Product recommendation method, apparatus, device, medium, and program product

By introducing regularization constraint parameters and gradient compression techniques into federated learning, the problems of model performance degradation and privacy leakage caused by non-independent and identically distributed data in the financial industry are solved, and efficient and secure product recommendation is achieved.

CN123118199APending Publication Date: 2026-11-17INDUSTRIAL AND COMMERCIAL BANK OF CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611025630.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-10
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In the financial industry, the data of each participant in federated learning is usually statistically heterogeneous and non-independently identically distributed data, which leads to a significant decline in the performance of the global model after training, model non-convergence, reduced accuracy, inaccurate product recommendations, and privacy leakage risks and communication overhead issues.

Method used

By introducing regularization constraint parameters, we can adapt to the non-independent and identically distributed data of each client, alleviate the global model update offset, reduce communication volume by using gradient compression technology, and perform aggregate training and local model fine-tuning on the server side to ensure data security and compliance.

Benefits of technology

It effectively avoids the global model from deviating from the optimal solution, improves the accuracy and security of product recommendations, meets compliance and data security requirements, and reduces communication overhead and privacy leakage risks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118199A_ABST
    Figure CN123118199A_ABST
Patent Text Reader

Abstract

This application provides a product recommendation method, apparatus, device, medium, and program product, which can be applied to the fields of artificial intelligence and fintech. The method includes: calling a trained local model to process the target user's business feature data and product feature data to obtain a product matching prediction result; wherein the trained local model is obtained by federated learning of a global model and a local model based on regularization constraint parameters and non-independent identically distributed data through m clients and a server; m is an integer greater than 1, and the m clients include the target client; and recommending target products to the target user based on the product matching prediction result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and fintech, specifically to a product recommendation method, apparatus, device, medium, and program product, and more specifically to a product recommendation method, apparatus, device, medium, and program product based on federated learning. Background Technology

[0002] Data has become a key new production factor, and the demand for data value mining and cross-institutional sharing in the financial industry continues to rise. Financial institutions are generally facing business bottlenecks such as low conversion rates for new customer acquisition, difficulty in activating dormant users, and insufficient accuracy in wealth management recommendations. Due to compliance and data security constraints, raw data cannot be directly shared, making federated learning a major solution for multi-party collaborative construction of wealth management product recommendation models. However, the local customer groups and business data of each participating party exhibit significant statistical heterogeneity, and the samples show typical non-independent identically distributed characteristics, which is an objective reality that cannot be avoided in cross-institutional / cross-domain joint modeling scenarios.

[0003] Currently, user product recommendation models are mainly built using federated learning technology. However, the data from each participant in federated learning is usually statistically heterogeneous and non-independently identically distributed data, which can easily lead to a significant decrease in the performance of the global model obtained after training, resulting in model non-convergence, reduced accuracy, and the non-independently identically distributed data causing the global model to deviate from the global optimum, resulting in inaccurate product recommendations. Summary of the Invention

[0004] In view of the above problems, embodiments of this application provide a product recommendation method, apparatus, device, medium, and program product based on federated learning.

[0005] According to a first aspect of this application, a product recommendation method based on federated learning is provided, applied to a target client. The method includes: calling a trained local model to process the target user's business feature data and product feature data to obtain a product matching prediction result; wherein the trained local model is obtained by federated learning on a global model and a local model based on regularization constraint parameters and non-independent identically distributed data through m clients and a server; m is an integer greater than 1, and the m clients include the target client; and recommending target products to the target user based on the product matching prediction result.

[0006] According to an embodiment of this application, the step of performing federated learning training on a global model based on regularization constraint parameters and non-independent and identically distributed data through m clients and a server includes: receiving a global model from the server through the m clients; performing local model gradient compression training through the m clients based on the local data of the m clients and the global model to obtain local models corresponding to the m clients, and determining the parameters of the local models corresponding to the m clients; wherein the local data of the m clients belongs to the non-independent and identically distributed data; and uploading the parameters of the local models corresponding to the m clients to the server through the m clients; so that the... The server receives the local model parameters corresponding to the m clients, performs aggregate training on the global model based on the regularization constraint parameters and the local model parameters corresponding to the m clients, obtains a trained global model, and distributes the trained global model to the m clients; the m clients receive the trained global model; and the m clients, based on the trained global model, fine-tune the local models corresponding to the m clients to obtain the trained local models corresponding to the m clients; wherein, the trained local models corresponding to the m clients are used by the m clients to recommend target products to target users.

[0007] According to a second aspect of this application, a product recommendation method based on federated learning is provided, applied to m clients, where m is an integer greater than 1. The method includes: receiving a global model from a server; performing gradient compression training on local models based on local data from the m clients and the global model to obtain local models corresponding to the m clients, and determining the parameters of the local models corresponding to the m clients; wherein the local data from the m clients is non-independent and identically distributed data; uploading the parameters of the local models corresponding to the m clients to the server; enabling the server to receive the parameters of the local models corresponding to the m clients, perform aggregate training on the global model based on regularization constraint parameters and the parameters of the local models corresponding to the m clients, obtain a trained global model, and distribute the trained global model to the m clients; receiving the trained global model; and fine-tuning the local models corresponding to the m clients based on the trained global model to obtain trained local models corresponding to the m clients; wherein the trained local models corresponding to the m clients are used by the m clients to recommend target products to target users.

[0008] According to an embodiment of this application, the step of fine-tuning the local models corresponding to the m clients based on the trained global model includes: calculating the Euclidean distance between the global model parameters corresponding to the trained global model and the local model parameters of the current client to obtain the Euclidean norm; determining a parameter offset penalty term based on the fine-tuning hyperparameters of the current client and the Euclidean norm; constructing a local objective function with the goal of minimizing the cross-entropy loss of the local model of the current client, and based on the parameter offset penalty term; and fine-tuning and updating the parameters of the local model of the current client based on the local data of the current client and the local objective function, until the local model of the current client converges.

[0009] According to an embodiment of this application, the step of performing local model gradient compression training based on the local data of the m clients and the global model includes: constructing a local model for the current client based on the global model; iteratively training the local model for the current client based on the local data of the current client; during the iterative training process, calculating the change in model parameters of the local model of the current client in two adjacent training rounds, and using a front-end truncation method to select and retain key gradients whose absolute value proportion of the change in model parameters meets a preset ratio, so as to update the local model parameters corresponding to the current client.

[0010] According to a third aspect of this application, a product recommendation method based on federated learning is provided, applied on a server side. The method includes: distributing a global model to m clients, enabling the m clients to receive the global model; performing gradient compression training on local models based on the local data of the m clients and the global model to obtain local models corresponding to the m clients, and determining the parameters of the local models corresponding to the m clients; and uploading the parameters of the local models corresponding to the m clients to the server; wherein the local data of the m clients is non-independent and identically distributed data; receiving the parameters of the local models corresponding to the m clients; performing aggregate training on the global model based on regularization constraint parameters and the parameters of the local models corresponding to the m clients to obtain a trained global model; and distributing the trained global model to the m clients, enabling the m clients to receive the trained global model, and fine-tuning the local models corresponding to the m clients based on the trained global model to obtain trained local models corresponding to the m clients; wherein the trained local models corresponding to the m clients are used by the m clients to recommend target products to target users.

[0011] According to an embodiment of this application, the step of performing aggregate training on the global model based on regularization constraint parameters and the local model parameters corresponding to the m clients includes: aggregating the local model parameters corresponding to the m clients based on the cross-entropy loss function, and calculating the mean of the aggregated local model parameters to obtain the local parameter mean; adopting a round-based dynamic adjustment strategy to dynamically determine the regularization constraint parameters based on the hyperparameters and the training rounds; determining the regularization constraint terms according to the regularization constraint parameters and the global model parameters; constructing a global objective function according to the local parameter mean and the regularization constraint terms; and performing model training and parameter updates on the global model based on the server-side business data and the global objective function.

[0012] According to an embodiment of this application, the method further includes: initializing the global model based on the learning rate, the hyperparameters, and the training epochs.

[0013] According to a fourth aspect of this application, a product recommendation device based on federated learning is provided, applied to a target client. The device includes: a product prediction module, used to call a trained local model to process the business feature data and product feature data of the target user to obtain a product matching prediction result; wherein the trained local model is obtained by federated learning training of a global model and a local model based on regularization constraint parameters and non-independent identically distributed data through m clients and a server; m is an integer greater than 1, and the m clients include the target client; and a product recommendation module, used to recommend target products to the target user based on the product matching prediction result.

[0014] According to a fifth aspect of this application, a product recommendation device based on federated learning is provided, applied to m clients, where m is an integer greater than 1. The device includes: a first global model receiving module for receiving a global model from a server; a gradient compression training module for performing gradient compression training on local models based on local data from the m clients and the global model, obtaining local models corresponding to the m clients, and determining the parameters of the local models corresponding to the m clients; wherein the local data from the m clients is non-independent and identically distributed data; and a local model parameter uploading module for uploading the local model parameters corresponding to the m clients to the server; so that the product recommendation is applied to m clients. The server receives the local model parameters corresponding to the m clients, performs aggregate training on the global model based on the regularization constraint parameters and the local model parameters corresponding to the m clients, obtains a trained global model, and distributes the trained global model to the m clients; a second global model receiving module is used to receive the trained global model; and a fine-tuning training module is used to fine-tune the local models corresponding to the m clients based on the trained global model, obtain trained local models corresponding to the m clients; wherein, the trained local models corresponding to the m clients are used by the m clients to recommend target products to target users.

[0015] According to a sixth aspect of this application, a product recommendation device based on federated learning is provided, applied on a server side. The device includes: a first global model distribution module, configured to distribute a global model to m clients, enabling the m clients to receive the global model, perform local model gradient compression training based on the local data of the m clients and the global model, obtain local models corresponding to the m clients, determine the local model parameters corresponding to the m clients, and upload the local model parameters corresponding to the m clients to the server side; wherein the local data of the m clients is non-independent and identically distributed data; and a local model parameter receiving module, configured to receive the parameters from the m clients. The system includes: local model parameters corresponding to each client; an aggregation training module for performing aggregation training on the global model based on regularization constraint parameters and the local model parameters corresponding to the m clients to obtain a trained global model; and a second global model distribution module for distributing the trained global model to the m clients so that the m clients receive the trained global model and, based on the trained global model, fine-tune the local models corresponding to the m clients to obtain trained local models corresponding to the m clients; wherein the trained local models corresponding to the m clients are used by the m clients to recommend target products to target users.

[0016] According to a seventh aspect of this application, an electronic device is provided, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0017] According to an eighth aspect of this application, a computer-readable storage medium is also provided, on which a computer program or instructions are stored, wherein the computer program or instructions, when executed by a processor, implement the steps of the above-described method.

[0018] According to a ninth aspect of this application, a computer program product is also provided, comprising a computer program or instructions that, when executed by a processor, implement the steps of the above-described method. Attached Figure Description

[0019] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0020] Figure 1 This illustration schematically depicts an application scenario diagram of product recommendation based on federated learning according to an embodiment of this application;

[0021] Figure 2 A flowchart illustrating a product recommendation method based on federated learning according to an embodiment of this application is shown schematically.

[0022] Figure 3 This illustration schematically shows a client flowchart of a product recommendation method based on federated learning according to an embodiment of this application;

[0023] Figure 4 This illustration schematically shows a model aggregation flowchart of a product recommendation method based on federated learning according to an embodiment of this application;

[0024] Figure 5 This illustration schematically shows a server-side flowchart of a product recommendation method based on federated learning according to an embodiment of this application;

[0025] Figure 6 This illustration schematically shows a federated learning initialization flowchart of a product recommendation method based on federated learning according to an embodiment of this application;

[0026] Figure 7 This illustration schematically depicts a federated learning architecture diagram of a personalized federated learning financial product recommendation method based on non-independent and identically distributed data, according to an embodiment of this application.

[0027] Figure 8 This schematic diagram illustrates a structural block diagram of a product recommendation apparatus based on federated learning according to an embodiment of this application;

[0028] Figure 9 This illustration schematically shows a client structure block diagram of a product recommendation apparatus based on federated learning according to an embodiment of this application;

[0029] Figure 10 This illustration schematically shows a server-side structural block diagram of a product recommendation apparatus based on federated learning according to an embodiment of this application;

[0030] Figure 11 A block diagram illustrating an electronic device suitable for implementing a federated learning-based product recommendation method according to an embodiment of this application is shown. Detailed Implementation

[0031] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0032] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0033] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0034] As used in this paper, the term "model" refers to a model that learns the relationship between inputs and outputs from training data, enabling it to generate corresponding outputs for a given input after training. Model generation can be based on machine learning techniques. Deep learning is a machine learning algorithm that processes inputs and provides corresponding outputs using multiple layers of processing units. A neural network model is an example of a deep learning-based model. In this paper, "model" may also be referred to as a "machine learning model," "learning model," "machine learning network," or "learning network," and these terms are used interchangeably.

[0035] It's important to note that the term "neural network" can refer to a machine learning network based on deep learning. A neural network processes input and provides corresponding output, typically consisting of an input layer, an output layer, and one or more hidden layers between them. Neural networks used in deep learning applications often include many hidden layers, increasing the network's depth. The layers of a neural network are connected sequentially, so that the output of the previous layer serves as the input to the next layer. The input layer receives the input to the neural network, while the output layer's output becomes the final output. Each layer of a neural network includes one or more nodes (also called processing nodes or neurons), each processing the input from the layer above.

[0036] It should be understood that machine learning generally includes three phases: training, testing, and application (also known as inference). In the training phase, a given model is trained using a large amount of training data, iteratively updating parameter values ​​until the model can consistently generate inferences that meet the expected goals from the training data. Through training, the model can be considered to have learned the relationship between inputs and outputs (also known as the input-output mapping) from the training data. The parameter values ​​of the trained model are determined. In the testing phase, test inputs are applied to the trained model to test whether it can provide the correct output, thus determining the model's performance. In the application phase, the model can be used to process actual inputs based on the trained parameter values ​​to determine the corresponding output.

[0037] Financial institutions are increasingly demanding data value mining and sharing, and are constantly exploring its application scenarios in the financial sector. Currently, institutions face deep-seated bottlenecks in customer acquisition (such as insufficient efficiency in reaching new customers, difficulty in activating dormant users, and limited accuracy in wealth management product recommendations), hindering new growth in wealth management product sales. The potential value contained in data urgently needs to be explored in depth. The problem of data silos still prevents the full utilization of data value, and a breakthrough needs to be achieved through data circulation, integration, and application. Therefore, personalized federated learning technology is currently being used to collaboratively build user wealth management product recommendation models with multiple parties, ensuring data security and compliance.

[0038] However, the data from each participant in federated learning is usually statistically heterogeneous and has non-independent and identically distributed characteristics. Training the model with this non-independent and identically distributed data can easily lead to a significant decrease in the performance of the global model obtained after training, which may result in problems such as model non-convergence and reduced accuracy.

[0039] Meanwhile, the training data contains sensitive information such as enterprise production and operations, posing a high risk of privacy breaches. Without adequate privacy protection mechanisms during training, the system is vulnerable to threats such as model inference attacks and gradient inversion attacks, seriously jeopardizing the data security and privacy of participating institutions. While homomorphic encryption can effectively improve security in existing research methods, its computational overhead is significantly increased, placing high demands on computational performance in practical deployments. Differential privacy, on the other hand, provides privacy at the expense of performance. Furthermore, due to its structural limitations, federated learning requires frequent communication between the client and server, generating significant communication overhead and potentially causing performance bottlenecks.

[0040] This application provides a product recommendation method based on federated learning, comprising: calling a trained local model to process the business feature data and product feature data of a target user to obtain a product matching prediction result; wherein, the trained local model is obtained by federated learning on a global model and a local model through m clients and a server, based on regularization constraint parameters and non-independent identically distributed data; m is an integer greater than 1, and the m clients include the target client; and recommending target products to the target user based on the product matching prediction result. In the embodiments of this application, during the model training process, regularization constraint parameters are introduced to adapt to the non-independent identically distributed data of each client, constraining the global model update offset, mitigating model bias caused by data heterogeneity, effectively avoiding the global model deviating from the optimal solution, and solving the problems of model non-convergence and inaccurate recommendations. Federated learning ensures that the original data of each client is retained locally without leaving the domain, meeting compliance and data security requirements, fully integrating multi-party data, and providing a secure and efficient service for data circulation. Under the premise of privacy compliance, multi-party data is integrated for joint modeling. After training, the client calls the trained local model to complete product prediction and recommendation, effectively predicting user purchase product preferences and accurately recommending products.

[0041] It should be noted that the product recommendation method and apparatus based on federated learning in this application can be used in the fields of artificial intelligence and fintech, as well as in any other fields. The application fields of the product recommendation method and apparatus based on federated learning in this application are not limited.

[0042] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user business information, user transaction information, user preference information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0043] In scenarios where personal information is used for automated decision-making, the methods, devices, and systems provided in this application all provide users with corresponding operation entry points for users to choose to agree to or reject the automated decision results; if the user chooses to reject, the process enters the expert decision-making process.

[0044] Figure 1 The illustration depicts an application scenario diagram of product recommendation based on federated learning according to an embodiment of this application. For example... Figure 1 As shown, application scenario 100 according to an embodiment of this application may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables. For example, a user can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send information, etc.

[0045] The first terminal device 101, the second terminal device 102, and the third terminal device 103 represent client devices, which can be electronic devices such as smartphones, wearable devices, personal computers, intelligent voice interaction devices, smart home appliances, intelligent vehicles, in-vehicle terminals, aircraft, unmanned vending terminals, and extended reality devices. Extended reality devices can include virtual reality devices, augmented reality devices, and mixed reality devices. The client device for the target application can be installed and running on the terminal device. This target application can include, but is not limited to, financial transaction applications, payment applications, shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (these are just examples). Furthermore, this application embodiment does not limit the form of the target application, including but not limited to applications, mini-programs, etc., installed on the terminal device, and can also be in the form of a webpage.

[0046] Server 105 can be a server providing various services, such as a backend management server supporting websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process received user requests and other data, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services such as cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks, and basic cloud computing services such as big data. The server can be the backend server of the aforementioned target application, used to provide backend services to the clients of the target application.

[0047] It should be noted that the product recommendation method based on federated learning provided in this application embodiment can generally be executed by server 105 and / or terminal devices 101-103. Accordingly, the product recommendation device based on federated learning provided in this application embodiment can generally be set in server 105 and / or terminal devices 101-103.

[0048] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0049] Figure 2 A flowchart illustrating a product recommendation method based on federated learning according to an embodiment of this application is shown schematically. Figure 2 As shown, the product recommendation method 200 based on federated learning according to the embodiments of this application is applied to the target client, and the method 200 may include steps S210 to S220.

[0050] In step S210, the trained local model is invoked to process the target user's business feature data and product feature data to obtain product matching prediction results. The trained local model is obtained by federated learning of the global model and the local model through m clients and the server, based on regularization constraint parameters and non-independent identically distributed data. m is an integer greater than 1, and the m clients include the target client.

[0051] Federated learning (FL) differs from traditional centralized learning. In federated learning, each client keeps the original data locally and only exchanges intermediate results for joint modeling, so that the model moves while the data remains stationary, and the data is available but not visible.

[0052] The regularization constraint parameter represents the dynamic weight coefficient of the regularization term, used to control the strength of the regularization term's constraint on the model, and directly determines the magnitude of the penalty for model parameters. This parameter, combined with fixed hyperparameters and dynamic changes in training progress, regulates the global overfitting suppression strength of the model.

[0053] Non-independent and identically distributed (Non-IID) data refers to data exhibiting Non-IID characteristics. A set of random variables or data does not satisfy the condition of being independently and identically distributed. These variables have different distributions (not from the same probability distribution), or the samples are not independent (there is a statistical dependency between them), or both. In financial scenarios, the data generation patterns of different parties are different, with completely different means, customer categories, and behavioral preferences, resulting in different data distributions. For example, branch A has more young wealth management users, while branch B has more middle-aged and elderly conservative users, leading to an unbalanced sample distribution. The data samples are not independent; a single sample can influence other samples, indicating a correlation between them. For example, continuous transactions by the same customer constitute time-series related behavioral data.

[0054] The global model employs a gradient boosting tree model, which consists of multiple hierarchical decision trees that are iteratively stacked. Residual optimization is used to fit the sample prediction bias layer by layer. The gradient boosting tree model is highly adaptable to the multidimensional structured features in financial product recommendation scenarios (such as static profiles like investment period, risk level, category preference, and account opening duration, as well as dynamic behavioral data like holding amount and transaction frequency), and can effectively capture the complex nonlinear mapping relationship between user attributes and product preferences.

[0055] The local model and the global model have the same model structure. The global model, initialized by the server, serves as the initial parameters, and each client independently trains its local model based on its local data. Since the local data of each client has non-independent and identically distributed characteristics, the parameters of the local model differ from those of the global model after training.

[0056] The target client refers to any one of the m clients participating in federated learning training, which can call itself to complete the training of its local model.

[0057] The target client reads locally stored business characteristic data, calls a pre-trained local model, and inputs the target user's business characteristic data (such as investment period, risk level, category preference, holding amount, transaction frequency, etc.) and product characteristic data (such as product type, investment term, minimum purchase amount range, product risk level, and return calculation method, etc.) into the pre-trained local model. The model learns the non-linear relationship between features and product preferences, outputs a target user customer-product matching probability score, and completes the product matching degree prediction. The target client can perform predictions offline, and the locally stored business characteristic data is not transmitted externally.

[0058] In step S220, based on the product matching prediction results, the target product is recommended to the target user.

[0059] The product matching prediction results include the matching probability scores of all products. The target client reads all matching probability scores, sorts them in descending order of scores, and constructs an initial product candidate set.

[0060] The initial product candidate set is filtered to remove invalid products whose risk level does not match the user's risk tolerance or whose investment period is incompatible with the user's capital turnover cycle.

[0061] Based on the filtered product candidate set, target products are recommended according to local operation rules, which include weighted reordering of new products and promotional products, and filtering out sold-out and discontinued products.

[0062] Recommend target products to target users, highlight target products on the wealth management homepage and personalized zones, and push related wealth management information and regular investment reminders.

[0063] In the embodiments of this application, during model training, regularization constraint parameters are introduced to adapt to the non-independent and identically distributed data of each client, constraining the global model update offset, mitigating model bias caused by data heterogeneity, effectively preventing the global model from deviating from the optimal solution, and solving the problems of model non-convergence and inaccurate recommendations. Federated learning ensures that the original data of each client is retained locally without leaving the domain, meeting compliance and data security requirements, fully integrating multi-party data, and providing a secure and efficient service for data circulation. Under the premise of privacy compliance, multi-party data is integrated for joint modeling. After training, the client calls the trained local model to complete product prediction and recommendation, effectively predicting user purchase preferences and accurately recommending products.

[0064] According to an embodiment of this application, a global model is federated learning trained using m clients and a server, based on regularization constraint parameters and non-independent and identically distributed (i.i.d.) data. This includes: receiving the global model from the server through the m clients; performing gradient compression training on the local models of the m clients based on their local data and the global model, obtaining local models corresponding to the m clients, and determining the parameters of the local models corresponding to the m clients; wherein the local data of the m clients is non-independent and identically distributed (i.i.d.) data; uploading the local model parameters corresponding to the m clients to the server through the m clients; enabling the server to receive the local model parameters corresponding to the m clients, perform aggregate training on the global model based on the regularization constraint parameters and the local model parameters corresponding to the m clients, obtain a trained global model, and distribute the trained global model to the m clients; receiving the trained global model through the m clients; and fine-tuning the local models corresponding to the m clients based on the trained global model, obtaining trained local models corresponding to the m clients; wherein the trained local models corresponding to the m clients are used by the m clients to recommend target products to target users.

[0065] Figure 3 A client flowchart illustrating a product recommendation method based on federated learning according to an embodiment of this application is shown schematically. Figure 3 As shown, the product recommendation method 300 based on federated learning according to the embodiment of this application is applied to m clients, where m is an integer greater than 1. The method 300 may include steps S310 to S360.

[0066] In step S310, m clients receive the global model sent by the server.

[0067] The server builds a gradient boosting tree ensemble model and completes parameter initialization, generates a global model, and distributes the global model (containing the complete global model structure and initialization parameters) to each client. Each client receives the global model distributed by the server.

[0068] In step S320, the m clients perform local model gradient compression training based on the local data of the m clients and the global model to obtain the local models corresponding to the m clients and determine the parameters of the local models corresponding to the m clients; wherein, the local data of the m clients are non-independent and identically distributed data.

[0069] Gradient compression is a communication optimization technique that reduces network bandwidth consumption and communication latency by decreasing the amount of gradient data transmitted between nodes in distributed learning.

[0070] Local data refers to proprietary business datasets that are stored independently on the client's local machine and not transmitted externally. These datasets include static user profiles, dynamic behavioral data, and financial product-related feature data. The local data of each client exhibits non-independent and identically distributed characteristics and is used only for local model training.

[0071] Each client trains its local model using local data based on the model parameters of the global model. During training, the local gradient is dynamically compressed using a top-K truncation method (such as the Top-K algorithm), retaining only the top K% gradient values ​​in absolute value to complete gradient compression training. Each client obtains its corresponding local model. Gradient compression can improve communication efficiency and effectively enhance the robustness of the model while ensuring communication efficiency.

[0072] In step S330, m clients upload their respective local model parameters to the server.

[0073] After each client completes local training, each client sends its own trained local model parameters to the server.

[0074] In step S340, the server receives the local model parameters corresponding to m clients, performs aggregate training on the global model based on the regularization constraint parameters and the local model parameters corresponding to m clients, obtains the trained global model, and distributes the trained global model to the m clients.

[0075] Figure 4 A schematic flowchart illustrating the model aggregation process of a federated learning-based product recommendation method according to an embodiment of this application is shown. Figure 4 As shown, the server receives and aggregates the local model parameters uploaded by each client. To avoid the impact of statistical heterogeneity in the local data of each client on model performance and to ensure that the model can provide more reliable decision support, regularization constraints are used to perform aggregation operations on these local model parameters to train the global model. This results in a globally trained model, which is then effectively constrained to prevent overfitting under statistically heterogeneous data while improving the model's generalization ability. The global model fits the local data distribution, improving the model accuracy for local data. Finally, the trained global model is distributed to each client.

[0076] In step S350, m clients receive the trained global model.

[0077] In step S360, based on the trained global model, the m clients fine-tune the local models corresponding to the m clients to obtain the trained local models corresponding to the m clients; wherein, the trained local models corresponding to the m clients are used by the m clients to recommend target products to the target users.

[0078] Since each client contains a different user group, the local data of the clients are not independent and identically distributed. If the training data of the clients are significantly different, the global model may learn common features across clients, but it may not be able to fully adapt to the distribution of local data of the clients. If the global model is used directly for inference locally, the model performance will decrease. In order to make more accurate recommendations of financial products to customers in different regions, each client needs to fine-tune the local model based on the received global model parameters. The trained local model can be used within the client to push suitable financial products to users.

[0079] The trained local model can be deployed to real-world business scenarios to provide personalized financial product recommendation services. While ensuring privacy and security, it can also obtain the purchase intentions of users who have never purchased financial products, enabling precise marketing.

[0080] In the embodiments of this application, multiple clients jointly perform closed-loop federated learning training with the server, eliminating the need for clients to share their original user data, thus meeting financial data compliance and privacy security requirements. Clients first perform gradient compression before uploading their local model parameters, and regularization constraints are introduced when the global model is aggregated and trained on the server side. This mitigates the local model bias caused by multi-party, non-independent, identically distributed data. After the global model is updated, the client's local model is then fine-tuned individually. This approach balances cross-client joint modeling, lightweight communication, and local adaptability, uniformly completing global model iteration and local model optimization, ultimately generating a local model adapted to customer groups in different regions, improving product matching and prediction accuracy.

[0081] According to an embodiment of this application, local model gradient compression training is performed based on local data from m clients and a global model, including: constructing a local model for the current client based on the global model; iteratively training the local model for the current client based on the local data of the current client; during the iterative training process, calculating the change in model parameters of the local model of the current client in two adjacent training rounds, and using a front-end truncation method to select and retain key gradients whose absolute value proportion of the change in model parameters meets a preset ratio, so as to update the local model parameters corresponding to the current client.

[0082] Each client generates a corresponding local model based on the distributed model structure and initial parameters, and then iteratively trains it using local data. Each client trains its local model using local data and performs gradient compression. The original training formula for the local model is as follows:

[0083]

[0084] in, for Round training client Model parameters, for Round training client Model parameters, For learning rate, For model parameters The gradient.

[0085] The local model is trained iteratively using local client data. During training, the parameter changes between two consecutive iterations are calculated. Then, the model is truncated and selected, retaining only the top 20% of gradients whose absolute values ​​of parameter changes reach a predetermined proportion. These selected gradients are used to update the local model parameters. The predetermined proportion is a pre-set hyperparameter k of the local model. By selecting only a certain percentage of gradients (e.g., 20%), only the top 20% of gradients with the highest absolute values ​​of parameter changes are retained, while the rest are discarded, thus compressing the amount of gradient data.

[0086] After gradient compression, the objective function for training the local model on the client side is adjusted to:

[0087]

[0088] in, The gradient is the compressed gradient of the local model; for Model parameters for each round of training; for Model parameters for each round of training; These are hyperparameters for the local model, used to set the compression ratio; To calculate the absolute value of the gradient between two adjacent training rounds of the local model; For gradient update magnitude, only the top k elements with the largest values ​​are retained, and the remaining elements are set to zero; This represents the change in model parameters between two consecutive training rounds, indicating the gradient update magnitude.

[0089] In the embodiments of this application, each client compresses and filters gradients before uploading local model parameters. On the one hand, based on the importance of gradients, parameters with higher value can be selected to participate in global model aggregation, and low-value redundant gradients can be filtered out, significantly reducing the amount of data transmitted between the client and the server, reducing communication bandwidth consumption, and improving interaction efficiency. On the other hand, during training iterations, gradients for parameter changes are calculated, and only the first few key gradients are retained to complete gradient sparsity compression. Gradient sparsity compression avoids directly exposing the original gradients, reducing the risk of privacy inference attacks. While significantly reducing communication overhead, it also takes into account communication efficiency and model performance, and enhances privacy and security.

[0090] According to an embodiment of this application, the global model is aggregated and trained based on regularization constraint parameters and local model parameters corresponding to m clients. This includes: aggregating the local model parameters corresponding to m clients based on the cross-entropy loss function, and calculating the mean of the aggregated local model parameters to obtain the local parameter mean; dynamically determining the regularization constraint parameters based on the hyperparameters and training epochs using a round-based dynamic adjustment strategy; determining the regularization constraint terms based on the regularization constraint parameters and global model parameters; constructing a global objective function based on the local parameter mean and the regularization constraint terms; and training and updating the parameters of the global model based on the server-side business data and the global objective function.

[0091] The server aggregates the model parameters uploaded by each client. Since the local data on each client is not independent and identically distributed, the aggregated global model may deviate from the global optimum, leading to slow convergence and decreased accuracy. Therefore, a loss function for global model training is added... Regularization constraints improve the performance of the global model after acquisition.

[0092] Aggregate the local model parameters from each client and calculate the aggregated parameter mean as the local parameter mean. Based on the preset hyperparameters, the current epoch, and the total number of training epochs, the regularization constraint parameters can be dynamically calculated through an epoch-based dynamic adjustment strategy. Using these regularization constraint parameters and the global model parameters, the regularization constraint term is solved. By superimposing the local parameter mean and the regularization constraint term, a global objective function can be constructed.

[0093] Business data refers to shared feature data used to guide the iterative optimization of the global model, and may include user profiles, user transaction behavior, and financial product attributes. The server-side completes global model training and parameter updates based on the business data and the global objective function.

[0094] The server obtains the global objective function of the model through aggregation, i.e., the global objective function, and the specific formula is as follows:

[0095]

[0096]

[0097] in, This is the local parameter mean. For regularization constraints, For the number of clients, These are global model parameters. Let cross-entropy be the loss function. For the client Model parameters, for Regularization term, These are the regularization constraint parameters, i.e., the dynamic parameters of the regularization term. These are preset hyperparameters. For the current round, The total number of training epochs for the model is defined. Due to model instability in the early stages of training, to prevent overfitting to noisy data, a dynamic epoch adjustment strategy combines fixed hyperparameters, the current iteration epoch, and the total number of training epochs to dynamically configure constraint weights. To control this, the regularization constraint strength is gradually increased as the training rounds increase, causing the model to gradually converge to a globally better model.

[0098] In the embodiments of this application, the server aggregates local parameters from the client, then dynamically adjusts the regularization hyperparameters based on the training epochs to generate constraint terms, constructing a global objective function with regularization constraints to complete the global model parameter update. Adding regularization constraints to the global model effectively constrains its complexity and parameter offset, preventing it from deviating from the optimal solution and improving its generalization ability while avoiding overfitting under statistically heterogeneous data. This avoids the impact of statistical heterogeneity of local data from different clients on the performance of the global model, ensuring that the local model can provide more reliable decision support for purchasing financial products.

[0099] According to an embodiment of this application, fine-tuning training is performed on the local models corresponding to m clients based on the trained global model, including: calculating the Euclidean distance between the global model parameters corresponding to the trained global model and the local model parameters of the current client to obtain the Euclidean norm; determining the parameter offset penalty term based on the fine-tuning hyperparameters and Euclidean norm of the current client; constructing a local objective function with the goal of minimizing the cross-entropy loss of the local model of the current client, based on the parameter offset penalty term; and fine-tuning training and parameter updating of the local model of the current client based on the local data and the local objective function of the current client, until the local model of the current client converges.

[0100] Each client uses local data to iteratively fine-tune and update its local model parameters based on the distributed global model parameters and local objective function. The construction process of the local objective function is as follows: the Euclidean distance between the global model parameters and the local model parameters is calculated to obtain the Euclidean norm. The client's fine-tuning hyperparameters are multiplied by the Euclidean norm to calculate the parameter offset penalty term. With minimizing the cross-entropy loss of the local model as the core, the parameter offset penalty term is added to construct the local objective function.

[0101] The client fine-tunes the local model based on the global model parameters, optimizing the local model's objective function. The formula for the local objective function is as follows:

[0102]

[0103] in, These are the local model parameters for client i; Let the cross-entropy loss of the local model of client i on the local data be denoted as . Cross-entropy loss when network parameters are used; These are global model parameters; Hyperparameters are used for fine-tuning hyperparameters, which control the personalized fine-tuning of the local model on the client side. It is the L2 norm (Euclidean norm); This represents the Euclidean distance between two parameter vectors; It is the square of the L2 norm, i.e., the square of the L2 distance; This is a penalty term for parameter offset.

[0104] In the embodiments of this application, the differences between the parameters of the local model and the global model are quantified by Euclidean norm, and a bias penalty term is constructed by combining fine-tuning hyperparameters to constrain the local update magnitude. A constrained local objective function is built with cross-entropy loss as the core. Based on local data, the local model is personalized for fine-tuning to alleviate the problem of the global model deviating from the local optimum due to non-independent and identically distributed data. This makes the model more consistent with the local data distribution, and the recommended financial products are more in line with the customer's financial investment preferences. The fine-tuning training is controllable and convergent, which not only retains the general customer features learned by global joint learning, but also fits the unique local user profile, greatly improving the relevance of regional financial product recommendations.

[0105] Figure 5 The diagram schematically illustrates a server-side flowchart of a product recommendation method based on federated learning according to an embodiment of this application. Figure 5 As shown, the product recommendation method 500 based on federated learning according to the embodiments of this application is applied to the server side, and the method 500 may include steps S510 to S560.

[0106] In step S510, the server sends the global model to m clients.

[0107] According to an embodiment of this application, before distributing the global model, method 500 further includes: initializing the global model based on the learning rate, hyperparameters, and training epochs.

[0108] Figure 6 The diagram schematically illustrates the federated learning initialization flowchart of a product recommendation method based on federated learning according to an embodiment of this application. Figure 6 As shown, the server initializes the global model and sets hyperparameters such as learning rate and training epochs. The server initializes and obtains the global model and distributes it to each client. Each client receives the global model distributed by the server, which can build a multi-party data circulation and fusion application platform to achieve initialization, so as to facilitate subsequent collaborative training.

[0109] In the embodiments of this application, the global model is initialized in conjunction with the learning rate, hyperparameters and training rounds. The initialization parameters are adapted to the dynamic adjustment strategy of subsequent rounds, which reduces the oscillation of the global model in the early stage, accelerates the overall training convergence speed, reduces the model bias caused by non-independent and identically distributed data in the early stage, makes the initial weights of the global model more consistent with the comprehensive distribution of multi-source data, reduces the iteration overhead of subsequent aggregation and fine-tuning, and improves the overall federated learning training efficiency.

[0110] In step S520, m clients receive the global model, perform gradient compression training of the local model based on the local data of the m clients and the global model, obtain the local model corresponding to the m clients, determine the parameters of the local model corresponding to the m clients, and upload the parameters of the local model corresponding to the m clients to the server; wherein, the local data of the m clients are non-independent and identically distributed data.

[0111] It should be noted that step S520 is the same as the method described in steps S310 to S330 above, and will not be repeated here.

[0112] In step S530, the server receives local model parameters corresponding to m clients.

[0113] In step S540, the server performs aggregate training on the global model based on the regularization constraint parameters and the local model parameters corresponding to m clients, and obtains the trained global model.

[0114] In step S550, the server distributes the trained global model to m clients.

[0115] It should be noted that steps S530 to S550 are the same as the method described in step S340 above, and will not be repeated here.

[0116] In step S560, m clients receive the trained global model and, based on the trained global model, fine-tune the local models corresponding to the m clients to obtain the trained local models corresponding to the m clients; wherein, the trained local models corresponding to the m clients are used by the m clients to recommend target products to the target users.

[0117] It should be noted that step S560 is the same as the method described in steps S350 to S360 above, and will not be repeated here.

[0118] For example, based on the above-described federated learning-based product recommendation method, embodiments of this application also provide a personalized federated learning-based financial product recommendation method based on non-independent and identically distributed data. Figure 7 The diagram illustrates the federated learning architecture of a personalized federated learning financial product recommendation method based on non-independent and identically distributed data, according to an embodiment of this application. Figure 7 As shown, the server initializes the global model and distributes it to the clients, achieving federated learning initialization. Clients can be of any number, and each client trains its local model using local data. Since the local data from multiple clients is not independent and identically distributed, to ensure compliance and prevent the leakage of sensitive user information, gradient compression is dynamically added locally on the client side before sending local model parameters to the server. This effectively defends against privacy-preserving inference attacks while reducing communication bandwidth and improving communication efficiency. To address the issue of the global model deviating from its global optimum due to heterogeneous statistical data, the server aggregates local model parameters, adds regularization constraints to the global model, and distributes the aggregated and trained global model to the clients. Because different regions and customers have different financial preferences, to better match the local data distribution and accurately characterize user profiles, the clients perform personalized fine-tuning training of their local models based on the distributed global model, achieving personalized learning from local data. By obtaining a finely tuned local model through a joint modeling client, the input of the model is the user's business characteristic data, and the output is the product matching prediction result corresponding to the user. This enables the prediction of customer matching degree for financial products, which can depict customer profiles from multiple dimensions and in depth, accurately identify potential buyers of products, and recommend financial products based on the product matching prediction results, thus significantly improving the success rate of financial product recommendations.

[0119] Based on the above-described product recommendation method based on federated learning, embodiments of this application also provide a product recommendation apparatus based on federated learning, applied to a target client. The following will combine... Figure 8 The device is described in detail.

[0120] Figure 8 A schematic block diagram of a product recommendation apparatus based on federated learning according to an embodiment of this application is shown.

[0121] like Figure 8 As shown, the product recommendation device 800 based on federated learning in this embodiment includes a product prediction module 810 and a product recommendation module 820.

[0122] The product prediction module 810 is used to call a pre-trained local model to process the target user's business feature data and product feature data to obtain product matching prediction results. The pre-trained local model is obtained by federated learning of the global model and the local model using m clients and a server, based on regularization constraint parameters and non-independent identically distributed data. m is an integer greater than 1, and the m clients include the target client. In one embodiment, the product prediction module 810 can be used to execute step S28 described above, which will not be repeated here.

[0123] The product recommendation module 820 is used to recommend target products to the target user based on the product matching prediction results. In one embodiment, the product recommendation module 820 can be used to perform step S220 described above, which will not be repeated here.

[0124] According to an embodiment of this application, the apparatus 800 further includes: a model training module, configured to receive a global model from a server through the m clients; perform local model gradient compression training based on the local data of the m clients and the global model through the m clients to obtain local models corresponding to the m clients, and determine the local model parameters corresponding to the m clients; wherein the local data of the m clients belongs to the non-independent and identically distributed data; and upload the local model parameters corresponding to the m clients to the server through the m clients, so that the server can receive the local model parameters corresponding to the m clients. The global model is aggregated and trained based on the local model parameters of the m clients, using regularization constraint parameters and the local model parameters corresponding to the m clients. The trained global model is then distributed to the m clients. The m clients receive the trained global model. Based on the trained global model, the local models corresponding to the m clients are fine-tuned and trained using the m clients to obtain the trained local models corresponding to the m clients. The trained local models corresponding to the m clients are used by the m clients to recommend target products to target users.

[0125] According to embodiments of this application, any multiple modules among the product prediction module 810, product recommendation module 820, and model training module can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the product prediction module 810, product recommendation module 820, and model training module can be at least partially implemented as hardware circuits, such as field-programmable gate arrays, programmable logic arrays, systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits, or any other reasonable means of integrating or packaging circuits, or implemented in software, hardware, or firmware, or in any appropriate combination of any of these three implementation methods. Alternatively, at least one of the product prediction module 810, product recommendation module 820, and model training module can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0126] Based on the above-described product recommendation method based on federated learning, embodiments of this application also provide a product recommendation device based on federated learning, with m clients, where m is an integer greater than 1. The following will combine... Figure 9 The device is described in detail.

[0127] Figure 9 The diagram illustrates a client-side structural block diagram of a product recommendation apparatus based on federated learning according to an embodiment of this application.

[0128] like Figure 9 As shown, the product recommendation device 900 based on federated learning in this embodiment includes a first global model receiving module 910, a gradient compression training module 920, a local model parameter uploading module 930, a second global model receiving module 940, and a fine-tuning training module 950.

[0129] The first global model receiving module 910 is used to receive the global model sent by the server. In one embodiment, the first global model receiving module 910 can be used to execute step S310 described above, which will not be repeated here.

[0130] The gradient compression training module 920 is used to perform local model gradient compression training based on the local data of the m clients and the global model, to obtain the local models corresponding to the m clients, and to determine the parameters of the local models corresponding to the m clients; wherein, the local data of the m clients are non-independent and identically distributed data. In one embodiment, the gradient compression training module 920 can be used to execute step S320 described above, which will not be repeated here.

[0131] The local model parameter upload module 930 is used to upload the local model parameters corresponding to the m clients to the server. This allows the server to receive the local model parameters from the m clients, perform aggregate training on the global model based on the regularization constraint parameters and the local model parameters from the m clients, obtain a trained global model, and then distribute the trained global model to the m clients. In one embodiment, the local model parameter upload module 930 can be used to execute steps S330-S340 described above, which will not be repeated here.

[0132] The second global model receiving module 940 is used to receive the trained global model. In one embodiment, the second global model receiving module 940 can be used to execute step S350 described above, which will not be repeated here.

[0133] The fine-tuning training module 950 is used to fine-tune the local models corresponding to the m clients based on the trained global model, thereby obtaining the trained local models corresponding to the m clients; wherein, the trained local models corresponding to the m clients are used by the m clients to recommend target products to target users. In one embodiment, the fine-tuning training module 950 can be used to execute step S360 described above, which will not be repeated here.

[0134] According to an embodiment of this application, the fine-tuning training module 950 is further configured to calculate the Euclidean distance between the global model parameters corresponding to the trained global model and the local model parameters of the current client, and obtain the Euclidean norm; determine a parameter offset penalty term based on the fine-tuning hyperparameters of the current client and the Euclidean norm; construct a local objective function with the goal of minimizing the cross-entropy loss of the local model of the current client, and based on the parameter offset penalty term; and perform fine-tuning training and parameter updates on the local model of the current client based on the local data of the current client and the local objective function, until the local model of the current client converges.

[0135] According to an embodiment of this application, the gradient compression training module 920 is further configured to construct a local model of the current client based on the global model; perform iterative training on the local model of the current client based on the local data of the current client; during the iterative training process, calculate the change in model parameters of the local model of the current client in two adjacent training rounds, and use a front-end truncation method to filter and retain key gradients whose absolute value proportion of the change in model parameters meets a preset ratio, so as to update the local model parameters corresponding to the current client.

[0136] According to embodiments of this application, any multiple modules among the first global model receiving module 910, gradient compression training module 920, local model parameter uploading module 930, second global model receiving module 940, and fine-tuning training module 950 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the first global model receiving module 910, gradient compression training module 920, local model parameter uploading module 930, second global model receiving module 940, and fine-tuning training module 950 can be at least partially implemented as hardware circuits, such as field-programmable gate arrays, programmable logic arrays, systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits, or any other reasonable means of integrating or packaging circuits, or implemented in software, hardware, or firmware, or in any appropriate combination of any of these three implementation methods. Alternatively, at least one of the first global model receiving module 910, gradient compression training module 920, local model parameter uploading module 930, second global model receiving module 940, and fine-tuning training module 950 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0137] Based on the above-described product recommendation method based on federated learning, embodiments of this application also provide a product recommendation apparatus based on federated learning, applied on the server side. The following will combine... Figure 10 The device is described in detail.

[0138] Figure 10 The diagram illustrates a server-side structural block diagram of a product recommendation apparatus based on federated learning according to an embodiment of this application.

[0139] like Figure 10 As shown, the product recommendation device 1000 based on federated learning in this embodiment includes a first global model distribution module 1010, a local model parameter receiving module 1020, an aggregation training module 1030, and a second global model distribution module 1040.

[0140] The first global model distribution module 1010 is used to distribute a global model to m clients, enabling the m clients to receive the global model, perform local model gradient compression training based on the local data of the m clients and the global model, obtain the local models corresponding to the m clients, determine the parameters of the local models corresponding to the m clients, and upload the parameters of the local models corresponding to the m clients to the server; wherein, the local data of the m clients is non-independent and identically distributed data. In one embodiment, the first global model distribution module 1010 can be used to execute steps S510~S520 described above, which will not be repeated here.

[0141] The local model parameter receiving module 1020 is used to receive the local model parameters corresponding to the m clients. In one embodiment, the local model parameter receiving module 1020 can be used to execute step S530 described above, which will not be repeated here.

[0142] The aggregation training module 1030 is used to perform aggregation training on the global model based on the regularization constraint parameters and the local model parameters corresponding to the m clients, to obtain a trained global model. In one embodiment, the aggregation training module 1030 can be used to execute step S540 described above, which will not be repeated here.

[0143] The second global model distribution module 1040 is used to distribute the trained global model to the m clients; so that the m clients receive the trained global model and, based on the trained global model, fine-tune and train the local models corresponding to the m clients to obtain the trained local models corresponding to the m clients; wherein, the trained local models corresponding to the m clients are used by the m clients to recommend target products to target users. In one embodiment, the second global model distribution module 1040 can be used to execute steps S550~S560 described above, which will not be repeated here.

[0144] According to an embodiment of this application, the aggregation training module 1030 is further configured to: aggregate the local model parameters corresponding to the m clients based on the cross-entropy loss function; calculate the mean of the aggregated local model parameters to obtain the local parameter mean; adopt a round-based dynamic adjustment strategy to dynamically determine the regularization constraint parameters based on the hyperparameters and training rounds; determine the regularization constraint terms based on the regularization constraint parameters and the global model parameters; construct a global objective function based on the local parameter mean and the regularization constraint terms; and perform model training and parameter updates on the global model based on the server-side business data and the global objective function.

[0145] According to an embodiment of this application, the device 1000 further includes an initialization module for initializing the global model based on the learning rate, the hyperparameters, and the training epochs.

[0146] According to embodiments of this application, any multiple modules among the first global model distribution module 1010, local model parameter receiving module 1020, aggregation training module 1030, second global model distribution module 1040, and initialization module can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the first global model distribution module 1010, local model parameter receiving module 1020, aggregation training module 1030, second global model distribution module 1040, and initialization module can be at least partially implemented as hardware circuits, such as field-programmable gate arrays, programmable logic arrays, systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits, or other reasonable means of integrating or packaging circuits, or implemented in hardware or firmware, or in any one of software, hardware, and firmware implementations, or in a suitable combination of any of these. Alternatively, at least one of the first global model distribution module 1010, the local model parameter receiving module 1020, the aggregated training module 1030, the second global model distribution module 1040, and the initialization module can be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0147] Figure 11 A block diagram illustrating an electronic device suitable for implementing a federated learning-based product recommendation method according to an embodiment of this application is shown.

[0148] like Figure 11 As shown, an electronic device 1200 according to an embodiment of this application includes a processor 1201, which can perform various appropriate actions and processes according to a program stored in a read-only memory 1202 or a program loaded from a storage portion 1208 into a random access memory 1203. The processor 1201 may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a dedicated microprocessor. The processor 1201 may also include onboard memory for caching purposes. The processor 1201 may include a single processing unit or multiple processing units for executing different steps of the method flow according to an embodiment of this application.

[0149] Random access memory 1203 stores various programs and data required for the operation of electronic device 1200. Processor 1201, read-only memory 1202, and random access memory 1203 are interconnected via bus 1204. Processor 1201 executes various steps of the method flow according to embodiments of this application by executing programs in read-only memory 1202 and / or random access memory 1203. It should be noted that the programs may also be stored in one or more memories other than read-only memory 1202 and random access memory 1203. Processor 1201 may also execute various steps of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0150] According to embodiments of this application, the electronic device 1200 may further include an input / output interface 1205, which is also connected to the bus 1204. The electronic device 1200 may also include one or more of the following components connected to the input / output interface 1205: an input section 1206 including a keyboard, mouse, etc.; an output section 1207 including a cathode ray tube, liquid crystal display, etc., and a speaker, etc.; a storage section 1208 including a hard disk, etc.; and a communication section 1209 including a network interface card, such as a local area network card, modem, etc. The communication section 1209 performs communication processing via a network such as the Internet. A drive 1210 is also connected to the input / output interface 1205 as needed. A removable medium 1211, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1210 as needed so that computer programs read from it can be installed into the storage section 1208 as needed.

[0151] Embodiments of this application also provide a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0152] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof. In embodiments of this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include the read-only memory 1202, and / or random access memory 1203, and / or one or more memories other than read-only memory 1202 and random access memory 1203 described above.

[0153] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the methods provided in the embodiments of this application.

[0154] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 1209, and / or installed from the removable medium 1211. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0155] In embodiments of this application, the computer program can be downloaded and installed from a network via communication section 1209, and / or installed from removable medium 1211. When the computer program is executed by processor 1201, it performs the functions defined in the system of this application embodiment. According to embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0156] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0157] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0158] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A product recommendation method based on federated learning, characterized in that, Applied to the target client, the method includes: The pre-trained local model is invoked to process the target user's business feature data and product feature data to obtain product matching prediction results. The pre-trained local model is obtained through federated learning of the global model and the local model using m clients and a server, based on regularization constraint parameters and non-independent identically distributed data; m is an integer greater than 1, and the m clients include the target client. Based on the product matching prediction results, target products are recommended to the target users.

2. The method according to claim 1, characterized in that, The process of training a global model through federated learning using m clients and servers, based on regularization constraint parameters and non-independent identically distributed data, includes: The global model is received from the server through the m clients; Using the m clients, local model gradient compression training is performed based on the local data of the m clients and the global model to obtain the local models corresponding to the m clients and determine the parameters of the local models corresponding to the m clients; wherein, the local data of the m clients belongs to the non-independent and identically distributed data; The m clients upload their local model parameters to the server; the server receives the local model parameters from the m clients, performs aggregation training on the global model based on the regularization constraint parameters and the local model parameters from the m clients, obtains a trained global model, and distributes the trained global model to the m clients. The trained global model is received through the m clients; and Using the m clients, and based on the trained global model, the local models corresponding to the m clients are fine-tuned and trained to obtain the trained local models corresponding to the m clients; wherein, the trained local models corresponding to the m clients are used by the m clients to recommend target products to target users.

3. A product recommendation method based on federated learning, characterized in that, The method, applied to m clients where m is an integer greater than 1, includes: Receive the global model sent by the server; Based on the local data of the m clients and the global model, local model gradient compression training is performed to obtain the local models corresponding to the m clients and determine the parameters of the local models corresponding to the m clients; wherein, the local data of the m clients are non-independent and identically distributed data; Upload the local model parameters corresponding to the m clients to the server; so that the server receives the local model parameters corresponding to the m clients, performs aggregate training on the global model based on the regularization constraint parameters and the local model parameters corresponding to the m clients, obtains the trained global model, and distributes the trained global model to the m clients. Receive the trained global model; and Based on the trained global model, the local models corresponding to the m clients are fine-tuned to obtain the trained local models corresponding to the m clients; wherein, the trained local models corresponding to the m clients are used by the m clients to recommend target products to target users.

4. The method according to claim 3, characterized in that, The step of fine-tuning the local models corresponding to the m clients based on the trained global model includes: Calculate the Euclidean distance between the global model parameters corresponding to the trained global model and the local model parameters of the current client to obtain the Euclidean norm; Based on the current client's fine-tuning hyperparameters and the Euclidean norm, determine the parameter offset penalty term; The optimization objective is to minimize the cross-entropy loss of the local model of the current client, and a local objective function is constructed based on the parameter offset penalty term; and Based on the local data of the current client and the local objective function, the local model of the current client is fine-tuned and its parameters are updated until the local model of the current client converges.

5. The method according to claim 4, characterized in that, The step of performing local model gradient compression training based on the local data from the m clients and the global model includes: Based on the global model, construct the local model for the current client; Based on the local data of the current client, the local model of the current client is iteratively trained; During iterative training, the changes in model parameters of the local model of the current client in two adjacent training rounds are calculated, and the key gradients whose absolute value proportion of the changes in model parameters meets the preset ratio are selected and retained by the first column truncation method, so as to update the local model parameters corresponding to the current client.

6. A product recommendation method based on federated learning, characterized in that, Applied to the server side, the method includes: A global model is distributed to m clients so that the m clients can receive the global model. Based on the local data of the m clients and the global model, local model gradient compression training is performed to obtain the local model corresponding to the m clients. The parameters of the local model corresponding to the m clients are determined and uploaded to the server. The local data of the m clients are non-independent and identically distributed data. Receive the local model parameters corresponding to the m clients; Based on the regularization constraint parameters and the local model parameters corresponding to the m clients, the global model is aggregated and trained to obtain a trained global model; and The trained global model is distributed to the m clients; so that the m clients receive the trained global model and, based on the trained global model, fine-tune and train the local models corresponding to the m clients to obtain the trained local models corresponding to the m clients; wherein, the trained local models corresponding to the m clients are used by the m clients to recommend target products to target users.

7. The method according to claim 6, characterized in that, The aggregation training of the global model based on regularization constraint parameters and the local model parameters corresponding to the m clients includes: Based on the cross-entropy loss function, the local model parameters corresponding to the m clients are aggregated, and the mean of the aggregated local model parameters is calculated to obtain the local parameter mean. A round-based dynamic adjustment strategy is adopted to dynamically determine the regularization constraint parameters based on hyperparameters and training rounds; The regularization constraint terms are determined based on the regularization constraint parameters and the global model parameters; Construct a global objective function based on the local parameter mean and the regularization constraint term; and Based on the server-side business data and the global objective function, the global model is trained and its parameters are updated.

8. The method according to claim 7, characterized in that, The method further includes: The global model is initialized based on the learning rate, the hyperparameters, and the number of training epochs.

9. A product recommendation device based on federated learning, characterized in that, Applied to a target client, the device includes: The product prediction module is used to call a pre-trained local model to process the target user's business feature data and product feature data to obtain product matching prediction results. The pre-trained local model is obtained by federated learning of a global model and a local model using m clients and a server, based on regularization constraint parameters and non-independent identically distributed data; m is an integer greater than 1, and the m clients include the target client. The product recommendation module is used to recommend target products to the target user based on the product matching prediction results.

10. A product recommendation device based on federated learning, characterized in that, The device, applicable to m clients where m is an integer greater than 1, comprises: The first global model receiving module is used to receive the global model sent by the server. The gradient compression training module is used to perform gradient compression training of the local model based on the local data of the m clients and the global model, to obtain the local model corresponding to the m clients, and to determine the parameters of the local model corresponding to the m clients; wherein, the local data of the m clients are non-independent and identically distributed data; The local model parameter upload module is used to upload the local model parameters corresponding to the m clients to the server so that the server receives the local model parameters corresponding to the m clients, performs aggregate training on the global model based on the regularization constraint parameters and the local model parameters corresponding to the m clients, obtains the trained global model, and distributes the trained global model to the m clients. The second global model receiving module is used to receive the trained global model; and The fine-tuning training module is used to fine-tune the local models corresponding to the m clients based on the trained global model, so as to obtain the trained local models corresponding to the m clients; wherein, the trained local models corresponding to the m clients are used by the m clients to recommend target products to target users.

11. A product recommendation device based on federated learning, characterized in that, The device, applied to the server side, includes: The first global model distribution module is used to distribute a global model to m clients, so that the m clients receive the global model, perform local model gradient compression training based on the local data of the m clients and the global model, obtain the local model corresponding to the m clients, determine the local model parameters corresponding to the m clients, and upload the local model parameters corresponding to the m clients to the server; wherein, the local data of the m clients is non-independent and identically distributed data; The local model parameter receiving module is used to receive the local model parameters corresponding to the m clients; The aggregation training module is used to perform aggregation training on the global model based on regularization constraint parameters and the local model parameters corresponding to the m clients, to obtain a trained global model; and The second global model distribution module is used to distribute the trained global model to the m clients; so that the m clients receive the trained global model and, based on the trained global model, fine-tune and train the local models corresponding to the m clients to obtain the trained local models corresponding to the m clients; wherein, the trained local models corresponding to the m clients are used by the m clients to recommend target products to target users.

12. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 7.

13. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

14. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.