early stages of learning to not match their actual needs, affecting user experience and learning outcomes. Summary of the invention

[0005] The purpose of the present invention is to provide a method for constructing an adaptive learning model. The method improves the learning ability of the global prediction model and the local prediction model during cold start by introducing a meta-learning model, thereby solving the technical problem that in a federated learning scenario, effective model training and optimization are difficult due to the small amount of online learning data.

[0006] A method for constructing an adaptive learning model, comprising: The central server builds a meta-learning model through online learning data and initializes the global prediction model through the meta-learning model; The client obtains the meta-learning model and the global prediction model parameters, and initializes the local prediction model according to the global prediction model parameters; The client uses the meta-learning model and local learning data to personalize and update the local prediction model. Integrate the local prediction model with the local recommendation service model; the local recommendation service model outputs recommended content based on the prediction results of the local prediction model; The local recommendation service model is updated based on historical recommendation content, local learning data, the probability of mastering each knowledge point, and the degree of perfection of the knowledge path.

[0007] Preferably, updating the local recommendation service model according to historical recommendation content, local learning data, mastery probability of each knowledge point and degree of perfection of knowledge path includes: Label the historical recommended content to obtain knowledge point labels; Obtain local learning data, the mastery probability of each knowledge point, the degree of perfection of the knowledge path, and the recommended evaluation indicators between the knowledge point labels; Update the local recommendation service model based on the recommendation evaluation indicators and knowledge point labels.

[0008] Preferably, the updating of the local recommendation service model according to the recommendation evaluation index and the knowledge point label comprises the steps of: Convert the model input parameters into gene sequences suitable for genetic algorithm processing to generate the initial population; Calculate the fitness value of each individual through the recommended evaluation indicators; Perform crossover on individuals to generate new individuals, perform mutation operations on the new individuals, and update the population based on the new individuals; Repeat the iteration until the stopping condition is reached, and select the individual with the highest fitness value as the new local recommendation service model.

[0009] Preferably, the training of the meta-learning model by online learning data comprises: Initialize meta-learning model parameters; Arrange the online learning data according to the user identifier to obtain a plurality of user data, and divide the user data into a training set and a test set; Randomly sample user data from the training set to generate several tasks, and the total number of tasks is recorded as N; Traverse each task in the batch task group, calculate the gradient of each meta-learning model parameter according to the support set of each task, and use the gradient descent method to obtain the first model update parameter group; Calculate the sum of the loss functions based on the query set of all tasks in the batch task group; Obtain a second model update parameter group using a gradient descent method according to the sum of the loss functions; updating the meta-learning model parameters according to the second model update parameter group; The iteration is repeated until the preset termination condition is reached.

[0010] Preferably, the method further comprises the steps of: the central server collecting local prediction model update parameters from all clients, updating the global prediction model according to all local prediction model update parameters, and distributing the updated global prediction model to each client.

[0011] Preferably, the central server collects local prediction model update parameters from all clients, updates the global prediction model according to all local prediction model update parameters, and distributes the updated global prediction model to each client, including: The central server generates a central private key and a central public key, and transmits the central public key to each client; The client encrypts the updated parameters of the local prediction model through homomorphic encryption technology and sends them to the central server; The central server collects the local prediction model update parameters of all clients, aggregates the local prediction model update parameters through homomorphic encryption, and decrypts the aggregate calculation results to obtain new global model parameters; The central server encrypts the new global model parameters using the central private key and sends them to each client; The client decrypts the received new global model parameters using the central public key.

[0012] Preferably, in the updating of the global prediction model according to all local prediction model update parameters, the global prediction model is expressed as: A new =sum(W i ) / N; Among them, A new is the global prediction model, W i is the local model parameter updated by the i-th client, and N is the total number of devices participating in federated learning.

[0013] Preferably, the online learning data is expressed as: X={x i │i=1,2, . . . n}, Where i represents the client number, x i represents the learning data transmitted by the i-th client, and n represents the total number of clients cooperating with the central server; The learning data is represented as: i ={x ij │j= 1,2, . . . m}, where j represents the category of the learning data, x ij represents the jth type of learning data transmitted by the i-th client, and m is the total number of types of learning data A system for building an adaptive learning model, comprising: Model training module, used to train the meta-learning model through online learning data; Model initialization module, used by the central server to initialize the global prediction model through the meta-learning model; A first data transmission module, used for sending the meta-learning model and the global prediction model to the client; A data processing module, used for the client to initialize the local prediction model through the global prediction model, and integrate the local prediction model with the local recommendation service model; The data update module is used by the client to perform personalized updates on the local prediction model through the meta-learning model and local learning data; The second data transmission module is used for the central server to collect local prediction model update parameters from all clients, update the global prediction model according to all local prediction model update parameters, and distribute the updated global prediction model to each client.

[0014] An electronic device comprises: a processor and a memory, wherein the memory is used to store computer program code, and the computer program code comprises computer instructions. When the processor executes the computer instructions, the electronic device executes the aforementioned method for constructing an adaptive learning model.

[0015] The beneficial effects of the present invention are: The present invention uses a meta-learning model to personalize and update the local prediction model according to local learning data, so that the local prediction model can quickly adapt to the specific needs and characteristics of local users, and can significantly improve the prediction accuracy of the local prediction model in the cold start phase; on the basis of the local prediction model, a local recommendation service model is integrated to provide personalized recommendations based on the data and needs of specific users; the local recommendation service model can be continuously and dynamically updated, so that the recommendation system can continuously optimize and improve the quality of recommendations. The present invention realizes accurate personalized recommendations and global teaching strategy optimization by constructing an adaptive learning model, which not only improves the learning experience and efficiency of individual users, but also improves the utilization efficiency of teaching resources and the teaching quality of the platform.

[0016] The embodiment of the present invention achieves accurate matching and personalized recommendation of recommended content by labeling historical recommended content, obtaining local learning data, the mastery probability of each knowledge point, the degree of perfection of the knowledge path and the recommendation evaluation index between the knowledge point label, and updating the local recommendation service model according to the recommendation evaluation index and the knowledge point label. At the same time, the weights of the local recommendation service model are continuously optimized through the genetic algorithm to ensure the improvement of the recommendation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0019] Figure 1 A flowchart of a method for constructing an adaptive learning model of the present invention; Figure 2 A schematic diagram of the process of constructing a meta-learning model through online learning data of the present invention; Figure 3 A flow chart of a central server of the present invention collecting local prediction model update parameters from all clients, updating a global prediction model according to all local prediction model update parameters, and distributing the updated global prediction model to each client; Figure 4 The figure is a schematic diagram of the hardware structure of an electronic device of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0022] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0023] When faced with the needs of personalized teaching and global teaching strategy optimization, the federated learning model demonstrates its unique applicability and advantages. Federated learning allows local learning models to be deployed on each user's device. These models can be trained based on the user's personal data and learning behavior without uploading personal data to a central server, thereby protecting user privacy. While maintaining personal data privacy, federated learning achieves global knowledge extraction and sharing by aggregating local model parameters or updates on each device rather than directly aggregating the data itself.

[0024] However, students exhibit huge differences in their behavior patterns during the learning process, including learning time, learning methods, and concentration. These individual differences require the system to be highly flexible and able to capture and analyze each student's unique behavior patterns in real time. For users who have just joined the platform (especially new users without historical learning data), it is difficult for the system to accurately predict their learning preferences and knowledge mastery, making it difficult to immediately provide effective personalized learning resources, which is the "cold start" problem. The cold start problem may cause the resources obtained by new users in the early stages of learning to not match their actual needs, affecting user experience and learning outcomes.

[0025] The present invention uses a meta-learning model to perform personalized updates on the local prediction model based on local learning data, so that the local prediction model can quickly adapt to the specific needs and characteristics of local users, and can significantly improve the prediction accuracy of the local prediction model in the cold start phase; on the basis of the local prediction model, a local recommendation service model is integrated to provide personalized recommendations based on the data and needs of specific users; the local recommendation service model can be continuously and dynamically updated, so that the recommendation system can continuously optimize and improve the quality of recommendations. The present invention realizes accurate personalized recommendations and global teaching strategy optimization, which not only improves the learning experience and efficiency of individual users, but also improves the utilization efficiency of teaching resources and the teaching quality of the platform.

[0026] Example 1 A method for constructing an adaptive learning model, reference Figure 1 , including the steps of: Step S1, the central server constructs a meta-learning model through online learning data, and initializes the global prediction model through the meta-learning model; Among them, online learning data refers to the data related to the learning process uploaded by users through the client. Its data source includes the real-time learning behavior data of all students on the online education platform, such as course clicks, homework submissions, test results, learning paths, interactive discussions, etc.; or facial expressions, posture information, learning material information, behavior information, etc. obtained through image acquisition during the learning process. The central server constructs a meta-learning model through online learning data. The model contains rich prior knowledge and can effectively initialize the global prediction model.

[0027] Step S2, the client obtains the meta-learning model and the global prediction model parameters, and initializes the local prediction model according to the global prediction model parameters; In the present invention, the central server and the client adopt a federated learning architecture. Federated learning is a distributed learning method that allows multiple entities (such as mobile devices or entire organizations) to cooperate in training models without directly sharing original data. In the setting of federated learning, the central server is a node with relatively large computing power and storage capacity, responsible for coordinating and controlling the entire federated learning process. Its main responsibilities include: initializing the global prediction model, coordinating the model training of each client, collecting the model updates of each client, aggregating these updates to update the global prediction model, and distributing the new global prediction model to each client. The client can be any kind of device, on which data is generated and stored. Each client performs model training based on the global prediction model provided by the central server and its own local data to obtain a local prediction model, and then sends the update of the local prediction model relative to the global prediction model back to the central server. Under the federated learning architecture, each client can use local data for model training, and does not need to send the original data to the central server or other clients, so that data privacy can be protected. At the same time, the central server aggregates the model updates of all clients to update the global prediction model, so that the global prediction model can learn from the data of all clients, not just the data of a single client, thereby improving the performance and generalization ability of the model.

[0028] Step S3, the client performs personalized update of the local prediction model through the meta-learning model and the local learning data; The meta-learning model can quickly learn and adapt to new tasks with a small amount of data. Then each client (that is, the user's device) has its own local learning data, which contains various data accumulated by the user during the learning process. When the local prediction model needs to be updated, the local learning data is input into the meta-learning model. Because the meta-learning model has the ability to quickly learn new tasks, it can understand this data and learn accordingly, and update the local prediction model based on the learning results. In this way, the local prediction model can be updated with a small amount of local data to adapt to the personalized needs of the local user of the client.

[0029] Step S4, integrating the local prediction model with the local recommendation service model; the local recommendation service model outputs recommended content according to the prediction result of the local prediction model; The local recommendation service model outputs recommended content according to the prediction results of the local prediction model. By integrating the local recommendation service model with the local prediction model in the client, the local recommendation service model can recommend content based on the prediction results output by the local prediction model.

[0030] The local prediction model focuses on understanding and predicting the user's learning status, while the local recommendation service model provides specific and personalized learning content recommendations based on the prediction results of the local prediction model, that is, the user's learning status. These two models work together to form an adaptive learning system that can provide users with a customized learning experience and help them achieve their learning goals in the most effective way.

[0031] Step S5, updating the local recommendation service model according to historical recommendation content, local learning data, mastery probability of each knowledge point and degree of perfection of the knowledge path.

[0032] The local recommendation service model updates itself based on the user's historical recommendation content, local learning data, the probability of mastering each knowledge point, and the degree of perfection of the knowledge path. This means that the system will continuously monitor the user's learning progress and behavior data, thereby continuously optimizing the recommendation strategy of the local recommendation service model.

[0033] In the present invention, local learning data is used to update and optimize the local model, provide exclusive recommendations for individual users, and realize localized adaptation of the global model. Online learning data is the learning data uploaded to the central server by several clients through relevant security protocols, which is used for global processing of the meta-learning model, realizes the integration and refinement of multiple user data, and provides a widely applicable global prediction model.

[0034] The present invention uses a meta-learning model to perform personalized updates on the local prediction model based on local learning data, so that the local prediction model can quickly adapt to the specific needs and characteristics of local users, and can significantly improve the prediction accuracy of the local prediction model in the cold start phase; on the basis of the local prediction model, a local recommendation service model is integrated to provide personalized recommendations based on the data and needs of specific users; the local recommendation service model can be continuously and dynamically updated, so that the recommendation system can continuously optimize and improve the quality of recommendations.

[0035] Preferably, updating the local recommendation service model according to historical recommendation content, local learning data, mastery probability of each knowledge point and degree of perfection of knowledge path includes: Step S51, labeling the historical recommended content to obtain knowledge point labels; Knowledge point labels are used to describe the knowledge point attributes of a historical recommendation. Labeling historical recommendation content facilitates subsequent data analysis and model training. For example, if a user's historical recommendation content is "Solve the equation (x^2 - 5x + 6 = 0)", and its solution methods include "factorization" and "root-finding formula", then the historical recommendation content will be associated with three knowledge point labels: "quadratic equation", "factorization" and "root-finding formula".

[0036] Step S52, obtaining local learning data, the mastering probability of each knowledge point, the degree of perfection of the knowledge path and the recommended evaluation index between the knowledge point label; Recommendation evaluation indicators include accuracy, coverage, and diversity. The accuracy indicator is used to measure the degree of match between knowledge point labels and actual user acceptance. For example, if the recommended content is a course with the knowledge point label "factorization", and the local learning data identifies that the user has actually completed the course with the knowledge point label "factorization", and the probability of mastering the knowledge point corresponding to the knowledge point label "factorization" increases, then this recommendation is considered accurate.

[0037] The coverage index is used to measure the increment of the degree of perfection of the knowledge path; it can be understood that the degree of perfection of the knowledge path can be quantified by a value ranging from 0% to 100%; if the degree of perfection of the knowledge path is improved, it means that the recommended content involves the content of a key point of a certain knowledge path; for example, the aforementioned historical recommended content is associated with the three knowledge point labels of "quadratic equation", "factorization" and "root-finding formula", then the historical recommended content is associated with the knowledge path label "polynomial equation". When a user has a low level of mastery of the knowledge points of "quadratic equation", "factorization" and / or "root-finding formula", after learning the recommended content, the more the degree of perfection of the knowledge path corresponding to the knowledge path label "polynomial equation" increases, the greater the value of the coverage index.

[0038] Diversity indicators are used to measure the number of knowledge points involved in the recommended content, so that the recommended content tends to include more knowledge points. For example, when it is identified that the user has a low level of mastery of the knowledge point "quadratic equation", it can not only match the user with "factorization" and "root formula" related to "quadratic equation" in mathematics, but also match the recommended content corresponding to related knowledge points such as "parabola" and "free fall motion" in physics.

[0039] Step S53: update the local recommendation service model according to the recommendation evaluation index and the knowledge point label.

[0040] The recommendation evaluation index can be regarded as a measure of the model recommendation effect. After determining the recommendation evaluation index corresponding to the recommended content, the values ​​of all weights of the local recommendation service model are updated through the genetic algorithm to improve the value of the recommendation evaluation index. After the model update is completed, the model needs to be verified to ensure that the updated model has indeed improved the recommendation effect. The verification method can be to compare the difference between the comprehensive evaluation indicators before and after the update. If the model verification result meets the preset effect improvement threshold, the current round of model update is terminated. If not, the genetic algorithm needs to be continued for the next round of model update. The process usually needs to be repeated to continuously improve the recommendation effect.

[0041] Furthermore, the local recommendation service model can be trained by methods such as gradient descent, but these methods usually rely on local information (such as gradients) and are prone to falling into local optimal solutions, while genetic algorithms can search for optimal solutions globally. When processing the matching of recommended content according to the user's learning status, the amount of training calculations for the neural network is very large, while the genetic algorithm can reduce unnecessary calculations by optimizing the network structure, and the genetic algorithm can be used to adaptively find the optimal parameter configuration. With respect to the process of updating the local service model by a genetic algorithm of the present invention, as a preferred embodiment, the updating of the local recommendation service model according to the recommendation evaluation index and the knowledge point label includes the steps of: Step S531, converting the model input parameters into gene sequences suitable for genetic algorithm processing; generating an initial population; Initializing the population; determining the encoding scheme of the local recommendation service model, and converting the model input parameters into a gene sequence suitable for genetic algorithm processing; the initial population includes a number of model individuals, each of which is composed of different parameter configurations; Step S532, calculating the fitness value of each individual by using the recommended evaluation index; Evaluate the population; evaluate each individual (i.e. each set of parameter configurations) using the previously defined recommended evaluation indicators, including accuracy, coverage, and diversity indicators; calculate the fitness value of the individual. The fitness value can be the weighted sum of the evaluation indicators, and the specific calculation formula can be flexibly adjusted according to actual needs.

[0042] Step S533, selecting individuals according to fitness values; Selection process: select excellent individuals according to their fitness values, with individuals with high fitness values ​​having a higher probability of being selected. Selection methods include roulette selection and tournament selection.

[0043] Step S534, crossover the individuals to generate new individuals, perform mutation operations on the new individuals, and update the population based on the new individuals; Crossover and mutation: Perform crossover operations to combine selected individuals to generate new individuals; crossover methods include single-point crossover and multi-point crossover. Perform mutation operations on new individuals to randomly change certain genes with low probability to ensure population diversity and explore new solution space.

[0044] Update the population; add new individuals generated by crossover and mutation to the population to form a new generation of population; retain the elite individuals (individuals with the highest fitness values) from previous generations of populations to avoid the optimal solution being discarded during the update process.

[0045] Step S535, repeat the iteration until the stop condition is reached, and select the individual with the highest fitness value as the new local recommendation service model; Repeat steps S532-S534 and determine whether the stopping condition is met. The stopping condition may be reaching a predetermined fitness value, running to a specified number of generations, or the fitness value of the population converges. The stopping conditions include: the fitness value reaches a preset threshold, reaches a maximum number of iterations, and the fitness value change is less than a preset value.

[0046] This embodiment achieves accurate matching and personalized recommendation of recommended content by labeling historical recommended content, obtaining local learning data, the probability of mastering each knowledge point, the degree of perfection of the knowledge path and the recommendation evaluation index between the knowledge point label, and updating the local recommendation service model according to the recommendation evaluation index and the knowledge point label. At the same time, the weights of the local recommendation service model are continuously optimized through the genetic algorithm to ensure the improvement of the recommendation effect.

[0047] Preferably, step S1, referring to Figure 2 , the construction of a meta-learning model through online learning data includes: Step S110, initializing meta-learning model parameters; Meta-learning refers to the process of improving learning algorithms in multiple learning stages. In the basic learning process, the internal (or lower layer / basic) learning algorithm solves the tasks defined by the data set and the target. In the meta-learning process, the external (or upper layer / meta) algorithm updates the internal learning algorithm so that the model it learns improves the external target. Meta-learning includes two layers: the first layer is to let the machine learn to learn, so that it has the ability to analyze and solve problems, and the machine gains experience by completing tasks and improves its ability to complete tasks; the second layer is to allow machine learning models to better generalize to new fields, so as to complete new tasks with great differences. Few-Shot Learning is the application of Meta-Learning in the field of supervised learning. In the Meta-training stage, the data set is decomposed into different tasks to learn the generalization ability of the model when the category changes. In the Meta-testing stage, in the face of a new category, there is no need to change the existing model, and only one or a few steps of training are needed to meet the requirements. The initialized parameters will be optimized through meta-training in the subsequent process to ensure that the meta-learning model can better adapt to the application scenarios in the present invention.

[0048] Step S120, sorting the online learning data according to the user identifier to obtain a number of user data, and dividing the user data into a training set and a test set; It can be understood that both the training set and the test set include multiple pieces of user data.

[0049] Step S130, randomly sampling user data from the training set to generate a number of tasks, and randomly selecting a number of tasks to form a batch task group; The task includes a support set and a query set, both of which include multiple pieces of user data, and the total number of tasks is recorded as N; a number of tasks are randomly selected to form a batch task group, each batch task group includes M tasks, 1≤M ≤N.

[0050] Step S140, traversing each task in the batch task group, calculating the gradient of each meta-learning model parameter according to the support set of each task, and obtaining a first model update parameter group using a gradient descent method; Among them, the support set of the task is used to calculate the gradient of each meta-learning model parameter. The gradient of the meta-learning model parameter is the partial derivative of the loss function with respect to the meta-learning model parameter, and the loss function adopts the MSE (mean error) function. In this embodiment, the gradient describes the direction in which the loss function changes fastest in the parameter space. The size of the gradient value indicates the rate of change of the loss function, and the direction indicates the direction in which the loss function changes steepest.

[0051] Use the gradient descent method to obtain the first model update parameter group, that is, adjust the meta-learning model parameters in the opposite direction of the gradient to make the value of the loss function decrease as quickly as possible. The adjusted meta-learning model parameters are the first model update parameter group. The first model update parameter group is not directly used to replace the original meta-learning model parameters, but is used to calculate the second model update parameter group.

[0052] Step S150, calculating the sum of loss functions according to the query sets of all tasks in the batch task group; Step S160, obtaining a second model update parameter group using a gradient descent method according to the sum of the loss functions; Step S170, updating the meta-learning model parameters according to the second model update parameter group; The purpose of step S140 is to calculate the first gradient update, and steps S150-S160 calculate the second gradient update by gradient by gradient (learning how to learn by gradient descent by gradient descent) based on the first model update parameter group obtained by the first gradient update. The second model update parameter group calculated during the second gradient update directly acts on the meta-learning model by gradient descent, that is, the meta-learning model parameters are updated according to the second model update parameter group.

[0053] Step S180, repeat the iteration until a preset termination condition is reached.

[0054] After repeating steps S130-170 until a preset termination condition is reached, the performance evaluation index of the meta-learning model is calculated using the test set.

[0055] Preferably, the online learning data is represented as: X={x i │i=1,2, . . . n}, Where i represents the client number, x i represents the learning data transmitted by the i-th client, and n represents the total number of clients cooperating with the central server; The learning data is represented as: x i ={x ij │j= 1,2, . . . m}, where j represents the category of the learning data, x ij represents the j-th type of learning data transmitted by the ith client, and m is the total number of categories of learning data.

[0056] In one embodiment, m=6, and the categories of learning data include total learning time, learning subjects, knowledge point labels, course progress, answering behavior data, and learning feedback data. It can be understood that the learning data can represent the learning data in the online learning data used to train the meta-learning model in the central server, and can also represent the local learning data used to update the local prediction model.

[0057] As a preferred embodiment, the method for constructing an adaptive learning model of the present invention further includes the steps of: Step S6, the central server collects local prediction model update parameters from all clients, updates the global prediction model according to all local prediction model update parameters, and distributes the updated global prediction model to each client; repeat steps S3-S6 until the preset termination condition is reached.

[0058] Specifically, the central server collects the local prediction model update parameters from all clients. The clients here can be regarded as the devices of each user. Each device has a local model for learning and inference. After processing individual user data and learning, these devices will form new model parameters, which are the results of local learning. The central server collects the parameters of the local models from all user devices. The global prediction model is updated according to all model update parameters. After collecting the local model parameters of all clients, the central server will summarize and analyze them and use these parameters to update the global prediction model. This global model represents the sum of the learning effects of all users and is a unified representation of the global situation. The updated global prediction model is distributed to each client. After the update is completed, the central server will distribute the new global model to each client. At this time, the global model has been updated according to the overall model update weight of each client. Distributing the updated global prediction model to each client can help each user get better prediction results.

[0059] Among them, the termination conditions may include various ones, such as reaching a predetermined number of iterations or time, reaching a predetermined level of model performance, meeting the requirements for model optimization, etc. Only when this condition is met will the repetition of steps S3 to S6 stop. This can ensure that the operation of the model or algorithm achieves the expected effect and avoids ineffective operation consumption.

[0060] Preferably, in the updating of the global prediction model according to all local prediction model update parameters, the global prediction model is expressed as: A new =sum(W i ) / N Among them, A new is the global prediction model, W i is the local model parameter updated by the i-th client, and N is the total number of devices participating in federated learning.

[0061] The present invention makes personalized recommendations based on the specific needs of each customer, collects local behavior data of users, builds a data set of user interaction records and attribute information features, and then extracts user preferences and behavior pattern features that are consistent with the user's personalized needs from the original data. Based on a deep learning model, the local data is divided into a training set and a validation set. The model is trained using the user's local data, and through iterative optimization, the model is made to learn the user's specific behavior patterns and demand characteristics. Finally, the model performance is evaluated on the validation set, and the trained model is used to predict the content that the user will be interested in in the future, generating a personalized recommendation list.

[0062] Preferably, reference Figure 3 , step S6, the central server collects local prediction model update parameters from all clients, updates the global prediction model according to all local prediction model update parameters, and distributes the updated global prediction model to each client, including: The central server generates a central private key and a central public key, and transmits the central public key to each client; The client encrypts the updated parameters of the local prediction model through homomorphic encryption technology and sends them to the central server; The central server collects the local prediction model update parameters of all clients, aggregates the local prediction model update parameters through homomorphic encryption, and decrypts the aggregate calculation results to obtain new global model parameters; The central server encrypts the new global model parameters using the central private key and sends them to each client; The client decrypts the received new global model parameters using the central public key.

[0063] This embodiment combines homomorphic encryption and asymmetric encryption technology to ensure data security during federated learning, while ensuring the integrity and privacy of data during transmission and calculation. This approach effectively improves the security, privacy protection level and calculation accuracy of the system.

[0064] Example 2 A system for building an adaptive learning model, comprising: Model training module, used to train the meta-learning model through online learning data; Model initialization module, used by the central server to initialize the global prediction model through the meta-learning model; A first data transmission module, used for sending the meta-learning model and the global prediction model to the client; A data processing module, used for the client to initialize the local prediction model through the global prediction model, and integrate the local prediction model with the local recommendation service model; The data update module is used by the client to perform personalized updates on the local prediction model through the meta-learning model and local learning data; The second data transmission module is used for the central server to collect local prediction model update parameters from all clients, update the global prediction model according to all local prediction model update parameters, and distribute the updated global prediction model to each client.

[0065] Example 3 An electronic device includes: a processor and a memory, the memory is used to store computer program code, the computer program code includes computer instructions, when the processor executes the computer instructions, the electronic device executes a method for constructing an adaptive learning model.

[0066] refer to Figure 4 , the electronic device 2 includes a processor 21, a memory 22, an input device 23, and an output device 24. The processor 21, the memory 22, the input device 23, and the output device 24 are coupled via a connector, and the connector includes various interfaces, transmission lines, or buses, etc., which are not limited in the embodiments of the present invention. It should be understood that in various embodiments of the present invention, coupling refers to mutual connection in a specific manner, including direct connection or indirect connection through other devices, for example, through various interfaces, transmission lines, buses, etc.

[0067] The processor 21 may be one or more graphics processing units (GPUs). When the processor 21 is a GPU, the GPU may be a single-core GPU or a multi-core GPU. Optionally, the processor 21 may be a processor group consisting of multiple GPUs, and the multiple processors are coupled to each other via one or more buses. Optionally, the processor may also be other types of processors, etc., which are not limited in the embodiments of the present invention.

[0068] The memory 22 can be used to store computer program instructions and various computer program codes including program codes for executing the scheme of the present invention. Optionally, the memory includes but is not limited to random access memory (RAM), read-only memory (ROM), erasable programmable read only memory (EPROM), or portable read only memory (CD-ROM), which is used for related instructions and data.

[0069] The input device 23 is used to input data and / or signals, and the output device 24 is used to output data and / or signals. The output device 24 and the input device 23 may be independent devices or an integrated device.

[0070] In the federated learning architecture of the present invention, each client can use local data for model training, and does not need to send the original data to the central server or other clients, so that data privacy can be protected. At the same time, the central server aggregates the model updates of all clients to update the global prediction model, so that the global prediction model can learn from the data of all clients, not just the data of a single client, thereby improving the performance and generalization ability of the model. The central server will distribute the new global model to each client. At this time, the global model has been updated according to the overall model update weights of each client. After distributing the updated global prediction model to each client, it can help each user get better prediction results.

[0071] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for constructing an adaptive learning model, characterized in that: include: The central server builds a meta-learning model through online learning data and initializes the global prediction model through the meta-learning model; The client obtains the meta-learning model and the global prediction model parameters, and initializes the local prediction model according to the global prediction model parameters; The client uses the meta-learning model and local learning data to personalize and update the local prediction model. Integrate the local prediction model with the local recommendation service model; the local recommendation service model outputs recommended content based on the prediction results of the local prediction model; The local recommendation service model is updated based on historical recommendation content, local learning data, the probability of mastering each knowledge point, and the degree of perfection of the knowledge path.

2. The method for constructing an adaptive learning model according to claim 1, characterized in that: The updating of the local recommendation service model according to the historical recommendation content, local learning data, the mastery probability of each knowledge point and the degree of perfection of the knowledge path includes: Label the historical recommended content to obtain knowledge point labels; Obtain local learning data, the mastery probability of each knowledge point, the degree of perfection of the knowledge path, and the recommended evaluation indicators between the knowledge point labels; Update the local recommendation service model based on the recommendation evaluation indicators and knowledge point labels.

3. The method for constructing an adaptive learning model according to claim 2, characterized in that: The updating of the local recommendation service model according to the recommendation evaluation index and the knowledge point label comprises the following steps: Convert the model input parameters into gene sequences suitable for genetic algorithm processing to generate the initial population; Calculate the fitness value of each individual through the recommended evaluation indicators; Perform crossover on individuals to generate new individuals, perform mutation operations on the new individuals, and update the population based on the new individuals; Repeat the iteration until the stopping condition is reached, and select the individual with the highest fitness value as the new local recommendation service model.

4. The method for constructing an adaptive learning model according to claim 1, characterized in that: The meta-learning model training by online learning data includes: Initialize meta-learning model parameters; Arrange the online learning data according to the user identifier to obtain a plurality of user data, and divide the user data into a training set and a test set; Randomly sample user data from the training set to generate several tasks, and the total number of tasks is recorded as N; Traverse each task in the batch task group, calculate the gradient of each meta-learning model parameter according to the support set of each task, and use the gradient descent method to obtain the first model update parameter group; Calculate the sum of the loss functions based on the query set of all tasks in the batch task group; Obtain a second model update parameter group using a gradient descent method according to the sum of the loss functions; updating the meta-learning model parameters according to the second model update parameter group; The iteration is repeated until the preset termination condition is reached.

5. The method for constructing an adaptive learning model according to claim 1, characterized in that: The method also includes the following steps: a central server collects local prediction model update parameters from all clients, updates the global prediction model according to all local prediction model update parameters, and distributes the updated global prediction model to each client.

6. The method for constructing an adaptive learning model according to claim 5, characterized in that: The central server collects local prediction model update parameters from all clients, updates the global prediction model according to all local prediction model update parameters, and distributes the updated global prediction model to each client, including: The central server generates a central private key and a central public key, and transmits the central public key to each client; The client encrypts the updated parameters of the local prediction model through homomorphic encryption technology and sends them to the central server; The central server collects the local prediction model update parameters of all clients, aggregates the local prediction model update parameters through homomorphic encryption, and decrypts the aggregate calculation results to obtain new global model parameters; The central server encrypts the new global model parameters using the central private key and sends them to each client; The client decrypts the received new global model parameters using the central public key.

7. The method for constructing an adaptive learning model according to claim 5, characterized in that: In the updating of the global prediction model according to the updating parameters of all local prediction models, the global prediction model is expressed as: A new =sum(W i ) / N; Among them, A new is the global prediction model, W i is the local model parameter updated by the i-th client, and N is the total number of devices participating in federated learning.

8. The method for constructing an adaptive learning model according to claim 1, characterized in that: The online learning data is represented as: X={x i │i=1,2, . . . n}, Where i represents the client number, x i represents the learning data transmitted by the i-th client, and n represents the total number of clients cooperating with the central server; The learning data is represented as: i ={x ij │j= 1,2, . . . m}, where j represents the category of the learning data, x ij represents the j-th type of learning data transmitted by the ith client, and m is the total number of categories of learning data.

9. A system for constructing an adaptive learning model, characterized in that: include: Model training module, used to train the meta-learning model through online learning data; Model initialization module, used by the central server to initialize the global prediction model through the meta-learning model; A first data transmission module, used for sending the meta-learning model and the global prediction model to the client; A data processing module, used for the client to initialize the local prediction model through the global prediction model, and integrate the local prediction model with the local recommendation service model; The data update module is used by the client to perform personalized updates on the local prediction model through the meta-learning model and local learning data; The second data transmission module is used for the central server to collect local prediction model update parameters from all clients, update the global prediction model according to all local prediction model update parameters, and distribute the updated global prediction model to each client.

10. An electronic device, characterized in that: include: A processor and a memory, wherein the memory is used to store computer program code, wherein the computer program code includes computer instructions, and when the processor executes the computer instructions, the electronic device executes a method for constructing an adaptive learning model as described in any one of claims 1 to 8.

Images