Data feature query methods, devices, equipment, and media based on federated learning

Abstract

This application provides a data feature query method, apparatus, device, and medium based on federated learning. The method includes: an active-side device determining at least one query data feature from multiple data features used to train the federated learning model based on the importance of the data features to the model; wherein the importance level indicates the degree of influence of the query data feature on the prediction result of the federated learning model; determining a passive-side device corresponding to each query data feature; sending a query request corresponding to each query data feature to the corresponding passive-side device; and when the active-side device has feature query permissions, receiving feature information of the corresponding query data feature returned by the passive-side device. In this way, the active-side device obtains feature information provided by the passive-side device by sending query requests for corresponding data features, improving the interpretability of the federated learning model and enhancing the user experience for users of the active-side modeling.

Description

Technical Field

[0001] This application relates to artificial intelligence technology, and more particularly to a data feature query method, apparatus, device, and medium based on federated learning. Background Technology

[0002] Vertical federated learning typically involves different participants jointly training a machine learning model. The labeled participant (usually only one) is called the active party, and the unlabeled participant is called the passive party. During training, participants only transmit encrypted model and gradient information, not other data, thus protecting user privacy. Because the feature information of passive party devices is completely invisible to active party devices, while this protects their feature information, it can also cause confusion for active party devices. For example, after completing modeling, the active party device can only see the number of input variables and the strength of predictive power, but has no understanding of the meaning of the input variables. In reality, some passive party devices are not so sensitive to the disclosure of feature information and can accept a certain level of disclosure. Therefore, when passive party devices can accept feature information disclosure, this complete shielding of feature information provided by passive party devices can negatively impact the user experience for active party users. Summary of the Invention

[0003] This application provides a data feature query method, apparatus, electronic device, computer-readable storage medium, and computer program product based on federated learning, which can obtain feature information provided by the passive party device, improve the interpretability of the federated learning model, and enhance the user experience of the active party modeling user.

[0004] The technical solution of this application embodiment is implemented as follows:

[0005] This application provides a data feature query method based on federated learning. The method is based on a federated learning system, which includes an active party device with labels and at least one passive party device. The method includes:

[0006] The active-side device determines at least one query data feature from multiple data features used to train the federated learning model based on the importance of the data features to the federated learning model.

[0007] The importance level is used to indicate the degree of influence of the data feature to be queried on the prediction result of the federated learning model;

[0008] Determine the passive party device corresponding to each of the features of the data to be queried;

[0009] Send the query request corresponding to each of the data features to be queried to the corresponding passive party device;

[0010] When the active device has feature query permission, it receives the feature information of the corresponding data feature to be queried returned by the passive device.

[0011] This application provides a data feature query device based on federated learning, including:

[0012] The first determining module is used by the active party device to determine at least one query data feature from multiple data features used to train the federated learning model based on the importance of the data feature to the federated learning model; wherein, the importance is used to indicate the degree of influence of the query data feature on the prediction result of the federated learning model;

[0013] The second determining module is used to determine the passive party device corresponding to each of the features of the data to be queried;

[0014] The sending module is used to send query requests corresponding to each of the data features to be queried to the corresponding passive party device;

[0015] The receiving module is used to receive feature information of the corresponding data feature to be queried returned by the passive device when the active device has feature query permission.

[0016] In the above scheme, the first determining module is further configured to obtain the importance of each data feature among multiple data features used to train the federated learning model; sort the multiple data features according to the importance order to obtain a sorting result; and select at least two data features as the data features to be queried, starting from the first data feature in the sorting result.

[0017] In the above scheme, the second determining module is further used to obtain the passive party identifier associated with each of the query data features; and to determine the passive party device corresponding to each of the query data features based on the passive party identifier associated with the query data features.

[0018] In the above scheme, the sending module is further configured to obtain authorization information sent by each of the passive party devices; and to send a query request carrying corresponding authorization information to each of the data features to be queried; wherein, the authorization information is used for the passive party device to determine the priority of the active party device based on the authorization information, and to query the data features to be queried based on the priority.

[0019] In the above scheme, the receiving module is further configured to receive the feature content of the corresponding data feature to be queried returned by the passive device when the active device has feature query permission and the feature query permission of the active device corresponds to the feature content of the data feature; wherein, the feature content is obtained by the passive device through feature content query based on the feature identifier carried in the query request.

[0020] In the above scheme, the receiving module is further configured to receive the feature category to which the corresponding data feature to be queried belongs when the active device has feature query permission and the feature query permission of the active device corresponds to the feature category of the data feature; wherein, the feature category is obtained by the passive device querying the tag carried by the data feature to be queried.

[0021] In the above scheme, the device further includes a rejection module, which is used to receive rejection information returned by the passive device for the query request when the active device does not have the feature query permission; wherein the rejection information is used to indicate that the active device does not have the query permission for the corresponding feature.

[0022] In the above scheme, the device further includes a training module, which is used to determine the importance of the feature information of the data feature to be queried; adjust the weights of the data feature used to train the federated learning model based on the importance of the feature information of the data feature to be queried; and retrain the federated learning model using the adjusted weights and the data feature.

[0023] This application provides an electronic device, including:

[0024] Memory, used to store executable instructions;

[0025] When the processor executes the executable instructions stored in the memory, it implements the data feature query method based on federated learning provided in the embodiments of this application.

[0026] This application provides a computer-readable storage medium storing executable instructions for inducing a processor to execute and implement the federated learning-based data feature query method provided in this application.

[0027] This application provides a computer program product, including a computer program that, when executed by a processor, implements the federated learning-based data feature query method provided in this application.

[0028] The embodiments of this application have the following beneficial effects:

[0029] To protect the data privacy of passive learning systems by completely shielding them from the feature information provided by the passive party device, the federated learning model remains a black box to the active party device, thus impacting the user experience. To address this, this application provides a data feature query method. The active party device sends query requests for corresponding data features to the passive party device based on the importance of those features to the federated learning model. When the active party device has feature query permissions, it receives the feature information returned by the passive party device. This method, by obtaining the feature information provided by the passive party device, improves the interpretability of the federated learning model and enhances the user experience for the active party modeling user. Attached Figure Description

[0030] Figure 1 This is a schematic diagram illustrating an implementation scenario of the data feature query method based on federated learning provided in this application embodiment;

[0031] Figure 2 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application;

[0032] Figure 3 This is a flowchart illustrating the data feature query method based on federated learning provided in an embodiment of this application;

[0033] Figure 4 This is an optional flowchart illustrating the process of determining the features of the data to be queried provided in an embodiment of this application;

[0034] Figure 5 This is an optional interactive diagram of the data feature query method based on federated learning provided in the embodiments of this application;

[0035] Figure 6 This is an optional flowchart illustrating the adjustment of multiple data features for training a federated learning model, provided in an embodiment of this application.

[0036] Figure 7 This is an optional federated learning interaction diagram of the data feature query method based on federated learning provided in the embodiments of this application;

[0037] Figure 8 This is an optional flowchart illustrating a data feature query method based on federated learning provided in an embodiment of this application.

[0038] Figure 9 This is an optional flowchart illustrating a data feature query method based on federated learning provided in an embodiment of this application.

[0039] Figure 10 This is a schematic diagram of the structure of the data feature query device based on federated learning provided in the embodiments of this application. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0041] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0042] In the following description, the terms "first, second, third" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0044] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0045] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0046] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0047] 1) Federated machine learning (also known as federated learning, joint learning, or consortium learning) is a machine learning framework that effectively helps multiple organizations use data and perform machine learning modeling while meeting user privacy, data security, and government regulations. As a distributed machine learning paradigm, federated learning effectively solves the data silo problem, allowing participants to jointly model data without sharing it, thus technically breaking down data silos and achieving collaborative intelligence.

[0048] Federated learning (aka Federated Machine Learning) can be divided into three categories: Horizontal Federated Learning, Vertical Federated Learning, and Federated Transfer Learning.

[0049] Horizontal federated learning, also known as feature-aligned federated learning, involves participants whose data features are aligned. It is suitable for situations where participants' data features overlap significantly, while their sample identifiers (IDs, identity documents) overlap less. Vertical federated learning, also known as sample-aligned federated learning, involves participants whose training samples are aligned. It is suitable for situations where participants' training sample IDs overlap significantly, while their data features overlap less.

[0050] 2) Homomorphic encryption (HE) is a symmetric encryption algorithm proposed by Craig Gentry of IBM in 2009. The goal of homomorphic encryption is to find an encryption algorithm that can perform addition and multiplication operations on ciphertext such that performing a certain operation on the encrypted ciphertext yields the same result as encrypting the plaintext again after performing the desired operation. Homomorphic encryption effectively ensures that the data processor can directly process the ciphertext without knowing the plaintext information being processed. This characteristic of homomorphic encryption provides adequate security for user data and privacy; therefore, it is applied in many real-world scenarios to ensure data security.

[0051] If an encryption function satisfies both additive and multiplicative homomorphism, it is called a fully homomorphic encryption function. This function can be used to perform various encrypted operations (addition, subtraction, multiplication, division, polynomial evaluation, exponentiation, logarithms, trigonometric functions, etc.).

[0052] 3) Risk control, also known as risk management, refers to the control of risks through risk assessment of business operations, such as risk control of bank loan funds, including pre-loan risk assessment (such as credit assessment), risk monitoring during the loan process (such as monitoring credit changes), and post-loan loan management (such as refinancing based on credit changes).

[0053] 4) IV (Information Value) is mainly used to assess the predictive power of input variables. The magnitude of the IV value of a feature variable indicates the strength of its predictive power.

[0054] In modeling problems such as risk control and recommendation systems, to ensure model effectiveness, it is often necessary to obtain multi-dimensional and comprehensive user features to characterize the user profile. This typically requires acquiring feature data from multiple sources. However, due to increasingly stringent user privacy protections, directly obtaining user data from data sources is often impractical. Federated learning systems offer a solution for multi-party collaborative modeling without disclosing user privacy data, making them highly suitable for risk control modeling scenarios.

[0055] The applicant discovered that during the modeling process of the federated learning system, the feature information of the passive party devices was completely shielded by the system. While this protected the data privacy of the passive party devices, in reality, some passive party devices were not so sensitive to the disclosure of feature information and could accept a certain degree of disclosure. Therefore, it is advisable to consider disclosing feature information with limitations, provided the passive party devices consent.

[0056] Based on this, embodiments of this application provide a data feature query method, apparatus, electronic device, computer-readable storage medium, and computer program product based on federated learning, which can obtain feature information provided by the passive party device, improve the interpretability of the federated learning model, and enhance the user experience of the active party modeling user.

[0057] Based on the above explanation of the nouns and terms used in the embodiments of this application, the following describes the implementation scenario of the data feature query method based on federated learning provided in the embodiments of this application. See [link to relevant documentation]. Figure 1 , Figure 1This is a schematic diagram illustrating an implementation scenario of the federated learning-based data feature query method provided in this application. To support an exemplary application, the active device 200-1 connects to the passive device 200-2 via network 300. The active device 200-1 can be an active device storing user tag information, such as a bank. The passive device 200-2 can be a passive device storing user feature data, such as a shopping mall or supermarket. The active device 200-1 sends query requests corresponding to each data feature to be queried to the passive device 200-2, and when it has feature query permissions, it receives the feature information of the corresponding data feature to be queried returned by the passive device 200-2. The network 300 can be a wide area network or a local area network, or a combination of both, and data transmission is achieved using wireless or wired links.

[0058] An active party device (including active party device 200-1) is used to determine at least one query data feature from multiple data features used to train the federated learning model based on the importance of the data feature to the federated learning model; wherein, the importance is used to indicate the degree of influence of the query data feature on the prediction result of the federated learning model; to determine the passive party device corresponding to each query data feature; and to send the query request corresponding to each query data feature to the corresponding passive party device.

[0059] The passive device (passive device 200-2) is used to obtain the query requests corresponding to each data feature to be queried sent by the active device; when the active device has the feature query permission, it queries the feature information of the corresponding data feature to be queried based on the obtained query requests corresponding to each data feature to be queried; and sends the queried feature information of the corresponding data feature to the active device.

[0060] The active device (including active device 200-1) is also used to receive feature information of the corresponding data features to be queried returned by the passive device.

[0061] In this embodiment, after obtaining the feature information provided by the passive device, the effectiveness of the federated learning model is enhanced by adjusting the feature information based on its impact on the predictive ability of the model. For example, the federated learning model in this embodiment can be applied to risk control management. For instance, it can predict a user's creditworthiness (e.g., excellent, average, or very poor credit), and then implement risk control measures (e.g., loan amount) based on the prediction results. When it is known that a certain app has a significant impact on a user's creditworthiness, the frequency of loan advertisements within that app is increased, thereby refining the user scope, improving the accuracy of advertising, and reducing resource consumption and waste. Another example is the use of a federated learning model to predict the risk level of a project (e.g., high, medium, or low risk), and then implementing risk control measures (implementation or non-implementation) based on the prediction results. When it is known that the implementation time of a project has a significant impact on its risk level, the project implementation time is changed, thereby reducing the project's risk level. This comprehensive prediction of project risk level improves the accuracy of risk control for the project.

[0062] In practical applications, the active device 200-1 and the passive device 200-2 can be independent physical servers, server clusters or distributed systems composed of multiple physical servers, or cloud servers providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The active device 200-1 and the passive device 200-2 can also be smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, etc., but are not limited to these. The active device 200-1 and the passive device 200-2 can be connected directly or indirectly through wired or wireless communication methods, which is not limited herein.

[0063] The hardware structure of the electronic device implementing the federated learning-based data feature query method provided in the embodiments of this application is described in detail below. The electronic device includes, but is not limited to, a server or a terminal. See also Figure 2 , Figure 2 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Figure 2The illustrated electronic device 200 includes at least one processor 210, a memory 250, at least one network interface 220, and a user interface 230. The various components in the electronic device 200 are coupled together via a bus system 240. It will be understood that the bus system 240 is used to implement communication between these components. In addition to a data bus, the bus system 240 also includes a power bus, a control bus, and a status signal bus. However, for clarity, ... Figure 2 The general labeled all buses as Bus System 240.

[0064] Processor 210 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0065] User interface 230 includes one or more output devices 231 that enable the presentation of media content, including one or more speakers and / or one or more visual displays. User interface 230 also includes one or more input devices 232, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0066] The memory 250 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state storage, hard disk drives, optical disk drives, etc. The memory 250 may optionally include one or more storage devices physically located away from the processor 210.

[0067] The memory 250 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), and the volatile memory may be random access memory (RAM). The memory 250 described in this application embodiment is intended to include any suitable type of memory.

[0068] In some embodiments, memory 250 is capable of storing data to support various operations, examples of which include programs, modules, and data structures or subsets or supersets thereof, as illustrated below.

[0069] Operating system 251 includes system programs for handling various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, driver layer, etc., for implementing various basic business functions and handling hardware-based tasks;

[0070] The network communication module 252 is used to reach other computing devices via one or more (wired or wireless) network interfaces 220, such as Bluetooth, WiFi, and Universal Serial Bus (USB).

[0071] The input processing module 253 is used to detect and translate one or more user inputs or interactions from one or more input devices 232.

[0072] In some embodiments, the data feature query device based on federated learning provided in this application can be implemented in software. Figure 2 A data feature query device 254 based on federated learning, stored in memory 250, is shown. It can be software in the form of programs and plug-ins, including the following software modules: a first determination module 2541, a second determination module 2542, a sending module 2543, and a receiving module 2544. These modules are logically related and can therefore be arbitrarily combined or further divided according to the functions they implement. The functions of each module will be described below.

[0073] In other embodiments, the data feature query device based on federated learning provided in this application can be implemented in a combination of hardware and software. As an example, the data feature query device based on federated learning provided in this application can be a processor in the form of a hardware decoding processor, which is programmed to execute the multi-class gradient boosting tree construction method provided in this application. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0074] Based on the above description of the implementation scenarios and electronic devices of the federated learning-based data feature query method in the embodiments of this application, the federated learning-based data feature query method provided in the embodiments of this application is described below. See also Figure 3 , Figure 3 This is a flowchart illustrating the data feature query method based on federated learning provided in this application embodiment. The data feature query method based on federated learning provided in this application embodiment includes:

[0075] Step 101: The active device determines at least one query data feature from multiple data features used to train the federated learning model based on the importance of the data feature to the federated learning model; wherein, the importance is used to indicate the degree of influence of the query data feature on the prediction result of the federated learning model.

[0076] In practice, the degree of influence on the prediction results of the federated learning model is specifically defined as the influence weight of the data features input to the model on the feature prediction results of the federated learning model. Here, the active party device acts as the label holder and feature provider of the training samples when training the federated learning model, while the passive party device only acts as the feature provider of the training samples.

[0077] For example, when a federated learning model is applied to risk management or risk prediction, the labels of the training samples used to train the model can be classification labels identifying user creditworthiness, user loyalty, or user quality. Here, the active party can be a platform that conducts credit assessments, such as a financial lending platform (e.g., a bank), and the passive party can be a platform holding user feature data, such as various types of consumer platforms (e.g., shopping malls, supermarkets, online shopping platforms, gaming platforms, mobile app stores, etc.) or internet companies. The multiple data features used to train the federated learning model can include, but are not limited to, user basic information, business data, behavioral data, and device information. Basic information includes, but is not limited to, the user's name, gender, age, education, occupation, mobile phone number, ID number, address, and income data (e.g., annual salary, monthly salary). Business data encompasses data from multiple business scenarios. For example, in a loan business scenario, the data includes loan information and repayment data. Loan data includes loan amount, loan term, and lending institution (e.g., XX Bank), while repayment data includes repayment amount, repayment date, lending institution, and overdue repayment data (e.g., number of overdue payments and number of overdue days). In a consumption scenario, the data includes the purpose of consumption (e.g., education, shopping, medical care, daily living expenses), consumption amount, consumption time, consumption institution (e.g., shopping malls, supermarkets, convenience stores, online stores), consumption method (e.g., credit card or debit card), and the user's address at the time of consumption. Behavioral data may include the user's frequently used payment institutions (e.g., WeChat Pay, bank card payment), frequently used payment methods (e.g., QR code payment, card payment), and operation frequency (e.g., frequency of payment operations within a certain period). Device information includes the model of the user's frequently used devices and the device's address at the time of the user's actions.

[0078] In some embodiments, the process of determining the data features to be queried from multiple data features used to train a federated learning model is described in [reference needed]. Figure 4 , Figure 4 This is an optional flowchart illustrating the process of determining the features of the data to be queried provided in an embodiment of this application, based on... Figure 4 Step 101 can also be achieved in the following way:

[0079] Step 1011: The active device acquires the importance of each data feature among multiple data features used to train the federated learning model.

[0080] In practice, after the federated learning model is trained, the active device determines the predictive power of the input model's data features based on its own label information. It should be noted that the input model's data features include features localized to the active device and features provided by the passive device, and the authority to determine the importance of each data feature belongs solely to the active device.

[0081] Step 1012: Sort the multiple data features according to their importance to obtain the sorting results.

[0082] In practice, besides sorting multiple data features in descending order of importance, they can also be sorted in ascending order of importance. It should be noted that whether the sorting of multiple feature data is based on descending or ascending order of importance can be chosen based on the needs of the active party; this application does not impose any restrictions on this.

[0083] Step 1013: Starting from the first data feature in the sorting results, select at least two data features as the data features to be queried.

[0084] In practice, the data features in the sorting results include features localized by the active device and features provided by the passive device, while the features of the data to be queried are those provided by the passive device. Therefore, when selecting data features, it is necessary to filter out the features localized by the active device in order to select the features of the data to be queried.

[0085] In some embodiments, when multiple data features are sorted in descending order of importance, at least two data features are selected as query features, starting from the first data feature in the sorting result. Here, the selected query features are those with high importance to the federated learning model. By identifying the data features with high importance to the federated learning model, the active party can enhance the effectiveness of the federated learning model. For example, when the federated learning model in this embodiment is applied to the field of risk control management, it predicts the creditworthiness of users and then performs risk control for those users based on the prediction results. When it is known that a certain app has a significant impact on a user's creditworthiness, the frequency of loan advertisements within that app is increased, thereby refining the user scope, improving the accuracy of advertisement delivery, and reducing resource consumption and waste.

[0086] In other embodiments, when multiple data features are sorted in ascending order of importance, at least two data features are selected as query features, starting from the first data feature in the sorting result. Here, the selected query features are those with low importance to the federated learning model. By identifying data features with low importance to the federated learning model, the active party can also enhance the effectiveness of the federated learning model. For example, when the federated learning model in this embodiment is applied to the field of risk control management, it predicts the creditworthiness of users and then performs risk control for those users based on the prediction results. When it is determined that gender has little impact on a user's creditworthiness, there is no need to design different lending schemes based on the user's gender, reducing resource waste.

[0087] It should be noted that the number of selected data features to be queried is determined based on the feature query permissions of the active device. In some embodiments, one data feature may be selected as the data feature to be queried. The process of determining the feature query permissions of the active device will be explained in detail in step 103. The process of determining the number of data features to be queried based on the feature query permissions of the active device will be explained in detail in step 104.

[0088] Step 102: Determine the passive device corresponding to each feature of the data to be queried.

[0089] In practice, since the federated learning model may be jointly trained by one active party device and multiple passive party devices, the features of each query data may be provided by multiple passive party devices. (See below for reference.) Figure 5 , Figure 5This is an optional interactive schematic diagram of the data feature query method provided in the embodiments of this application. After determining each data feature to be queried, it is necessary to obtain the passive party identifier associated with each data feature to be queried; based on the passive party identifier associated with each data feature to be queried, the passive party device corresponding to each data feature to be queried is determined.

[0090] Here, the passive party identifier can be the device identifier of the passive party device, used to uniquely identify the passive party device, such as the serial number of the passive party device. By using the passive party identifier associated with each data feature to be queried, the passive party device corresponding to each data feature to be queried can be determined.

[0091] Step 103: Send the query request corresponding to each data feature to be queried to the corresponding passive device.

[0092] Because different passive devices have varying degrees of sensitivity to feature information disclosure, only passive devices that are not sensitive to feature information disclosure will accept a certain level of feature information disclosure. Therefore, feature information will only be disclosed in a limited manner with the authorization of the passive device. Based on this, in actual implementation, it is necessary to first obtain the authorization information sent by each passive device, such as authorization codes and query codes; then, for each data feature to be queried, a query request carrying the corresponding authorization information is sent to the passive device. It should be noted that the authorization information here is used by the passive device to determine the priority of the active device based on the authorization information, and to query the data feature to be queried based on the priority.

[0093] In practice, the priority of the active party's equipment can be related to the active party's social credibility. For example, government agencies have a higher priority than hospitals and banks, while hospitals and banks have a higher priority than shopping malls and supermarkets. Alternatively, the priority of the active party can be determined by the passive party. For example, it can be determined by the number of times the active party and the passive party have cooperated, or by the active party's high-quality points with the passive party. These high-quality points can be determined by the amount of money given by the active party, or by the passive party's satisfaction rating of the active party after the joint training is completed.

[0094] Step 104: When the active device has feature query permission, it receives the feature information of the corresponding data feature to be queried returned by the passive device.

[0095] Here, the feature query permission corresponds to the priority of the active device. Due to the difference in priority, when the active device has the feature query permission, the feature information of the corresponding data feature returned by the passive device will also differ based on the different feature query permissions.

[0096] Next, we will explain in detail the process of receiving different feature information of the corresponding data features to be queried from the passive device based on different feature query permissions.

[0097] In some embodiments, when the active device has feature query permission and the feature query permission of the active device corresponds to the feature content of the data feature, it receives the feature content of the corresponding data feature to be queried returned by the passive device; wherein, the feature content is obtained by the passive device through feature content query based on the feature identifier carried in the query request.

[0098] For example, when the active device has a higher priority, its feature query permission allows it to retrieve the feature content of the corresponding data feature. In this case, the passive device determines the feature content of the corresponding feature based on the feature identifier carried in the query request. For instance, if the high-priority active device wants to query three features, the passive device determines the feature content of these three features based on the feature identifier carried in the query request. If the three features are determined to be gender, age, and education level, the passive device directly sends the feature content of gender, age, and education level to the active device. In other words, when the active device has higher query permission, the feature information returned by the passive device can be the specific feature content.

[0099] In other embodiments, when the active device has feature query permission and the feature query permission of the active device corresponds to the feature category of the data feature, it receives the feature category to which the corresponding data feature to be queried belongs, returned by the passive device; wherein, the feature category is obtained by the passive device querying the tag carried by the data feature to be queried.

[0100] For example, when the active device has a low priority, its feature query permission allows it to retrieve the feature category to which the corresponding data feature belongs. In this case, the passive device determines the feature category to which the corresponding feature belongs based on the tags carried by the data feature to be queried. For instance, if an active device with a moderate priority wants to query three features, the passive device determines the feature category to which these three features belong based on the tags carried by the data features to be queried. If the tag for these three features is determined to be population attribute, the population attribute feature category is directly sent to the active device. In other words, when the active device has a moderate query permission, the feature information returned by the passive device can be the feature category of the corresponding feature.

[0101] It should be noted that the tags (i.e., feature categories) carried by the features to be queried are obtained by the passive party device during the system deployment phase, before providing the features. For example, the passive party device can tag features such as gender, age, and education level as "demographic attributes," or users of a certain app as "xx user," etc. This application embodiment does not impose excessive limitations on the tagging method of the passive party device. When the tags carried by the features to be queried belong to multiple feature categories, the union of the multiple tags contained in the features to be queried is taken, and the multiple feature categories are sent to the active party device.

[0102] Following the example above, when multiple data features to be queried carry the labels "demographic attribute" and "xx user", the union of the two labels is taken, and the two feature categories "demographic attribute" and "xx user" are sent to the initiating device.

[0103] It should be noted that after the passive device obtains the feature information of the data to be queried, it performs homomorphic encryption on the feature information and sends the encrypted feature information to the active device. The active device then decrypts the encrypted feature information upon receiving it from the passive device to obtain the corresponding feature information. In this way, by encrypting the feature information, privacy is protected and the security of data transmission is improved.

[0104] In practice, the feature query permissions of the active party can determine whether the feature information returned by the passive party is feature content or feature category, and can also determine the number of features of the data to be queried.

[0105] Next, the process of determining the number of features in the data to be queried will be explained in detail.

[0106] In some embodiments, when the initiating device has a higher priority, i.e., greater feature query authority, a data feature can be selected as the data feature to be queried, that is, at least one data feature is selected as the data feature to be queried from the sorting results of multiple data features.

[0107] In other embodiments, when the priority of the active device is generally low (i.e., its feature query privileges are limited), at least two data features are selected as the data features to be queried from the ranking results of multiple data features. Here, since the feature categories received by the active device are obtained by taking the union of the tags carried by the data features to be queried, when the tags carried by the data features to be queried belong to multiple feature categories, in order to prevent the active device from sending query requests for each data feature to be queried one by one, the correspondence between the tags carried by the data features to be queried and the data features to be queried can be determined. Therefore, when the priority of the active device is generally low (i.e., its feature query privileges are limited), the active device is limited to selecting at least two data features as the data features to be queried from the ranking results of multiple data features.

[0108] It should be noted that when the priority of the active device is generally low, i.e., the feature query permission is relatively small, if a data feature is selected as the data feature to be queried from the sorting results of multiple data features, and the query requests corresponding to each data feature to be queried are sent to the corresponding passive device, a prompt message will be received from the passive device to remind the active device to select at least two data features as the data features to be queried.

[0109] Step 104 assumes that the active device has the permission to query features. However, in actual implementation, when the active device does not have the permission to query features, it receives a rejection message from the passive device regarding the query request. The rejection message indicates that the active device does not have the permission to query the corresponding feature.

[0110] In practice, the lack of feature query permissions for the active device can stem from situations where the passive device is overly sensitive to feature information disclosure and has not authorized it, or from a low priority setting. For example, if the active device receives a low amount of economic incentive from the active party, resulting in a low quality score for the active device on the passive party, the passive device, upon receiving query requests for each feature from the active device, will return a rejection message to the active device. This allows the active device to adjust its priority in a timely manner, i.e., to allocate an appropriate amount of economic incentive, thereby promptly obtaining the feature information of the corresponding data to be queried.

[0111] It should be noted that, in some embodiments, after determining the feature information of the data features that the active device wants to know, it will also adjust multiple data features used to train the federated learning model based on the determined feature information.

[0112] Here, the process of adjusting multiple data features for training the federated learning model based on determined feature information is described in [reference needed]. Figure 6 , Figure 6This is a schematic diagram of an optional process for adjusting multiple data features used to train a federated learning model, provided in an embodiment of this application. Figure 3 Following step 104, the data feature query method based on federated learning provided in this application embodiment further includes:

[0113] Step 201: The active device determines the importance of the feature information of the data to be queried.

[0114] Here, since there is a correspondence between the features of the data to be queried and the feature information, the importance of the feature information is determined after the importance of the features of the data to be queried is determined.

[0115] As an example, when there are 10 features to be queried, and the active device receives the feature content of the corresponding features to be queried from the passive device, that is, the active device receives 10 specific feature contents. Based on the importance of the 10 features to be queried, it can be determined that the 10 feature contents are features with high importance to the federated learning model.

[0116] As another example, when there are 10 features to be queried, and the active device receives the feature categories of the corresponding features returned by the passive device, following the example above, that is, the active device receives the feature categories "demographic attributes" and "xx user". Based on the importance of the 10 features to be queried, it can be determined that the feature categories "demographic attributes" and "xx user" are features with high importance to the federated learning model.

[0117] Step 202: Adjust the weights of the data features used to train the federated learning model based on the importance of the feature information of the data to be queried.

[0118] Following the example above, when the active device determines that 10 features are of high importance to the federated learning model, the weights of the 10 features used to train the federated learning model are adjusted.

[0119] Following the example above, when the active party determines that the two feature categories "demographic attributes" and "xx user" are features with high importance to the federated learning model, the weights of the features related to "demographic attributes" and "xx user" used to train the federated learning model are adjusted.

[0120] Step 203: Retrain the federated learning model using the adjusted weights and data features.

[0121] In practice, after adjusting the weights of the data features used to train the federated learning model, the active device can use the adjusted weights and local features to train the new federated learning model for a preset number of rounds to obtain the target federated learning model; or the active and passive devices can transmit encrypted model information, and then use the adjusted weights, local features of the active device, and features provided by the passive device to jointly train the new federated learning model, and update the new federated learning model by transmitting encrypted gradient information. In this way, the target federated learning model is obtained after a preset number of rounds.

[0122] In this way, by adjusting the weights of features that are highly important to the federated learning model, the predictive power and effectiveness of the federated learning model are improved.

[0123] Applying the embodiments described above, the active device sends a query request for the corresponding data feature to the passive device based on the importance of the data feature to the model. When the active device has feature query permissions, it receives the feature information of the corresponding data feature returned by the passive device. In this way, obtaining the feature information provided by the passive device improves the interpretability of the federated learning model and enhances the user experience for users of the active modeling device.

[0124] Next, taking the application scenario of vertical federated learning as an example, the data feature query method based on federated learning provided in this application embodiment will be described. In the scenario of vertical federated learning, there are usually at least two participants. That is, the training of the federated learning model and the data feature query after training are jointly implemented by at least two participating devices. Among them, the participant with label information is the active party, and the participant without label information is called the passive party. There is only one active party, while there can be multiple passive parties. See [link to relevant documentation]. Figure 7 , Figure 7 This is an optional federated learning interaction diagram for the data feature query method based on federated learning provided in this application embodiment. In actual implementation, both the active party device and the passive party device can be servers or terminals. It should be noted that the specific implementation process of the data feature query method based on federated learning provided in this application embodiment differs due to the differences in feature query permissions possessed by the active party device. The following will describe in detail the different data feature query processes performed by the active party device based on the differences in its feature query permissions.

[0125] When the initiating device has feature query permissions and the feature query permissions are high, see [link to relevant documentation]. Figure 8 , Figure 8 This is an optional flowchart illustrating a data feature query method based on federated learning provided in an embodiment of this application. See also... Figure 8The data feature query method based on federated learning provided in this application includes:

[0126] Step 301: The active device acquires the trained federated learning model.

[0127] Here, the completed federated learning model is a federated learning model obtained by joint training of the active party device and the passive party device.

[0128] Step 302: Obtain the importance of each data feature among the multiple data features used to train the federated learning model.

[0129] Here, after the federated learning model is trained, the active device determines the predictive power of the data features input to the model based on its own label information. It should be noted that the data features input to the model include the features local to the active device and the features provided by the passive device, and the authority to obtain the importance of each data feature belongs only to the active device.

[0130] The importance of each data feature refers to the degree of influence of each data feature on the prediction result of the federated learning model, that is, the influence weight of the data features input into the model on the feature prediction result of the federated learning model.

[0131] Step 303: Sort the multiple data features according to their importance from largest to smallest to obtain the sorting result.

[0132] Step 304: Starting from the first data feature in the sorting results, select the target number of data features as the data features to be queried.

[0133] In practice, the target number is determined by the importance of the active device. When the active device is highly important, at least one data feature is selected as the data feature to be queried from the ranking results of multiple data features; when the active device has a lower priority, i.e., less feature query privileges, at least two data features are selected as the data features to be queried from the ranking results of multiple data features.

[0134] When the importance of the active device is average, if it selects a data feature as the data feature to be queried from the ranking results of multiple data features and sends the query requests corresponding to each data feature to the corresponding passive device, it will receive a prompt message from the passive device to remind the active device to select at least two data features as the data features to be queried.

[0135] Step 305: Determine the passive device corresponding to each feature of the data to be queried.

[0136] In practice, since the federated learning model may be jointly trained by one active party device and multiple passive party devices, each data feature to be queried may be provided by multiple passive party devices. Therefore, after determining each data feature to be queried, it is necessary to obtain the passive party identifier associated with each data feature to be queried; based on the passive party identifier associated with each data feature to be queried, the passive party device corresponding to each data feature to be queried is determined.

[0137] Here, the passive party identifier can be the device identifier of the passive party device, used to uniquely identify the passive party device, such as the serial number of the passive party device. By using the passive party identifier associated with each data feature to be queried, the passive party device corresponding to each data feature to be queried can be determined.

[0138] Step 306: Obtain the authorization information sent by the passive device.

[0139] Here, because different passive devices have varying degrees of sensitivity to the disclosure of feature information, only passive devices that are not sensitive to the disclosure of feature information will accept a certain degree of disclosure. Therefore, feature information will only be disclosed in a limited manner with the authorization of the passive device. Based on this, in actual implementation, it is necessary to first obtain the authorization information sent by each passive device, such as authorization codes and query codes.

[0140] Step 307: For each feature of the data to be queried, send a query request carrying the corresponding authorization information to the passive party device.

[0141] Step 308: The passive device determines the priority of the active device based on the authorization information carried in the received query request.

[0142] Here, the priority of the active party's equipment refers to the importance of the active party's users to the passive party's equipment. In practice, the priority of the active party's equipment can be related to the active party's social credibility. For example, government agencies have a higher priority than hospitals and banks, while hospitals and banks have a higher priority than shopping malls and supermarkets. Alternatively, the priority of the active party can be determined by the passive party. For example, it can be determined by the number of times the active party and the passive party have cooperated, or by the active party's high-quality points with the passive party. These high-quality points can be determined by the amount of money provided by the active party, or by the passive party's satisfaction rating of the active party after the joint training is completed.

[0143] Step 309: Based on the priority of the active device, query the features of the data to be queried to obtain the feature information of the corresponding data to be queried.

[0144] In practice, when the active device has a high priority, the passive device determines the feature content of the corresponding feature based on the feature identifier carried in the query request; when the active device has a low priority, it determines the feature category to which the corresponding feature belongs based on the label carried by the feature of the data to be queried.

[0145] For example, during the deployment phase of the federated learning system, passive-party devices label the features that can be disclosed, such as gender, age, and education level with demographic attribute labels. Then, during the data feature query phase, the passive-party devices disclose the label information in a tiered manner based on the priority (importance) of the active-party devices. For the features gender, age, and education level, if the active-party device has high importance (high priority), the information is disclosed at the gender, age, and education level based on the feature identifier carried in the query request. If the active-party device has moderate importance (moderate priority), the information is disclosed at the population attribute level based on the demographic attribute label carried by the feature in the query request.

[0146] It should be noted that when the importance of the active device is generally low, i.e., the priority is generally low, the union of the tags contained in at least two features queried by the active device is taken as the corresponding feature information.

[0147] Step 310: Send the retrieved feature information to the active device.

[0148] In actual implementation, after the passive device obtains the feature information of the corresponding data to be queried, it performs homomorphic encryption on the feature information and sends the encrypted feature information to the active device. After receiving the encrypted feature information returned by the passive device, the active device decrypts the encrypted feature information to obtain the corresponding feature information.

[0149] When the initiating device does not have feature query permissions, see [link to relevant documentation]. Figure 9 , Figure 9 This is an optional flowchart illustrating a data feature query method based on federated learning provided in an embodiment of this application. See also... Figure 9 The data feature query method based on federated learning provided in this application includes:

[0150] Step 401: The active device acquires the trained federated learning model.

[0151] Here, the completed federated learning model is a federated learning model obtained by joint training of the active party device and the passive party device.

[0152] Step 402: Obtain the importance of each data feature among the multiple data features used to train the federated learning model.

[0153] Step 403: Sort the multiple data features according to their importance from largest to smallest to obtain the sorting result.

[0154] Step 404: Starting from the first data feature in the sorting results, select at least one data feature as the data feature to be queried.

[0155] Step 405: Determine the passive device corresponding to each feature of the data to be queried.

[0156] Step 406: Obtain the authorization information sent by the passive device.

[0157] Step 407: For each feature of the data to be queried, send a query request carrying the corresponding authorization information to the passive party device.

[0158] Step 408: Based on the authorization information carried in the received query request, the passive device determines that the active device does not have the query permission with the corresponding characteristics.

[0159] Here, the lack of feature query permissions for the active device can specifically mean that the passive device is sensitive to the disclosure of feature information and has not authorized it, or that the feature has a low priority. For example, when the active device receives a small amount of economic benefit from the active party, resulting in a low quality score for the active party with the passive party, the passive device, upon receiving query requests from the active party for each feature to be queried, determines, based on the authorization information carried in the received query requests, that the active device does not have query permissions for the corresponding feature.

[0160] Step 409: The passive device sends a rejection message for the query request to the active device.

[0161] In practice, when the passive device determines that the active device does not have the query permission for the corresponding feature, it sends the corresponding rejection information to the active device. Here, the rejection information sent by the passive device for the query request may be a message reminding the active device to adjust its own priority, or it may be a message indicating that the passive device is sensitive to the disclosure of feature information and therefore cannot perform feature queries.

[0162] Thus, when the initiating device lacks feature query permissions because its priority is low, the content of the rejection information can enable the initiating device to adjust its own priority in a timely manner, i.e., to be given an appropriate economic amount, so as to obtain the feature information of the corresponding data to be queried in a timely manner.

[0163] Applying the embodiments described above, the active device sends a query request for the corresponding data feature to the passive device based on the importance of the data feature to the model. When the active device has feature query permissions, it receives the feature information of the corresponding data feature returned by the passive device. In this way, obtaining the feature information provided by the passive device improves the interpretability of the federated learning model and enhances the user experience for users of the active modeling device.

[0164] The following continues to describe the federated learning-based data feature query device 254 provided in the embodiments of this application. See also... Figure 10 , Figure 10 This is a schematic diagram of the structure of the data feature query device 254 based on federated learning provided in this application embodiment. The data feature query device 254 based on federated learning provided in this application embodiment includes:

[0165] The first determining module 2541 is used by the active-side device to determine at least one query data feature from multiple data features used to train the federated learning model based on the importance of the data feature to the federated learning model; wherein, the importance is used to indicate the degree of influence of the query data feature on the prediction result of the federated learning model;

[0166] The second determining module 2542 is used to determine the passive party device corresponding to each of the features of the data to be queried;

[0167] Sending module 2543 is used to send query requests corresponding to each of the data features to be queried to the corresponding passive party device;

[0168] The receiving module 2544 is used to receive the feature information of the corresponding data feature to be queried returned by the passive device when the active device has feature query permission.

[0169] In some embodiments, the first determining module 2541 is further configured to obtain the importance of each data feature among a plurality of data features used to train the federated learning model; sort the plurality of data features according to the order of importance to obtain a sorting result; and select at least two data features as the data features to be queried, starting from the first data feature in the sorting result.

[0170] In some embodiments, the second determining module 2542 is further configured to obtain the passive party identifier associated with each of the query data features; and determine the passive party device corresponding to each of the query data features based on the passive party identifier associated with the query data features.

[0171] In some embodiments, the sending module 2543 is further configured to obtain authorization information sent by each of the passive party devices; and to send a query request carrying corresponding authorization information to each of the data features to be queried; wherein the authorization information is used for the passive party device to determine the priority of the active party device based on the authorization information, and to query the data features to be queried based on the priority.

[0172] In some embodiments, the receiving module 2544 is further configured to receive the feature content of the corresponding data feature to be queried returned by the passive device when the active device has feature query permission and the feature query permission of the active device corresponds to the feature content of the data feature; wherein the feature content is obtained by the passive device through feature content query based on the feature identifier carried in the query request.

[0173] In some embodiments, the receiving module 2544 is further configured to receive the feature category to which the corresponding data feature to be queried belongs when the active device has feature query permission and the feature query permission of the active device corresponds to the feature category of the data feature; wherein, the feature category is obtained by the passive device querying the tag carried by the data feature to be queried.

[0174] In some embodiments, the apparatus further includes a rejection module, which is configured to receive rejection information for the query request returned by the passive device when the initiating device does not have the feature query permission; wherein the rejection information is used to indicate that the initiating device does not have the query permission for the corresponding feature.

[0175] In some embodiments, the apparatus further includes a training module, which is configured to determine the importance of feature information of the data feature to be queried; adjust the weights of the data feature used to train the federated learning model based on the importance of the feature information of the data feature to be queried; and retrain the federated learning model using the adjusted weights and the data feature.

[0176] Applying the embodiments described above, the active device sends a query request for the corresponding data feature to the passive device based on the importance of the data feature to the model. When the active device has feature query permissions, it receives the feature information of the corresponding data feature returned by the passive device. In this way, obtaining the feature information provided by the passive device improves the interpretability of the federated learning model and enhances the user experience for users of the active modeling device.

[0177] This application provides an electronic device, including:

[0178] Memory, used to store executable instructions;

[0179] When the processor executes the executable instructions stored in the memory, it implements the data feature query method based on federated learning provided in the embodiments of this application.

[0180] This application provides a computer-readable storage medium storing executable instructions for inducing a processor to execute and implement the federated learning-based data feature query method provided in this application.

[0181] This application provides a computer program product, including a computer program that, when executed by a processor, implements the federated learning-based data feature query method provided in this application.

[0182] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a variety of devices including one or any combination of the above-mentioned memories.

[0183] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0184] As an example, executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborating files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0185] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0186] In summary, the embodiments of this application enable the acquisition of feature information provided by the passive party device, thereby improving the interpretability of the federated learning model and enhancing the user experience for the active party modeling user.

[0187] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. A data feature query method based on federated learning, characterized in that, The method is based on a federated learning system, which includes labeled active party devices and at least one passive party device. The method includes: The active-side device determines at least one query data feature from multiple data features used to train the federated learning model based on the importance of the data features to the federated learning model. The importance level is used to indicate the degree of influence of the data feature to be queried on the prediction result of the federated learning model; Determine the passive party device corresponding to each of the features of the data to be queried; Obtain the authorization information sent by each of the passive party devices; For each of the aforementioned data characteristics to be queried, a query request carrying the corresponding authorization information is sent to the passive party device; The authorization information is used by the passive device to determine the priority of the active device based on the authorization information, and to query the data features to be queried based on the priority. When the active device has feature query permission, it receives the feature information of the corresponding data feature to be queried returned by the passive device; When the active device has feature query permission, it receives feature information of the corresponding data feature to be queried returned by the passive device, including: When the initiating device has feature query permission, and the feature query permission of the initiating device corresponds to the feature content of the data feature. Receive the feature content of the corresponding data feature to be queried returned by the passive device; The feature content is obtained by the passive device querying the feature content based on the feature identifier carried in the query request; When the active device has feature query permission, it receives feature information of the corresponding data feature to be queried returned by the passive device, including: When the initiating device has feature query permission, and the feature query permission of the initiating device corresponds to the feature category of the data feature. Receive the feature category to which the corresponding query data feature belongs, returned by the passive device; The feature category is obtained by the passive device querying the tags carried by the features of the data to be queried.

2. The method according to claim 1, characterized in that, The process of determining at least one query data feature from multiple data features used to train the federated learning model based on the importance of the data features to the model includes: To determine the importance of each data feature among multiple data features used to train a federated learning model; Based on the order of importance, the multiple data features are sorted to obtain a sorting result; Starting from the first data feature in the sorting results, at least two data features are selected as the data features to be queried.

3. The method according to claim 1, characterized in that, The step of determining the passive party device corresponding to each of the features of the data to be queried includes: Obtain the passive party identifier associated with each of the features of the data to be queried; Based on the passive party identifier associated with the data features to be queried, the passive party device corresponding to each of the data features to be queried is determined.

4. The method according to claim 1, characterized in that, The method further includes: When the active device does not have feature query permission, it receives a rejection message for the query request returned by the passive device. The rejection information indicates that the initiating device does not have the query permission for the corresponding feature.

5. The method according to claim 1, characterized in that, The method further includes: Determine the importance of the feature information of the data to be queried; The weights of the data features used to train the federated learning model are adjusted based on the importance of the feature information of the data to be queried. The federated learning model is retrained using the adjusted weights and the data features.

6. A data feature query device based on federated learning, the device comprising: The first determining module is used by the active party device to determine at least one query data feature from multiple data features used to train the federated learning model based on the importance of the data feature to the federated learning model; wherein, the importance is used to indicate the degree of influence of the query data feature on the prediction result of the federated learning model; The second determining module is used to determine the passive party device corresponding to each of the features of the data to be queried; The sending module is used to obtain authorization information sent by each of the passive party devices; and to send a query request carrying the corresponding authorization information to each of the data features to be queried; wherein, the authorization information is used for the passive party device to determine the priority of the active party device based on the authorization information, and to query the data features to be queried based on the priority; The receiving module is configured to: receive feature information of the corresponding data feature to be queried returned by the passive device when the active device has feature query permission; receive feature content of the corresponding data feature to be queried returned by the passive device when the active device has feature query permission and the feature query permission of the active device corresponds to the feature content of the data feature; wherein the feature content is obtained by the passive device through feature content query based on the feature identifier carried in the query request; and receive the feature category to which the corresponding data feature to be queried belongs returned by the passive device when the active device has feature query permission and the feature query permission of the active device corresponds to the feature category of the data feature; wherein the feature category is obtained by the passive device querying the tag carried by the data feature to be queried.

7. An electronic device, characterized in that, The electronic device includes: Memory, used to store executable instructions; A processor, when executing executable instructions stored in the memory, implements the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores executable instructions for implementing the method of any one of claims 1 to 5 when executed by a processor.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 5.

Images