Data processing method and related device

Abstract

The present application proposes a data processing method and related devices, which include: obtaining first feature data provided by a first business party, the first feature data is obtained by homomorphically encrypting the first access data owned by the first business party; obtaining second feature data provided by a second business party, the second feature data is obtained by homomorphically encrypting the second access data owned by the second business party; performing computational processing on the first feature data and the second feature data in a homomorphic space to obtain sample feature data; using the sample feature data to perform federated learning training on a multi-task model, and the trained multi-task model is used to predict one or more task indicators corresponding to the target advertising data provided by the first business party. Through this application, the accuracy of task indicators can be improved, thereby optimizing the effect of attracting new customers through advertising.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a data processing method, a data processing device, a computer equipment, a computer-readable storage medium, and a computer program product. Background Art

[0002] In daily life, advertising acquisition is a hot topic of research in advertising recommendation scenarios. The so-called advertising acquisition refers to a method in which the first business party (such as an advertiser) attracts new users for the second business party (such as an advertising platform) by delivering advertising data, in order to maintain a continuous growth in the number of users visiting the advertising platform.

[0003] At present, the common way to optimize the effect of attracting new users through advertising is that advertisers use their own data (such as user's gender, age and other attribute data) to connect to the advertising platform, and the advertising platform uses the data provided by the advertiser to estimate the task indicators of the advertising data (such as retention rate indicators, which can be used to reflect the ability of the advertising platform to retain users), thereby optimizing the effect of attracting new users through advertising. However, due to data privacy considerations, advertisers will not connect complete user data to the advertising platform, resulting in inaccurate estimation of the task indicators of the advertising data and poor new user acquisition effect. Therefore, how to improve the accuracy of the predicted task indicators is an important issue that needs to be solved urgently. Summary of the invention

[0004] The embodiments of the present application propose a data processing method, apparatus, computer equipment, computer-readable storage medium, and computer program product, which can improve the accuracy of task indicators and thus optimize the effect of attracting new customers through advertising.

[0005] On the one hand, an embodiment of the present application provides a data processing method, the method comprising:

[0006] Acquire first characteristic data provided by the first business party, where the first characteristic data is obtained by homomorphically encrypting first access data owned by the first business party, and the first access data includes data generated when the advertisement data provided by the first business party is accessed within a preset time period;

[0007] Obtaining second characteristic data provided by the second business party, where the second characteristic data is obtained by homomorphically encrypting second access data owned by the second business party, and the second access data includes data generated when the advertisement data is accessed within the preset time period during the process of exposure of the advertisement data on the platform corresponding to the second business party;

[0008] Performing computation on the first feature data and the second feature data in the homomorphic space to obtain sample feature data;

[0009] The multi-task model is trained through federated learning using sample feature data, and the trained multi-task model is used to predict one or more task indicators corresponding to the target advertising data provided by the first business party.

[0010] On the one hand, an embodiment of the present application provides a data processing device, the device comprising:

[0011] an acquisition unit, configured to acquire first characteristic data provided by the first business party, the first characteristic data being obtained by homomorphically encrypting first access data owned by the first business party, the first access data including data generated when the advertisement data provided by the first business party is accessed within a preset time period;

[0012] The acquisition unit is further used to acquire second characteristic data provided by the second business party, where the second characteristic data is obtained after homomorphic encryption processing is performed on second access data owned by the second business party, and the second access data includes data generated when the advertisement data is accessed within the preset time period during the exposure of the advertisement data on the platform corresponding to the second business party;

[0013] A processing unit, used for performing operation processing on the first feature data and the second feature data in the homomorphic space to obtain sample feature data;

[0014] The processing unit is also used to perform federated learning training on the multi-task model using sample feature data, and the trained multi-task model is used to predict one or more task indicators corresponding to the target advertising data provided by the first business party.

[0015] In a possible implementation, the first access data includes attribute data of m objects that access the advertisement data provided by the first business party within a preset time period; the second access data includes browsing data of n objects that access the advertisement data within a preset time period during exposure of the advertisement data on a platform corresponding to the second business party; m and n are both positive integers; and the processing unit is further configured to perform the following operations:

[0016] Performing a federated encryption intersection process on the first access data and the second access data to obtain a federated encryption intersection process result;

[0017] Among them, the federated encrypted intersection processing result is used to indicate that the first business party needs to filter out the attribute data of p objects from the first access data, and generate first feature data based on the attribute data of the p objects; and the federated encrypted intersection processing result is also used to indicate that the second business party needs to filter out the browsing data of p objects from the second access data, and generate second feature data based on the browsing data of the p objects; p is a positive integer, p≤m and p≤n.

[0018] In a possible implementation, the processing unit performs a federated encryption intersection process on the first access data and the second access data to obtain a federated encryption intersection process result, which is used to perform the following operations:

[0019] Encrypting the attribute data of m objects to obtain encrypted m attribute data;

[0020] Encrypting the browsing data of n objects to obtain n encrypted browsing data;

[0021] According to the encrypted m attribute data and the encrypted n browsing data, an intersection process is performed on the m attribute data and the n browsing data in the encryption space to obtain a federated encryption intersection process result.

[0022] In a possible implementation manner, the process of the first service party generating the first feature data includes:

[0023] Perform feature extraction processing on the attribute data of the p objects respectively to obtain the attribute features of each object in the p objects;

[0024] Based on the attribute features of the p objects, determining first feature data; and,

[0025] The process of the second business party generating the second characteristic data includes:

[0026] Perform feature extraction processing on the browsing data of p objects respectively to obtain browsing features of each object in the p objects;

[0027] Based on the browsing features of the p objects, second feature data is determined.

[0028] In a possible implementation, the processing unit performs computation on the first feature data and the second feature data in the homomorphic space to obtain sample feature data for performing the following operations:

[0029] Performing homomorphic encryption operation on the first feature data and the second feature data in the homomorphic space to obtain third feature data after the homomorphic encryption operation;

[0030] The third characteristic data is homomorphically decrypted in the homomorphic space to obtain sample characteristic data.

[0031] In a possible implementation, the task indicator includes a retention rate indicator; the sample feature data also includes k retention rate labels of each object in p objects, where p is a positive integer; the processing unit uses the sample feature data to perform federated learning training on the multi-task model to perform the following operations:

[0032] Input the sample feature data into the fully connected layer of the multi-task model and output k predicted retention rates for each of the p objects;

[0033] Based on the k predicted retention rates for each object and the k retention rate labels for each object, the model parameters of the multi-task model are adjusted.

[0034] In a possible implementation, the k predicted retention rates of each object include a first predicted retention rate and a second predicted retention rate; the k retention rate labels of each object include a first label and a second label;

[0035] The processing unit adjusts the model parameters of the multi-task model based on the k predicted retention rates of each object and the k retention rate labels of each object to perform the following operations:

[0036] Calculate the first loss of the multi-task model based on the first predicted retention rate of each object and the first label of each object;

[0037] Calculating a second loss of the multi-task model based on a second predicted retention rate of each object and a second label of each object;

[0038] A target loss is determined according to the first loss and the second loss, and a model parameter of the multi-task model is adjusted based on the target loss.

[0039] In a possible implementation, the processing unit is further configured to perform the following operations:

[0040] Test the trained multi-task model to obtain the model evaluation index value of the trained multi-task model;

[0041] If the model evaluation index value of the trained multi-task model meets the model evaluation index condition, the trained multi-task model is called to predict one or more task indicators corresponding to the target advertising data provided by the first business party.

[0042] In a possible implementation, the processing unit is further configured to perform the following operations:

[0043] Call the trained multi-task model to perform prediction processing on the target object accessing the target advertising data, and obtain k predicted retention rates of the target object;

[0044] The value of the target advertising data pushed to the target object is determined based on k predicted retention rates of the target object.

[0045] In a possible implementation, the task indicator also includes an active duration indicator; and the processing unit is further configured to perform the following operations:

[0046] Call the trained multi-task model to perform prediction processing on the set of objects that access the target advertising data within the target time period, and obtain k predicted retention rates and active durations for each object in the object set;

[0047] During the target time period, additional advertising data is pushed to target objects that meet the retention rate threshold and active duration conditions.

[0048] On the one hand, an embodiment of the present application provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the above-mentioned data processing method.

[0049] On the one hand, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is read and executed by a processor of a computer device, the computer device executes the above-mentioned data processing method.

[0050] On the one hand, an embodiment of the present application provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the above-mentioned data processing method.

[0051] In an embodiment of the present application, first, the first characteristic data provided by the first business party can be obtained. The first characteristic data is obtained after homomorphic encryption processing of the first access data owned by the first business party, and the first access data includes data generated by accessing the advertising data provided by the first business party within a preset time period. Secondly, the second characteristic data provided by the second business party can also be obtained. The second characteristic data is obtained after homomorphic encryption processing of the second access data owned by the second business party, and the second access data includes data generated by accessing the advertising data within a preset time period during the exposure of the advertising data on the platform corresponding to the second business party. Then, the first characteristic data and the second characteristic data are operated and processed in the homomorphic space to obtain sample characteristic data. Finally, the sample characteristic data is used to perform federated learning training on the multi-task model, and the trained multi-task model is used to predict one or more task indicators corresponding to the target advertising data provided by the first business party. It can be seen that in the embodiment of the present application, both the first feature data provided by the first business party and the second feature data provided by the second business party can be obtained. Therefore, the training samples of the multi-task model are richer, which can improve the accuracy of the multi-task model; further, the determined sample feature data is a data interaction carried out in the homomorphic encryption space, and therefore the original data of both the first business party and the second business party will not be leaked, thereby ensuring the security of the original data of both parties; in addition, the trained multi-task model can predict multiple task indicators, thereby improving the accuracy of the predicted task indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 It is a schematic diagram of the principle of a data processing solution provided in an embodiment of the present application;

[0054] Figure 2 It is a structural diagram of a data processing system provided in an embodiment of the present application;

[0055] Figure 3 It is a flowchart of a data processing method provided in an embodiment of the present application;

[0056] Figure 4 It is a flowchart of another data processing method provided in an embodiment of the present application;

[0057] Figure 5 It is a flowchart of a training multi-task model provided in an embodiment of the application;

[0058] Figure 6 It is a schematic diagram of a feature extraction process provided by an embodiment of the present application;

[0059] Figure 7 It is a schematic diagram of an application scenario of a multi-task model provided in an embodiment of the present application;

[0060] Figure 8 is a structural schematic diagram of a data processing device provided in an embodiment of the present application;

[0061] Fig. 9 It is a structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0062] Here, exemplary embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application.

[0063] The embodiment of the present application proposes a data processing solution, which can improve the accuracy of the task indicators corresponding to the predicted target advertising data. The solution can be applied to various recommendation scenarios such as advertising recommendation and public account recommendation. Based on the encrypted first feature data provided by the first business party and the encrypted second feature data provided by the second business party, a more accurate multi-task model can be trained to optimize the new customer acquisition effect of the target advertising data. Figure 1 , Figure 1 is a schematic diagram of a data processing solution provided in an embodiment of the present application, such as Figure 1 As shown, the general principle of the data processing scheme is as follows: First, obtain the first feature data provided by the first business party (such as an advertiser), wherein the first business party has the first access data (which may include the data generated by the advertising data provided by the first business party being accessed within a preset time period, such as the user's nickname, avatar, gender, age and other data), and the first business party can obtain the first feature data after homomorphic encryption of the first access data; secondly, obtain the second feature data provided by the second business party (such as an advertising platform), wherein the second business party has the second access data (which may include the data generated by the advertising data being accessed within a preset time period during the exposure of the advertising data on the platform corresponding to the second business party, such as the user's interests, hobbies, browsing behavior and other data), and the second business party can obtain the first feature data after homomorphic encryption of the second access data. Next, the first feature data and the second feature data can also be processed in the homomorphic space to obtain sample feature data. Finally, the sample feature data can be used to perform federated learning training on the multi-task model. In one possible implementation, the federated learning training of the multi-task model using sample feature data can be performed by the first business party, that is, Figure 1 The advertiser shown in the figure performs the federated learning training of the multi-task model; in another possible implementation, the federated learning training of the multi-task model using sample feature data can also be performed by the second business party, that is, Figure 1 The advertising platform shown in the figure performs federated learning training on the multi-task model; in another possible implementation, a third-party federated learning platform may use sample feature data to perform federated learning training on the multi-task model, which is not specifically limited in the embodiments of the present application. It is understandable that both the advertiser and the advertising platform can call the trained multi-task model for model prediction. For example, the advertiser can call the trained multi-task model to predict one or more task indicators (such as retention rate indicators, active duration indicators, etc.) corresponding to the target advertising data provided by the advertiser.

[0064] It can be seen that in the embodiment of the present application, both the first feature data provided by the first business party and the second feature data provided by the second business party can be obtained, so the training samples of the multi-task model are richer, which can improve the accuracy of the multi-task model; further, the determined sample feature data is a data interaction carried out in the homomorphic encryption space, and therefore the original data of both the first business party and the second business party will not be leaked, thereby ensuring the security of the original data of both parties; in addition, the trained multi-task model can predict multiple task indicators, thereby improving the accuracy of the predicted task indicators.

[0065] Next, the above-mentioned data processing solution is introduced in combination with the technical terms involved in this application:

[0066] 1. Artificial Intelligence

[0067] 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 respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making. Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level technology and software-level technology. Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0068] Machine Learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0069] In one possible implementation, the data processing scheme of the present application can be combined with the machine learning technology in the field of artificial intelligence. Specifically, the data processing scheme of the present application can be combined with the federated learning technology in the machine learning technology. For example, federated learning technology can be used to train multi-task models, etc. Among them, the so-called federated learning (Federated machine learning / Federated Learning), also known as federated machine learning, federated learning, and alliance learning, is a machine learning framework that can effectively help multiple institutions to use data and machine learning modeling while meeting the requirements of user privacy protection, data security and government regulations. According to the distribution characteristics of data, federated learning can be divided into three categories: horizontal federated learning, vertical federated learning, and federated transfer learning. As a distributed machine learning paradigm, federated learning can effectively solve the "data island" problem, allowing participants to jointly model without sharing data, thereby technically breaking the "data island" and realizing AI collaboration.

[0070] 2. Blockchain:

[0071] The blockchain network is a network composed of a peer-to-peer network (P2P network) and a blockchain. The blockchain refers to a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithm. It is essentially a decentralized database, a string of data blocks (or blocks) generated by cryptographic methods.

[0072] In a possible implementation, the data processing scheme of the present application can be combined with blockchain technology. For example, the first characteristic data provided by the first business party, the second characteristic data provided by the second business party, and the sample characteristic data can be uploaded to the blockchain of the blockchain network for storage to prevent the internal data of the computer device from being tampered with, thereby improving the security and privacy of the log data.

[0073] 3. Cloud Technology:

[0074] Cloud computing refers to the delivery and use model of IT infrastructure, which means obtaining required resources through the network in an on-demand and easily scalable manner; in a broad sense, cloud computing refers to the delivery and use model of services, which means obtaining required services through the network in an on-demand and easily scalable manner. This service can be IT and software, Internet-related, or other services. Cloud computing is the product of the development and integration of traditional computer and network technologies such as grid computing, distributed computing, parallel computing, utility computing, network storage technologies, virtualization, and load balancing.

[0075] In one possible implementation, when executing the data processing scheme of the present application, the first feature data and the second feature data are processed in the homomorphic space to obtain sample feature data. This process involves large-scale calculations and requires large computing power and storage space. Therefore, in one possible implementation of the present application, a computer device can obtain sufficient computing power and storage space through cloud computing technology to execute the steps of calculating sample feature data involved in the present application.

[0076] It is understandable that the data processing scheme provided by the embodiment of the present application can be applied to various recommendation scenarios such as advertising recommendation and public account recommendation. For example, taking the advertising recommendation scenario as an example, the data processing scheme provided by the embodiment of the present application can be used, with the advertiser as the first business party and the advertising platform (such as a media recommendation platform) as the second business party, and based on the first feature data provided by the advertiser and the second feature data provided by the advertising platform, a multi-task model is jointly trained, and the trained multi-task model can be used to predict one or more task indicators (such as retention rate indicators, activity indicators, etc.) corresponding to the target advertising data provided by the advertiser. For advertisers, the target advertising data can be estimated based on the predicted task indicators; for advertising platforms, the data of both advertisers and advertising platforms can be fully utilized to recommend advertising data, thereby optimizing the recommendation effect of advertising data; for users, they can see advertising data that is more in line with their interests, improve user experience, and is beneficial to user activity and retention.

[0077] It is particularly important to note that in the subsequent specific implementation methods of this application, related data such as user information is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0078] Based on the above description, next, see Figure 2 , Figure 2 Schematic diagram of a data processing system provided in an embodiment of the present application. Figure 2 As shown, the structural diagram of the data processing system may include: a server 240 and a terminal device cluster, wherein the terminal device cluster may include: a terminal device 210, a terminal device 220, ..., a terminal device 230, etc. Any terminal device in the terminal device cluster may be directly or indirectly connected to the server 240 via a wired or wireless communication, and the present application does not limit this; in addition, the types of any two terminal devices in the terminal device cluster may be the same or different, and the present application does not limit this either.

[0079] Figure 2 The server 240 shown can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be 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), as well as big data and artificial intelligence platforms, etc.

[0080] Figure 2 The terminal device 210, terminal device 220, and terminal device 230 shown may include, but are not limited to: mobile phones, tablet computers, laptop computers, PDAs, mobile Internet devices (MIDs), intelligent voice interaction devices, vehicle-mounted terminals, roadside devices, aircraft, wearable devices, smart home appliances, or wearable devices with data processing functions such as smart watches, smart bracelets, pedometers, etc.

[0081] In a possible implementation, when the terminal device 210, the terminal device 220 and the server 240 jointly execute the data processing scheme in the present application, the terminal device 210 can be used as a terminal device used by the first business party, the terminal device 220 can be used as a terminal device used by the second business party, and the server 240 can be used as a third-party federated learning platform. The server 240 can obtain the first feature data provided by the terminal device 210, wherein the first feature data is obtained after the terminal device 210 performs homomorphic encryption processing on the first access data it owns, and the first access data includes the data generated by the advertisement data provided by the terminal device 210 being accessed within a preset time period. The server 240 can also obtain the second feature data provided by the terminal device 220, wherein the second feature data is obtained after the terminal device 220 performs homomorphic encryption processing on the second access data it owns, and the second access data includes the data generated by the advertisement data provided by the terminal device 220 being accessed within a preset time period. Next, the server 240 can perform computational processing on the first feature data and the second feature data in the homomorphic space to obtain sample feature data; finally, the server 240 can use the sample feature data to perform federated learning training on the multi-task model. The trained multi-task model may be used to predict one or more task indicators corresponding to the target advertisement data provided by the first business party.

[0082] Subsequently, the server 240 may send the trained multi-task model to the terminal device 210 and the terminal device 220. Then, the terminal device 210 and the terminal device 220 may respectively call the trained multi-task model to perform prediction operations. For example, the terminal device 210 may call the trained multi-task model to predict the retention rate index corresponding to the target advertising data provided by the first business party; for another example, the terminal device 220 may call the trained multi-task model to predict the activity index corresponding to the target advertising data provided by the first business party, and so on.

[0083] It should be understood that the above is only an exemplary description of the various steps that the terminal device 210, the terminal device 220 and the server 240 are responsible for performing, and the embodiments of the present application do not limit this. In another possible implementation, for example, the first feature data provided by the terminal device 210 is obtained, and the second feature data provided by the terminal device 220 is obtained; the above process can also be performed by the terminal device 210 or the terminal device 220, and then the terminal device 210 or the terminal device 220 sends the first feature data and the second feature data to the server 240, and then the server 240 performs subsequent operations. For another example, the first feature data provided by the terminal device 210 is obtained, and the second feature data provided by the terminal device 220 is obtained; the first feature data and the second feature data are processed in the homomorphic space to obtain sample feature data; these processes can be performed by the terminal device 210 or the terminal device 220, and then the sample feature data is sent to the server 240, and the server 240 performs the steps of using the sample feature data to perform federated learning training on the multi-task model. For example, first feature data provided by terminal device 210 is obtained, and second feature data provided by terminal device 220 is obtained; the first feature data and the second feature data are processed in a homomorphic space to obtain sample feature data; the sample feature data is used to perform federated learning training on a multi-task model; these processes can be executed by terminal device 210 or terminal device 220, and so on.

[0084] Furthermore, we can Figure 2 The provided data processing system is deployed on the nodes of the blockchain. For example, the terminal device 210, the terminal device 220 and the server 240 can all be regarded as node devices of the blockchain to form a blockchain network. Therefore, the data processing process involved in this application can be executed on the blockchain, which can not only ensure the fairness and justice of the data processing process, but also make the data processing process traceable, thereby improving the security of the data processing process.

[0085] It can be understood that the system architecture diagram described in the embodiment of the present application is for more clearly illustrating the technical solution of the embodiment of the present application, and does not constitute a limitation on the technical solution provided in the embodiment of the present application. A person of ordinary skill in the art can know that with the evolution of the system architecture and the emergence of new business scenarios, the technical solution provided in the embodiment of the present application is also applicable to similar technical problems.

[0086] Based on the above description of the data processing scheme, the present application embodiment proposes a data processing method. Figure 3 As shown, Figure 3This is a flow chart of a data processing method provided in an embodiment of the present application. The data processing method can be executed by a terminal device or a server, or can be executed by the terminal device and the server mentioned above. For the convenience of explanation, the following description will be given by taking a computer device executing the data processing method as an example. Among them, the data processing method may include the following steps S301 to S304:

[0087] S301: Acquire first characteristic data provided by the first business party, where the first characteristic data is obtained after homomorphic encryption processing is performed on first access data owned by the first business party. The first access data includes data generated when the advertisement data provided by the first business party is accessed within a preset time period.

[0088] In an embodiment of the present application, the first business party may refer to an advertiser who produces advertising data, and the first access data may include data generated when the advertising data provided by the advertiser is accessed by an object (such as a user) within a preset time period (for example, within a month). In a possible implementation, the first access data may include attribute data of m objects that access the advertising data provided by the first business party within a preset time period, where m is a positive integer. Among them, the attribute data of each object may include, but is not limited to: the object's identity document (Id), nickname, avatar, gender, age and other basic attributes. In this embodiment, users who access advertising data all permit or agree to have their user information obtained, and the user information includes, but is not limited to, the user's identity, nickname, avatar, gender, age, etc., and the collection, use and processing of relevant information need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0089] In another possible implementation, the first access data may also include a retention rate label of each of the m objects. The retention rate may include, but is not limited to: the next-day retention rate (referring to the probability that the target object that accessed the advertising data provided by the advertiser on the i-th day still accesses the advertising data provided by the advertiser on the i+1th day), the two-day retention rate (referring to the probability that the target object that accessed the advertising data provided by the advertiser on the i-th day still accesses the advertising data provided by the advertiser on the i+2th day), the three-day retention rate refers to the probability that the target object that accessed the advertising data provided by the advertiser on the i-th day still accesses the advertising data provided by the advertiser on the i+3rd day, and so on. Where i is a positive integer. Then, the retention rate label of each object may be 1 or 0.

[0090] In addition, the so-called preset time period may refer to a historical time period, that is, a time period corresponding to a period of time from the current moment, for example, the preset time period may be 24 hours, 48 ​​hours, etc. It is understandable that the preset time period may be customized according to scene requirements or user requirements, and the embodiments of the present application do not specifically limit this. For example, the first access data owned by the advertiser may include: attribute data of 100 users who accessed the advertising data provided by the advertiser in the past 24 hours.

[0091] S302: Acquire second characteristic data provided by the second business party, where the second characteristic data is obtained by homomorphically encrypting second access data owned by the second business party. The second access data includes data generated when the advertising data is accessed within a preset time period during exposure of the advertising data on the platform corresponding to the second business party.

[0092] In the embodiment of the present application, the second business party may refer to an advertising platform that exposes advertising data. It should be noted that the first business party (e.g., an advertiser) may be responsible for customizing or producing advertising data, and then the advertiser sends the produced advertising data to the advertising platform through a real-time application programming interface (RTA), and the advertising platform may expose the advertising data provided by the advertiser, so that users can browse the advertising data provided by the advertiser by accessing the advertising platform.

[0093] Among them, the second access data may include data generated by the advertising data being accessed within a preset time period during the exposure of the advertising data on the platform (advertising platform) corresponding to the second business party. In a possible implementation, the second access data may include browsing data of n objects that accessed the advertising data within a preset time period during the exposure of the advertising data on the platform corresponding to the second business party, where n is a positive integer. Among them, the browsing data of each object may include, but is not limited to: the object's browsing behavior, click behavior, negative feedback behavior, interests and hobbies and other portrait data. For example, the second access data owned by the advertising platform may include: browsing data of 120 objects that accessed the advertising data in the past 24 hours during the exposure of the advertising data on the advertising platform.

[0094] It should be noted that the order of obtaining the first characteristic data and obtaining the second characteristic data is not specifically limited in the embodiments of the present application. Specifically, the computer device may first execute the step of "obtaining the first characteristic data provided by the first business party" shown in S301, and then execute the step of "obtaining the second characteristic data provided by the second business party" shown in S302; it may also first execute the step of "obtaining the second characteristic data provided by the second business party" shown in S302, and then execute the step of "obtaining the first characteristic data provided by the first business party" shown in S301; it may also simultaneously execute the steps of "obtaining the first characteristic data provided by the first business party" shown in S301 and "obtaining the second characteristic data provided by the second business party" shown in S301, and the embodiments of the present application do not specifically limit this.

[0095] S303: Performing calculations on the first feature data and the second feature data in the homomorphic space to obtain sample feature data.

[0096] As can be seen from the foregoing, the first feature data and the second feature data are both data obtained through homomorphic encryption processing, wherein the principle of homomorphic encryption includes: processing the homomorphically encrypted data to obtain an output, decrypting the output, and the result is the same as the output result obtained by processing the unencrypted original data in the same way. In an embodiment of the present application, the operation processing performed on the homomorphically encrypted first feature data and the homomorphically encrypted second feature data in the homomorphic space may include but is not limited to: additive homomorphic operation processing, multiplicative homomorphic operation processing, and the like. It should be noted that homomorphic encryption can be divided into semi-homomorphic encryption and fully homomorphic encryption. If a cryptographic algorithm only satisfies multiplicative homomorphism or additive homomorphism, it is called semi-homomorphic encryption (SHE (Somewhat Homomorphic Encryption) or PHE (Partially Homomorphic Encryption)); and if a cryptographic algorithm satisfies both multiplicative homomorphism and additive homomorphism, it is called fully homomorphic encryption (FHE, Fully Homomorphic Encryption). In an embodiment of the present application, semi-homomorphic encryption technology that satisfies the additive property can be used to perform privacy operations, such as the Paillier homomorphic encryption scheme, the Okamoto-Uchiyama homomorphic encryption scheme, and the like.

[0097] In one possible implementation, a computer device performs homomorphic encryption operations on first feature data and second feature data in a homomorphic space to obtain third feature data after the homomorphic encryption operations; then, the computer device performs homomorphic decryption processing on the third feature data in the homomorphic space to obtain sample feature data.

[0098] For example, in the embodiment of the present application, the first service provider and the second service provider may jointly train a multi-task learning model. In the process of the first service provider and the second service provider jointly training the multi-task learning model, for example Figure 1 As shown, the first business party (such as an advertiser) can be responsible for the guest side model of the multi-task learning model, and the second business party (such as an advertising platform) can be responsible for the host side model of the multi-task learning model. Among them, the first feature data provided by the first business party can be expressed as: ENCRYPT (guest_embedding), and the second feature data provided by the second business party can be expressed as: ENCRYPT (host_embedding). Then, the first feature data (ENCRYPT (guest_embedding)) and the second feature data (ENCRYPT (host_embedding)) are processed in the homomorphic space to obtain the third feature data after homomorphic encryption operation: ENCRYPT (host_embedding) + ENCRYPT (guest_embedding). Finally, the third feature data is homomorphically decrypted in the homomorphic space to obtain sample feature data: DECRYPT (ENCRYPT (host_embedding) + ENCRYPT (guest_embedding)) = host_emb edding + guest_embedding.

[0099] Through the above method, the first feature data and the second feature data are processed in the homomorphic space, so that the original data owned by the first business party and the original data owned by the second business party are not directly obtained, and the calculation is only performed in the homomorphic encryption space, thereby ensuring the privacy and security of the original data owned by the first business party and the original data owned by the second business party.

[0100] S304: Perform federated learning training on the multi-task model using sample feature data, and the trained multi-task model is used to predict one or more task indicators corresponding to the target advertising data provided by the first business party.

[0101] In an embodiment of the present application, a specific scenario in which a computer device uses sample feature data to perform federated learning training on a multi-task model may include: in one possible implementation, a first business party (e.g., an advertiser) uses sample feature data to perform federated learning training on a multi-task model; in another possible implementation, a second business party uses sample feature data to perform federated learning training on a multi-task model; in yet another possible implementation, the first business party sends the first feature data to a third-party federated learning platform, and the second business party sends the second feature data to a third-party federated learning platform, and the third-party federated learning platform uses the sample feature data to perform federated learning training on the multi-task model. The embodiments of the present application do not specifically limit this.

[0102] It is understandable that the so-called multi-task model means that during the training process of the model, the task indicators of the model can be one or more. Then, the trained multi-task model can be used to predict one or more task indicators. In one possible implementation, the training of the multi-task model in the embodiment of the present application can be based on multiple retention rates as task indicators for model training. For example, the task indicators of the multi-task model trained by the sample feature data in the embodiment of the present application may include but are not limited to: next-day retention rate, two-day retention rate, three-day retention rate and other task indicators. In another possible implementation, the task indicators of the training of the multi-task model in the embodiment of the present application may also include but are not limited to: exposure, active time, number of new objects and other multiple task indicators, which can be customized according to user needs or scenario needs. The embodiment of the present application does not specifically limit the type and number of task indicators of the multi-task model. Subsequently, both the first business party and the second business party can call the trained multi-task model for model prediction. It is understandable that if the task indicators set in the training process of the multi-task model are N, the trained multi-task model can be used to predict N corresponding task indicators, and N is a positive integer. For example, if the task indicators in the training process of the multi-task model are: next-day retention rate, two-day retention rate, and three-day retention rate, then the trained multi-task model can be used to predict the next-day retention rate, two-day retention rate, and three-day retention rate corresponding to the target object.

[0103] In an embodiment of the present application, first, the first characteristic data provided by the first business party can be obtained. The first characteristic data is obtained after homomorphic encryption processing of the first access data owned by the first business party, and the first access data includes data generated by accessing the advertising data provided by the first business party within a preset time period. Secondly, the second characteristic data provided by the second business party can also be obtained. The second characteristic data is obtained after homomorphic encryption processing of the second access data owned by the second business party, and the second access data includes data generated by accessing the advertising data within a preset time period during the exposure of the advertising data on the platform corresponding to the second business party. Then, the first characteristic data and the second characteristic data are operated and processed in the homomorphic space to obtain sample characteristic data. Finally, the sample characteristic data is used to perform federated learning training on the multi-task model, and the trained multi-task model is used to predict one or more task indicators corresponding to the target advertising data provided by the first business party. It can be seen that in the embodiment of the present application, both the first feature data provided by the first business party and the second feature data provided by the second business party can be obtained. Therefore, the training samples of the multi-task model are richer, which can improve the accuracy of the multi-task model; further, the determined sample feature data is a data interaction carried out in the homomorphic encryption space, and therefore the original data of both the first business party and the second business party will not be leaked, thereby ensuring the security of the original data of both parties; in addition, the trained multi-task model can predict multiple task indicators, thereby improving the accuracy of the predicted task indicators.

[0104] based on Figure 3 For a description of the embodiments, see Figure 4 , Figure 4 4 is a flow chart of another data processing method provided in an embodiment of the present application, and the data processing method can be executed by a computer device. The data processing method can include the following steps S401 to S404:

[0105] S401: Perform a federated encryption intersection process on the first access data and the second access data to obtain a federated encryption intersection process result.

[0106] The first access data may include the attribute data of m objects that access the advertising data provided by the first business party within a preset time period; the second access data may include the browsing data of n objects that access the advertising data within a preset time period during the exposure of the advertising data on the platform corresponding to the second business party. In a possible implementation, the computer device performs a federal encryption intersection process on the first access data and the second access data to obtain a federal encryption intersection result. The federal encryption intersection result is used to indicate that the first business party needs to filter out the attribute data of p objects from the first access data, and generate first feature data based on the attribute data of the p objects; and the federal encryption intersection result is also used to indicate that the second business party needs to filter out the browsing data of p objects from the second access data, and generate second feature data based on the browsing data of the p objects; p is a positive integer, p≤m and p≤n.

[0107] Specifically, since it is necessary to protect the privacy and security of the data of both the first business party and the second business party (i.e., the first access data and the second access data), the computer device can perform a federal encryption intersection process on the first access data and the second access data. For example, the computer device can use the privacy protection set intersection (Private Set Intersection, PSI) technology to perform a federal encryption intersection process, so as to obtain the intersection of the first access data owned by the first business party and the second access data owned by the second business party. It can be understood that the federated learning training adopted in the embodiment of the present application can be a vertical federated learning training, so the data intersection here can refer to an object-based intersection process, that is, the computer device can determine the common objects in the first access data and the second access data based on the PSI privacy intersection technology.

[0108] In a possible implementation, the computer device performs a federated encryption intersection process on the first access data and the second access data to obtain a federated encryption intersection process result, which may include: the computer device may encrypt the attribute data of m objects to obtain the encrypted m attribute data; and the computer device may encrypt the browsing data of n objects to obtain the encrypted n browsing data. Next, the computer device performs an intersection process on the m attribute data and the n browsing data in the encryption space according to the encrypted m attribute data and the encrypted n browsing data to obtain a federated encryption intersection process result. For example, the computer device may use the RSA asymmetric encryption algorithm to encrypt the attribute data of m objects, and use the RSA asymmetric encryption algorithm to encrypt the browsing data of n objects. It can be understood that in the embodiments of the present application, the encryption algorithm is not specifically limited. In other feasible embodiments, the Elgamal algorithm (an asymmetric encryption algorithm), the Rabin algorithm (an asymmetric encryption algorithm), the Diddie-Hellman algorithm (an asymmetric encryption algorithm), the ECC algorithm (elliptic curve encryption algorithm), and the like encryption algorithms may also be used. In this way, by performing intersection processing on the m attribute data of the first business party and the n browsing data of the second business party in the encrypted space, the privacy security of the data of both parties can be guaranteed.

[0109] See also Figure 5 , Figure 5 1 is a flowchart of a multi-task model training method provided by the application embodiment. Figure 5 As shown, the first business party can filter out the attribute data (such as ID, nickname, avatar, gender, age, etc.) of p objects from the first access data according to the result of the federated encryption intersection processing, and generate the first feature data based on the attribute data of the p objects. Specifically, the process of the first business party generating the first feature data may include: first, the computer device performs feature extraction processing on the attribute data of the p objects respectively to obtain the attribute features of each of the p objects; then, the computer device determines the first feature data based on the attribute features of the p objects. Among them, the feature extraction processing may include word embedding processing, and the word embedding processing may specifically include embedding processing. The so-called embedding processing can map large sparse vectors to low-dimensional space, thereby solving the feature representation problem of sparse features. That is, more effective features (i.e., attribute features) can be extracted from the attribute data of the object based on the embedding processing.

[0110] For example, in the embodiment of the present application, an artificial intelligence model can be used to extract features from the attribute data of each object. Figure 6 , Figure 6Schematic diagram of a feature extraction process provided by an embodiment of the present application. Figure 6 As shown, in the scenario of feature extraction of attribute data, the input data (x1, x2, x3) can be the attribute data of the object, for example, the nickname of the object "Love China". After obtaining the nickname of the object, the computer device preprocesses the nickname of the object (for example, word segmentation), inputs the preprocessed attribute data into the artificial intelligence model of the encoding and decoding structure, and uses the artificial intelligence model of the encoding and decoding structure to extract key features of the attribute data of the object, thereby obtaining the attribute features of the object, such as Figure 6 The output attribute features (y1, y2, y3) are shown in .

[0111] In addition, in a possible implementation, the computer device may perform a weighted average operation on the attribute features of the p objects to obtain the attribute features of the p objects after the weighted average, and then perform homomorphic encryption processing on the attribute features of the p objects after the weighted average, thereby obtaining the first feature data. For example, in the embodiment of the present application, the first business party and the second business party may jointly train a multi-task learning model, such as Figure 1 As shown, in the process of the first business party and the second business party jointly training the multi-task learning model, the first business party (such as an advertiser) can be responsible for the guest side model of the multi-task learning model. Then, the attribute features of the p objects after weighted average obtained by the first business party can be expressed as: guest_embedding, then, the first feature data obtained by the first business party after homomorphic encryption of guest_embedding can be expressed as: ENCRYPT(guest_embedding).

[0112] Similarly, the second business party can also filter out the browsing data of p objects (such as the browsing behavior, click behavior, negative feedback behavior, interests and hobbies of the objects) from the second access data according to the results of the federated encryption intersection processing, and generate second feature data based on the browsing data of the p objects. Specifically, the process for the second business party to generate the second feature data may include: first, the computer device performs feature extraction processing on the attribute browsing data of the p objects respectively to obtain the browsing features of each of the p objects; then, the computer device determines the second feature data based on the browsing features of the p objects. Among them, the computer device can also extract browsing features from the browsing data of the object based on embedding processing. In addition, in a possible implementation method, the computer device can perform a weighted average operation on the browsing features of the p objects to obtain the browsing features of the p objects after the weighted average, and then perform homomorphic encryption processing on the browsing features of the p objects after the weighted average, so as to obtain the second feature data. For example, in an embodiment of the present application, the first business party and the second business party can jointly train a multi-task learning model, such as Figure 1 As shown, in the process of the first business party and the second business party jointly training the multi-task learning model, the second business party (such as an advertising platform) can be responsible for the host side model of the multi-task learning model. Then, the browsing features of the p objects after weighted average obtained by the second business party can be expressed as: host_embedding, then, the second feature data obtained by the second business party after homomorphic encryption of host_embedding can be expressed as: ENCRYPT(host_embedding).

[0113] S402: Performing calculations on the first feature data and the second feature data in the homomorphic space to obtain sample feature data.

[0114] In one possible implementation, performing computational processing on the first feature data and the second feature data in a homomorphic space to obtain sample feature data may include: first, a computer device may perform a homomorphic encryption operation on the first feature data and the second feature data in the homomorphic space to obtain third feature data after the homomorphic encryption operation; then, the computer device may perform a homomorphic decryption operation on the third feature data in the homomorphic space to obtain sample feature data.

[0115] For example, if Figure 5 As shown, after the first business party provides the first feature data and the second business party provides the second feature data, the third-party federated learning platform (such as Figure 5The federated learning framework shown in the figure) performs federated learning training on the multi-task model. It is understandable that the process of data interaction between the first business party and the second business party is carefully designed and fully encrypted so that the first business party cannot guess the private data content of the second business party, and the second business party cannot guess the private data content of the first business party. However, the purpose of joint modeling can be achieved. That is to say, except for the model parameters of the final output, the data information of any party should not be leaked in the above process, nor can the intermediate results of the data information be inferred. In addition, the multi-task model provided in the embodiment of the present application can be specifically a "decentralized" system architecture. After the data related to the model training is encrypted by an encryption algorithm, it is directly communicated between the first business party and the second business party without relying on a third party to participate in "forwarding", realizing "decentralization". The entire training process only needs to coordinate the progress of the first business party and the second business party, so it can enhance the security in practical applications.

[0116] S403: Input the sample feature data into the fully connected layer of the multi-task model, and output k predicted retention rates for each of the p objects.

[0117] Among them, the multi-task model can refer to a neural network model of any structure. For example, the neural network model can include but is not limited to: DNN (Deep Neural Networks) model, LSTM (Long Short-Term Memory) model, GRU (Gated Recurrent Neural Network) model, etc. It should be noted that the embodiment of the present application does not specifically limit the model structure of the multi-task model. In a possible implementation method, considering the operating efficiency of the model, the embodiment of the present application can adopt the DNN model as the basic architecture of the multi-task model.

[0118] It is understandable that, in the process of performing federated learning training on the multi-task model, the task indicators of the multi-task model may be task indicators set for the first business party; may also be task indicators set for the second business party; or may be task indicators set jointly by the first business party and the second business party. In one possible implementation, the task indicators of the multi-task model during training may include task indicators of multiple tasks constructed based on retention rates, and the retention rate may be a task indicator set for the first business party (i.e., advertiser). Then, the first access data owned by the first business party may also include k retention rate labels for each of the m objects. After the aforementioned PSI privacy intersection processing and homomorphic encryption processing, the obtained sample feature data may include k retention rate labels for each of the p objects. The k retention rate labels may specifically include but are not limited to: next-day retention rate (referring to the probability that the target object that accessed the advertising data provided by the advertiser on the i-th day still accesses the advertising data provided by the advertiser on the i+1th day), two-day retention rate (referring to the probability that the target object that accessed the advertising data provided by the advertiser on the i-th day still accesses the advertising data provided by the advertiser on the i+2th day), three-day retention rate refers to the probability that the target object that accessed the advertising data provided by the advertiser on the i-th day still accesses the advertising data provided by the advertiser on the i+3rd day, etc., wherein i is a positive integer. Then, the retention rate label of each object can be 1 or 0.

[0119] Next, the k retention rate labels and sample feature data of each of the p objects can be used as training data for the multi-task model. Then, the multi-task model can also predict the k predicted retention rates of each of the p objects.

[0120] S404: Adjust model parameters of the multi-task model according to the k predicted retention rates of each object and the k retention rate labels of each object.

[0121] In one possible implementation, the k predicted retention rates of each object include a first predicted retention rate and a second predicted retention rate; the k retention rate labels of each object include a first label and a second label. The computer device adjusts the model parameters of the multi-task model according to the k predicted retention rates of each object and the k retention rate labels of each object, which may include: first, the computer device may calculate the first loss of the multi-task model according to the first predicted retention rate of each object and the first label of each object; and the computer device may also calculate the second loss of the multi-task model according to the second predicted retention rate of each object and the second label of each object. Then, the computer device determines the target loss according to the first loss and the second loss, and adjusts the model parameters of the multi-task model based on the target loss.

[0122] For example, if Figure 1As shown, in an embodiment of the present application, two task indicators, the next-day retention rate and the seven-day retention rate, can be set for the multi-task model. Then, the first label of each object can be a next-day retention rate label (1 or 0), and the second label can be a seven-day retention rate label (1 or 0); and the sample feature data is input into the fully connected layer of the multi-task model, and the next-day predicted retention rate (first predicted retention rate) and the seven-day predicted retention rate (second predicted retention rate) of each object in the p objects can be output. In a possible implementation, in an embodiment of the present application, different loss functions can be set for different tasks (each task corresponds to a task indicator, for example, task 1 can correspond to the next-day retention rate; task 2 can correspond to the seven-day retention rate). For example, for task indicator 1, the loss function set can be an absolute loss function (absoluteloss function); for another example, for task indicator 2, the loss function set can be a square loss function (quadraticloss function). Then, the computer device can calculate the loss according to the respective loss functions, that is, the computer device can obtain the first loss of the multi-task model according to the next-day predicted retention rate of each object and the next-day retention rate label of each object; and the computer device can obtain the second loss of the multi-task model according to the seven-day predicted retention rate of each object and the seven-day retention rate label of each object. Next, when the multi-task model back-propagates the gradient, the loss of each task can be calculated to obtain the summarized loss (i.e., the target loss), where the calculation here can include but is not limited to: summation or weighted average calculation. Finally, the computer device can update the model parameters of the multi-task model based on the target loss.

[0123] In one possible implementation, when the adjusted multi-task model meets the model convergence condition, the training of the multi-task model is stopped. The so-called model convergence condition may include but is not limited to: when the number of training times of the multi-task model reaches a preset training threshold, such as 100 times, the multi-task model can be considered to meet the model convergence condition. Alternatively, when the difference data between the predicted retention rate and the retention rate label predicted by the model is less than the error threshold, the multi-task model can be considered to meet the model convergence condition. Alternatively, when the change between the predicted retention rates obtained by two consecutive trainings of the multi-task model is less than the change threshold, the multi-task model can be considered to meet the model convergence condition.

[0124] In one possible implementation, Figure 5As shown, after training the multi-task model, the computer device can also test the trained multi-task model to obtain the model evaluation index value of the trained multi-task model, for example, the model evaluation index value of the multi-task model may include the area under the ROC curve (AUC), wherein AUC can be used to evaluate the training effect of the multi-task model, and when AUC=1, the multi-task model can be considered to be a perfect multi-task model (under ideal conditions). If the model evaluation index value of the trained multi-task model meets the model evaluation index condition, the computer device can call the trained multi-task model to predict one or more task indicators corresponding to the target advertising data provided by the first business party. Among them, the model evaluation index condition may include an index threshold, and if the model evaluation index value of the trained multi-task model (for example, 0.9) is greater than or equal to the index threshold (0.88), it can be determined that the model evaluation index value of the trained multi-task model meets the model evaluation index condition. It can be understood that the index threshold here can be a manually customized setting, or it can be obtained by the computer device based on an empirical value, and the embodiment of the present application does not specifically limit this.

[0125] In a possible implementation, the task indicators of the multi-task model may include a retention rate indicator, the retention rate indicator may include k retention rates, the k retention rates may include a first retention rate (e.g., a next-day retention rate) and a second retention rate (e.g., a seven-day retention rate), and the computer device may call the trained multi-task model to perform prediction processing on the target object accessing the target advertising data to obtain the first predicted retention rate and the second predicted retention rate of the target object. Then, the computer device may determine the value of the target advertising data pushed to the target object based on the first predicted retention rate and the second predicted retention rate of the target object.

[0126] For example, see Figure 7 , Figure 7 : is a schematic diagram of an application scenario of a multi-task model provided in an embodiment of the present application. The application scenario of the multi-task model includes a terminal device and a server used by a target object. In a possible implementation, Figure 7The server shown may be a server where the advertiser is located (hereinafter referred to as a computer device), and then the advertiser can predict one or more business indicators of the target advertising data provided by the advertiser through the server to obtain the prediction result of the multi-task model. In the embodiment of the present application, the first business party may be an advertiser, and the second business party may be an advertising platform. Then, after the first business party and the second business party jointly learn and train a multi-task model, the advertiser can call the trained multi-task model to predict the retention rate (e.g., the next-day retention rate and the seven-day retention rate) of the target object accessing the target advertising data provided by the advertiser, and obtain the next-day retention prediction rate of the target object (e.g., 0.9) and the seven-day retention prediction rate of the target object (e.g., 0.85). It should be noted that the advertiser can send the target advertising data to the advertising platform through the RTA interface, and the advertising platform can expose the target advertising data provided by the advertiser, so that the target object can browse the target advertising data provided by the advertiser by accessing the advertising platform. Among them, the target advertising data provided by the advertiser may include but is not limited to: sports type advertising data, food type advertising data, beauty type advertising data, etc., and the embodiment of the present application does not specifically limit the type of target advertising data. Then, based on the model trained by the embodiment of the present application, the advertiser can call the trained multi-task model to predict the next-day retention prediction rate of the target object (for example, 0.9) and the seven-day retention prediction rate of the target object (for example, 0.85). Further, the computer device determines the value of the target advertising data based on the next-day retention prediction rate of the target object (for example, 0.9) and the seven-day retention prediction rate of the target object (for example, 0.85), which may include: for example, the next-day retention prediction rate of the target object (for example, 0.9) is higher than the next-day retention rate threshold (for example, 0.88), and the seven-day retention prediction rate of the target object (for example, 0.85) is higher than the next-day retention rate threshold (for example, 0.80), then it can be determined that the target object has a high degree of interest in the target advertising data.

[0127] It is understandable that the target object may refer to any new user brought by the target advertising data delivered by the advertiser to the advertising platform through the RTA interface within a certain period of time (for example, one day). During this period of time, if the number of target objects whose next-day retention prediction rate is greater than the next-day retention rate threshold, or whose seven-day retention prediction rate is greater than the seven-day retention prediction rate threshold, reaches the threshold, it can be determined that the target advertising data delivered by the advertiser has a good publicity effect, and the user acceptance and recognition are high. Subsequently, the value of the target advertising data can be increased, thereby bringing higher profits to the advertiser.

[0128] In another possible implementation, the task indicators of the multi-task model may include a retention rate indicator (a first retention rate and a second retention rate) and an active duration indicator. The computer device may call the trained multi-task model to perform prediction processing on the set of objects accessing the target advertising data within the target time period, and obtain k predicted retention rates and active durations for each object in the object set; and push additional advertising data to the target objects that meet the retention rate threshold condition and the active duration condition within the target time period.

[0129] For example, based on the above description, more task indicators can be set according to the scenario requirements, such as the active duration indicator of the object. Then call the trained multi-task model to predict the next-day retention rate and seven-day retention rate of each object in the set of objects (which may include q objects) that access the target advertising data within the target time period (for example, one week), and the active duration of each object can also be predicted. Subsequently, additional advertising data can be pushed to one or more target objects that meet the retention rate threshold conditions and active duration conditions (for example, the predicted active duration can reach the duration threshold, etc.) within a week, further increasing the advertiser's revenue and increasing the number of visitors to the advertising platform. It should be noted that if the task indicators include retention rate indicators, and the number of retention rate indicators is k, then the retention rate threshold conditions here may include: each of the k predicted retention rates predicted by the multi-task model needs to be greater than or equal to the corresponding retention rate threshold; specifically, the number of retention rate indicators is 3 (next-day retention rate, three-day retention rate, seven-day retention rate), then, if the next-day predicted retention rate predicted by the multi-task model needs to be greater than or equal to the next-day retention rate threshold, the three-day predicted retention rate predicted by the multi-task model needs to be greater than or equal to the three-day retention rate threshold, and the next-seven-day predicted retention rate predicted by the multi-task model needs to be greater than or equal to the seven-day retention rate threshold, then it is determined that the retention rate threshold conditions are met.

[0130] It should be noted that if the retention rate indicator includes the next-day retention rate, the retention rate threshold condition may include: the next-day predicted retention rate of the target object is greater than or equal to the next-day retention rate threshold; if the retention rate indicator includes the next-day retention rate and the seven-day retention rate, the retention rate threshold condition may include: the next-day predicted retention rate of the target object is greater than or equal to the next-day retention rate threshold, and the seven-day predicted retention rate of the target object is greater than or equal to the seven-day retention rate threshold. In addition, the task indicators of the multi-task model can be customized according to the scene requirements; in addition, when the multi-task indicators include the retention rate indicators, the number of retention rate indicators can also be customized according to the scene requirements, and the embodiments of the present application do not specifically limit this.

[0131] In summary, in the embodiment of the present application, a multi-task model can be trained by federated learning based on the first feature data provided by the advertiser and the second feature data provided by the advertising platform, and both the advertiser and the advertising platform can call the trained multi-task model to predict the task indicators. For users, users can see advertising data that is more in line with their interests, which can improve the user experience and is conducive to user activity and retention; for advertising platforms, the embodiments of the present application can effectively solve the problem of data isolation, and on the basis of protecting the data privacy and security of both advertisers and advertising platforms, make full use of the data of both parties to jointly optimize the recommendation effect of advertising data; for advertisers, the recommendation effect of advertising data is more accurate, which can reduce the advertising cost of advertisers and improve the retention rate of users.

[0132] See also Figure 8 , Figure 8 800 is a structural diagram of a data processing device provided in an embodiment of the present application. The data processing device 800 can be applied to a computer device corresponding to the above embodiment. The data processing device 800 can be a computer program (including program code) running in a computer device, for example, the data processing device 800 is an application software; the device can be used to execute the corresponding steps in the method provided in an embodiment of the present application. The data processing device 800 may include:

[0133] An acquisition unit 801 is used to acquire first characteristic data provided by a first business party, where the first characteristic data is obtained by homomorphically encrypting first access data owned by the first business party, and the first access data includes data generated when the advertisement data provided by the first business party is accessed within a preset time period;

[0134] The acquisition unit 801 is further used to acquire second characteristic data provided by the second business party, where the second characteristic data is obtained by homomorphically encrypting second access data owned by the second business party, and the second access data includes data generated when the advertisement data is accessed within the preset time period during the exposure of the advertisement data on the platform corresponding to the second business party;

[0135] A processing unit 802 is used to perform operation processing on the first feature data and the second feature data in the homomorphic space to obtain sample feature data;

[0136] The processing unit 802 is further used to perform federated learning training on the multi-task model using the sample feature data, and the trained multi-task model is used to predict one or more task indicators corresponding to the target advertising data provided by the first business party.

[0137] In a possible implementation, the first access data includes attribute data of m objects that access the advertisement data provided by the first business party within a preset time period; the second access data includes browsing data of n objects that access the advertisement data within a preset time period during the exposure of the advertisement data on the platform corresponding to the second business party; m and n are both positive integers; the processing unit 802 is further configured to perform the following operations:

[0138] Performing a federated encryption intersection process on the first access data and the second access data to obtain a federated encryption intersection process result;

[0139] Among them, the federated encrypted intersection processing result is used to indicate that the first business party needs to filter out the attribute data of p objects from the first access data, and generate first feature data based on the attribute data of the p objects; and the federated encrypted intersection processing result is also used to indicate that the second business party needs to filter out the browsing data of p objects from the second access data, and generate second feature data based on the browsing data of the p objects; p is a positive integer, p≤m and p≤n.

[0140] In a possible implementation, the processing unit 802 performs a federated encryption intersection process on the first access data and the second access data to obtain a federated encryption intersection process result, which is used to perform the following operations:

[0141] Encrypting the attribute data of m objects to obtain encrypted m attribute data;

[0142] Encrypting the browsing data of n objects to obtain n encrypted browsing data;

[0143] According to the encrypted m attribute data and the encrypted n browsing data, an intersection process is performed on the m attribute data and the n browsing data in the encryption space to obtain a federated encryption intersection process result.

[0144] In a possible implementation manner, the process of the first service party generating the first feature data includes:

[0145] Perform feature extraction processing on the attribute data of the p objects respectively to obtain the attribute features of each object in the p objects;

[0146] Based on the attribute features of the p objects, determining first feature data; and,

[0147] The process of the second business party generating the second characteristic data includes:

[0148] Perform feature extraction processing on the browsing data of p objects respectively to obtain browsing features of each object in the p objects;

[0149] Based on the browsing features of the p objects, second feature data is determined.

[0150] In a possible implementation, the processing unit 802 performs computation on the first feature data and the second feature data in the homomorphic space to obtain sample feature data for performing the following operations:

[0151] Performing homomorphic encryption operation on the first feature data and the second feature data in the homomorphic space to obtain third feature data after the homomorphic encryption operation;

[0152] The third characteristic data is homomorphically decrypted in the homomorphic space to obtain sample characteristic data.

[0153] In a possible implementation, the task indicator includes a retention rate indicator; the sample feature data also includes k retention rate labels of each of the p objects, where p is a positive integer; the processing unit 802 uses the sample feature data to perform federated learning training on the multi-task model to perform the following operations:

[0154] Input the sample feature data into the fully connected layer of the multi-task model and output k predicted retention rates for each of the p objects;

[0155] Based on the k predicted retention rates for each object and the k retention rate labels for each object, the model parameters of the multi-task model are adjusted.

[0156] In a possible implementation, the k predicted retention rates of each object include a first predicted retention rate and a second predicted retention rate; the k retention rate labels of each object include a first label and a second label;

[0157] The processing unit 802 adjusts the model parameters of the multi-task model according to the k predicted retention rates of each object and the k retention rate labels of each object to perform the following operations:

[0158] Calculate the first loss of the multi-task model based on the first predicted retention rate of each object and the first label of each object;

[0159] Calculating a second loss of the multi-task model based on a second predicted retention rate of each object and a second label of each object;

[0160] A target loss is determined according to the first loss and the second loss, and a model parameter of the multi-task model is adjusted based on the target loss.

[0161] In a possible implementation, the processing unit 802 is further configured to perform the following operations:

[0162] Test the trained multi-task model to obtain the model evaluation index value of the trained multi-task model;

[0163] If the model evaluation index value of the trained multi-task model meets the model evaluation index condition, the trained multi-task model is called to predict one or more task indicators corresponding to the target advertising data provided by the first business party.

[0164] In a possible implementation, the processing unit 802 is further configured to perform the following operations:

[0165] Call the trained multi-task model to perform prediction processing on the target object accessing the target advertising data, and obtain k predicted retention rates of the target object;

[0166] The value of the target advertising data pushed to the target object is determined based on k predicted retention rates of the target object.

[0167] In a possible implementation, the task indicator further includes an active duration indicator; and the processing unit 802 is further configured to perform the following operations:

[0168] Call the trained multi-task model to perform prediction processing on the set of objects that access the target advertising data within the target time period, and obtain k predicted retention rates and active durations for each object in the object set;

[0169] During the target time period, additional advertising data is pushed to target objects that meet the retention rate threshold and active duration conditions.

[0170] In an embodiment of the present application, first, the first characteristic data provided by the first business party can be obtained. The first characteristic data is obtained after homomorphic encryption processing of the first access data owned by the first business party, and the first access data includes data generated by accessing the advertising data provided by the first business party within a preset time period. Secondly, the second characteristic data provided by the second business party can also be obtained. The second characteristic data is obtained after homomorphic encryption processing of the second access data owned by the second business party, and the second access data includes data generated by accessing the advertising data within a preset time period during the exposure of the advertising data on the platform corresponding to the second business party. Then, the first characteristic data and the second characteristic data are operated and processed in the homomorphic space to obtain sample characteristic data. Finally, the sample characteristic data is used to perform federated learning training on the multi-task model, and the trained multi-task model is used to predict one or more task indicators corresponding to the target advertising data provided by the first business party. It can be seen that in the embodiment of the present application, both the first feature data provided by the first business party and the second feature data provided by the second business party can be obtained. Therefore, the training samples of the multi-task model are richer, which can improve the accuracy of the multi-task model; further, the determined sample feature data is a data interaction carried out in the homomorphic encryption space, and therefore the original data of both the first business party and the second business party will not be leaked, thereby ensuring the security of the original data of both parties; in addition, the trained multi-task model can predict multiple task indicators, thereby improving the accuracy of the predicted task indicators.

[0171] See also Fig. 9 , Fig. 9 9 is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. The computer device 900 is used to execute the steps performed by the computer device in the aforementioned method embodiment, and the computer device 900 includes: one or more processors 910; one or more input devices 920, one or more output devices 930 and a memory 940. The above-mentioned processors 910, input devices 920, output devices 930 and memory 940 are connected via a bus 950. The memory 940 is used to store a computer program, and the computer program includes program instructions. The processor 910 is used to call the program instructions stored in the memory 940 to perform the following operations:

[0172] Acquire first characteristic data provided by the first business party, where the first characteristic data is obtained by homomorphically encrypting first access data owned by the first business party, and the first access data includes data generated when the advertisement data provided by the first business party is accessed within a preset time period;

[0173] Obtaining second characteristic data provided by the second business party, where the second characteristic data is obtained by homomorphically encrypting second access data owned by the second business party, and the second access data includes data generated when the advertisement data is accessed within the preset time period during the process of exposure of the advertisement data on the platform corresponding to the second business party;

[0174] Performing computation on the first feature data and the second feature data in the homomorphic space to obtain sample feature data;

[0175] The multi-task model is trained through federated learning using sample feature data, and the trained multi-task model is used to predict one or more task indicators corresponding to the target advertising data provided by the first business party.

[0176] In a possible implementation, the first access data includes attribute data of m objects that access the advertisement data provided by the first business party within a preset time period; the second access data includes browsing data of n objects that access the advertisement data within a preset time period during exposure of the advertisement data on a platform corresponding to the second business party; m and n are both positive integers; the processor 910 is further configured to perform the following operations:

[0177] Performing a federated encryption intersection process on the first access data and the second access data to obtain a federated encryption intersection process result;

[0178] Among them, the federated encrypted intersection processing result is used to indicate that the first business party needs to filter out the attribute data of p objects from the first access data, and generate first feature data based on the attribute data of the p objects; and the federated encrypted intersection processing result is also used to indicate that the second business party needs to filter out the browsing data of p objects from the second access data, and generate second feature data based on the browsing data of the p objects; p is a positive integer, p≤m and p≤n.

[0179] In a possible implementation, the processor 910 performs a federated encryption intersection process on the first access data and the second access data to obtain a federated encryption intersection process result, which is used to perform the following operations:

[0180] Encrypting the attribute data of m objects to obtain encrypted m attribute data;

[0181] Encrypting the browsing data of n objects to obtain n encrypted browsing data;

[0182] According to the encrypted m attribute data and the encrypted n browsing data, an intersection process is performed on the m attribute data and the n browsing data in the encryption space to obtain a federated encryption intersection process result.

[0183] In a possible implementation manner, the process of the first service party generating the first feature data includes:

[0184] Perform feature extraction processing on the attribute data of the p objects respectively to obtain the attribute features of each object in the p objects;

[0185] Based on the attribute features of the p objects, determining first feature data; and,

[0186] The process of the second business party generating the second characteristic data includes:

[0187] Perform feature extraction processing on the browsing data of p objects respectively to obtain browsing features of each object in the p objects;

[0188] Based on the browsing features of the p objects, second feature data is determined.

[0189] In a possible implementation, the processor 910 performs computation processing on the first feature data and the second feature data in the homomorphic space to obtain sample feature data for performing the following operations:

[0190] Performing homomorphic encryption operation on the first feature data and the second feature data in the homomorphic space to obtain third feature data after the homomorphic encryption operation;

[0191] The third characteristic data is homomorphically decrypted in the homomorphic space to obtain sample characteristic data.

[0192] In a possible implementation, the task indicator includes a retention rate indicator; the sample feature data also includes k retention rate labels of each of the p objects, where p is a positive integer; the processor 910 uses the sample feature data to perform federated learning training on the multi-task model to perform the following operations:

[0193] Input the sample feature data into the fully connected layer of the multi-task model and output k predicted retention rates for each of the p objects;

[0194] Based on the k predicted retention rates for each object and the k retention rate labels for each object, the model parameters of the multi-task model are adjusted.

[0195] In a possible implementation, the k predicted retention rates of each object include a first predicted retention rate and a second predicted retention rate; the k retention rate labels of each object include a first label and a second label;

[0196] The processor 910 adjusts the model parameters of the multi-task model according to the k predicted retention rates of each object and the k retention rate labels of each object to perform the following operations:

[0197] Calculate the first loss of the multi-task model based on the first predicted retention rate of each object and the first label of each object;

[0198] Calculating a second loss of the multi-task model based on a second predicted retention rate of each object and a second label of each object;

[0199] A target loss is determined according to the first loss and the second loss, and a model parameter of the multi-task model is adjusted based on the target loss.

[0200] In a possible implementation, the processor 910 is further configured to perform the following operations:

[0201] Test the trained multi-task model to obtain the model evaluation index value of the trained multi-task model;

[0202] If the model evaluation index value of the trained multi-task model meets the model evaluation index condition, the trained multi-task model is called to predict one or more task indicators corresponding to the target advertising data provided by the first business party.

[0203] In a possible implementation, the processor 910 is further configured to perform the following operations:

[0204] Call the trained multi-task model to perform prediction processing on the target object accessing the target advertising data, and obtain k predicted retention rates of the target object;

[0205] The value of the target advertising data pushed to the target object is determined based on k predicted retention rates of the target object.

[0206] In a possible implementation, the task indicator further includes an active duration indicator; and the processor 910 is further configured to perform the following operations:

[0207] Call the trained multi-task model to perform prediction processing on the set of objects that access the target advertising data within the target time period, and obtain k predicted retention rates and active durations for each object in the object set;

[0208] During the target time period, additional advertising data is pushed to target objects that meet the retention rate threshold and active duration conditions.

[0209] In an embodiment of the present application, first, the first characteristic data provided by the first business party can be obtained. The first characteristic data is obtained after homomorphic encryption processing of the first access data owned by the first business party, and the first access data includes data generated by accessing the advertising data provided by the first business party within a preset time period. Secondly, the second characteristic data provided by the second business party can also be obtained. The second characteristic data is obtained after homomorphic encryption processing of the second access data owned by the second business party, and the second access data includes data generated by accessing the advertising data within a preset time period during the exposure of the advertising data on the platform corresponding to the second business party. Then, the first characteristic data and the second characteristic data are operated and processed in the homomorphic space to obtain sample characteristic data. Finally, the sample characteristic data is used to perform federated learning training on the multi-task model, and the trained multi-task model is used to predict one or more task indicators corresponding to the target advertising data provided by the first business party. It can be seen that in the embodiment of the present application, both the first feature data provided by the first business party and the second feature data provided by the second business party can be obtained. Therefore, the training samples of the multi-task model are richer, which can improve the accuracy of the multi-task model; further, the determined sample feature data is a data interaction carried out in the homomorphic encryption space, and therefore the original data of both the first business party and the second business party will not be leaked, thereby ensuring the security of the original data of both parties; in addition, the trained multi-task model can predict multiple task indicators, thereby improving the accuracy of the predicted task indicators.

[0210] In addition, it should be pointed out here that: the embodiment of the present application also provides a computer storage medium, and a computer program is stored in the computer storage medium, and the computer program includes program instructions. When the processor executes the above program instructions, the method in the corresponding embodiment of the foregoing text can be executed, so it will not be repeated here. For technical details not disclosed in the computer storage medium embodiment involved in this application, please refer to the description of the method embodiment of this application. As an example, the program instructions can be deployed on a computer device, or executed on multiple computer devices located in one place, or, executed on multiple computer devices distributed in multiple locations and interconnected by a communication network.

[0211] According to one aspect of the present application, a computer program product or a computer program is provided, the computer program product or the computer program includes a computer instruction, and the computer instruction is stored in a computer-readable storage medium. The processor of the computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device can execute the method in the above corresponding embodiment, so it will not be repeated here.

[0212] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the above-mentioned program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The above-mentioned storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0213] The above disclosure is only the preferred embodiment of the present application, which certainly cannot be used to limit the scope of rights of the present application. Therefore, equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. A data processing method, characterized in that: include: Acquire first characteristic data provided by the first business party, where the first characteristic data is obtained by homomorphically encrypting first access data owned by the first business party, and the first access data includes data generated when the advertisement data provided by the first business party is accessed within a preset time period; Acquire second characteristic data provided by the second business party, where the second characteristic data is obtained by homomorphically encrypting second access data owned by the second business party, and the second access data includes data generated when the advertisement data is accessed within the preset time period during the process of exposure of the advertisement data on the platform corresponding to the second business party; Performing homomorphic encryption operation on the first feature data and the second feature data in a homomorphic space to obtain third feature data after the homomorphic encryption operation; Performing homomorphic decryption processing on the third characteristic data in the homomorphic space to obtain sample characteristic data; The sample feature data is used to perform federated learning training on a multi-task model, and the trained multi-task model is used to predict one or more task indicators corresponding to the target advertising data provided by the first business party; the task indicators include a retention rate indicator, and the retention rate indicator includes k retention rates; Calling the trained multi-task model to perform prediction processing on the target object accessing the target advertisement data, and obtaining k predicted retention rates of the target object; the target object refers to: any new object brought by the target advertisement data placed by the first business party to the second business party within a preset time period; the k predicted retention rates include: a next-day predicted retention rate and a seven-day predicted retention rate; If the number of target objects whose next-day predicted retention rate is greater than the next-day retention rate threshold, or whose seven-day predicted retention rate is greater than the seven-day predicted retention rate threshold, reaches a threshold, the value of the target advertising data placed by the first business party is increased; Among them, the federated learning training includes any one of the following methods: the first business party uses the sample feature data to perform federated learning training on the multi-task model; the second business party uses the sample feature data to perform federated learning training on the multi-task model; the first business party sends the first feature data to a third-party federated learning platform, and the second business party sends the second feature data to the third-party federated learning platform, and the third-party federated learning platform uses the sample feature data to perform federated learning training on the multi-task model.

2. The method according to claim 1, characterized in that The first access data includes attribute data of m objects that access the advertisement data provided by the first business party within a preset time period; the second access data includes browsing data of n objects that access the advertisement data within the preset time period during the exposure of the advertisement data on the platform corresponding to the second business party; m and n are both positive integers; The method further comprises: Performing a federal encryption intersection process on the first access data and the second access data to obtain a federal encryption intersection process result; Among them, the federal encryption intersection processing result is used to indicate that the first business party needs to filter out the attribute data of p objects from the first access data, and generate the first feature data based on the attribute data of the p objects; and the federal encryption intersection processing result is also used to indicate that the second business party needs to filter out the browsing data of the p objects from the second access data, and generate the second feature data based on the browsing data of the p objects; p is a positive integer, p≤m and p≤n.

3. The method according to claim 2, characterized in that The performing a federated encryption intersection process on the first access data and the second access data to obtain a federated encryption intersection process result includes: Encrypting the attribute data of the m objects to obtain m encrypted attribute data; Encrypting the browsing data of the n objects to obtain n encrypted browsing data; According to the encrypted m attribute data and the encrypted n browsing data, intersection processing is performed on the m attribute data and the n browsing data in the encryption space to obtain a federated encryption intersection processing result.

4. The method according to claim 2, characterized in that The process of the first business party generating the first characteristic data includes: Performing feature extraction processing on the attribute data of the p objects respectively to obtain attribute features of each object in the p objects; Based on the attribute characteristics of the p objects, determining the first feature data; and, The process of the second business party generating the second characteristic data includes: Performing feature extraction processing on the browsing data of the p objects respectively to obtain browsing features of each object in the p objects; The second feature data is determined based on the browsing features of the p objects.

5. The method according to claim 2, characterized in that The sample feature data also includes k retention rate labels of each object in p objects, where p is a positive integer; and the method of using the sample feature data to perform federated learning training on the multi-task model includes: Inputting the sample feature data into the fully connected layer of the multi-task model, and outputting k predicted retention rates for each of the p objects; Adjusting model parameters of the multi-task model based on the k predicted retention rates for each object and the k retention rate labels for each object.

6. The method according to claim 5, characterized in that The k predicted retention rates of each object include a first predicted retention rate and a second predicted retention rate; the k retention rate labels of each object include a first label and a second label; The adjusting the model parameters of the multi-task model according to the k predicted retention rates of each object and the k retention rate labels of each object includes: Calculating a first loss of the multi-task model based on a first predicted retention rate of each object and a first label of each object; Calculating a second loss of the multi-task model based on a second predicted retention rate of each object and a second label of each object; A target loss is determined according to the first loss and the second loss, and a model parameter of the multi-task model is adjusted based on the target loss.

7. The method according to claim 6, characterized in that The method further comprises: Testing the trained multi-task model to obtain a model evaluation index value of the trained multi-task model; If the model evaluation index value of the trained multi-task model meets the model evaluation index condition, the trained multi-task model is called to predict one or more task indicators corresponding to the target advertising data provided by the first business party.

8. The method according to claim 7, characterized in that The task indicator also includes an active duration indicator; the method also includes: Calling the trained multi-task model to perform prediction processing on the set of objects accessing the target advertising data within the target time period, and obtaining k predicted retention rates and active durations of each object in the set of objects; During the target time period, additional advertisement data is pushed to the objects to be pushed that meet the retention rate threshold condition and the active duration condition.

9. A data processing device, characterized in that: include: an acquisition unit, configured to acquire first characteristic data provided by a first business party, wherein the first characteristic data is obtained by homomorphically encrypting first access data owned by the first business party, and the first access data includes data generated when the advertisement data provided by the first business party is accessed within a preset time period; The acquisition unit is further used to acquire second feature data provided by the second business party, where the second feature data is obtained by homomorphically encrypting second access data owned by the second business party, and the second access data includes data generated when the advertisement data is accessed within the preset time period during the exposure of the advertisement data on the platform corresponding to the second business party; A processing unit, configured to perform homomorphic encryption operation on the first feature data and the second feature data in a homomorphic space to obtain third feature data after the homomorphic encryption operation; The processing unit is further used to perform homomorphic decryption processing on the third feature data in the homomorphic space to obtain sample feature data; The processing unit is further used to perform federated learning training on a multi-task model using the sample feature data, and the trained multi-task model is used to predict one or more task indicators corresponding to the target advertising data provided by the first business party; the task indicators include a retention rate indicator, and the retention rate indicator includes k retention rates; Calling the trained multi-task model to perform prediction processing on the target object accessing the target advertisement data, and obtaining k predicted retention rates of the target object; the target object refers to: any new object brought by the target advertisement data placed by the first business party to the second business party within a preset time period; the k predicted retention rates include: a next-day predicted retention rate and a seven-day predicted retention rate; If the number of target objects whose next-day predicted retention rate is greater than the next-day retention rate threshold, or whose seven-day predicted retention rate is greater than the seven-day predicted retention rate threshold, reaches a threshold, the value of the target advertising data placed by the first business party is increased; Among them, the federated learning training includes any one of the following methods: the first business party uses the sample feature data to perform federated learning training on the multi-task model; the second business party uses the sample feature data to perform federated learning training on the multi-task model; the first business party sends the first feature data to a third-party federated learning platform, and the second business party sends the second feature data to the third-party federated learning platform, and the third-party federated learning platform uses the sample feature data to perform federated learning training on the multi-task model.

10. A computer device, characterized in that: include: Memory and processor; a memory storing one or more computer programs; A processor, used to load the one or more computer programs to implement the data processing method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor and executing the data processing method according to any one of claims 1 to 8.

12. A computer program product, characterized in that The computer program product comprises a computer program, and the computer program is suitable for being loaded by a processor and executing the data processing method according to any one of claims 1 to 8.

Images