User Feature Acquisition Method, Device, Computer Equipment and Storage Medium

Abstract

An embodiment of the present application discloses a method, device, computer device and storage medium for obtaining user characteristics, belonging to the field of computer technology. The method includes: invoking a first feature extraction model to extract features from the first user information of a stored target user identifier to obtain first user features, receiving second user features sent by a first device, and obtaining first combined user features of the target user identifier according to the first user features and the second user features. The first device provides the user features extracted by the first device for a second device at its own end, without providing the original user information stored by the first device, avoiding the leakage of user information. And the second device combines the user features extracted by the second device and the first device to obtain combined features. Since the combined features include features in the user information stored by the second device and the first device, the amount of information of the user features is enriched, and the accuracy of the combined user features is improved.

Description

Technical Field

[0001] Embodiments of the present application relate to the field of computer technology, and particularly to a method, apparatus, computer device, and storage medium for obtaining user characteristics. Background Art

[0002] With the development of computer technology, user information has become increasingly complex and diverse. In order to accurately describe a user, user characteristics can usually be obtained based on user information, and the user can be described by these user characteristics.

[0003] In related technologies, a method for obtaining user characteristics is provided. The user information of a target user is obtained, and a feature extraction model is called to process the user information to obtain the user characteristics of the target user. Since the amount of user information used in the above method is small, the accuracy of the obtained user characteristics is poor. Summary of the Invention

[0004] Embodiments of the present application provide a method, apparatus, computer device, and storage medium for obtaining user characteristics, which can improve the accuracy of the obtained combined user characteristics. The technical solutions are as follows:

[0005] On the one hand, a method for obtaining user characteristics is provided. The method includes:

[0006] Call a first feature extraction model to extract features from the first user information of the stored target user identifier to obtain a first user characteristic;

[0007] Receive a second user characteristic sent by a first device, where the second user characteristic is obtained by the first device calling a second feature extraction model to process the second user information of the stored target user identifier, and the first feature extraction model and the second feature extraction model are different models for extracting user characteristics;

[0008] Obtain a first combined user characteristic of the target user identifier according to the first user characteristic and the second user characteristic.

[0009] In a possible implementation, the method further includes:

[0010] Encrypt the first weight according to a first public key to obtain a second weight;

[0011] Send the second weight to the first device, and the first device is used to obtain a second adjustment parameter according to the second weight and the second sample user information;

[0012] Receive the second adjustment parameter sent by the first device;

[0013] Decrypt the second adjustment parameter according to the first private key corresponding to the first public key to obtain a third adjustment parameter;

[0014] Send the third adjustment parameter to the first device, and the first device is configured to adjust the fifth feature extraction model according to the third adjustment parameter.

[0015] In another possible implementation, the first device is configured to obtain a fourth adjustment parameter according to the second weight and the second sample user information, and perform a fusion process on the fourth adjustment parameter and the fifth noise feature to obtain the second adjustment parameter;

[0016] The first device is configured to perform a fusion process on the third adjustment parameter and the sixth noise feature, and adjust the fifth feature extraction model according to the fused adjustment parameter, where the sixth noise feature is opposite to the fifth noise feature.

[0017] In another possible implementation, the method further includes:

[0018] Obtain a loss value of the first feature extraction model according to the predicted user label and the sample user label;

[0019] In response to the loss value being not greater than a preset threshold, stop training the first feature extraction model.

[0020] In another possible implementation, the method further includes;

[0021] In response to the loss value being not greater than a preset threshold, send a stop training notification to the first device, and the first device is configured to stop training the fifth feature extraction model according to the stop training notification.

[0022] On the other hand, a user feature acquisition device is provided, and the device includes:

[0023] A feature extraction module, configured to call a first feature extraction model to extract features from first user information of a target user identifier stored, to obtain a first user feature;

[0024] A feature receiving module, configured to receive a second user feature sent by a first device, where the second user feature is obtained by the first device calling a second feature extraction model to process second user information of the stored target user identifier, and the first feature extraction model and the second feature extraction model are different models for extracting user features;

[0025] A combined feature acquisition module, configured to obtain a first combined user feature of the target user identifier according to the first user feature and the second user feature.

[0026] In a possible implementation manner, the second feature extraction model is a model after being encrypted according to a first public key. The combined feature acquisition module includes:

[0027] A decryption processing unit, configured to decrypt the second user feature according to a first private key corresponding to the first public key to obtain a decrypted user feature;

[0028] A first combination processing unit, configured to perform a combination process on the first user feature and the decrypted user feature to obtain the first combined user feature.

[0029] In another possible implementation manner, the device further includes:

[0030] An encryption processing module, configured to encrypt a third feature extraction model according to the first public key to obtain the second feature extraction model;

[0031] A model sending module, configured to send the second feature extraction model to the first device.

[0032] In another possible implementation manner, the device further includes:

[0033] An information processing module, configured to call a fourth feature extraction model to process the first user information to obtain a third user feature;

[0034] A feature sending module, configured to send the third user feature to the first device. The first device is configured to obtain a second combined user feature according to a fourth user feature and the third user feature, where the fourth user feature is obtained by the first device calling a fifth feature extraction model to perform feature extraction on the second user information;

[0035] The feature receiving module is configured to receive the second combined user feature sent by the first device.

[0036] In another possible implementation manner, before calling the fourth feature extraction model to perform feature extraction on the first user information to obtain a third user feature, the device further includes:

[0037] A model receiving module, configured to receive the fourth feature extraction model sent by the first device.

[0038] In another possible implementation manner, the first device is configured to encrypt a sixth feature extraction model according to a second public key to obtain the fourth feature extraction model;

[0039] The first device is used to decrypt the third user feature according to the second private key corresponding to the second public key to obtain the decrypted user feature; and combine the fourth user feature and the decrypted user feature to obtain the second combined user feature.

[0040] In another possible implementation, the second user feature is obtained by the first device through fusing a fifth user feature and a first noise feature, and the fifth user feature is obtained by the first device invoking the second feature extraction model to extract features from the second user information.

[0041] The second combined user feature is obtained by the first device through fusing a third combined user feature and a second noise feature, and the third combined user feature is obtained by the first device combining the fourth user feature and the third user feature, where the first noise feature and the second noise feature are opposite to each other.

[0042] The device further includes:

[0043] A combination processing module, configured to combine the first combined user feature and the second combined user feature to obtain a fourth combined user feature.

[0044] In another possible implementation, the information processing module includes:

[0045] A feature extraction unit, configured to invoke the fourth feature extraction model to extract features from the first user information to obtain a sixth user feature.

[0046] A first fusion processing unit, configured to fuse the sixth user feature and a third noise feature to obtain the third user feature.

[0047] The combined feature acquisition module includes:

[0048] A second combination processing unit, configured to combine the first user feature and the second user feature to obtain a fifth combined user feature.

[0049] A second fusion processing unit, configured to fuse the fifth combined user feature and a fourth noise feature to obtain the first combined user feature, where the third noise feature and the fourth noise feature are opposite to each other.

[0050] In another possible implementation, the device further includes:

[0051] A sample acquisition module, configured to acquire first sample user information.

[0052] The feature extraction module is further configured to call the first feature extraction model to extract features from the first sample user information, so as to obtain first sample user features;

[0053] The feature receiving module is further configured to receive second sample user features, which are obtained by the first device calling the second feature extraction model to process second sample user information, and the first sample user information and the second sample user information belong to the same sample user identifier;

[0054] The combined feature acquisition module is further configured to obtain first sample combined user features according to the first sample user features and the second sample user features;

[0055] The model training module is configured to train the first feature extraction model according to the first sample combined user features and the first sample user information.

[0056] In another possible implementation manner, the device further includes:

[0057] The information processing module is further configured to call the fourth feature extraction model to process the first sample user information, so as to obtain third sample user features;

[0058] The feature sending module is further configured to send the third sample user features to the first device, and the first device is configured to obtain second sample combined user features according to fourth sample user features and the third sample user features, and the fourth sample user features are obtained by the first device calling the fifth feature extraction model to extract features from the second sample user information;

[0059] The feature receiving module is further configured to receive the second sample combined user features sent by the first device;

[0060] The model training module includes:

[0061] The model training unit is configured to train the first feature extraction model according to the first sample combined user features, the second sample combined user features and the first sample user information.

[0062] In another possible implementation manner, the model training unit is configured to obtain a predicted user label of the sample user identifier according to the first sample combined user features and the second sample combined user features; determine a difference between the predicted user label and a sample user label corresponding to the first sample user information as a first weight; obtain a first adjustment parameter of the first feature extraction model according to the first weight and the first sample user information; and adjust the first feature extraction model according to the first adjustment parameter.

[0063] In another possible implementation, the device further includes:

[0064] A weight encryption module, configured to encrypt the first weight according to a first public key to obtain a second weight;

[0065] A weight sending module, configured to send the second weight to the first device, where the first device is configured to obtain a second adjustment parameter according to the second weight and the second sample user information;

[0066] A parameter receiving module, configured to receive the second adjustment parameter sent by the first device;

[0067] A parameter sending module, configured to decrypt the second adjustment parameter according to a first private key corresponding to the first public key to obtain a third adjustment parameter;

[0068] A model adjustment module, configured to send the third adjustment parameter to the first device, where the first device is configured to adjust the fifth feature extraction model according to the third adjustment parameter.

[0069] In another possible implementation, the first device is configured to obtain a fourth adjustment parameter according to the second weight and the second sample user information, and perform a fusion process on the fourth adjustment parameter and a fifth noise feature to obtain the second adjustment parameter;

[0070] The first device is configured to perform a fusion process on the third adjustment parameter and a sixth noise feature, and adjust the fifth feature extraction model according to the fused adjustment parameter, where the sixth noise feature is opposite to the fifth noise feature.

[0071] In another possible implementation, the device further includes:

[0072] A loss value obtaining module, configured to obtain a loss value of the first feature extraction model according to the predicted user label and the sample user label;

[0073] The model training unit is further configured to stop training the first feature extraction model in response to the loss value being not greater than a preset threshold.

[0074] In another possible implementation, the device further includes;

[0075] A notification sending module, configured to send a stop training notification to the first device in response to the loss value being not greater than a preset threshold, where the first device is configured to stop training the fifth feature extraction model according to the stop training notification.

[0076] On the other hand, a method for obtaining user characteristics is provided, characterized in that the method includes:

[0077] The second device calls the first feature extraction model to extract features from the first user information of the stored target user identifier to obtain the first user feature;

[0078] The first device calls the second feature extraction model to process the second user information of the stored target user identifier to obtain the second user feature, and sends the second user feature to the second device, where the first feature extraction model and the second feature extraction model are different models for extracting user features;

[0079] The second device receives the second user feature; according to the first user feature and the second user feature, the first combined user feature of the target user identifier is obtained.

[0080] On the other hand, a computer device is provided, the computer device includes a processor and a memory, and at least one instruction is stored in the memory, and the at least one instruction is loaded and executed by the processor to implement the user characteristic acquisition method as described in the above aspect.

[0081] On the other hand, a computer-readable storage medium is provided, and at least one instruction is stored in the computer-readable storage medium, and the at least one instruction is loaded and executed by a processor to implement the user characteristic acquisition method as described in the above aspect.

[0082] In still another aspect, a computer program product is provided, the computer program product includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided in the various optional implementation manners of the above aspect.

[0083] The beneficial effects brought by the technical solutions provided in the embodiments of the present application at least include:

[0084] The method, device, computer device, and storage medium provided by the embodiments of the present application split the model for extracting user features and store them separately in the second device and the first device at the local end. The second device and the first device each call the stored feature extraction model to extract features from the user information of the target user identifier stored by each of them. The first device provides the user features extracted by the first device to the second device without providing the original user information stored by the first device, avoiding the leakage of user information. Moreover, the second device combines the user features extracted by the second device and the first device to obtain combined features. Since the combined features include the features in the user information stored by the second device and the first device, the amount of information of the user features is enriched, and the accuracy of the combined user features is improved. Compared with the solution of storing different user information in a central device and performing feature extraction by the central device, the embodiments of the present application store user information separately in different devices and perform feature extraction separately, realizing a decentralized feature extraction method, avoiding information leakage caused by storing user information in a central device, and improving the security of user information. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] To more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0086] Figure 1 is a schematic diagram of a feature extraction system provided by an embodiment of the present application;

[0087] Figure 2 is a schematic diagram of the distribution of a feature extraction model provided by an embodiment of the present application;

[0088] Figure 3 is a schematic diagram of the distribution of a feature extraction model provided by an embodiment of the present application;

[0089] Figure 4 is a schematic diagram of the distribution of a feature extraction model provided by an embodiment of the present application;

[0090] Figure 5 is a flowchart of a method for obtaining user features provided by an embodiment of the present application;

[0091] Figure 6 is a flowchart of a method for obtaining user features provided by an embodiment of the present application;

[0092] Figure 7 is a flowchart of a method for training a feature extraction model provided by an embodiment of the present application;

[0093] Figure 8 It is a flowchart of a method for obtaining user features provided by an embodiment of the present application;

[0094] Figure 9 It is a flowchart of a method for training a feature extraction model provided by an embodiment of the present application;

[0095] Figure 10 It is a flowchart of a method for obtaining user features provided by an embodiment of the present application;

[0096] Figure 11 It is a flowchart of a method for training a feature extraction model provided by an embodiment of the present application;

[0097] Figure 12 It is a flowchart of a method for obtaining user features provided by an embodiment of the present application;

[0098] Figure 13 It is a schematic structural diagram of a device for obtaining user features provided by an embodiment of the present application;

[0099] Figure 14 It is a schematic structural diagram of a device for obtaining user features provided by an embodiment of the present application;

[0100] Figure 15 It is a schematic structural diagram of a terminal provided by an embodiment of the present application;

[0101] Figure 16 It is a schematic structural diagram of a server provided by an embodiment of the present application. Detailed implementation manners

[0102] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will further describe the embodiments of the present application in detail with reference to the accompanying drawings.

[0103] The terms "first", "second", "third", "fourth", "fifth", "sixth", etc. used in the present application may be used herein to describe various concepts, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the present application, the first user feature may be referred to as a user feature, and similarly, the second user feature may be referred to as the first user feature.

[0104] As used in this application, the terms "at least one", "multiple", "each", and "any one" are defined as follows: "at least one" includes one, two, or more than two; "multiple" includes two or more than two; "each" refers to each one of the corresponding multiple; and "any one" refers to any one of the multiple. For example, if there are three elements in a multiple, "each" refers to each of these three elements, and "any one" refers to any one of these three elements, which could be the first, the second, or the third.

[0105] It should be noted that when collecting and processing relevant data (such as user information, user characteristics, user tags, model data, etc.) in the practical application of this application, the informed consent or separate consent of the personal information subject should be obtained in strict accordance with the requirements of relevant national laws and regulations, and subsequent data use and processing activities should be carried out within the scope authorized by laws and regulations and the personal information subject.

[0106] Artificial Intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a way similar to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines to enable machines to have the functions of perception, reasoning, and decision-making.

[0107] Artificial intelligence technology is an interdisciplinary subject with a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. The software technologies of artificial intelligence mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0108] Machine Learning (ML) is an interdisciplinary subject that involves multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve their own performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent, and its applications cover all fields of artificial intelligence. Machine learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and learning from demonstration.

[0109] The solution provided in the embodiment of the present application can train a feature extraction model based on artificial intelligence machine learning technology, and implement a user feature acquisition method using the trained feature extraction model.

[0110] In the era of artificial intelligence, machine learning, especially deep learning models, requires a large amount of training data as a prerequisite. However, in many business scenarios, the training data of the model is often scattered across different business teams, departments, and even different companies. Due to user privacy, these data cannot be used directly, forming the so-called "data island". In the past two years, Federated Learning technology has developed rapidly, providing new solutions for cross-team data collaboration and breaking the "data island", and has begun to move from theoretical research to the implementation stage of batch application.

[0111] One of the core differences between federated learning and ordinary machine learning tasks is that the training participants have changed from one party to two or even multiple parties. Federated learning allows multiple parties to participate in the same model training task together, complete the model training task without releasing data and protecting data privacy, and break the "data island". Therefore, a core issue is: how to coordinate two or more parties to complete a model training task together. We call this coordination method "federated algorithm protocol". When two or more parties participate in a training task together, each party runs according to the pre-set algorithm protocol to ensure the correct operation of the algorithm.

[0112] The user feature acquisition method provided in the embodiment of the present application can be used in a computer device, which can be a terminal or a server. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited to this. The terminal and the server can be directly or indirectly connected via wired or wireless communication, and this application does not limit this.

[0113] Figure 1 is a structural diagram of a feature extraction system provided in an embodiment of the present application, such as Figure 1 As shown, the system includes a first device 101 and a second device 102. The first device 101 may be a terminal or a server, and the second device 102 may be a terminal or a server.

[0114] The second device 102 invokes the first feature extraction model to extract features from the first user information of the stored target user identifier, obtaining the first user feature. The first device 101 invokes the second feature extraction model to process the second user information of the stored target user identifier, obtaining the second user feature, and sends the second user feature to the second device 102. The second device 102 receives the second user feature sent by the first device 101, and obtains the first combined user feature of the target user identifier according to the first user feature and the second user feature.

[0115] The method provided by the embodiments of the present application can be used in various scenarios.

[0116] For example, in the scenario of item recommendation:

[0117] After the terminal determines the target user identifier, it uses the user feature acquisition method provided by the embodiments of the present application to obtain the first combined user feature of the target user identifier. Subsequently, it can determine the user label of the target user identifier. Through the user label of the target user identifier, it can determine the items that match the target user identifier, and recommend the matching items to the target user, so as to be able to recommend items that meet the target user's preferences to the target user.

[0118] For another example, in the scenario of friend recommendation:

[0119] After the terminal determines the target user identifier, it uses the user feature acquisition method provided by the embodiments of the present application to obtain the first combined user feature of the target user identifier. Subsequently, it can determine the user label of the target user identifier, and recommend other user identifiers that match the user label to the target user according to the user label of the target user identifier, so as to be able to recommend friends with similar preferences to the target user.

[0120] For another example, in the scenario of risk level assessment:

[0121] After the management terminal determines the target user identifier, it uses the user feature acquisition method provided by the embodiments of the present application. Through the user information of the target user identifier in multiple devices, it obtains the first combined user feature of the target user identifier, and evaluates the target user identifier according to the first combined user feature of the target user identifier, obtaining the risk level of the target user identifier. According to the risk level, it can determine the risk of the target user's repayment overdue, or set the fund usage limit for the target user, etc.

[0122] Before elaborating on the method provided by the embodiments of the present application in detail, first, the following explanations are made for the multiple feature extraction models involved in the embodiments of the present application:

[0123] As Figure 2 shown, the second device includes the first feature extraction model, and the first device includes the second feature extraction model.

[0124] Both the first feature extraction model and the second feature extraction model are models used to extract features from user information, and their functions are similar. The difference is that the devices they are stored on are different, and the user information stored on the devices may also be different, which may lead to different user features extracted by the two models based on the user information on their respective devices. For example, the first feature extraction model is used to extract features from the user information stored in the second device, and the second feature extraction model is used to extract features from the user information stored in the first device. Through the first feature extraction model and the second extraction model, the user features of the user information stored in different devices for the same user identifier can be extracted, and the user features extracted by the two models can be combined to obtain combined user features.

[0125] Furthermore, as Figure 3 shown, the second device may further include a fourth feature extraction model, and the first device may further include a fifth feature extraction model.

[0126] Both the fourth feature extraction model and the fifth feature extraction model are models used to extract features from user information, and their functions are similar to those of the first feature extraction model and the second feature extraction model.

[0127] The difference is that the devices where the fourth feature extraction model and the fifth feature extraction model are stored are different, and the user information stored on the devices may also be different, which may lead to different user features extracted by the two models based on the user information on their respective devices.

[0128] The fourth feature extraction model is used to extract features from the user information stored in the second device, and the fifth feature extraction model is used to extract features from the user information stored in the first device. Through the fourth feature extraction model and the fifth extraction model, the user features of the user information stored in different devices for the same user identifier can be extracted, and the user features extracted by the two models can be combined to obtain combined user features.

[0129] Moreover, the first feature extraction model is provided by the second device itself, while the fourth feature extraction model is provided by the first device to the second device. Both of these models can extract features from the user information stored on the second device. Since these two models may be different, different user features may be extracted based on the same user information.

[0130] Similarly, the fifth feature extraction model is provided by the first device itself, while the second feature extraction model is provided by the second device to the first device. Both models can extract features from the user information stored on the first device. Since these two models may be different, they may extract different user features based on the same user information.

[0131] Furthermore, as Figure 4 shown, the second device further includes a third feature extraction model, and the first device further includes a sixth feature extraction model.

[0132] Both the third feature extraction model and the sixth feature extraction model are models for extracting features from user information, and their functions are similar to those of the first feature extraction model and the second feature extraction model.

[0133] The third feature extraction model is stored in the second device and is the original model of the second feature extraction model. That is, the second device encrypts the third feature extraction model to obtain the second feature extraction model and provides the second feature extraction model to the first device.

[0134] The sixth feature extraction model is stored in the first device and is the original model of the fourth feature extraction model. That is, the first device encrypts the sixth feature extraction model to obtain the fourth feature extraction model and provides the fourth feature extraction model to the second device.

[0135] Figure 5 is a flowchart of a method for obtaining user features provided by an embodiment of the present application, which is applied to the second device. As Figure 5 shown, the method includes:

[0136] 501. The second device calls the first feature extraction model to extract features from the first user information of the stored target user identifier, obtaining the first user feature.

[0137] Among them, the first feature extraction model is a model for extracting user features, and the first feature extraction model is stored in the second device.

[0138] The target user identifier is used to represent the unique identifier of the target user, and the target user identifier can be an ID number, a mobile phone number, a user account, a user nickname, etc. The first user information is the user information of the target user corresponding to the target user identifier, and the first user information can include user information in multiple dimensions. For example, the first user information includes the age, salary, height, etc. of the user. The first user feature belongs to the user feature of the target user identifier, and the first user feature can be represented by a vector or a matrix.

[0139] In the embodiment of the present application, the target user identifier and the first user information are stored in the second device correspondingly. Through the target user identifier, the first user information stored in the second device can be determined. Therefore, the second device can call the first feature extraction model stored locally to extract features from the first user information stored locally, and obtain the first user feature, which can describe the target user and reflect the preferences of the target user.

[0140] 502. The second device receives the second user feature sent by the first device.

[0141] In the embodiment of the present application, the first device and the second device are different devices. The user information including the same user identifier is included in the first device and the second device. A communication connection is established between the first device and the second device. Through the established communication connection, the first device and the second device can interact with each other. Therefore, in the embodiment of the present application, through the established communication connection, the first device can send the second user feature to the second device, and the second device receives the second user feature.

[0142] Wherein, the second user feature is obtained by the first device calling the second feature extraction model to process the second user information of the stored target user identifier. The first feature extraction model and the second feature extraction model are different models for extracting user features.

[0143] The second user information is the user information of the target user corresponding to the target user identifier. The second user information may include user information in multiple dimensions. For example, the second user information includes the consumption record, salary, age, etc. of the user. The second user feature belongs to the user feature of the target user identifier, and the second user feature can be represented by a vector or a matrix.

[0144] In the embodiment of the present application, the second user information and the target user identifier are stored in the first device correspondingly. The first device can determine the second user information stored in the first device through the target user identifier. Therefore, the first device can call the second feature extraction model stored locally to extract features from the second user information stored locally, and obtain the second user feature, which can describe the target user and reflect the preferences of the target user.

[0145] Both the second user information and the first user information belong to the user information of the target user. The first user information and the second user information may include user information in the same dimension, or may include at least one different dimension of user information.

[0146] For example, if the first device is a store terminal and the second device is an online shopping server, and the store terminal stores the consumption records of a user, then the user information generated based on the user's consumption records may include: bank account number, consumption amount, consumption time, names of purchased items, etc. The online shopping server stores the user's online shopping records, and the user information that can be generated based on the user's online shopping records may include: bank account number, consumption amount, consumption time, names of purchased items, etc. Then, the user information in the store terminal and the user information in the online shopping server include user information of the same dimension; the user information in the store terminal may include: bank account number, consumption amount, consumption time, names of purchased items, and the user information in the online shopping server may further include: bank account number, consumption amount, consumption time, names of purchased items, user account, user address. Then, the user information in the store terminal and the user information in the online shopping server include user information of at least one different dimension.

[0147] 503. The second device obtains the first combined user feature of the target user identifier according to the first user feature and the second user feature.

[0148] Since both the first user feature and the second user feature belong to the features of the target user identifier, the second device can obtain the first combined user feature of the target user identifier through the first user feature and the second user feature, and the first combined user feature contains the first user feature and the second user feature.

[0149] In the method provided by the embodiments of the present application, by splitting the model for extracting user features and storing them separately in the second device and the first device at the local end, the second device and the first device respectively call the stored feature extraction models to extract features from the user information of the target user identifier stored by each of them. The first device provides the user features extracted by the first device for the second device, without providing the original user information stored by the first device, thus avoiding the leakage of user information. Moreover, the second device combines the user features extracted by the second device and the first device to obtain a combined feature. Since the combined feature includes the features in the user information stored by the second device and the first device, it enriches the information volume of the user features and improves the accuracy of the combined user feature. Compared with the solution of storing different user information in a central device and extracting features by the central device, the embodiments of the present application store user information separately in different devices and extract features separately, realizing a decentralized feature extraction method, avoiding information leakage caused by storing user information in a central device, and improving the security of user information.

[0150] In Figure 5Based on the illustrated embodiments, the second device may also receive the second combined user features sent by the first device, and obtain the user tags of the target user identifier according to the first combined user features and the second combined user features. For the specific process, please refer to the following embodiments.

[0151] Figure 6 FIG. is a flowchart of a user feature acquisition method provided by an embodiment of the present application, which is applied to a first device and a second device. As Figure 6 shown, the method includes:

[0152] 601. The first device sends the fourth feature extraction model to the second device.

[0153] In the embodiment of the present application, the second device stores a first feature extraction model and a second feature extraction model, and the first device stores a fifth feature extraction model and a fourth feature extraction model. The first feature extraction model, the second feature extraction model, the fourth feature extraction model, and the fifth feature extraction model are all different from each other.

[0154] Both the first device and the second device store the user information of the target user identifier. To improve the accuracy of the user tags of the target user identifier obtained by the second device, the combined user features of the target user identifier are obtained through the first user information in the second device and the second user information in the first device, so as to obtain the user tags of the target user identifier. In the process of obtaining the combined user features of the target user identifier, to ensure that the third user features sent by the second device to the first device match the first device, and the second user features sent by the first device to the second device match the second device. Therefore, before obtaining the third user features and the second user features, the first device needs to send the fourth feature extraction model to the second device, and the second device sends the second feature extraction model to the first device, so that both the first device and the second device can call the feature extraction models sent by the other device to extract features from the user information in their own devices, and obtain user features that match the other device.

[0155] Among them, the fourth feature extraction model is a model for obtaining user features, and the fourth feature extraction model is generated by the first device.

[0156] In the embodiment of the present application, the first device and the second device belong to different devices. The first device and the second device include user information with the same user identifier, and a communication connection is established between the first device and the second device. Through the established communication connection, the first device and the second device can interact with each other. Through the communication connection established between the first device and the second device, the first device can send the fourth feature extraction model to the second device.

[0157] Since both the first device and the second device contain user information of the target user identifier, in order to extract features from the user information of the target user identifier stored in the first device and the second device when obtaining the user tags of the target user identifier, therefore, the first device sends the fourth feature extraction model to the second device so that the second device can subsequently call the fourth feature extraction model to obtain the user features of the target user identifier.

[0158] 602. The second device receives the fourth feature extraction model sent by the first device.

[0159] The second device receives the fourth feature extraction model sent by the first device, stores the fourth feature extraction model in the second device, and can subsequently call the fourth feature extraction model to extract features from the first user information in the second device to obtain user features matching the first device.

[0160] 603. The second device sends the second feature extraction model to the first device.

[0161] Among them, the second feature extraction model is a model for obtaining user features, and the second feature extraction model is generated by the second device. Through the communication connection established between the first device and the second device, the second device sends the second feature extraction model to the first device.

[0162] 604. The first device receives the second feature extraction model sent by the second device.

[0163] The first device receives the second feature extraction model sent by the second device, stores the second feature extraction model in the first device, and can subsequently call the second feature extraction model to extract features from the second user information in the first device to obtain user features matching the first device.

[0164] By the first device sending the fourth feature extraction model to the second device and the second device sending the second feature extraction model to the first device, so that subsequently both the first device and the second device can call the feature extraction models sent by the other device to extract features from the user information in their own devices to obtain user features matching the other device, and after obtaining the user features matching the other device, send the user features matching the other device to the other device to facilitate the other device to process the received user features, thus realizing a way of jointly using the user information in multiple devices.

[0165] 605. The second device calls the first feature extraction model to extract features from the stored first user information of the target user identifier to obtain the first user features.

[0166] Among them, the first feature extraction model is a model for extracting user features from the first user information, and the first feature extraction model is stored in the second device.

[0167] In a possible implementation, the first feature extraction model is a first feature extraction model matrix, and the first user information is a first user information matrix; then step 605 may include: taking the product between the first feature extraction model matrix and the first user information as the first user feature.

[0168] For example, if the first feature extraction model matrix is W1 and the first user information is X1, then the first user feature may be W1×X1.

[0169] 606. The first device calls the second feature extraction model to process the second user information of the stored target user identifier, obtains the second user feature, and sends the second user feature to the second device.

[0170] Among them, the second feature extraction model is a model for extracting user features from the second user information. Since the second feature extraction model, the fifth feature extraction model, the first feature extraction model, and the fourth feature extraction model are all different, the second user feature obtained by calling the second feature extraction model is different from the first user feature, the third user feature, and the fourth user feature.

[0171] In a possible implementation, the second feature extraction model is a second feature extraction model matrix, and the second user information is a second user information matrix; then step 606 may include: taking the product between the second feature extraction model matrix and the second user information as the second user feature.

[0172] For example, if the second feature extraction model matrix is R1 and the second user information is X2, then the second user feature may be R1×X2.

[0173] Since the second feature extraction model is sent by the second device, the second user feature obtained by the first device calling the second feature extraction model matches the second device. The second device can use the second user feature. Then, through the communication connection between the first device and the second device, the second user feature is sent to the second device so that the second device can process the second user feature subsequently.

[0174] 607. The second device receives the second user feature sent by the first device, and obtains the first combined user feature of the target user identifier according to the first user feature and the second user feature.

[0175] The second device receives the second user feature sent by the first device. Since the second user feature is obtained by the first device invoking the second feature extraction model to process the second user information, the second user feature matches the second device. The second device can process the first user feature and the second user feature to obtain the first combined user feature.

[0176] In a possible implementation, step 607 may include: The second device combines and processes the first user feature and the second user feature to obtain the first combined user feature. For example, if the first user feature is W1×X1 and the second user feature is R1×X2, then the first combined user feature is W1×X1+R1×X2.

[0177] By sending the second feature extraction model to the first device, the first device invokes the second feature extraction model to extract features from the second user information to obtain the second user feature. There is no need for the first device to send the second user information to the second device, which avoids the leakage of user information, improves security, and also makes the obtained second user feature match the second device. The second device can directly process the received second user feature. Moreover, the first combined user feature obtained by the second device contains the first user feature corresponding to the first user information and the second user feature corresponding to the second user information, making the user features more comprehensive, improving the accuracy of the first combined user feature, and making the subsequent obtained user tags more accurate.

[0178] 608. The second device invokes the fourth feature extraction model to process the first user information to obtain the third user feature, and sends the third user feature to the first device.

[0179] Among them, the fourth feature extraction model is different from the first feature extraction model. The third user feature is the user feature of the target user identifier. Since the fourth feature extraction model is different from the first feature extraction model, by invoking the fourth feature extraction model to process the first user information, the obtained third user feature is different from the first user feature.

[0180] In a possible implementation, the fourth feature extraction model is the fourth feature extraction model matrix, and the first user information is the first user information matrix; then step 608 may include: taking the product between the fourth feature extraction model matrix and the first user information as the third user feature.

[0181] For example, if the fourth feature extraction model matrix is R2 and the first user information is X1, then the third user feature may be R2×X1.

[0182] Since the fourth feature extraction model is sent by the first device, the third user feature obtained by the second device invoking the fourth feature extraction model matches the first device. The first device can use the third user feature. After the second device obtains the third user feature, the third user feature is sent to the first device through the communication connection between the first device and the second device, so that the first device can process the third user feature subsequently.

[0183] 609. The first device receives the third user feature sent by the second device.

[0184] After receiving the third user feature, the first device stores the third user feature, so as to obtain the second combined user feature through the third user feature subsequently.

[0185] 610. The first device invokes the fifth feature extraction model to extract features from the second user information, and obtains the fourth user feature.

[0186] The fifth feature extraction model is a model for extracting user features from the second user information, and the fifth feature extraction model is stored in the second device. Since the fifth feature extraction model, the first feature extraction model, and the fourth feature extraction model are all different, the fourth user feature, the first user feature, and the third user feature obtained by invoking the fifth feature extraction model are all different.

[0187] In the embodiments of the present application, the first user information and the second user information may include user information of the same dimension, or may include at least one different dimension of user information, that is, the first user information is different from the second user information. The user information in the first device and the second device is both generated through interaction with the user. Since the first device and the second device are different, for the target user identifier, the first user information stored in the first device and the second user information stored in the second device may be different.

[0188] For example, the first device is a storage device of a store in the target area, and the second device is a server of a bank in the target area. The bank is used to provide asset storage services to users. Then the first device generates the first user information through the consumption records of the user purchasing items in the store, and the second device generates the second user information through the storage services of the user in the bank. Then the first user information and the second user information may include the same information or may include different information.

[0189] In a possible implementation manner, the fifth feature extraction model is a fifth feature extraction model matrix, and the second user information is a second user information matrix; then step 610 may include: taking the product of the fifth feature extraction model matrix and the second user information as the fourth user feature.

[0190] For example, if the fifth feature extraction model matrix is W2 and the second user information is X2, then the fourth user feature can be W2 × X2.

[0191] 611. The first device obtains the second combined user feature of the target user identifier based on the fourth user feature and the third user feature, and sends the second combined user feature to the second device.

[0192] Since the fourth user feature is obtained through the second user information and the third user feature is obtained through the first user information, the second combined user feature obtained by the first device through the fourth user feature and the third user feature incorporates the user features of different user information, improving the accuracy of the obtained second combined user feature. Through the communication connection between the first device and the second device, the first device sends the second combined user feature to the second device so that the second device can subsequently obtain the user label through the second combined user feature.

[0193] In a possible implementation manner, step 611 may include: The second device combines the fourth user feature and the third user feature to obtain the second combined user feature. For example, if the fourth user feature is W2 × X2 and the third user feature is R2 × X1, then the second combined user feature is W2 × X2 + R2 × X1.

[0194] The first device sends the fourth feature extraction model to the second device. The second device calls the fourth feature extraction model to extract features from the first user information to obtain the third user feature, without the second device sending the first user information to the first device, avoiding the leakage of user information, improving security, and also making the obtained third user feature match the first device. The first device can directly process the received third user feature. Moreover, the second combined user feature obtained by the first device includes the third user feature corresponding to the first user information and the fourth user feature corresponding to the second user information, and the user features are more sufficient, improving the accuracy of the second combined user feature and making the subsequently obtained user label more accurate.

[0195] 612. The second device receives the second combined user feature sent by the first device, and obtains the user label of the target user identifier based on the first combined user feature and the second combined user feature.

[0196] The user label is used to represent the user's preferences. For example, the user label can be a high-consumption user, a low-consumption user, a user who loves traveling, etc.

[0197] Since the first feature extraction model, the second feature extraction model, the fourth feature extraction model, and the fifth feature extraction model are different from each other, the user features obtained by extracting features from the user information through the first feature extraction model, the second feature extraction model, the fourth feature extraction model, and the fifth feature extraction model are also different. Through the interaction between the first device and the second device, the first combined user features and the second combined user features both include the user features corresponding to different user information, and the user features are more sufficient. Therefore, by processing the first combined user features and the second combined user features, the user label of the target user identifier can be obtained, improving the accuracy of the obtained user label.

[0198] In a possible implementation manner, step 612 may include: The second device combines and processes the first combined user features and the second combined user features, and obtains the user label of the target user identifier according to the combined combined user features.

[0199] It should be noted that the embodiments of the present application are described by taking the second device obtaining the user label of the target user identifier according to the first combined user features and the second combined user features. In another embodiment, steps 601-602 and 608-612 do not need to be executed, and the first combined user features obtained by the second device are sufficient. Subsequently, the second device can store the first combined user features.

[0200] It should be noted that the embodiments of the present application are only described by taking the second device obtaining the user label of the target user identifier. The first device obtaining the user label of the target user identifier may include the following two methods:

[0201] The first method: After the second device obtains the user label of the target user identifier, the second device sends the user label of the target user identifier to the first device, and the first device receives the user label of the target user identifier sent by the second device.

[0202] The second method: After the second device obtains the first combined user features, the second device sends the first combined user features to the first device. The first device receives the first combined user features sent by the second device, and obtains the user label of the target user identifier according to the first combined user features and the second combined user features.

[0203] It should be noted that the embodiments of the present application are described by taking the second device obtaining the user label of the target user identifier. In another embodiment, the step in 611 where the first device sends the second combined user features to the second device and step 612 do not need to be executed. Instead, the second device can send the first combined user features to the first device, and the first device obtains the user label of the target user identifier according to the first combined user features and the second combined user features.

[0204] It should be noted that this application is described in terms of obtaining user tags of the target user identifier. After the second device obtains the first combined user feature and the second combined user feature, other information of the target user identifier can also be obtained according to the first combined user feature and the second combined user feature, such as obtaining the category to which the target user identifier belongs, obtaining the risk level of the target user identifier, etc. This application does not make any limitations in this regard.

[0205] The method provided by the embodiments of this application splits the model for extracting user features and stores them separately in the second device and the first device at the local end. The second device and the first device respectively call the stored feature extraction models to extract features from the user information of the target user identifier stored by each of them. The first device provides the user features extracted by the first device for the second device, without providing the original user information stored by the first device, thus avoiding the leakage of user information. Moreover, the second device combines the user features extracted by the second device and the first device to obtain a combined feature. Since this combined feature includes features in the user information stored by the second device and the first device, it enriches the amount of information of the user features and improves the accuracy of the combined user features. Compared with the solution of storing different user information in a central device and performing feature extraction by this central device, the embodiments of this application store user information separately in different devices and perform feature extraction separately, realizing a decentralized feature extraction method, avoiding information leakage caused by storing user information in a central device, and improving the security of user information.

[0206] Furthermore, through the fifth feature extraction model and the fourth feature extraction model, features are respectively extracted from the user information in the first device and the second device to obtain the second combined user feature. Since this combined feature includes features in the user information stored by the first device and the second device, it enriches the amount of information of the user features of the target user identifier and improves the accuracy of the combined user features, so that the user tag can be obtained according to the obtained first combined user feature and the second combined user feature subsequently, thereby improving the accuracy of the user tag.

[0207] In Figure 6 Based on the embodiment shown, before calling the first feature extraction model and the fifth feature extraction model for feature extraction, the first feature extraction model and the fifth feature extraction model need to be trained. The specific process is detailed in the following embodiments.

[0208] Figure 7 is a flowchart of a feature extraction model training method provided by the embodiments of this application, which is applied to the first device and the second device. As Figure 7 shown, this method includes:

[0209] 701. The first device sends the fourth feature extraction model to the second device.

[0210] In an embodiment of the present application, a fifth feature extraction model and a fourth feature extraction model are stored in the first device, and a first feature extraction model and a second feature extraction model are stored in the second device. Through the first sample user information in the second device and the second sample user information in the first device, the fifth feature extraction model in the first device and the first feature extraction model in the second device are jointly trained, and the fourth feature extraction model and the second feature extraction model do not need to be trained.

[0211] Among them, the fourth feature extraction model can be generated by the first device or sent by other devices. The fourth feature extraction model can be randomly initialized by the first device and can be directly used without training; or, the fourth feature extraction model is a model that has been trained. The second feature extraction model can be generated by the second device or sent by other devices. The second feature extraction model can be randomly initialized by the second device and can be directly used without training; or, the second feature extraction model is a model that has been trained.

[0212] 702. The second device receives the fourth feature extraction model sent by the first device.

[0213] 703. The second device sends the second feature extraction model to the first device.

[0214] 704. The first device receives the second feature extraction model sent by the second device.

[0215] Steps 701-704 of the embodiment of the present application are similar to steps 601-604 of the above embodiment, and will not be described in detail here.

[0216] By sending the fourth feature extraction model from the first device to the second device and sending the second feature extraction model from the second device to the first device, so that both the first device and the second device can call the feature extraction models sent by the other device in the subsequent process, perform feature extraction on the sample user information in their own devices, obtain sample user features that match the other device, and train the first feature extraction model and the fifth feature extraction model through the obtained sample user features. After training, the first feature extraction model, the second feature extraction model, the fifth feature extraction model, and the fourth feature extraction model can be used jointly to obtain combined user features.

[0217] It should be noted that during the training process of the first feature extraction model and the fifth feature extraction model, the first device has sent the fourth feature extraction model to the second device, and the second device has sent the second feature extraction model to the first device. Then, after training, through the trained first feature extraction model and fifth feature extraction model, according to Figure 6When obtaining user tags in the illustrated embodiment, steps 601-604 do not need to be repeatedly executed. Subsequently, the first device can directly call the stored second feature extraction model, and the second device can directly call the stored fourth feature extraction model.

[0218] 705. The second device obtains first sample user information.

[0219] Among them, the first sample user information is the user information of the sample user corresponding to the sample user identifier, and the first sample user information is stored in the second device. The first sample user information may include user information in multiple dimensions.

[0220] In a possible implementation manner, obtaining the first sample user information may include: obtaining a plurality of first user identifiers stored in the first device and a plurality of second user identifiers stored in the second device, selecting the same user identifier from the plurality of first user identifiers and the plurality of second user identifiers as the sample user identifier, and using the user information of the sample user identifier already stored in the second device as the first sample user information.

[0221] In a possible implementation manner, obtaining the first sample user information may include: the first device sends the stored plurality of first user identifiers to the second device, the second device receives the plurality of first user identifiers, the second device obtains the stored plurality of second user identifiers, selects the same user identifier from the plurality of first user identifiers and the plurality of second user identifiers as the sample user identifier, uses the user information of the stored sample user identifier as the first sample user information, sends the sample user identifier to the first device, and the first device obtains second sample user information according to the sample user identifier.

[0222] In a possible implementation manner, the second device obtains a plurality of first sample user information.

[0223] 706. The second device calls the first feature extraction model to perform feature extraction on the first sample user information to obtain first sample user features.

[0224] 707. The first device calls the second feature extraction model to process the second sample user information to obtain second sample user features, and sends the second sample user features to the second device.

[0225] 708. The second device receives the second sample user features sent by the first device, and obtains first sample combined user features according to the first sample user features and the second sample user features.

[0226] 709. The second device calls the fourth feature extraction model to process the first sample user information to obtain third sample user features, and sends the third sample user features to the first device.

[0227] 710. The first device receives the third sample user feature sent by the second device.

[0228] 711. The first device invokes the fifth feature extraction model to extract features from the second sample user information, obtaining the fourth sample user feature.

[0229] Wherein, the second sample user information and the first sample user information belong to the same sample user identifier, and the second sample user information is stored in the first device.

[0230] 712. The first device obtains the second sample combined user feature according to the fourth sample user feature and the third sample user feature, and sends the second sample combined user feature to the second device.

[0231] Steps 706 - 712 in the embodiments of this application are similar to steps 605 - 611 in the above embodiments, and will not be elaborated here.

[0232] 713. The second device receives the second sample combined user feature sent by the first device, and trains the first feature extraction model according to the first sample combined user feature, the second sample combined user feature, and the first sample user information.

[0233] Since the first sample combined user feature and the second sample combined user feature are obtained through the first sample user information and the second sample user information, the inaccuracy of the first feature extraction model can be obtained through the first sample combined user feature, the second sample combined user feature, and the first sample user information, and the first feature extraction model is adjusted according to this inaccuracy to improve the accuracy of the first feature extraction model.

[0234] During the process of training the first feature extraction model, the above steps 706 - 713 are repeatedly executed through multiple first sample user information in the first device and multiple second sample user information in the second device to perform iterative training on the first feature extraction model, thereby obtaining the trained first feature extraction model.

[0235] In a possible implementation manner, in response to the number of iterations being equal to the preset number, the training of the first feature extraction model is stopped. Wherein, the preset number can be any set number, such as 20 times, 50 times, etc.

[0236] It should be noted that the embodiments of this application are only described in the process of training the first feature extraction model. The process of training the fifth feature extraction model in the first device may include: after the second device obtains the first sample combination user features, the second device sends the first sample combination user features to the first device. The first device receives the first sample combination user features sent by the second device and trains the fifth feature extraction model according to the first sample combination user features, the second sample combination user features, and the second sample user information.

[0237] In a possible implementation manner, step 713 may include the following steps 7131-7134:

[0238] 7131. The second device obtains the predicted user label of the sample user identifier according to the first sample combination user features and the second sample combination user features.

[0239] The predicted user label is the user label of the sample user corresponding to the sample user identifier and is predicted through the feature extraction model and the sample user information.

[0240] Since both the first sample combination user features and the second sample combination user features include the user features contained in different user information and are obtained after feature extraction by different feature extraction models, the predicted user label of the sample user identifier can be obtained by processing the first sample combination user features and the second sample combination user features.

[0241] In a possible implementation manner, step 7131 may include: obtaining the predicted user label Q of the sample user identifier according to the first sample combination user features P1 and the second sample combination user features P2. The first sample combination user features P1, the second sample combination user features P2, and the predicted user label Q satisfy the following relationship:

[0242] Q = sigmoid(P1 + P2)

[0243]

[0244] where e represents the base of the natural logarithm function and sigmoid() represents the logistic regression function.

[0245] 7132. The second device determines the difference between the predicted user label and the sample user label corresponding to the first sample user information as the first weight.

[0246] Among them, the sample user label is the true user label of the sample user corresponding to the first sample user information, which is used to represent the preferences of the sample user. For example, the sample user label indicates that the sample user belongs to a high-consumption user, a low-consumption user, a travel-loving user, etc. This sample user label can be obtained through manual annotation. The sample user label can be obtained through manual annotation or sent by the second device.

[0247] The first weight is used to represent the difference between the predicted user label of the sample user identifier and the sample user label. Since the predicted user label is obtained through the feature extraction model and the sample user information, and the sample user label is the true user label corresponding to the sample user identifier, there will be a difference between the predicted user label and the sample user label. By determining the difference between the predicted user label and the sample user label as the first weight, it is convenient to adjust the first feature extraction model according to the first weight subsequently, so as to improve the accuracy of the first feature extraction model.

[0248] In a possible implementation manner, this step 7132 may include: determining the difference between the predicted user label Q and the sample user label Y as the first weight m. The predicted user label Q, the sample user label Y, and the first weight m satisfy the following relationship:

[0249] m = Q - Y

[0250] 7133. The second device obtains the first adjustment parameter of the first feature extraction model according to the first weight and the first sample user information.

[0251] Among them, the first adjustment parameter is a parameter used to adjust the first feature extraction model. Since the first feature extraction model extracts features from the first sample user information in the process of obtaining the predicted user label of the sample user identifier, the first adjustment parameter of the first feature extraction model can be obtained through the first weight and the first sample user information, and the first feature extraction model can be adjusted through the first adjustment parameter subsequently.

[0252] In a possible implementation manner, the first weight is the first weight matrix, and the first sample user information is the first sample user information matrix. This step 7133 may include: taking the product between the first weight matrix and the first sample user information matrix as the first adjustment parameter of the first feature extraction model.

[0253] In a possible implementation manner, according to the first weight m and the first sample user information X1, the first adjustment parameter g1 of the first feature extraction model is obtained. The first weight m, the first sample user information X1, and the first adjustment parameter g1 satisfy the following relationship:

[0254] g1 = m × X1

[0255] 7134. The second device adjusts the first feature extraction model according to the first adjustment parameter.

[0256] The first feature extraction model is adjusted by the first adjustment parameter to reduce the difference between the predicted user label obtained by the first feature extraction model and the sample user label, so that the trained first feature extraction model is accurate.

[0257] In a possible implementation manner, the first feature extraction model is the first feature extraction model matrix W1, and this step 7134 may include: adjusting the first feature extraction model W1 according to the first adjustment parameter g1 to satisfy the following relationship:

[0258] W1 = W1 - g1

[0259] In a possible implementation manner, after step 7134, the method further includes: obtaining the loss value of the first feature extraction model according to the predicted user label and the sample user label, and stopping training the first feature extraction model in response to the loss value not being greater than a preset threshold.

[0260] Wherein, the preset threshold is any value set in advance, such as 0.3 or 0.4, etc. The loss value of the first feature extraction model is used to represent the similarity difference between the predicted user label and the sample user label. The smaller the loss value, the more accurate the first feature extraction model. Responding to the loss value of the first feature extraction model not being greater than the preset threshold indicates that the currently trained first feature extraction model already meets the requirements, and the iterative training of the first feature extraction model can be stopped.

[0261] In a possible implementation manner, after step 7132, training the fifth feature extraction model in the first device may include the following steps 7135 - 7138:

[0262] 7135. The second device sends the first weight to the first device.

[0263] This step is similar to the above step 601 and will not be elaborated here.

[0264] 7136. The first device receives the first weight sent by the second device, and obtains the second adjustment parameter of the fifth feature extraction model according to the first weight and the second sample user information.

[0265] This step is similar to the above step 7133 and will not be elaborated here.

[0266] 7137. The first device adjusts the fifth feature extraction model according to the second adjustment parameter.

[0267] This step is similar to the above step 7134 and will not be elaborated here.

[0268] Since the second device combines user features of the first sample, user features of the second sample, and sample user tags to obtain a first weight, and sends the first weight to the second device, enabling the second device to train the fifth feature extraction model according to the first weight, a joint training method between the first device and the second device is achieved, ensuring synchronous training of the first feature extraction model of the first device and the fifth feature extraction model in the second device. Moreover, the second device does not need to send sample user tags to the first device, avoiding the leakage of sample user tags and providing the security of sample user tags.

[0269] In a possible implementation manner, after step 7137, the method further includes:

[0270] 7138. In response to the loss value of the first feature extraction model being greater than a preset threshold, the second device sends a stop training notification to the first device.

[0271] The stop training notification is used to instruct the first device to stop training the fifth feature extraction model.

[0272] 7139. The first device receives the stop training notification sent by the second device and stops training the fifth feature extraction model according to the stop training notification.

[0273] Since the embodiments of the present application perform joint training on the first feature extraction model of the first device and the fifth feature extraction model of the second device through the first device and the second device, when the loss value of the first feature extraction model is greater than the preset threshold, it indicates that the first feature extraction model has met the requirements and the training of the first feature model needs to be stopped, and it also indicates that the fifth feature extraction model has met the requirements and the training of the fifth feature extraction model needs to be stopped. Therefore, the second device sends a stop training notification to the first device to enable the first device to stop training the fifth feature extraction model.

[0274] It should be noted that the embodiments of the present application are described by the first device training the first feature extraction model according to user features of the first sample, user features of the second sample, and first sample user information. In another embodiment, steps 701 - 702, 709 - 713 do not need to be executed, and the second device can train the first feature extraction model according to user features of the first sample and first sample user information.

[0275] It should be noted that the embodiments of the present application are described based on the first device obtaining sample user tags. In another embodiment, the steps in 712 where the first device sends the second sample combined user features to the second device and step 713 are not required. Instead, the second device can send the first sample combined user features to the first device, and the first device receives the first sample combined user features sent by the second device. The first device trains the fifth feature extraction model based on the first sample combined user features, the second sample combined user features, and the second sample user information.

[0276] The method provided by the embodiments of the present application jointly trains the first feature extraction model through the first sample user information in the first device and the second sample user information in the second device, enriching the sample user information and improving the accuracy of the trained first feature extraction model. Moreover, during the training process, there is no need to transmit the first sample user information or the second sample user information, avoiding the leakage of the first sample user information or the second sample user information and improving the security of user information.

[0277] In Figure 6 Based on the embodiments shown, when the first device sends the feature extraction model to the second device, the feature extraction model can be encrypted and then the encrypted feature extraction model is sent. When the second device sends the feature extraction model to the first device, the feature extraction model can be encrypted and then the encrypted feature extraction model is sent. The specific process is detailed in the following embodiments.

[0278] Figure 8 It is a flowchart of a user feature acquisition method provided by the embodiments of the present application, which is applied to the first device and the second device. As Figure 8 shown, the method includes:

[0279] 801. The first device encrypts the sixth feature extraction model according to the second public key to obtain the fourth feature extraction model, and sends the fourth feature extraction model to the second device.

[0280] Among them, the first device includes the second public key and the second private key corresponding to the second public key. The second public key is the key used to encrypt data, and the second private key is the key used to decrypt the data encrypted by the second public key. The second public key and the second private key are used as a key pair. The second public key and the second public key can be sent to the first device by other devices or randomly generated by the first device. The sixth feature extraction model is a model used to extract user features, and the sixth feature extraction model is stored in the first device. When encrypting the sixth feature extraction model with the second public key, the Paillier (The Paillier Cryptosystem, a homomorphic encryption) algorithm or other homomorphic encryption algorithms can be used.

[0281] The first device encrypts the sixth feature extraction model using the second public key to obtain the fourth feature extraction model, and then transmits the encrypted fourth feature extraction model, avoiding the leakage of the sixth feature extraction model and improving the security of the sixth feature extraction model.

[0282] 802. The second device receives the fourth feature extraction model sent by the first device.

[0283] This step is similar to step 602 above and will not be elaborated here.

[0284] 803. The second device encrypts the third feature extraction model according to the first public key to obtain the second feature extraction model, and sends the second feature extraction model to the first device.

[0285] Among them, the second device includes the first public key and the first private key corresponding to the first public key. The first public key is the key used to encrypt data, and the first private key is the key used to decrypt the data encrypted by the second public key. The first public key and the first private key are used as a key pair. The first public key and the first private key can be sent to the second device by other devices, or can be randomly generated by the second device. The third feature extraction model is a model used to extract user features, and the third feature extraction model is stored in the first device.

[0286] The first device encrypts the third feature extraction model according to the first public key to obtain the second feature extraction model, and then transmits the encrypted second feature extraction model, avoiding the leakage of the third feature extraction model and improving the security of the sixth feature extraction model.

[0287] 804. The first device receives the second feature extraction model sent by the second device.

[0288] This step is similar to step 604 above and will not be elaborated here.

[0289] 805. The second device calls the first feature extraction model to extract features from the first user information of the stored target user identifier to obtain the first user feature.

[0290] 806. The first device calls the second feature extraction model to process the second user information of the stored target user identifier to obtain the second user feature, and sends the second user feature to the second device.

[0291] This step is similar to step 606 above and will not be elaborated here.

[0292] 807. The second device receives the second user feature sent by the first device, and decrypts the second user feature according to the first private key corresponding to the first public key to obtain the decrypted user feature.

[0293] Since the second user feature is obtained by the first device processing the second user information using the second feature extraction model, and the second feature extraction model is obtained by encrypting with the first public key, the second user feature is also an encrypted user feature and needs to be decrypted by the second device according to the first private key corresponding to the first public key.

[0294] 808. The second device combines the first user feature and the decrypted user feature to obtain the first combined user feature of the target user identifier.

[0295] This step is similar to step 607 above and will not be elaborated here.

[0296] 809. The second device calls the fourth feature extraction model to process the first user information to obtain the third user feature, and sends the third user feature to the first device.

[0297] This step is similar to step 608 above and will not be elaborated here.

[0298] 810. The first device receives the third user feature sent by the second device, and decrypts the third user feature according to the second private key corresponding to the second public key to obtain the decrypted user feature.

[0299] Since the third user feature is obtained by the second device processing the first user information using the fourth feature extraction model, and the fourth feature extraction model is obtained by encrypting with the second public key, the third user feature is also an encrypted user feature and needs to be decrypted by the first device according to the second private key corresponding to the second public key.

[0300] 811. The first device calls the fifth feature extraction model to extract features from the second user information to obtain the fourth user feature.

[0301] This step is similar to step 610 above and will not be elaborated here.

[0302] 812. The first device combines the fourth user feature and the decrypted user feature to obtain the second combined user feature of the target user identifier, and sends the second combined user feature to the second device.

[0303] 813. The second device receives the second combined user feature sent by the first device, and obtains the user label of the target user identifier based on the first combined user feature and the second combined user feature.

[0304] Steps 812-813 in the embodiments of the present application are similar to steps 611-612 in the above embodiments, and will not be described in detail herein.

[0305] It should be noted that the present application is described by obtaining user tags of the target user identifier. After the second device obtains the first combined user feature and the second combined user feature, other information of the target user identifier can also be obtained according to the first combined user feature and the second combined user feature, such as obtaining the category to which the target user identifier belongs, obtaining the risk level of the target user identifier, etc. The present application does not limit this.

[0306] The method provided by the embodiments of the present application splits the model for extracting user features and stores them separately in the second device and the first device at the local end. The second device and the first device each call the stored feature extraction model to extract features from the user information of the target user identifier stored by each of them. The first device provides the user features extracted by the first device for the second device, without providing the original user information stored by the first device, thus avoiding the leakage of user information. And the second device combines the user features extracted by the second device and the first device to obtain a combined feature. Since this combined feature includes features in the user information stored by the second device and the first device, it enriches the amount of information of the user features and improves the accuracy of the combined user features. Compared with the solution of storing different user information in a central device and extracting features by the central device, the embodiments of the present application store user information separately in different devices and extract features separately, realizing a decentralized feature extraction method, avoiding information leakage caused by storing user information in a central device, and improving the security of user information.

[0307] Moreover, when the feature extraction model is sent between the first device and the second device, the sent feature extraction model is encrypted, avoiding the leakage of the feature extraction model, ensuring the security of the feature extraction model, and thus improving the security of the sent user features.

[0308] In Figure 8 Based on the embodiment shown, before calling the first feature extraction model and the fifth feature extraction model for feature extraction, the first feature extraction model and the fifth feature extraction model need to be trained. The specific process is shown in the following embodiments.

[0309] Figure 9 is a flowchart of a feature extraction model training method provided by the embodiments of the present application, which is applied to a computer device, such as Figure 9 shown, and the method includes:

[0310] 901. The first device encrypts the sixth feature extraction model according to the second public key to obtain the fourth feature extraction model, and sends the fourth feature extraction model to the second device.

[0311] 902. The second device receives the fourth feature extraction model sent by the first device.

[0312] 903. The second device encrypts the third feature extraction model according to the first public key to obtain the second feature extraction model, and sends the second feature extraction model to the first device.

[0313] 904. The first device receives the second feature extraction model sent by the second device.

[0314] Steps 901-904 of the embodiments of this application are similar to the above steps 801-804, and will not be elaborated here.

[0315] It should be noted that during the training process of the first feature extraction model and the fifth feature extraction model, the first device has sent the fourth feature extraction model to the second device, and the second device has sent the second feature extraction model to the first device. After the training is completed, when obtaining user tags according to the Figure 8 embodiment shown, there is no need to repeat steps 801-804. Subsequently, the first device can directly call the stored second feature extraction model, and the second device can directly call the stored fourth feature extraction model.

[0316] 905. The second device obtains the first sample user information.

[0317] This step is similar to the above step 705, and will not be elaborated here.

[0318] 906. The second device calls the first feature extraction model to extract features from the first sample user information to obtain the first sample user features.

[0319] 907. The first device calls the second feature extraction model to process the second sample user information to obtain the second sample user features, and sends the second user features to the second device.

[0320] 908. The second device receives the second user features sent by the first device, and decrypts the second sample user features according to the first private key corresponding to the first public key to obtain the decrypted sample user features.

[0321] 909. The second device combines the first sample user features and the decrypted sample user features to obtain the first sample combined user features.

[0322] 910. The second device calls the fourth feature extraction model to process the first sample user information to obtain the third sample user features, and sends the third sample user features to the first device.

[0323] 911. The first device receives the third user feature sent by the second device, and decrypts the third sample user feature according to the second private key corresponding to the second public key to obtain the decrypted sample user feature.

[0324] 912. The first device invokes the fifth feature extraction model to extract features from the second sample user information to obtain the fourth sample user feature.

[0325] 913. The first device combines the fourth sample user feature and the decrypted sample user feature to obtain the second sample combined user feature, and sends the second sample combined user feature to the second device.

[0326] Steps 906-913 of the embodiments of this application are similar to the above steps 805-812 and will not be elaborated here.

[0327] 914. The second device receives the second sample combined user feature sent by the first device, and trains the first feature extraction model according to the first sample combined user feature, the second sample combined user feature, and the first sample user information.

[0328] Since the first sample combined user feature and the second sample combined user feature are obtained from the first sample user information and the second sample user information, the inaccuracy of the first feature extraction model can be obtained through the first sample combined user feature, the second sample combined user feature, and the first sample user information, and the first feature extraction model is adjusted according to the inaccuracy to improve the accuracy of the first feature extraction model.

[0329] During the training process of the first feature extraction model, the above steps 906-914 are repeatedly executed through multiple first sample user information in the first device and multiple second sample user information in the second device to perform iterative training on the first feature extraction model, so as to obtain the trained first feature extraction model.

[0330] In a possible implementation manner, in response to the iteration number being equal to the preset number, the training of the first feature extraction model is stopped. Wherein, the preset number can be any set number, such as 20 times, 50 times, etc.

[0331] It should be noted that the embodiments of the present application are only described by taking the process of training the first feature extraction model as an example. The process of training the fifth feature extraction model in the first device may include: after the second device obtains the first sample combined user features, the second device sends the first sample combined user features to the first device, and the first device receives the first sample combined user features sent by the second device, and trains the fifth feature extraction model according to the first sample combined user features, the second sample combined user features, and the second sample user information.

[0332] In a possible implementation manner, step 914 may include the following steps:

[0333] 9141. The second device obtains the predicted user label of the sample user identifier according to the first sample combined user features and the second sample combined user features.

[0334] 9142. The second device determines the difference between the predicted user label and the sample user label corresponding to the first sample user information as the first weight.

[0335] 9143. The second device obtains the first adjustment parameter of the first feature extraction model according to the first weight and the first sample user information.

[0336] 9144. The second device adjusts the first feature extraction model according to the first adjustment parameter.

[0337] Steps 9141-9144 in the embodiments of the present application are similar to the above steps 7131-7134 and will not be elaborated here.

[0338] In another possible implementation manner, after step 9142, training the fifth feature extraction model in the first device may include the following steps 9145-9153:

[0339] 9145. The second device encrypts the first weight according to the first public key to obtain the second weight and sends the second weight to the first device.

[0340] This step is similar to the above step 803 and will not be elaborated here.

[0341] 9146. The first device receives the second weight sent by the second device, obtains the second adjustment parameter according to the second weight and the second sample user information, and sends the second adjustment parameter to the second device.

[0342] This step is similar to the above step 7133 and will not be elaborated here.

[0343] 9147. The second device receives the second adjustment parameter sent by the first device, decrypts the second adjustment parameter according to the first private key corresponding to the first public key to obtain a third adjustment parameter, and sends the third adjustment parameter to the first device.

[0344] Since the second adjustment parameter is obtained through the second weight and the second sample user information, and the second weight is obtained by the second device encrypting the first weight according to the first public key, the second adjustment parameter is also an encrypted parameter. Therefore, the second device needs to decrypt the second adjustment parameter according to the first private key corresponding to the first public key to obtain the decrypted third adjustment parameter.

[0345] This step is similar to step 807 above and will not be elaborated here.

[0346] 9148. The first device receives the third adjustment parameter sent by the second device and adjusts the fifth feature extraction model according to the third adjustment parameter.

[0347] This step is similar to step 7134 above and will not be elaborated here.

[0348] Since the second device obtains the first weight through the first sample combined user features, the second sample combined user features, and the sample user label, in order to ensure that the second device can train the fifth feature extraction model according to the first weight and avoid the leakage of the first weight, the second device sends the second weight obtained by encrypting the first weight to the first device. The first device obtains the encrypted second adjustment parameter according to the second weight, and then the second device decrypts the second adjustment parameter and sends the decrypted third adjustment parameter to the first device. The first device trains the fifth feature extraction model according to the third adjustment parameter, thus realizing the joint training process between the first device and the second device, avoiding the leakage of the first weight, and improving the security of the first weight.

[0349] In a possible implementation manner, after step 9148, the method further includes:

[0350] 9149. In response to the loss value of the first feature extraction model being greater than a preset threshold, the second device sends a stop training notification to the first device.

[0351] 9150. The first device receives the stop training notification sent by the second device and stops training the fifth feature extraction model according to the stop training notification.

[0352] The stop training notification is used to instruct the first device to stop training the fifth feature extraction model.

[0353] Since the embodiments of the present application jointly train the first feature extraction model of the first device and the fifth feature extraction model of the second device through the first device and the second device, when the loss value of the first feature extraction model is greater than the preset threshold, it indicates that the first feature extraction model has met the requirements and the training of the first feature model needs to be stopped. It also indicates that the fifth feature extraction model has met the requirements and the training of the fifth feature extraction model needs to be carried out. Therefore, the second device sends a stop training notification to the first device so that the first device stops training the fifth feature extraction model.

[0354] The method provided by the embodiments of the present application jointly trains the first feature extraction model through the first sample user information in the first device and the second sample user information in the second device, enriching the sample user information and improving the accuracy of the trained first feature extraction model. And during the training process, there is no need to transmit the first sample user information or the second sample user information, avoiding the leakage of the first sample user information or the second sample user information and improving the security of user information.

[0355] Moreover, during the training process of the feature extraction model, when the feature extraction model is sent between the first device and the second device, the sent feature extraction model is encrypted to avoid the leakage of the feature extraction model and ensure the security of the feature extraction model. The first weight of the adjustment model is encrypted to avoid the leakage of the first weight and improve the security of the first weight.

[0356] In Figure 6 Based on the shown embodiments, during the process of transmitting user features and combined user features between the first device and the second device, noise features can be added to the user features and combined user features to avoid the leakage of user features. For the specific process, please refer to the following embodiments.

[0357] Figure 10 is a flowchart of a user feature acquisition method provided by the embodiments of the present application, which is applied to the first device and the second device. As Figure 10 shown, the method includes:

[0358] 1001. The first device sends the fourth feature extraction model to the second device.

[0359] 1002. The second device receives the fourth feature extraction model sent by the first device.

[0360] 1003. The second device sends the second feature extraction model to the first device.

[0361] 1004. The first device receives the second feature extraction model sent by the second device.

[0362] 1005. The second device invokes the first feature extraction model to extract features from the first user information of the stored target user identifier, obtaining the first user feature.

[0363] 1006. The first device invokes the second feature extraction model to extract features from the second user information, obtaining the fifth user feature.

[0364] In the embodiments of this application, steps 1001 - 1006 are similar to the above steps 601 - 606 and will not be elaborated here.

[0365] 1007. The first device performs a fusion process on the fifth user feature and the first noise feature to obtain the second user feature, and sends the second user feature to the second device.

[0366] Among them, the first noise feature is randomly generated by the first device. The first noise can be represented by a vector, a matrix, or other forms.

[0367] Since the fifth user feature is obtained by extracting features from the second user information, the first device fuses the fifth user feature with the first noise feature, so that the obtained second user feature contains the first noise feature. Even if the second user feature is leaked, the second user information cannot be obtained through the second user feature, reducing the risk of leakage of the second user information and improving the security of the second user information.

[0368] 1008. The second device receives the second user feature sent by the first device, and performs a combination process on the first user feature and the second user feature to obtain the fifth combined user feature.

[0369] This step is similar to the above step 607 and will not be elaborated here.

[0370] 1009. The second device performs a fusion process on the fifth combined user feature and the fourth noise feature to obtain the first combined user feature of the target user identifier.

[0371] Among them, the fourth noise feature is randomly generated by the second device. The fourth noise can be represented by a vector, a matrix, or other forms. The third noise feature is opposite to the fourth noise feature. By fusing the first combined user feature with the fourth noise feature, the obtained first combined user feature contains the fourth noise feature, which is convenient for subsequent removal of the third noise in the second combined user feature.

[0372] 1010. The second device invokes the fourth feature extraction model to extract features from the first user information, obtaining the sixth user feature.

[0373] This step is similar to the above step 608 and will not be elaborated here.

[0374] 1011. The second device performs fusion processing on the sixth user feature and the third noise feature to obtain a third user feature, and sends the third user feature to the first device.

[0375] Among them, the third noise feature is randomly generated by the second device. The third noise can be represented by a vector, a matrix, or other forms. The third noise feature is opposite to the fourth noise feature.

[0376] Since the sixth user feature is obtained by performing feature extraction on the first user information, the second device fuses the sixth user feature with the third noise feature, so that the obtained third user feature contains the third noise feature. Even if the third user feature is leaked, the first user information cannot be obtained through the third user feature, reducing the risk of leakage of the first user information and improving the security of the first user information.

[0377] 1012. The first device receives the third sample user feature sent by the second device.

[0378] 1013. The first device calls the fifth feature extraction model to perform feature extraction on the second user information to obtain a fourth user feature.

[0379] 1014. The first device performs combination processing on the fourth user feature and the third user feature to obtain a third combined user feature.

[0380] Steps 1012-1014 in the embodiments of the present application are similar to the above steps 609-611, and will not be described in detail here.

[0381] 1015. The first device performs fusion processing on the third combined user feature and the second noise feature to obtain a second combined user feature of the target user identifier, and sends the second combined user feature to the second device.

[0382] Among them, the first noise feature is opposite to the second noise feature. By fusing the third combined user feature with the second noise feature, the obtained second combined user feature contains the second noise feature, which is convenient for subsequent removal of the first noise in the first combined user feature.

[0383] 1016. The second device receives the second combined user feature sent by the first device, and performs combination processing on the first combined user feature and the second combined user feature to obtain a fourth combined user feature.

[0384] Since the first noise feature is incorporated into the second user feature, the first user feature and the second user feature are combined and processed, so that the first noise feature is also incorporated into the obtained fifth combined user feature. The first combined user feature is obtained by fusing the fifth combined user feature with the fourth noise feature. Therefore, the first noise feature and the fourth noise feature are incorporated into the first combined user feature.

[0385] Since the third noise feature is incorporated into the third user feature, the third user feature and the fourth user feature are combined and processed, so that the third noise feature is also incorporated into the obtained third combined user feature. The second combined user feature is obtained by fusing the third combined user feature with the second noise feature. Therefore, the third noise feature and the second noise feature are incorporated into the second combined user feature.

[0386] Since the first noise feature is opposite to the second noise feature, and the third noise feature is opposite to the fourth noise feature, the first combined user feature and the second combined user feature are combined and processed, so that the first noise feature and the second noise feature are cancelled out, and the third noise feature and the fourth noise feature are cancelled out, realizing the denoising process of the combined user feature, so that the obtained fourth combined user feature does not contain noise features.

[0387] 1017. The second device obtains the user label of the target user identifier according to the fourth combined user feature.

[0388] This step is similar to step 612 above and will not be elaborated here.

[0389] The method provided by the embodiment of the present application splits the model for extracting user features and stores them separately in the second device and the first device at the local end. The second device and the first device respectively call the stored feature extraction models to extract features from the user information of the target user identifier stored by each of them. The first device provides the user features extracted by the first device for the second device, without providing the original user information stored by the first device, avoiding the leakage of user information. And the second device combines the user features extracted by the second device and the first device to obtain a combined feature. Since the combined feature includes the features in the user information stored by the second device and the first device, the amount of information of the user features is enriched, and the accuracy of the combined user features is improved. Compared with the solution of storing different user information in a central device and extracting features by the central device, the embodiment of the present application stores user information separately in different devices and extracts features separately, realizing a decentralized feature extraction method, avoiding information leakage caused by storing user information in a central device, and improving the security of user information.

[0390] Moreover, during the interaction between the first device and the second device, noise is added to the user features and combined user features transmitted between the first device and the second device, avoiding the leakage of user features and combined user features and thus preventing the leakage of user information, and improving the security of user information.

[0391] Based on the Figure 10 embodiment shown, before calling the first feature extraction model and the fifth feature extraction model for feature extraction, the first feature extraction model and the fifth feature extraction model need to be trained. The specific process is described in detail in the following embodiments.

[0392] Figure 11 is a flowchart of a user feature acquisition method provided by an embodiment of the present application, which is applied to the first device and the second device. As Figure 11 shown, the method includes:

[0393] 1101. The first device sends the fourth feature extraction model to the second device.

[0394] 1102. The second device receives the fourth feature extraction model sent by the first device.

[0395] 1103. The second device sends the second feature extraction model to the first device.

[0396] 1104. The first device receives the second feature extraction model sent by the second device.

[0397] Steps 1101 - 1104 in the embodiment of the present application are similar to steps 601 - 604 above and will not be elaborated here.

[0398] It should be noted that during the training process of the first feature extraction model and the fifth feature extraction model, the first device has sent the fourth feature extraction model to the second device, and the second device has sent the second feature extraction model to the first device. Then, after the training is completed, when obtaining user tags according to the Figure 10 embodiment shown, there is no need to repeat steps 1001 - 1004. Subsequently, the first device can directly call the stored second feature extraction model, and the second device can directly call the stored fourth feature extraction model.

[0399] 1105. The second device obtains the first sample user information.

[0400] This step is similar to step 705 above and will not be elaborated here.

[0401] 1106. The second device calls the first feature extraction model to perform feature extraction on the first sample user information to obtain the first sample user features.

[0402] 1107. The first device invokes the second feature extraction model to extract features from the second user information to obtain the fifth sample user features.

[0403] 1108. The first device performs a fusion process on the fifth sample user features and the first noise features to obtain the second sample user features, and sends the second sample user features to the second device.

[0404] 1109. The second device receives the second sample user features sent by the first device, and performs a combination process on the first sample user features and the second sample user features to obtain the fifth sample combined user features.

[0405] 1110. The second device performs a fusion process on the fifth sample combined user features and the fourth noise features to obtain the first sample combined user features.

[0406] 1111. The second device invokes the fourth feature extraction model to extract features from the first user information to obtain the sixth sample user features.

[0407] 1112. The second device performs a fusion process on the sixth sample user features and the third noise features to obtain the third sample user features, and sends the third sample user features to the first device.

[0408] 1113. The first device receives the third sample user features sent by the second device.

[0409] 1114. The first device invokes the fifth feature extraction model to extract features from the second user information to obtain the fourth sample user features.

[0410] 1115. The first device performs a combination process on the fourth sample user features and the third sample user features to obtain the third sample combined user features.

[0411] 1116. The first device performs a fusion process on the third sample combined user features and the second noise features to obtain the second sample combined user features, and sends the second sample combined user features to the second device.

[0412] 1117. The second device receives the second sample combined user features sent by the first device, and performs a combination process on the first sample combined user features and the second sample combined user features to obtain the fourth sample combined user features.

[0413] Steps 1106 - 1117 in the embodiments of this application are similar to the above steps 1005 - 1016, and will not be elaborated here.

[0414] 1118. The second device trains the first feature extraction model according to the fourth sample combined user features and the first sample user information.

[0415] This step is similar to the above-mentioned 713 and will not be elaborated here.

[0416] The method provided in the embodiment of the present application jointly trains the first feature extraction model through the first sample user information in the first device and the second sample user information in the second device, enriches the sample user information, and improves the accuracy of the trained first feature extraction model. And during the training process, there is no need to transmit the first sample user information or the second sample user information, avoiding the leakage of the first sample user information or the second sample user information and improving the security of user information.

[0417] Moreover, during the training process of the feature extraction model, noise features are added to the user features and combined user features transmitted between the first device and the second device, avoiding the leakage of user information caused by the leakage of user features and combined user features, and improving the security of user information.

[0418] Will pair Figure 8 The solution for obtaining user tags by model encryption in the embodiment, and Figure 10 The solution for obtaining user tags by adding noise features are combined. During the process of jointly obtaining user tags by the first device and the second device, the transmitted feature extraction model can be encrypted, and noise features can be added to the transmitted user features and combined user features. For the specific process, please refer to the following embodiments.

[0419] Figure 12 is a flowchart of a user feature acquisition method provided by an embodiment of the present application, which is applied to the first device and the second device. As Figure 12 shown, the method includes:

[0420] 1201. The first device encrypts the sixth feature extraction model according to the second public key to obtain the fourth feature extraction model, and sends the fourth feature extraction model to the second device.

[0421] 1202. The second device receives the fourth feature extraction model sent by the first device.

[0422] 1203. The second device encrypts the third feature extraction model according to the first public key to obtain the second feature extraction model, and sends the second feature extraction model to the first device.

[0423] 1204. The first device receives the second feature extraction model sent by the second device.

[0424] 1205. The second device calls the first feature extraction model to extract features from the first user information of the stored target user identifier to obtain the first user feature.

[0425] 1206. The first device invokes the second feature extraction model to extract features from the second user information, obtaining the fifth user feature.

[0426] 1207. The first device performs a fusion process on the fifth user feature and the first noise feature to obtain the second user feature, and sends the second user feature to the second device.

[0427] 1208. The second device receives the second user feature sent by the first device, and decrypts the second user feature according to the first private key corresponding to the first public key, obtaining the decrypted user feature.

[0428] 1209. The second device performs a combination process on the first user feature and the decrypted user feature, obtaining the fifth combined user feature.

[0429] 1210. The second device performs a fusion process on the fifth combined user feature and the fourth noise feature, obtaining the first combined user feature of the target user identifier.

[0430] 1211. The second device invokes the fourth feature extraction model to extract features from the first user information, obtaining the sixth user feature.

[0431] 1212. The second device performs a fusion process on the sixth user feature and the third noise feature, obtaining the third user feature, and sends the third user feature to the first device.

[0432] 1213. The first device receives the third sample user feature sent by the second device, and decrypts the third user feature according to the second private key corresponding to the second public key, obtaining the decrypted user feature.

[0433] 1214. The first device invokes the fifth feature extraction model to extract features from the second user information, obtaining the fourth user feature.

[0434] 1215. The first device performs a combination process on the fourth user feature and the decrypted user feature, obtaining the third combined user feature.

[0435] 1216. The first device performs a fusion process on the third combined user feature and the second noise feature, obtaining the second combined user feature of the target user identifier, and sends the second combined user feature to the second device.

[0436] 1217. The second device receives the second combined user feature sent by the first device, and performs a combination process on the first combined user feature and the second combined user feature, obtaining the fourth combined user feature.

[0437] 1218. The second device obtains the user label of the target user identifier according to the fourth combined user feature.

[0438] For example, according to the solution in Figure 12 the embodiment, the specific process of obtaining user tags is as follows. The first device includes a fifth feature extraction model W2 and a sixth feature extraction model R1, the second device includes a first feature extraction model W1 and a third feature extraction model R2, the first device generates a second public key PK2 and a second private key SK2, the second device generates a first public key PK1 and a first private key SK1, the first device stores second user information X2 of the target user identifier, and the second device stores first user information X1 of the target user identifier. The second device sends the second feature extraction model PK1(R2) to the first device, and the first device stores it. The first device sends the fourth feature extraction model PK2(R1) to the second device, and the second device stores it.

[0439] The second device obtains the first user feature W1*X1 through the first feature extraction model W1 and the first user information X1.

[0440] The first device obtains the fifth user feature PK1(R2)*X2 through the second feature extraction model PK1(R2) and the second user information X2, sends the fifth user feature PK1(R2)*X2 to the second device, fuses the fifth user feature PK1(R2)*X2 with the first noise feature -noise2 to obtain the second user feature PK1(R2)*X2-noise2, and sends the second user feature PK1(R2)*X2-noise2 to the second device.

[0441] The second device receives the second user feature PK1(R2)*X2-noise2 sent by the first device, decrypts the second user feature PK1(R2)*X2-noise2 according to the first private key SK1 to obtain the decrypted user feature R2*X2-noise2, performs a fusion process on the first user feature W1*X1 and the decrypted user feature R2*X2-noise2 to obtain the fifth combined user feature W1*X1+R2*X2-noise2, and fuses the fifth combined user feature with the fourth noise feature noise1 to obtain the first combined user feature W1*X1+R2*X2-noise2+noise1.

[0442] The second device calls the fourth feature extraction model PK2(R1) to perform feature extraction on the first user information X1 to obtain the sixth user feature PK2(R1)*X1, fuses the sixth user feature PK2(R1)*X1 with the third noise -noise1 to obtain the third user feature PK2(R1)*X1-noise1, and sends the third user feature PK2(R1)*X1-noise1 to the first device.

[0443] The first device decrypts the third user feature PK2(R1)*X1-noise1 according to the second private key SK2 to obtain the decrypted user feature R1*X1-noise1, invokes the fifth feature extraction model W2 to extract features from the second user information X2 to obtain the fourth user feature W2*X2, combines the fourth user feature W2*X2 with the decrypted user feature R1*X1-noise1 to obtain the third combined user feature W2*X2+R1*X1-noise1, fuses the third combined user feature W2*X2+R1*X1-noise1 with the second noise noise2 to obtain the second combined user feature W2*X2+R1*X1-noise1+noise2, and sends the second combined user feature to the second device.

[0444] The second device fuses the first combined user feature W1*X1+R2*X2-noise2+noise1 with the second combined user feature W2*X2+R1*X1-noise1+noise2 to obtain the fourth combined user feature W1*X1+R2*X2+W2*X2+R1*X1, and obtains the first user label of the target user identifier according to the fourth combined user feature W1*X1+R2*X2+W2*X2+R1*X1

[0445] Based on Figure 12 the embodiments shown, in the process of jointly training the first feature extraction model and the fifth feature extraction model, it can be obtained by combining the joint training solutions in the above Figure 9 embodiments with Figure 11 the joint training solutions in the embodiments.

[0446] It should be noted that based on the above embodiments, in the process of training the first feature extraction model and the fifth feature extraction model, it can be obtained by combining steps 901-914 and 1101-1118 in the above embodiments.

[0447] In addition, it should be noted that after the second device obtains the first weight, in order to enable the first device to train the fifth feature extraction model, the method may include:

[0448] 1219. The second device encrypts the first weight according to the first public key to obtain the second weight, and sends the second weight to the first device.

[0449] 1220. The first device receives the second weight sent by the second device, and obtains the fourth adjustment parameter according to the second weight and the second sample user information.

[0450] 1221. The first device fuses the fourth adjustment parameter and the fifth noise feature to obtain a second adjustment parameter, and sends the second adjustment parameter to the second device.

[0451] Among them, the fifth noise feature is randomly generated by the first device. The first noise can be represented by a vector, a matrix, or other forms.

[0452] 1222. The second device receives the second adjustment parameter sent by the first device, decrypts the second adjustment parameter according to the first private key corresponding to the first public key to obtain a third adjustment parameter, and sends the third adjustment parameter to the first device.

[0453] Since the second adjustment parameter is obtained through the second weight, and the second weight is obtained by the second device encrypting the first weight according to the first public key, the second adjustment parameter is also an encrypted parameter. Therefore, the second device needs to decrypt the second adjustment parameter according to the first private key corresponding to the first public key to obtain the decrypted third adjustment parameter. The second device sends the decrypted third adjustment parameter to the first device so that the first device can adjust the fifth feature extraction model according to the third adjustment parameter.

[0454] 1223. The first device receives the third adjustment parameter sent by the second device, fuses the third adjustment parameter and the sixth noise feature, and adjusts the fifth feature extraction model according to the fused adjustment parameter.

[0455] Among them, the fifth noise feature is opposite to the sixth noise feature. Since the second adjustment parameter contains the fifth noise, the third adjustment parameter contains the fifth noise feature. By fusing the third adjustment parameter and the sixth noise feature, the resulting fused adjustment parameter does not contain noise features, realizing the denoising process of the adjustment parameter.

[0456] Since the fourth adjustment parameter is obtained through the second weight and the second sample user information, by fusing the fourth adjustment parameter and the fifth noise feature, the transmitted second adjustment parameter contains noise features, and the second device cannot obtain the second sample user information according to the second adjustment parameter, improving the security of the second sample user information.

[0457] Regarding the specific process of adjusting the fifth feature extraction model according to the above steps 1219 - 1223. For example, if the first weight is m and the first public key is PK1, the second weight obtained by the second device is PK1(m), and the second weight PK1(m) is sent to the first device.

[0458] The first device obtains the fourth adjustment parameter PK1(m)*X2 through the second weight PK1(m) and the second sample user information X2, fuses the fourth adjustment parameter PK1(m)*X2 with the fifth noise σ to obtain the second adjustment parameter PK1(m)*X2+σ, and sends the second adjustment parameter to the second device.

[0459] The second device decrypts the second adjustment parameter according to the first private key SK1 to obtain the third adjustment parameter m*X2+σ, and sends the third adjustment parameter m*X2+σ to the first device.

[0460] The first device fuses the third adjustment parameter m*X2+σ with the sixth noise feature -σ to obtain the fused adjustment parameter g2 as m*X2, and adjusts the fifth feature extraction model W2 according to the fused adjustment parameter g2. The adjusted fifth feature model can be expressed as W2 = W2 - g2.

[0461] Figure 13 It is a schematic structural diagram of a user feature acquisition device provided by an embodiment of the present application. As Figure 13 shown, the device includes:

[0462] A feature extraction module 1301, configured to call a first feature extraction model to extract features from the first user information of the stored target user identifier to obtain a first user feature;

[0463] A feature receiving module 1302, configured to receive a second user feature sent by the first device. The second user feature is obtained by the first device calling a second feature extraction model to process the second user information of the stored target user identifier. The first feature extraction model and the second feature extraction model are different models for extracting user features;

[0464] A combined feature acquisition module 1303, configured to obtain a first combined user feature of the target user identifier according to the first user feature and the second user feature.

[0465] In a possible implementation manner, as Figure 14 shown, the second feature extraction model is a model after being encrypted according to the first public key. The combined feature acquisition module 1303 includes:

[0466] A decryption processing unit 1331, configured to decrypt the second user feature according to the first private key corresponding to the first public key to obtain the decrypted user feature;

[0467] A first combination processing unit 1332, configured to perform combination processing on the first user feature and the decrypted user feature to obtain a first combined user feature.

[0468] In another possible implementation manner, asFigure 14 As shown in the figure, the device further includes:

[0469] An encryption processing module 1304, configured to encrypt the third feature extraction model according to the first public key to obtain a second feature extraction model;

[0470] A model sending module 1305, configured to send the second feature extraction model to the first device.

[0471] In another possible implementation manner, as Figure 14 shown in the figure, the device further includes:

[0472] An information processing module 1306, configured to call a fourth feature extraction model to process the first user information to obtain a third user feature;

[0473] A feature sending module 1307, configured to send the third user feature to the first device, where the first device is configured to obtain a second combined user feature according to the fourth user feature and the third user feature, and the fourth user feature is obtained by the first device calling a fifth feature extraction model to extract features from the second user information;

[0474] A feature receiving module 1302, configured to receive the second combined user feature sent by the first device.

[0475] In another possible implementation manner, as Figure 14 shown in the figure, the device further includes:

[0476] A model receiving module 1308, configured to receive the fourth feature extraction model sent by the first device.

[0477] In another possible implementation manner, the first device is configured to encrypt the sixth feature extraction model according to the second public key to obtain a fourth feature extraction model;

[0478] The first device is configured to decrypt the third user feature according to the second private key corresponding to the second public key to obtain a decrypted user feature; and perform a combination process on the fourth user feature and the decrypted user feature to obtain a second combined user feature.

[0479] In another possible implementation manner, the second user feature is obtained by the first device fusing a fifth user feature and a first noise feature, and the fifth user feature is obtained by the first device calling a second feature extraction model to extract features from the second user information;

[0480] The second combined user feature is obtained by the first device fusing a third combined user feature and a second noise feature, and the third combined user feature is obtained by the first device combining the fourth user feature and the third user feature, and the first noise feature is opposite to the second noise feature;

[0481] The apparatus further includes:

[0482] A combined processing module 1309, configured to perform combined processing on the first combined user feature and the second combined user feature to obtain a fourth combined user feature.

[0483] In another possible implementation, as Figure 14 shown, the information processing module 1306 includes:

[0484] A feature extraction unit 1361, configured to call a fourth feature extraction model to perform feature extraction on the first user information to obtain a sixth user feature;

[0485] A first fusion processing unit 1362, configured to perform fusion processing on the sixth user feature and the third noise feature to obtain a third user feature;

[0486] A combined feature acquisition module 1303 includes:

[0487] A second combined processing unit 1333, configured to perform combined processing on the first user feature and the second user feature to obtain a fifth combined user feature;

[0488] A second fusion processing unit 1334, configured to perform fusion processing on the fifth combined user feature and the fourth noise feature to obtain a first combined user feature, where the third noise feature is opposite to the fourth noise feature.

[0489] In another possible implementation, as Figure 14 shown, the apparatus further includes:

[0490] A sample acquisition module 1310, configured to acquire first sample user information;

[0491] A feature extraction module 1301 is further configured to call a first feature extraction model to perform feature extraction on the first sample user information to obtain a first sample user feature;

[0492] A feature receiving module 1302 is further configured to receive a second sample user feature, where the second sample user feature is obtained by a first device calling a second feature extraction model to process second sample user information, and the first sample user information and the second sample user information belong to the same sample user identifier;

[0493] A combined feature acquisition module 1303 is further configured to obtain a first sample combined user feature according to the first sample user feature and the second sample user feature;

[0494] A model training module 1311, configured to train the first feature extraction model according to the first sample combined user feature and the first sample user information.

[0495] In another possible implementation, as Figure 14 shown, the apparatus further includes:

[0496] An information processing module 1306, further configured to call a fourth feature extraction model to process the first sample user information to obtain a third sample user feature;

[0497] A feature sending module 1307, further configured to send the third sample user feature to a first device, where the first device is configured to obtain a second sample combined user feature according to the fourth sample user feature and the third sample user feature, and the fourth sample user feature is obtained by the first device calling a fifth feature extraction model to perform feature extraction on the second sample user information;

[0498] A feature receiving module 1302, further configured to receive the second sample combined user feature sent by the first device;

[0499] A model training module 1311 includes:

[0500] A model training unit 13111, configured to train a first feature extraction model according to the first sample combined user feature, the second sample combined user feature, and the first sample user information.

[0501] In another possible implementation, the model training unit 13111 is configured to obtain a predicted user label of a sample user identifier according to the first sample combined user feature and the second sample combined user feature; determine a difference between the predicted user label and the sample user label corresponding to the first sample user information as a first weight; obtain a first adjustment parameter of the first feature extraction model according to the first weight and the first sample user information; and adjust the first feature extraction model according to the first adjustment parameter.

[0502] In another possible implementation, as Figure 14 shown, the apparatus further includes:

[0503] A weight encryption module 1312, configured to perform encryption processing on the first weight according to a first public key to obtain a second weight;

[0504] A weight sending module 1313, configured to send the second weight to the first device, where the first device is configured to obtain a second adjustment parameter according to the second weight and the second sample user information;

[0505] A parameter receiving module 1314, configured to receive the second adjustment parameter sent by the first device;

[0506] A parameter sending module 1315, configured to perform decryption processing on the second adjustment parameter according to a first private key corresponding to the first public key to obtain a third adjustment parameter;

[0507] A model adjustment module 1316, configured to send a third adjustment parameter to a first device, where the first device is configured to adjust a fifth feature extraction model according to the third adjustment parameter.

[0508] In another possible implementation, the first device is configured to obtain a fourth adjustment parameter according to a second weight and second sample user information, perform a fusion process on the fourth adjustment parameter and fifth noise features, and obtain a second adjustment parameter.

[0509] The first device is configured to perform a fusion process on a third adjustment parameter and sixth noise features, and adjust the fifth feature extraction model according to the fused adjustment parameter, where the sixth noise features are opposite to the fifth noise features.

[0510] In another possible implementation, as Figure 14 shown, the apparatus further includes:

[0511] A loss value acquisition module 1317, configured to obtain a loss value of a first feature extraction model according to a predicted user label and a sample user label.

[0512] A model training unit 13111 is further configured to stop training the first feature extraction model in response to the loss value being not greater than a preset threshold.

[0513] In another possible implementation, as Figure 14 shown, the apparatus further includes;

[0514] A notification sending module 1318, configured to send a stop training notification to the first device in response to the loss value being not greater than a preset threshold, where the first device is configured to stop training a fifth feature extraction model according to the stop training notification.

[0515] Figure 15 FIG. shows a schematic structural diagram of a terminal 1500 provided by an exemplary embodiment of the present application. The terminal 1500 is configured to execute the steps performed by the terminal in the above user feature acquisition method.

[0516] Generally, the terminal 1500 includes a processor 1501 and a memory 1502.

[0517] The processor 1501 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. The processor 1501 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array). The processor 1501 may also include a main processor and a coprocessor. The main processor is a processor used to process data in the wake state, also known as the CPU (Central Processing Unit); the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 1501 may be integrated with a GPU (Graphics Processing Unit), and the GPU is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 1501 may further include an AI (Artificial Intelligence) processor, and the AI processor is used to process computational operations related to machine learning.

[0518] The memory 1502 may include one or more computer-readable storage media, and the computer-readable storage media may be non-transitory. The memory 1502 may further include high-speed random access memory and non-volatile memory, such as one or more disk storage devices and flash storage devices. In some embodiments, the non-transitory computer-readable storage media in the memory 1502 is used to store at least one instruction, and the at least one instruction is used to be possessed by the processor 1501 to implement the user feature acquisition method provided in the method embodiments of the present application.

[0519] In some embodiments, the terminal 1500 may further optionally include: a peripheral device interface 1503 and at least one peripheral device. The processor 1501, the memory 1502, and the peripheral device interface 1503 may be connected by a bus or signal lines. Each peripheral device may be connected to the peripheral device interface 1503 through a bus, signal lines, or a circuit board. Specifically, the peripheral devices include at least one of a radio frequency circuit 1504, a display screen 1505, a camera assembly 1506, an audio circuit 1507, and a power supply 1509.

[0520] The peripheral device interface 1503 can be used to connect at least one I / O (Input / Output) related peripheral device to the processor 1501 and the memory 1502. In some embodiments, the processor 1501, the memory 1502, and the peripheral device interface 1503 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 1501, the memory 1502, and the peripheral device interface 1503 can be implemented on separate chips or circuit boards, and this embodiment does not limit this.

[0521] The radio frequency circuit 1504 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 1504 communicates with a communication network and other communication devices through electromagnetic signals. The radio frequency circuit 1504 converts an electrical signal into an electromagnetic signal for transmission, or converts the received electromagnetic signal into an electrical signal. Optionally, the radio frequency circuit 1504 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, and so on. The radio frequency circuit 1504 can communicate with other terminals through at least one wireless communication protocol. The wireless communication protocol includes but is not limited to: metropolitan area network, generations of mobile communication networks (2G, 3G, 4G, and 5G), wireless local area network, and / or WiFi (Wireless Fidelity) network. In some embodiments, the radio frequency circuit 1504 may further include a circuit related to NFC (Near Field Communication), and this application does not limit this.

[0522] The display screen 1505 is used to display the UI (User Interface). The UI may include graphics, text, icons, videos, and any combination thereof. When the display screen 1505 is a touch display screen, the display screen 1505 also has the ability to collect touch signals on or above the surface of the display screen 1505. The touch signals can be input to the processor 1501 as control signals for processing. At this time, the display screen 1505 can also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards. In some embodiments, there may be one display screen 1505, which is disposed on the front panel of the terminal 1500; in other embodiments, there may be at least two display screens 1505, which are respectively disposed on different surfaces of the terminal 1500 or are in a folding design; in other embodiments, the display screen 1505 may be a flexible display screen, which is disposed on the curved surface or folding surface of the terminal 1500. Even, the display screen 1505 can also be set to an irregular non-rectangular shape, that is, a special-shaped screen. The display screen 1505 can be prepared using materials such as LCD (Liquid Crystal Display) and OLED (Organic Light-Emitting Diode).

[0523] The camera module 1506 is used to capture images or videos. Optionally, the camera module 1506 includes a front camera and a rear camera. Generally, the front camera is disposed on the front panel of the terminal 1500, and the rear camera is disposed on the back of the terminal 1500. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth camera, a wide-angle camera, and a telephoto camera respectively, so as to realize the function of background blurring by fusing the main camera and the depth camera, panoramic shooting by fusing the main camera and the wide-angle camera, and VR (Virtual Reality) shooting function or other fusion shooting functions. In some embodiments, the camera module 1506 may further include a flash. The flash can be a single-color temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to the combination of a warm light flash and a cold light flash, which can be used for light compensation under different color temperatures.

[0524] The audio circuit 1507 may include a microphone and a speaker. The microphone is used to collect sound waves of the user and the environment, and convert the sound waves into electrical signals for input to the processor 1501 for processing, or input to the radio frequency circuit 1504 to achieve voice communication. For the purpose of stereo collection or noise reduction, there may be multiple microphones, which are respectively arranged at different parts of the terminal 1500. The microphone may also be an array microphone or an omnidirectional collection microphone. The speaker is used to convert the electrical signal from the processor 1501 or the radio frequency circuit 1504 into sound waves. The speaker may be a traditional thin film speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert the electrical signal into sound waves audible to humans, but also convert the electrical signal into sound waves inaudible to humans for uses such as ranging. In some embodiments, the audio circuit 1507 may further include a headphone jack.

[0525] The power supply 1509 is used to supply power to each component in the terminal 1500. The power supply 1509 may be alternating current, direct current, a primary battery or a rechargeable battery. When the power supply 1509 includes a rechargeable battery, the rechargeable battery may support wired charging or wireless charging. The rechargeable battery may also be used to support fast charging technology.

[0526] In some embodiments, the terminal 1500 further includes one or more sensors 1510. The one or more sensors 1510 include but are not limited to: an acceleration sensor 1511, a gyroscope sensor 1512, a pressure sensor 1513, an optical sensor 1515, and a proximity sensor 1516.

[0527] The acceleration sensor 1511 can detect the magnitude of acceleration on the three coordinate axes of the coordinate system established with the terminal 1500. For example, the acceleration sensor 1511 can be used to detect the components of the gravitational acceleration on the three coordinate axes. The processor 1501 can control the display screen 1505 to display the user interface in a landscape view or a portrait view according to the gravitational acceleration signal collected by the acceleration sensor 1511. The acceleration sensor 1511 can also be used to collect motion data of applications or users.

[0528] The gyroscope sensor 1512 can detect the body direction and rotation angle of the terminal 1500. The gyroscope sensor 1512 can cooperate with the acceleration sensor 1511 to collect 3D actions of the user on the terminal 1500. According to the data collected by the gyroscope sensor 1512, the processor 1501 can achieve the following functions: motion sensing (such as changing the UI according to the user's tilt operation), image stabilization during shooting, application control, and inertial navigation.

[0529] The pressure sensor 1513 can be disposed on the side frame of the terminal 1500 and / or the lower layer of the display screen 1505. When the pressure sensor 1513 is disposed on the side frame of the terminal 1500, it can detect the holding signal of the user on the terminal 1500, and the processor 1501 can perform left / right hand recognition or quick operation according to the holding signal collected by the pressure sensor 1513. When the pressure sensor 1513 is disposed on the lower layer of the display screen 1505, the processor 1501 can control the operable controls on the UI interface according to the pressure operation of the user on the display screen 1505. The operable controls include at least one of a button control, a scroll bar control, an icon control, and a menu control.

[0530] The optical sensor 1515 is used to collect the ambient light intensity. In one embodiment, the processor 1501 can control the display brightness of the display screen 1505 according to the ambient light intensity collected by the optical sensor 1515. Specifically, when the ambient light intensity is high, the display brightness of the display screen 1505 is increased; when the ambient light intensity is low, the display brightness of the display screen 1505 is decreased. In another embodiment, the processor 1501 can also dynamically adjust the shooting parameters of the camera module 1506 according to the ambient light intensity collected by the optical sensor 1515.

[0531] The proximity sensor 1516, also known as a distance sensor, is usually disposed on the front panel of the terminal 1500. The proximity sensor 1516 is used to collect the distance between the user and the front of the terminal 1500. In one embodiment, when the proximity sensor 1516 detects that the distance between the user and the front of the terminal 1500 is gradually decreasing, the processor 1501 controls the display screen 1505 to switch from the lit state to the off state; when the proximity sensor 1516 detects that the distance between the user and the front of the terminal 1500 is gradually increasing, the processor 1501 controls the display screen 1505 to switch from the off state to the lit state.

[0532] Those skilled in the art can understand that Figure 15 the structure shown in does not constitute a limitation on the terminal 1500, and may include more or fewer components than shown in the figure, or combine some components, or adopt a different component layout.

[0533] Figure 16It is a schematic structural diagram of a server provided by an embodiment of the present application. The server 1600 may vary greatly due to different configurations or performances, and may include one or more processors (Central Processing Units, CPUs) 1601 and one or more memories 1602. Among them, at least one instruction is stored in the memory 1602, and the at least one instruction is loaded and executed by the processor 1601 to implement the methods provided by the above various method embodiments. Of course, the server may also have components such as wired or wireless network interfaces, keyboards, and input / output interfaces for input / output. The server may also include other components for implementing the functions of the device, which will not be elaborated here.

[0534] The server 1600 can be used to execute the above user feature acquisition method.

[0535] An embodiment of the present application also provides a computer device, which includes a processor and a memory. At least one instruction is stored in the memory, and the at least one instruction is loaded and executed by the processor to implement the user feature acquisition method of the above embodiment.

[0536] An embodiment of the present application also provides a computer-readable storage medium, in which at least one instruction is stored, and the at least one instruction is loaded and executed by the processor to implement the user feature acquisition method of the above embodiment.

[0537] An embodiment of the present application also provides a computer program product, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided in the various optional implementation manners of the above embodiment.

[0538] Those of ordinary skill in the art can understand that all or part of the steps of implementing the above embodiment can be completed by hardware, or can be completed by a program instructing relevant hardware. The program can be stored in a computer-readable storage medium, and the above-mentioned storage medium can be a read-only memory, a disk, an optical disc, etc.

[0539] The above are only optional embodiments of the embodiments of the present application, and are not intended to limit the embodiments of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the embodiments of the present application shall be included in the protection scope of the present application.

Claims

1. A method for obtaining user characteristics, characterized in that, Performed by a second device; the second device includes a first feature extraction model and a fourth feature extraction model, and the fourth feature extraction model is provided by the first device and is a model after being encrypted and processed according to a second public key; the fourth feature extraction model is used to obtain user features matching the first device; the first device includes a second feature extraction model and a fifth feature extraction model, and the second feature extraction model is provided by the second device and is a model after being encrypted and processed according to a first public key; the second feature extraction model is used to obtain user features matching the second device; the first feature extraction model, the second feature extraction model, the fourth feature extraction model, and the fifth feature extraction model are different models for extracting user features; the second device correspondingly stores a target user identifier and first user information, and the first device can store the target user identifier and second user information. The method includes: Invoking the first feature extraction model to perform feature extraction on the first user information to obtain first user features; receiving second user features sent by the first device, where the second user features are obtained by the first device invoking the second feature extraction model to process the second user information; Performing decryption processing on the second user features according to a first private key corresponding to the first public key to obtain first decrypted features, and performing combination processing on the first user features and the first decrypted features to obtain first combined user features, where the first decrypted features are user features after decrypting the second user features; Invoking the fourth feature extraction model to process the first user information to obtain third user features, and sending the third user features to the first device; the first device is used to perform decryption processing on the third user features according to a second private key corresponding to the second public key to obtain second decrypted features, and performing combination processing on fourth user features and the second decrypted features to obtain second combined user features, where the second decrypted features are user features after decrypting the third user features, and the fourth user features are obtained by the first device invoking the fifth feature extraction model to perform feature extraction on the second user information; Receiving the second combined user features sent by the first device; obtaining a user label of the target user identifier according to the first combined user features and the second combined user features.

2. The method according to claim 1, characterized in that, Before receiving the second user features sent by the first device, the method further includes: Performing encryption processing on a third feature extraction model according to the first public key to obtain the second feature extraction model; Sending the second feature extraction model to the first device.

3. The method according to claim 1, characterized in that, Before invoking the fourth feature extraction model to process the first user information to obtain third user features, the method further includes: Receiving the fourth feature extraction model sent by the first device.

4. The method according to claim 3, characterized in that, The first device is used to perform encryption processing on a sixth feature extraction model according to the second public key to obtain the fourth feature extraction model.

5. The method according to claim 1, characterized in that, The second user feature is obtained by the first device through fusing a fifth user feature and a first noise feature, where the fifth user feature is obtained by the first device invoking the second feature extraction model to perform feature extraction on the second user information; The second combined user feature is obtained by the first device through fusing a third combined user feature and a second noise feature, where the third combined user feature is obtained by the first device through combining a fourth user feature and a third user feature, and the first noise feature is opposite to the second noise feature; Obtaining a user label of the target user identifier according to the first combined user feature and the second combined user feature includes: Combining the first combined user feature and the second combined user feature to obtain a fourth combined user feature; Obtaining the user label according to the fourth combined user feature.

6. The method according to claim 1, characterized in that, Invoking the fourth feature extraction model to process the first user information to obtain a third user feature includes: Invoking the fourth feature extraction model to perform feature extraction on the first user information to obtain a sixth user feature; Fusing the sixth user feature and a third noise feature to obtain the third user feature; The method further includes: Combining the first user feature and the second user feature to obtain a fifth combined user feature; Fusing the fifth combined user feature and a fourth noise feature to obtain the first combined user feature, where the third noise feature is opposite to the fourth noise feature.

7. The method according to claim 1, characterized in that, The method further includes: Obtaining first sample user information; Invoking the first feature extraction model to perform feature extraction on the first sample user information to obtain a first sample user feature; Receiving a second sample user feature, where the second sample user feature is obtained by the first device invoking the second feature extraction model to process second sample user information, and the first sample user information and the second sample user information belong to the same sample user identifier; Obtaining a first sample combined user feature according to the first sample user feature and the second sample user feature; Training the first feature extraction model according to the first sample combined user feature and the first sample user information.

8. The method according to claim 7, characterized in that, The method further includes: Invoking the fourth feature extraction model to process the first sample user information to obtain a third sample user feature; Sending the third sample user feature to the first device, where the first device is configured to obtain a second sample combined user feature according to a fourth sample user feature and the third sample user feature, and the fourth sample user feature is obtained by the first device invoking the fifth feature extraction model to perform feature extraction on the second sample user information; Receiving the second sample combined user feature sent by the first device; Training the first feature extraction model according to the first sample combined user feature and the first sample user information includes: Train the first feature extraction model according to the user characteristics of the first sample combination, the user characteristics of the second sample combination, and the first sample user information.

9. The method according to claim 8, wherein The training of the first feature extraction model according to the user characteristics of the first sample combination, the user characteristics of the second sample combination, and the first sample user information includes: Obtain the predicted user label of the sample user identifier according to the user characteristics of the first sample combination and the user characteristics of the second sample combination; Determine the difference between the predicted user label and the sample user label corresponding to the first sample user information as the first weight; Obtain the first adjustment parameter of the first feature extraction model according to the first weight and the first sample user information; Adjust the first feature extraction model according to the first adjustment parameter.

10. A method for obtaining user characteristics, wherein The second device includes a first feature extraction model and a fourth feature extraction model. The fourth feature extraction model is provided by the first device and is a model processed by encryption according to the second public key. The fourth feature extraction model is used to obtain user characteristics matching the first device. The first device includes a second feature extraction model and a fifth feature extraction model. The second feature extraction model is provided by the second device and is a model processed by encryption according to the first public key. The second feature extraction model is used to obtain user characteristics matching the second device. The first feature extraction model, the second feature extraction model, the fourth feature extraction model, and the fifth feature extraction model are different models for extracting user characteristics. The second device correspondingly stores a target user identifier and first user information. The first device can store the target user identifier and second user information. The method includes: The second device calls the first feature extraction model to extract features from the first user information to obtain first user characteristics; The first device calls the second feature extraction model to process the second user information to obtain second user characteristics, and sends the second user characteristics to the second device; The second device receives the second user characteristics; decrypts the second user characteristics according to the first private key corresponding to the first public key to obtain first decrypted characteristics; performs a combination process on the first user characteristics and the first decrypted characteristics to obtain first combined user characteristics. The first decrypted characteristics are the user characteristics after decrypting the second user characteristics; The second device calls the fourth feature extraction model to process the first user information to obtain third user characteristics, and sends the third user characteristics to the first device; The first device decrypts the third user feature according to the second private key corresponding to the second public key to obtain a second decrypted feature, combines the fourth user feature and the second decrypted feature to obtain a second combined user feature, and sends the second combined user feature to the second device. The second decrypted feature is the user feature after decrypting the third user feature, and the fourth user feature is obtained by the first device calling the fifth feature extraction model to extract features from the second user information. The second device receives the second combined user feature sent by the first device, and obtains the user label of the target user identifier according to the first combined user feature and the second combined user feature.

11. A device for obtaining user characteristics, wherein It is set in the second device. The second device includes a first feature extraction model and a fourth feature extraction model. The fourth feature extraction model is provided by the first device and is a model encrypted according to the second public key. The fourth feature extraction model is used to obtain user features matching the first device. The first device includes a second feature extraction model and a fifth feature extraction model. The second feature extraction model is provided by the second device and is a model encrypted according to the first public key. The second feature extraction model is used to obtain user features matching the second device. The first feature extraction model, the second feature extraction model, the fourth feature extraction model, and the fifth feature extraction model are different models for extracting user features. The second device correspondingly stores the target user identifier and the first user information, and the first device can store the target user identifier and the second user information. The device includes: A feature extraction module, configured to call the first feature extraction model to extract features from the first user information to obtain first user features. A feature receiving module, configured to receive the second user features sent by the first device, where the second user features are obtained by the first device calling the second feature extraction model to process the second user information. A combined feature acquisition module, including: A decryption processing unit, configured to decrypt the second user feature according to the first private key corresponding to the first public key to obtain a first decrypted feature, where the first decrypted feature is the user feature after decrypting the second user feature. A first combination processing unit, configured to combine the first user feature and the first decrypted feature to obtain a first combined user feature. An information processing module, configured to call the fourth feature extraction model to process the first user information to obtain third user features. A feature sending module, configured to send the third user feature to the first device; the first device is configured to decrypt the third user feature according to the second private key corresponding to the second public key to obtain a second decrypted feature, and perform a combination process on the fourth user feature and the second decrypted feature to obtain a second combined user feature, where the second decrypted feature is the user feature after decrypting the third user feature, and the fourth user feature is obtained by the first device invoking the fifth feature extraction model to extract features from the second user information; The feature receiving module is configured to receive the second combined user feature sent by the first device; A module for performing the following steps: obtaining a user tag of the target user identifier according to the first combined user feature and the second combined user feature.

12. A computer device, wherein The computer device includes a processor and a memory, and at least one instruction is stored in the memory, and the at least one instruction is loaded and executed by the processor to implement the user feature acquisition method according to any one of claims 1 to 9.

13. A computer-readable storage medium, wherein At least one instruction is stored in the computer-readable storage medium, and the at least one instruction is loaded and executed by a processor to implement the user feature acquisition method according to any one of claims 1 to 9.

14. A computer program product, wherein The computer program product includes computer instructions, the computer instructions are stored in a computer-readable storage medium, a processor of a computer device reads the computer instructions from the computer-readable storage medium, and when the processor executes the computer instructions, the user feature acquisition method according to any one of claims 1 to 9 is implemented.

Images