Data processing method and related device

Abstract

The invention discloses a data processing method and a related device.The method comprises the steps that first data of a target object in target software is obtained, and the first data is numeric data used for identifying operation implemented by the target object through the target software; noise is added to the first data in a differential privacy mode, noise-added interaction data is obtained, and the consistency difference between second numerical value distribution of numerical values in the noise-added interaction data and first numerical value distribution of numerical values in the first data meets the consistency condition. And finally, sending noise-added interaction data to the promotion platform, wherein the noise-added interaction data is used for carrying out strategy adjustment on advertisement putting of the promotion platform for the target software. According to the method, the numerical value in the first data can be fuzzified, the account privacy is prevented from being known by a promotion platform, meanwhile, the operation characteristics of the target object in the target software can be reflected, and the effect of correctly guiding the advertisement putting strategy is achieved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a data processing method and related devices. Background Art

[0002] In related art, a promotion platform promotes target software by delivering advertisements to target users, hoping that users will operate the target software.

[0003] In the process of advertising, it is necessary to obtain the first data of the target object and adjust the advertising strategy according to the first data. In related technologies, the method of directly transmitting the real first data directly to the promotion platform is often used for subsequent strategy adjustment.

[0004] From this, it can be seen that the relevant technology will expose the real first data of the target object, and cannot meet the needs of adjusting the advertising strategy while reasonably protecting the privacy data. Summary of the Invention

[0005] In order to solve the above technical problems, the present application provides a data processing method and related devices, which can blur the numerical values ​​in the first data to prevent the account privacy from being known by the promotion platform. At the same time, it can reflect the operational characteristics of the target object in the target software, and play a correct guiding role in the advertising delivery strategy.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] In one aspect, an embodiment of the present application provides a data processing method, the method comprising:

[0008] Acquire first data of a target object in target software, where the first data is numerical data used to identify an operation performed by the target object through the target software;

[0009] adding noise to the first data in a differential privacy manner to obtain noisy interaction data, wherein a consistency difference between a second numerical distribution of values ​​in the noisy interaction data and a first numerical distribution of values ​​in the first data meets a consistency condition;

[0010] The noise-added interaction data is sent to a promotion platform, where the noise-added interaction data is used to adjust a strategy for the promotion platform to place advertisements for the target software.

[0011] On the other hand, an embodiment of the present application provides a data processing device, the device comprising: an acquisition module, an adding module, and a sending module;

[0012] The acquisition module is configured to acquire first data of a target object in the target software, wherein the first data is numerical data used to identify an operation performed by the target object through the target software;

[0013] The adding module is configured to add noise to the first data in a differential privacy manner to obtain noisy interaction data, wherein a consistency difference between a second numerical distribution of values ​​in the noisy interaction data and a first numerical distribution of values ​​in the first data meets a consistency condition;

[0014] The sending module is used to send the noise-added interaction data to the promotion platform, and the noise-added interaction data is used to adjust the promotion platform's advertising strategy for the target software.

[0015] In another aspect, an embodiment of the present application provides a computer device, comprising a processor and a memory:

[0016] Memory is used to store computer programs;

[0017] The processor is configured to execute the above-described method according to the computer program.

[0018] In another aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method described in the above aspects.

[0019] On the other hand, an embodiment of the present application provides a computer program product including a computer program, which, when executed on a computer device, enables the computer device to execute the method described in the above aspects.

[0020] It can be seen from the above technical solution that when advertising for the target software through the promotion platform, the first data obtained through the target software needs to be returned to the promotion platform to guide the promotion platform to adjust the delivery strategy. In order to avoid the real values ​​involved in the first data from being leaked to the promotion platform, while maintaining the overall distribution of the values ​​in the first data to play a guiding role, before returning, noise can be added to the first data through differential privacy to obtain noisy interaction data that is relatively consistent with the first data in terms of numerical distribution. Since the values ​​in the first data are blurred by noise, the account privacy is prevented from being known by the promotion platform. At the same time, since the noisy interaction data is similar to the first data in terms of overall numerical distribution, it can also reflect the operational characteristics of the target object in the target software, so that the promotion platform can also determine the target object of the target software based on the noisy interaction data. The target information is closer to the real interaction information, which can play a correct guidance effect on the advertising delivery strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0022] Figure 1 A schematic diagram of a data processing scenario provided in an embodiment of the present application;

[0023] Figure 2 A flowchart of a data processing method provided in an embodiment of the present application;

[0024] Figure 3 A schematic diagram of adding a noise sampling range provided in an embodiment of the present application;

[0025] Figure 4 A schematic diagram of a strategy adjustment scenario provided in an embodiment of the present application;

[0026] Figure 5 A schematic diagram of another scenario of policy adjustment provided in an embodiment of the present application;

[0027] Figure 6 A schematic diagram of a data processing device provided in an embodiment of the present application;

[0028] Figure 7 A structural diagram of a terminal device provided in an embodiment of the present application;

[0029] Figure 8 A structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0030] The embodiments of the present application are described below with reference to the accompanying drawings.

[0031] During the promotion process for the target software, the promotion platform will advertise the target software to the target audience, hoping to attract new users through advertising. To improve the advertising's ability to attract new users for the target software, the first data generated by the target audience in the target software needs to be returned to the promotion platform as a criterion for adjusting the advertising strategy. This allows the promotion platform to adjust the advertising strategy based on the first data, thereby improving the effectiveness of attracting new target audiences.

[0032] In related technologies, the first data actually generated by the target object is often directly transmitted back to the promotion platform. This method can ensure the efficiency of advertising strategy adjustment due to the authenticity of the data. However, this method will expose the privacy data of the target object and cannot effectively protect the privacy of the target object. It is impossible to meet the needs of advertising strategy adjustment while reasonably protecting the privacy data.

[0033] To this end, embodiments of the present application provide a data processing method and related apparatus that, through differential privacy, adds noise to acquired first data to generate noisy interaction data, while ensuring that the noisy interaction data and the first data have similar numerical distributions. This allows the numerical values ​​in the first data to be obscured, preventing the account's privacy from being discovered by the promotion platform. Furthermore, because the overall numerical distribution of the noisy interaction data is similar to that of the first data, it can also reflect the operational characteristics of the target object in the target software, providing accurate guidance for advertising delivery strategies.

[0034] The data processing method provided in the embodiments of the present application can be implemented by a computer device, which can be a terminal device or a server. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. Terminal devices include but are not limited to mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals, aircraft, etc. The terminal device and the server can be directly or indirectly connected via wired or wireless communication, which is not limited in this application.

[0035] It is understandable that in the specific implementation of this application, when the above embodiments of this application are applied to specific products or technologies regarding the first data generated by the target object and other related data, any one of them needs to obtain separate user permission or consent, and the collection, use and processing of relevant data need to comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0036] First, several noun terms that may be involved in the embodiments below in this application are explained.

[0037] JS divergence: used to evaluate the similarity between the first and second numerical distributions. The value range is 0 to 1. If the first and second numerical distributions are more similar, the JS divergence is smaller; if the first and second numerical distributions are more different, the JS divergence is larger.

[0038] Differential privacy: A privacy definition described in mathematical language. A function that satisfies differential privacy is generally called a mechanism. It is defined as follows: if for all adjacent datasets x and x′ and all possible outputs S, ∈ is called the privacy parameter or privacy budget. If the mechanism F satisfies:

[0039]

[0040] Mechanism F is said to satisfy differential privacy. Simply put, when using mechanism F to interfere with a user's first data, if F satisfies differential privacy, then the output of F is always nearly identical, regardless of whether the input contains any specific data. Suppose, for example, that the first data of a target object A is in x but not in x'. If other parties cannot determine whether the input of F is x or x', then they cannot even determine whether the input contains the target object A, let alone what the first data of the target object A is.

[0041] Artificial Intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also studies the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0042] Artificial intelligence (AI) technology is a comprehensive discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Foundational AI technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0043] Machine learning (ML) is a multidisciplinary field that encompasses probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is at the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications span all areas of AI. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0044] With the research and advancement of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, autonomous driving, drones, robots, smart medical care, smart customer service, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0045] The solution provided in the embodiment of the present application involves machine learning and big data processing technology of artificial intelligence. For example, when determining the predicted data of the target object in the future time period, a prediction model is required. The initial prediction model needs to be trained in advance with training samples to obtain the prediction model. The solution provided in the embodiment of the present application also needs to be based on big data processing technology, and noise is added to the first data in a differential privacy manner to obtain noisy interaction data. This is specifically illustrated by the following embodiments:

[0046] Figure 1 A schematic diagram of a data processing scenario provided in an embodiment of the present application, wherein the aforementioned computer device is a server.

[0047] When the promotion platform places advertisements for the target software, it first needs to obtain the first data of the target object in the target software. In order to ensure the privacy of the first data, noise is added to the first data through differential privacy to obtain noisy interaction data. This can avoid the exposure of the first data and, to a certain extent, ensure the privacy of the first data. The numerical values ​​in the first data satisfy the first numerical distribution, and the numerical values ​​in the noisy interaction data satisfy the second numerical distribution. Although there are differences between the first numerical distribution and the second numerical distribution, the consistency difference between the two meets the consistency condition, that is, there needs to be a certain consistency between the second numerical distribution of the noisy interaction data and the first numerical distribution of the first data. In this way, it can be ensured that even if the noisy interaction data partially obscures the first data, the numerical distribution characteristics of the first data can be reflected, that is, the distribution characteristics of the numerical data of the target object implemented through the target software can be retained.

[0048] After obtaining the noisy interaction data, the noisy interaction data is sent to the promotion platform so that the promotion platform can adjust the advertising strategy of the target software based on the noisy interaction data. Since the noisy interaction data can reflect the numerical distribution of the first data, it can protect the privacy of the first data of the target object while ensuring the effectiveness of the advertising strategy adjustment to a certain extent.

[0049] Figure 2 This is a flowchart of a data processing method provided in an embodiment of the present application. The method can be executed by a computer device. In this embodiment, the computer device is described as a server.

[0050] S201: Acquire first data of a target object in target software.

[0051] Target software refers to software designed and developed for specific needs and scenarios, such as game applications, social applications, etc. Target software includes but is not limited to applications installed and run in mobile terminals, software installed and run in non-mobile terminals (such as desktop computers), etc. In the embodiments of the present application, it is expected that advertisements for the target software will be delivered through the promotion platform, so that new target objects can perform certain operations on the target software under the influence of the delivered advertisements. The target object refers to a virtual account registered or created by the user in the target software, which is used to represent the user's virtual incarnation in the target software. The user can obtain the services of the target software in the target software through the target object, and perform operations based on demand, etc.

[0052] The first data is numerical data used to identify the operations performed by the target subject through the target software. Specifically, the first data is data corresponding to interactive behaviors that can be represented numerically. Therefore, the first data can represent numerical data such as the target subject's click frequency, usage duration, transferred feature values, and virtual equivalents spent on the target software. The first data can reflect the target subject's interactive characteristics with the target software, as well as the target subject's level of interest in the target software.

[0053] S202: Add noise to the first data using a differential privacy method to obtain noisy interaction data.

[0054] Differential privacy refers to a method of protecting the privacy data of the target object involved in the first data by adding noise or disturbance during the query or analysis of the first data. Specific differential privacy methods may include using Laplace noise, Gaussian noise, random response technology, etc. In the embodiment of this application, the method of adding Laplace noise to the first data is mainly used to achieve privacy protection for the first data. Laplace noise can be used Represents , where s represents the maximum degree of change of the value in the first data after noise is added (that is, the sampling range of the added noise), ∈ represents the privacy parameter (Privacy Parameter) or privacy budget (Privacy Budget), and ∈ is generally set to a value approximately equal to 1 or smaller to provide sufficient privacy.

[0055] When privacy protection for first data is achieved by adding Laplace noise to the first data, the noisy interaction data primarily consists of two parts: the original first data and the added Laplace noise. The values ​​of the noisy interaction data satisfy the second numerical distribution, while the values ​​of the first data satisfy the first numerical distribution. Numerical distribution refers to the distribution of the values ​​of the first data or noisy interaction data. Specifically, it may include the number of first data or noisy behavior data corresponding to different numerical values, as well as the sparse distribution of the values ​​within a certain numerical range within the first data or noisy behavior data. When placing advertisements, the promotion platform can obtain the account characteristics of the target subject and, through the numerical distribution, can determine the overall situation of the target subject's first data. Combining the account characteristics with the overall situation of the first data of the target subject's operations on the target software, it can determine the proportion of target subjects involved in the numerical distribution that have operated on the target software, and also determine the level of interest in the target software among target subjects corresponding to different account characteristics.

[0056] The consistency difference between the second numerical distribution of the values ​​in the noisy interactive data and the first numerical distribution of the values ​​in the first data meets the consistency condition. The consistency difference is used to describe the degree of difference between the first numerical distribution and the second numerical distribution. When the consistency difference is smaller, it proves that the consistency between the first numerical distribution and the second numerical distribution is stronger, that is, the more the noisy interactive data can express the numerical distribution law of the first data, the more it can reflect the operational characteristics of the target object in the target software. JS divergence can be used as an evaluation criterion when evaluating the consistency difference between the first numerical distribution and the second numerical distribution. Table 1 shows the difference between the first numerical distribution and the second numerical distribution obtained based on the differential privacy method provided in the embodiment of the present application, as shown in the following table:

[0057] Table 1. Differences between the first and second numerical distributions

[0058]

[0059] It can be seen from the above table that the difference between the second numerical distribution of the noisy interaction data obtained after processing by the differential privacy method in the embodiment of the present application and the first numerical distribution of the first data is small.

[0060] The consistency condition refers to the condition that corresponds to the degree of similarity that needs to be achieved between the second numerical distribution of the values ​​in the noisy interactive data and the first numerical distribution of the values ​​in the first data. When the consistency difference of the overall numerical distribution between the second numerical distribution of the values ​​in the noisy interactive data and the first numerical distribution of the values ​​in the first data meets the consistency condition, it means that the second numerical distribution of the values ​​in the noisy interactive data can reflect the numerical distribution characteristics of the first numerical distribution of the values ​​in the first data.

[0061] Although the noisy interaction data adds noise compared to the first data, since the overall numerical distribution between the second numerical distribution and the first numerical distribution is highly consistent, the noisy interaction data can reflect the proportion of target objects involved in the first data that have operated the target software, as well as the degree of interest of target objects corresponding to different account characteristics in the target software, which can help the promotion platform adjust its advertising delivery strategy based on the noisy interaction data.

[0062] S203: Sending the noise-added interaction data to a promotion platform.

[0063] After acquiring the noisy interaction data, it needs to be sent to the promotion platform, which uses it as guidance for adjusting its advertising strategy for the target software. The noisy interaction data includes the target's account characteristics. Specific account characteristics may include, but are not limited to, account identifier, account location, account online time, and account login count. It should be noted that while the noisy interaction data contains both the target's account characteristics and the first data after adding noise, there is no correspondence between the target's account characteristics and the first data after adding noise. For example, suppose the actual first data is 90 and 25. The noisy interaction data includes two account identifiers: "Universe Invincible Girl" and "Pig Head 123," and the first data after adding noise are 100 and 30. Because the noise is added, the actual first data is not present in the noisy interaction data, and there is no true correspondence between the target and the first data, thus ensuring privacy protection for the target.

[0064] Because the second numerical distribution of the values ​​in the noisy interaction data satisfies the consistency condition with the first numerical distribution of the values ​​in the first data, the noisy interaction data can reflect the distribution characteristics of the first numerical distribution of the values ​​in the first data, and thus can reflect the operational characteristics of the target object in the target software. The promotion platform that receives the noisy interaction data can then determine based on the noisy interaction data that the target object of the target software is more realistic in terms of interaction information. This allows the platform to ensure the privacy of the target object's first data while also, to a certain extent, ensuring the effectiveness of advertising strategy adjustments.

[0065] When the target software is a game application, the first data of the target object obtained at this time may be the number of times the target object has logged into the game application, the number of times the interface has been clicked, the number of game skins owned, etc. By adding noise to the first data in a differential privacy manner to obtain noisy interaction data, the first data of the target object can be blurred, thereby preventing the privacy of the target object (the privacy includes: the true correspondence between the target object and the first data, and the true value of the first data) from being known by the promotion platform. At the same time, the consistency difference between the first numerical distribution of the values ​​in the first data and the second numerical distribution of the values ​​in the noisy interaction data meets the consistency condition. Therefore, the noisy interaction data can reflect the proportion of target objects that operate the game application, and the degree of interest of different target objects in the game application can also be determined. Therefore, sending the noisy interaction data to the promotion platform can play a correct guiding role in the promotion platform's advertising for the target software.

[0066] It can be seen from the above technical solution that when advertising for the target software through the promotion platform, the first data obtained through the target software needs to be returned to the promotion platform to guide the promotion platform to adjust the delivery strategy. In order to avoid the real values ​​involved in the first data from being leaked to the promotion platform, while maintaining the overall distribution of the values ​​in the first data to play a guiding role, before returning, noise can be added to the first data in a differential privacy manner to obtain noisy interaction data that is relatively consistent with the first data in terms of numerical distribution. Since the values ​​in the first data are blurred by noise, the privacy of the target object is prevented from being known by the promotion platform. At the same time, since the noisy interaction data is similar to the first data in terms of overall numerical distribution, it can also reflect the operational characteristics of the target object in the target software, so that the promotion platform can also determine the target object of the target software based on the noisy interaction data. The target information is closer to the real interaction information, which can play a correct guidance effect on the advertising delivery strategy.

[0067] The aforementioned description states that "the first data is numerical data used to identify the operation performed by the target object via the target software." In other words, in the embodiments of the present application, the first data is a data type represented by a numerical value. Specifically, the numerical identifier of the first data is at least one of a feature value of a feature value transfer performed via the target software and the number of interactions performed.

[0068] The characteristic value transferred through the target software can be understood as a virtual resource with a certain value, such as game coins, points, etc. The characteristic value transfer can be understood as a process in which a certain characteristic value is transferred to the provider of the target software through the instruction of the target object, so that the target object obtains the corresponding service of the target software. For example, the characteristic value a in the user's first account is transferred to the second account of the provider of the target software through the instruction of the target object in the target software. At this time, the numerical value of the first data represents the numerical value of the virtual resource transfer. For example, assuming that the target object transfers 10 game coins to the provider of the target software, the corresponding characteristic value of the characteristic value transferred through the target software is 10.

[0069] Interaction counts refer to the number of times a target object performs actions on the target software. Specific actions include, but are not limited to, clicking on controls within the target software, entering text, and browsing pages. Interaction counts are the number of times a target object performs these actions within the target software. Each time a target object performs an action, the corresponding interaction count increases by one.

[0070] The aforementioned numerical value of the first data can be identified as at least one of the characteristic value of characteristic value transfer performed by the target software and the number of interactions performed. That is to say, the numerical value of the first data can only identify the characteristic value of characteristic value transfer performed by the target software or the number of interactions performed, and can also simultaneously identify the characteristic value of characteristic value transfer performed by the target software and the number of interactions performed. When performing simultaneous identification, the characteristics of the aforementioned two numerical values ​​can be effectively combined so that the numerical value of the first data includes data information related to multiple operations, which is convenient for fully grasping the situation of the first data.

[0071] Through the numerical identification of the first data provided above, it is explained that different types of numerical first data can be applied to the differential privacy method to add noise to obtain noisy interactive data, thereby realizing the privacy protection of multiple types of first data, avoiding the exposure of the first data, and ensuring the privacy of the first data to a certain extent.

[0072] The aforementioned S202 mentions "adding noise to the first data in a differential privacy manner to obtain noisy interaction data". In the process of adding noise to the first data, due to the influence of objective conditions, the first data to which the noise is added may not meet the reasonable numerical conditions. When it is found that the value of the first data to which the noise is added does not meet the reasonable numerical conditions, in order to ensure the rationality of the data, the data that does not meet the reasonable numerical conditions needs to be removed. Therefore, in one possible implementation method, the method for obtaining noisy interaction data is: first, adding noise to the first data in a differential privacy manner to obtain noisy data to be processed. Then, in response to the fact that the noisy data to be processed includes target noisy data that does not meet the reasonable numerical conditions, the target noisy data is removed from the noisy data to be processed to obtain noisy interaction data.

[0073] By adding noise to the first data using differential privacy, we can obtain the noisy data to be processed. Since the processing causes the first data to undergo numerical changes, some abnormal data may be present in the noisy data to be processed. Abnormal data here can be understood as noisy data to be processed whose values ​​do not meet the numerical rationality conditions. The numerical rationality conditions refer to the value patterns of the first data. Based on the numerical rationality conditions, we can identify the target noisy data in the noisy data to be processed that do not meet the first data's value patterns.

[0074] The aforementioned reasonable numerical conditions are related to the value pattern of the first data, and the numerical identifier of the first data can be at least one of the characteristic value of the characteristic value transfer through the target software and the number of interactions. Then the corresponding reasonable numerical conditions will also be different according to the different contents of the numerical identifier. For example, when the numerical identifier of the first data is the number of interactions of the target object with respect to the target software, since the minimum number of interactions is 0, the number of interactions cannot be a negative value. Then the corresponding value pattern of the first data (i.e., reasonable numerical conditions) at this time is that the value of the first data is not less than 0. If there is a value of the to-be-processed noisy data obtained by adding noise to the first data in a differential privacy manner that is less than 0, it proves that the to-be-processed noisy data is the target noisy data that does not meet the reasonable numerical conditions, and it needs to be eliminated to obtain the noisy interaction data.

[0075] As another example, assume that the numerical value of the first data is identified as an eigenvalue transferred by the target software, and that the eigenvalue is identified as a negative value. At this time, the maximum value of the numerical value of the first data is 0, so the eigenvalue cannot be positive. In this case, the corresponding numerical reasonableness condition is that the value of the first data is not greater than 0. If the value of the unprocessed noisy data obtained by adding noise to the first data using a differential privacy method is greater than 0, it proves that the unprocessed noisy data does not meet the numerical reasonableness condition and needs to be eliminated.

[0076] Using the aforementioned method for obtaining noisy interaction data, after adding noise to the first data using differential privacy to obtain the noisy data to be processed, target noisy data within the noisy data to be processed is determined based on numerical rationality conditions. Because the target noisy data does not meet the numerical rationality conditions, it is removed from the noisy data to be processed. This ensures, to a certain extent, that the resulting noisy interaction data is reasonable, thereby improving the accuracy of the promotion platform's advertising strategy adjustments for the target software.

[0077] It is mentioned above that "the target object refers to the identifier registered or created by the user in the target software, which is used to identify and verify the identity of the user." In the embodiment of the present application, the advertising strategy of the promotion platform is adjusted based on the noisy interaction data whose numerical distribution is relatively consistent with the first data, in the hope that more new users will pay attention to the target software. Therefore, it is necessary to pay more attention to the characteristics of the first data of new users in order to improve the effectiveness of the adjustment of the advertising strategy. In one possible implementation method, the target object mentioned above is a new account of the target software.

[0078] Target objects are divided into new accounts and old accounts based on the frequency and duration of use of the target software. New accounts include at least one of: target objects whose registration duration with the target software is less than a first duration threshold, or target objects whose re-login duration is less than a second duration threshold after being offline for a preset number of days. It should be noted that the values ​​of the first and second duration thresholds mentioned above can be the same or different. The specific values ​​of the first and second duration thresholds can be set by those skilled in the art based on actual conditions and application scenarios and are not limited here. For example, the first duration threshold can be set to 3 hours and the second duration threshold to 6 hours; alternatively, both the first and second duration thresholds can be set to 3 hours. Furthermore, the preset number of days can also be set by those skilled in the art based on actual conditions and application scenarios. For example, the preset number of days can be 30 days and the second duration threshold can be 1 hour. That is, if a target object has been offline for more than 30 days and has logged back into the target software for less than 1 hour, the target object is considered a new account.

[0079] From the above description, it can be seen that the new account of the target software refers to a target object with a shorter registration time for the target software, or a lower frequency of online access to the target software. When the target object is determined to be a new account of the aforementioned target software, the first data corresponding to the new account will be obtained in the embodiment of the present application, and then the first data will be noise-processed, and the second numerical distribution of the numerical values ​​in the obtained noisy interaction data and the first numerical distribution of the numerical values ​​in the first data will meet the consistency condition. The noisy interaction data corresponding to the new account is then sent to the promotion platform, which enables the promotion platform to adjust the advertising strategy of the target software based on the noisy interaction data of the new account obtained. Since the noisy interaction data of the new account has the distribution characteristics of the first data of the user corresponding to the new account, the adjusted advertising strategy can be made more in line with the operational requirements of the new account, which can also improve the efficiency of advertising.

[0080] Setting the target object as a new account of the target software as described above ensures that both the acquired first data and the noise-added interaction data reflect the characteristics of the first data of the new account of the target software. At this time, the noise-added interaction data is sent to the promotion platform, which adjusts the advertising strategy based on the characteristics of the new account's operations in the noise-added interaction data. This allows the adjusted strategy to be more closely aligned with the characteristics of the new account's operations, which is conducive to improving the efficiency of advertising in attracting new users to the target software. At the same time, since the object of the acquired first data is a new account, adjusting the advertising strategy based on the characteristics of the new account's first data is conducive to increasing the activity of the target software.

[0081] Since the numerical range involved in the values ​​in the first data often has a large span, if a differential privacy method is used to add uniform noise to the first data for different values, a smaller value of the first data will be added with noise much larger than the value, causing the value of the first data to be buried in the noise. The resulting noisy interactive data cannot accurately reflect the numerical distribution of the first data. Therefore, when it is necessary to add noise to the first data using a differential privacy method, it is necessary to first divide the numerical range into N different numerical intervals based on the numerical range of the values ​​in the first data. Then, the first data is divided into N numerical sets based on the N numerical intervals.

[0082] By dividing the numerical range into different numerical intervals, the numerical values ​​in the first data can be grouped according to their values. For example, assuming the numerical range of the numerical values ​​in the first data is 1 to 100, it can be divided into three intervals, specifically: [1, 10], [11, 50], and [51, 100]. Numerical intervals and numerical sets correspond one-to-one. A numerical set includes first data in the same numerical interval. In this case, the corresponding first numerical set includes first data in the numerical interval [1, 10], the second numerical set includes first data in the numerical interval [11, 50], and the third numerical set includes first data in the numerical interval [51, 100].

[0083] For example, assume there are three numerical intervals with a total of 90 first data, specifically: the first numerical interval [1, 10], the second numerical interval [11, 50], and the third numerical interval [51, 100]. When the first numerical distribution is uniform, the corresponding three numerical intervals: the first numerical interval [1, 10], the second numerical interval [11, 50], and the third numerical interval [51, 100] will each have 30 first data. When the first numerical distribution is uneven, the first numerical interval [1, 10] will have 10 first data, the second numerical interval [11, 50] will have 50 first data, and the third numerical interval [51, 100] will have 30 first data.

[0084] Once the numerical sets in the first data are divided, noise can be added to each data set according to its specific characteristics. This allows appropriate noise to be added to the first data of each numerical set, while preserving the numerical characteristics of the original first data and concealing the specific numerical values ​​of the first data. A specific method for obtaining noisy interaction data is to add noise to each of N numerical sets using differential privacy to obtain noisy interaction data, wherein the noise addition mechanism for the first data in each numerical set is different.

[0085] The first data is divided into multiple data sets through the aforementioned numerical intervals, and different data sets contain first data whose values ​​belong to the corresponding numerical intervals. The difference in the numerical values ​​between the first data in the same numerical interval is small, and the difference in the numerical values ​​between the first data in different numerical intervals is large. Therefore, in the process of adding noise to the first data to obtain noisy interactive data, it is necessary to perform targeted noise addition in units of numerical sets, and the noise adding mechanism corresponding to the first data in different numerical sets is different. For example, a smaller noise can be added to the first data in a numerical set with a smaller value; a larger noise can be added to the first data in a numerical set with a larger value. By adding noise to the first data of different numerical sets in a targeted and specific manner, the value characteristics of the corresponding numerical set can be retained for different first data, and the trend of the numerical difference between the first data of different numerical sets can also be retained.

[0086] For example, for a first data set with a numerical interval of [1, 10] included in a first numerical set, and a first data set with a numerical interval of [51, 100] included in a third numerical set, after noise is added to the first data in the first numerical set and the third numerical set according to the corresponding adding mechanism, the value of the first noisy interaction data of the first numerical set and the third noisy interaction data of the third numerical set should generally still be smaller than the value of the third noisy interaction data. Furthermore, within the first noisy interaction data and the third noisy interaction data, the value distribution characteristics of each first noisy interaction data should be generally consistent with the first first data, and the value distribution characteristics of each third noisy interaction data should be generally consistent with the third first data.

[0087] It should be noted that, in the aforementioned method for obtaining noisy interaction data, since the noise addition process causes the first data to undergo numerical changes, the resulting processed noisy data may contain some abnormal data, i.e., target noisy data whose values ​​do not meet the reasonable numerical requirements. In this case, this target noisy data needs to be removed to obtain the noisy interaction data.

[0088] Using the aforementioned method for obtaining noisy interaction data, numerical intervals are divided according to the numerical range of the values ​​in the first data. Then, corresponding numerical sets are determined based on the numerical intervals. Using differential privacy, noise is added to the first data in different numerical sets using different addition mechanisms. This interval division ensures that similar first data values ​​will not differ significantly when noise is added. This, to a certain extent, ensures that the noisy interaction data after noise addition accurately reflects the numerical distribution characteristics of the first data.

[0089] The aforementioned S202 mentioned that “noise is added to the first data in a differential privacy manner to obtain noisy interaction data”. Since the numerical values ​​in the first data themselves conform to the first numerical distribution, when adding noise to the first data, in addition to the numerical range, the distribution of the numerical values ​​of the first data also needs to be considered. According to the distribution of the numerical values, the corresponding noise is added, so that the added noise can be more targeted and the noisy interaction data after noise addition can be more consistent with the numerical distribution, and there will not be a large difference between the second numerical distribution of the numerical values ​​in the noisy interaction data and the first numerical distribution of the numerical values ​​in the first data due to the addition of noise. Therefore, in a possible implementation method, the aforementioned method of “dividing the numerical range into N different numerical intervals according to the numerical range of the numerical values ​​in the first data” is: dividing the numerical range into N different numerical intervals according to the numerical range of the numerical values ​​in the first data and the first numerical distribution.

[0090] In the foregoing description, it is mentioned that the numerical range needs to be divided into N different numerical intervals based on the numerical range of the numerical values ​​in the first data. In the process of dividing the numerical intervals, in order to improve the efficiency of adding noise to the first data and at the same time ensure that the second numerical distribution of the numerical values ​​in the noisy interactive data can have a certain consistency with the first numerical distribution of the numerical values ​​in the first data, when dividing the numerical range of the numerical values ​​in the first data into numerical intervals, it is necessary to consider both the numerical range and the first numerical distribution of the numerical values ​​in the first data.

[0091] The first numerical distribution refers to the distribution of the numerical values ​​in the first data, and the specific distribution can be divided into uniform distribution and uneven distribution. The numerical distribution mentioned above refers to the distribution of the numerical values ​​of the first data or noisy interaction data, and can specifically include the number of first data or noisy behavior data corresponding to different numerical values, and the sparseness of the distribution of the numerical values ​​in the first data or noisy behavior data within a certain numerical range.

[0092] Here, uniform distribution can be understood as the number of first data corresponding to each numerical interval being roughly the same, or the sparsity of the first data corresponding to each numerical interval being roughly the same. For example, assuming there are 90 first data in total, and the numerical range involved in the values ​​of each first data is 1 to 100, the specific first numerical distribution corresponding to the uniform distribution can be: [1,10], [11,20], [21,30], [31,40], [41,50], [51,60], [61,70], [71,80], [81,90], [91,100], each of which has 9 first data. At this time, according to the numerical range and the first numerical distribution, the numerical range can be divided into 10 different numerical intervals, namely: [1,10], [11,21], [21,30], [31,40], [41,50], [51,60], [61,70], [71,80], [81,90], [91,100], thereby evenly dividing the numerical range into 10 different numerical intervals.

[0093] The aforementioned uneven distribution can be understood as a significant difference in the number of first data points corresponding to different numerical intervals, or in the sparsity of the first data points within each numerical interval. For example, assume there are 90 first data points, and the values ​​in each first data point range from 1 to 100. The actual distribution of the first data points is: there are 3 first data points in [1,10], 1 first data point in [11,21], 20 first data points in [31,40], 4 first data points in [41,50], 2 first data points in [51,60], 30 first data points in [61,70], 10 first data points in [71,80], 10 first data points in [81,90], and 10 first data points in [91,100]. At this time, since there are fewer first data distributed in [1,10], [11,21], [41,50] and [51,60], when dividing the numerical intervals, [1,10] and [11,21] can be merged into [1,21] and regarded as one numerical interval, [41,50] and [51,60] can be merged into [41,60] and regarded as one numerical interval, [31,40], [61,70], [71,80], [81,90] and [91,100] can each be regarded as a numerical interval separately, thereby dividing the numerical range into 7 different numerical intervals.

[0094] By combining the numerical range of the values ​​in the first data and the distribution of the first values, the numerical intervals can be divided more reasonably according to the distribution of the values ​​in the first data. For the numerical range with less distributed first data, the corresponding numerical interval can be appropriately expanded, thereby reducing the number of divided numerical intervals. This reduces the complexity of the process of adding noise through differential privacy, and can improve the efficiency of adding noise to the first data to obtain noisy interactive data.

[0095] In the above, the process of obtaining the noisy interaction data is specifically introduced. During the process, the numerical intervals of the numerical values ​​in the first data need to be divided so that noise can be added specifically according to the division of the numerical intervals. When dividing the numerical intervals, the numerical intervals divided corresponding to the first numerical distribution of the first data can be uniform or uneven. Combining the numerical span of the numerical intervals can achieve more reasonable noise addition. Therefore, in one possible implementation method, the method for obtaining the noisy interaction data in the "adding noise to the N numerical sets respectively by differential privacy to obtain the noisy interaction data" mentioned in the above S202 is:

[0096] A1: Determine the value span of the value interval corresponding to the value set.

[0097] The span of a numerical interval refers to the difference or range between the maximum and minimum values ​​within the interval. The span can also be simply understood as the difference between the maximum and minimum values ​​within the interval. Depending on how the intervals are divided within a set of values, the spans of the corresponding intervals will vary. Specifically, when the intervals are evenly divided, the spans of each interval are consistent. When the intervals are unevenly divided, the spans of each interval will not be completely consistent.

[0098] Referring to the above example, when the numerical range involved is 1 to 100, based on the numerical range and the first numerical distribution, the numerical range is evenly divided into 10 different numerical intervals: [1,10], [11,20], [21,30], [31,40], [41,50], [51,60], [61,70], [71,80], [81,90] and [91,100], and the numerical spans corresponding to the ten numerical intervals are all 9. When the numerical range is unevenly divided into 7 different numerical intervals based on the numerical range and the first numerical distribution: [1,21], [31,40], [41,60], [61,70], [71,80], [81,90] and [91,100], it can be seen that different numerical intervals correspond to different numerical spans, among which [1,21] corresponds to a numerical span of 20, [41,60] corresponds to a numerical span of 19, and [31,40], [61,70], [71,80], [81,90] and [91,100] correspond to a numerical span of 9.

[0099] A2: Determine a sampling range for adding noise according to the numerical span.

[0100] The sampling range of added noise can refer to the degree or amplitude of change allowed when the noise is added to the first data. A larger sampling range means that the noise has a wider range of variation, which will have a greater impact on the first data in the numerical set. A smaller sampling range means that the noise has a narrower range of variation, which has a smaller impact on the first data. When the numerical span is larger, the corresponding sampling range is larger. When the added noise is Laplace noise, s can be used to represent the sampling range of the added noise. Lap(S) represents sampling with a Laplace distribution with a mean of 0 and a scaling factor of S. The Laplace distribution is expressed as:

[0101]

[0102] Where b is the scale parameter (when b = 1, the Laplace distribution is the standard Laplace distribution), Figure 3 A schematic diagram of adding a noise sampling range provided in an embodiment of the present application is provided. Figure 3 As shown in the figure, the x-axis is used to represent the value of the sampling range of the added noise, and the y-axis is used to represent the probability density. The sampling ranges of the added noise corresponding to a, b, c, and d decrease in sequence. For example, when a corresponds to b=4, then b, c, and d can correspond to b=3, b=2, and b=1, respectively. It can be seen from the figure that the larger the value of s, the wider the sampling range of the added noise. When b=1, the corresponding sampling range of the added noise is roughly (-6, +6), and when b=2, the corresponding sampling range of the added noise is roughly (-10, +10).

[0103] The larger the value span, the larger the sampling range of the added noise. That is, when the difference between the maximum and minimum values ​​in the value interval is larger, the corresponding b value is also larger. For example, when the value interval includes the value interval A[1,10] and the value interval B[11,30], the sampling range of the added noise corresponding to the value interval A will be smaller than the sampling range of the added noise corresponding to the value interval B. Figure 3 For reference, when the noise-added sampling range corresponding to the numerical interval A corresponds to the case of b=2 in the figure, the noise-added sampling range corresponding to the numerical interval B may correspond to the case of b=3 or b=4 in the figure.

[0104] A3: Using differential privacy, noise is added to the N value sets within the corresponding sampling range to obtain the noisy interaction data.

[0105] After determining the sampling ranges for adding noise corresponding to different numerical intervals based on the numerical spans of the numerical intervals corresponding to the numerical sets, noise can be added to the sampling ranges corresponding to each numerical set using differential privacy. The appropriate sampling ranges for adding noise corresponding to different numerical intervals can be determined based on the numerical spans of the numerical intervals, which can make the addition of noise more rational. For the first data in the same numerical set, adding noise within the same sampling range ensures that the noise-added interactive data obtained after adding noise does not excessively lose the characteristics of the data distribution compared to the original first data corresponding to each numerical set. On the contrary, because specific noise is added to different data sets, it can better ensure that the second numerical distribution of the values ​​in the obtained noisy interactive data is consistent with the first numerical distribution of the values ​​in the corresponding first data.

[0106] It should be noted that since adding noise will cause the values ​​in the original first data to change, this change may cause the first data that originally belonged to the same numerical interval to not belong to the same numerical interval after adding noise based on the sampling range, but to other adjacent numerical intervals. This situation where the numerical intervals corresponding to the noisy interactive data and the corresponding first data are inconsistent after adding noise is not a common phenomenon. Generally speaking, the method of adding noise based on the numerical span to determine the sampling range for adding noise can ensure that the first data in the same numerical interval have a roughly distributed distribution in the noisy interactive data obtained after adding noise.

[0107] When the added noise is Laplace noise, the corresponding formula for obtaining the noisy interaction data can be expressed as: Where F(x) is used to represent the obtained noisy interaction data, f(x) is used to represent the first data, is the added Laplace noise.

[0108] Taking the payment amount of the target object as an example, based on the numerical range of the values ​​in the first data, assume that the numerical range is divided into three different numerical intervals: the first numerical interval [1, 10], the second numerical interval [11, 50], and the third numerical interval [51, 100]. The value of ∈ is 1, and s represents the sampling range of the noise added to the first data. When ∈ = 1, the value of s is related to the numerical span of the numerical interval. The first numerical interval corresponds to s = 10 - 1 = 9, the second numerical interval corresponds to s 39, and the third numerical interval corresponds to s 49.

[0109] There is a corresponding relationship between the numerical interval and the numerical set. The Laplace noise to be added can be determined according to the sampling range of the noise added corresponding to each numerical interval. Specifically, the added noise corresponding to the first numerical interval is Lap(9 / 1), the added noise corresponding to the second numerical interval is Lap(39 / 1), and the added noise corresponding to the third numerical interval is Lap(49 / 1). Finally, the noise-added interaction data corresponding to each numerical set are: the noise-added interaction data corresponding to the first numerical interval is F(x)=f(x)+Lap(9 / 1), the noise-added interaction data corresponding to the second numerical interval is F(x)=f(x)+Lap(39 / 1), and the noise-added interaction data corresponding to the third numerical interval is F(x)=f(x)+Lap(49 / 1).

[0110] Using the aforementioned method for obtaining noisy interaction data, a sampling range for adding noise is determined based on the numerical span corresponding to the numerical interval. A differential privacy algorithm is then used to add noise to the first data based on the determined sampling range to obtain the noisy interaction data. This ensures that the noise addition meets the characteristics of the numerical values ​​in each numerical interval, allowing for specific noise addition based on the values ​​in different numerical intervals. This improves the rationality of the noise addition and prevents the added noise from excessively affecting the original first data.

[0111] It is mentioned above that "according to the numerical range of the numerical values ​​in the first data, the numerical range is divided into N different numerical intervals". In the process of dividing the numerical intervals, since there are cases where the values ​​in some first data are much larger than the values ​​in other first data and the number is small, the numerical intervals divided at this time will have a larger numerical span. When the numerical span is larger, the corresponding sampling range for adding noise is also larger. Then, when adding noise to the numerical set corresponding to the numerical interval, the first data with lower numerical values ​​in the numerical interval will be excessively affected by the noise. Therefore, for the above-mentioned situation where there is a large difference in values, the corresponding method of dividing the numerical interval is: first, when it is determined that the numerical value of the first data is greater than the numerical value threshold, the numerical value threshold is used as the numerical value of the first data, and the numerical range of the numerical value in the first data is determined according to the numerical value threshold. According to the numerical range of the numerical value in the determined first data, the numerical range is divided into N different numerical intervals.

[0112] The above-mentioned data value threshold is used to identify the maximum value in the numerical range involved in the numerical interval. That is to say, when the numerical value in the first data is greater than the numerical value threshold, the numerical value threshold needs to replace the original numerical value of the first data and serve as the numerical value of the first data. For example, assuming the numerical value threshold is 600, then when the numerical value of the first data is 1000, the numerical value threshold 600 will replace the original numerical value 1000 of the first data and serve as the numerical value of the first data. After the above-mentioned processing method, the numerical value of the first data will not exceed 600, thereby avoiding the situation where the numerical differences in the first data are too large.

[0113] When there is a first data whose value is much larger than the value of other first data, the value span of a certain value interval will be too large. The reason is that the number of first data with values ​​much larger than other first data is often small, and in the process of dividing the value interval, the value interval will be expanded for the case where the number of first data is small, so as to reduce the number of divided value intervals and thus improve the efficiency of adding noise. The expansion of the value interval will lead to the situation where the value span is too large, such as [500, 10000]. When the value span is too large, the corresponding sampling range of added noise will be too large. At this time, when adding noise to the value set [500, 10000] based on the sampling range, the corresponding added noise will affect the first data with smaller values ​​in the value set due to its large value, such as the first data with a value of 500, which will then affect the second value distribution of the noisy interaction data obtained after adding noise.

[0114] Therefore, before dividing the numerical range, the numerical values ​​of the first data involved can be counted and processed to determine whether there is any first data with a value greater than the numerical threshold. If it is determined that the value of the first data is greater than the numerical threshold, the numerical threshold is determined to be the value of the first data. Then, the numerical range of the values ​​in the first data needs to be re-determined, and the numerical threshold is now the maximum value in the numerical range.

[0115] It should be noted that the determination of the numerical value threshold can be determined in combination with the overall distribution of the numerical values ​​of the first data. Assuming that after statistics, 95% of the numerical values ​​of the first data are less than 100, then it can be determined that the number of first data greater than 100 is small. At this time, it can be considered to set the numerical value threshold to 100, that is, change the numerical values ​​of all first data greater than 100 to 100.

[0116] After processing the values ​​of the first data, the numerical range of the values ​​in the first data is re-determined based on the numerical threshold, where the maximum value of the numerical range is the numerical threshold. Based on the determined numerical range of the values ​​in the first data, the numerical range can be divided into different numerical intervals. The specific method for dividing the numerical intervals has been described in detail in the previous description and will not be repeated here.

[0117] By using the above-mentioned method of dividing the numerical interval, when it is determined that the numerical value in the first data is greater than the numerical value threshold, the numerical value threshold is used as the numerical value of the first data. This can avoid the numerical value in the first data being too large, so that the numerical span of the numerical interval is too large when the numerical interval is divided, and thus the sampling range of the noise added corresponding to the numerical interval is very large, resulting in the first data with smaller numerical values ​​in the numerical interval being excessively affected by the noise. By "truncating" the values ​​in the first data according to the numerical value threshold, it can be made more reasonable to add noise to the first data, and it can also be avoided to a certain extent that the second numerical distribution of the numerical values ​​in the noisy interactive data and the first numerical distribution of the numerical values ​​in the first data are significantly different. At the same time, by "truncating" the values ​​in the first data with excessive numerical values, target objects with high value (high numerical values ​​in the first data) can be hidden.

[0118] Figure 4 A schematic diagram of a strategy adjustment scenario provided in an embodiment of the present application is shown in FIG. Figure 4As shown, the target software requires targeted advertising by the promotion platform. This advertising can achieve exposure for the target software and, in turn, conversions at the user level. Within the target software, the converted target object will perform an operation (i.e., pay) through the target software to obtain first data. This privacy protection processing involves processing the first data. The specific processing process is as follows: noise is added to the first data using differential privacy to obtain noisy interaction data. Finally, the noisy interaction data obtained after the privacy protection processing is sent to the promotion platform, which allows the promotion platform to adjust its advertising strategy for the target software based on this noisy interaction data.

[0119] In the aforementioned S203, "sending the noisy interaction data to the promotion platform" is mentioned. The noisy interaction data can also include predicted data for the target object in a future time period. The purpose of adding predicted data is to provide the promotion platform with more data and information about the target object, thereby improving the effectiveness of advertising strategy adjustments. At the same time, the addition of new data can make the noisy interaction data more ambiguous. Therefore, in one possible implementation, the predicted data for the target object in the future time period can be first determined based on the target object's account characteristics in the target software. The predicted data is then added to the noisy interaction data, and the noisy interaction data with the added predicted data is sent to the promotion platform.

[0120] The target object's account characteristics may specifically include operational characteristics and information such as the target object's city. The operational characteristics may include but are not limited to the following: the target object's online time, number of logins, level, and task execution status for the target software.

[0121] Based on the aforementioned target object's account characteristics, predicted data for the target object in a future time period can be determined. This predicted data is numerical data used to identify the operations performed by the target object via the target software in the future time period. In other words, based on the target object's account characteristics in the current or historical time period, data on operations performed by the target object on the target software in the future time period can be obtained. Because the predicted data is obtained based on the target object's account characteristics, and the target object's account characteristics have not been subjected to data obfuscation operations such as adding noise, the predicted data is determined based on real data, has a high degree of credibility, and can well match the characteristics of the target object's interactive behavior.

[0122] Adding the predicted data to the noisy interaction data and sending it to the promotion platform together can increase the effective information in the noisy interaction data, and thus improve the guiding role of the sent noisy interaction data in helping the promotion platform adjust its advertising strategy for the target software. At the same time, since the predicted data is added to the original noisy interaction data, the blurring effect on the first data can be further enhanced, thereby improving the privacy protection effect of the first data.

[0123] Through the aforementioned method of sending noisy interaction data to the promotion platform, predicted data for the target object in a future time period is determined based on the target object's account characteristics in the target software. Numerical data representing the operations performed by the target object in this determined future time period is then added to the noisy interaction data and sent to the promotion platform. In this way, because the noisy interaction data obtained by the promotion platform incorporates the predicted data directly determined based on the account characteristics, the predicted data is more accurately aligned with the characteristics of the target object, providing a more effective guide for the promotion platform to adjust its strategies. Furthermore, the addition of the predicted data to the noisy interaction data further obscures the original first data.

[0124] The aforementioned "adding predicted data to the noisy interaction data" is mentioned. Since the amount of predicted data of the target object in the future time period determined based on the account characteristics of the target object is large, if all the determined predicted data are added to the noisy interaction data, then when the promotion platform adjusts the advertising strategy for the target software based on the noisy interaction data, it will face greater data processing pressure, which will in turn affect the efficiency of the strategy adjustment. Therefore, after the predicted data is determined, some of the data can be selected and added to the noisy interaction data for the adjustment of the advertising strategy. In one possible implementation method, the data addition process of the noisy interaction data is: first, the predicted data with a value greater than the associated threshold is selected from the predicted data as the target predicted data, and then the target predicted data is added to the noisy interaction data.

[0125] The correlation threshold can be understood as a threshold for the degree of correlation with the target software's advertising strategy adjustments. It is generally believed that larger values ​​in the predicted data correspond to a greater degree of correlation with the target software's advertising strategy adjustments. In other words, the correlation threshold is related to the values ​​in the predicted data. Once the predicted data are determined, they can be sorted by value. The target predicted data within the predicted data can be determined based on the correlation threshold, and only the target predicted data is added to the noisy interaction data.

[0126] Specifically, the association threshold itself can be a specific value or a percentage. For example, when the association threshold is a specific value, such as 100, all predicted data with values ​​greater than 100 are the target predicted data. When the association threshold is a percentage, such as topk% (where k ranges from 0.05 to 0.2), the predicted data that are ranked in the topk% of the total predicted data are the target predicted data. Assuming the association threshold is top0.2%, 1000 predicted data are calculated and sorted from largest to smallest by value. Since the association threshold is top0.2%, the top two predicted data in the sort are used as the target predicted data. The target predicted data are then added to the noisy interaction data to provide guidance for strategic adjustments to the target software's advertising, thereby improving the effectiveness of these adjustments.

[0127] Through the aforementioned data addition process for the noisy interaction data, after determining the predicted data, the predicted data with a value greater than the correlation threshold is selected as the target predicted data, and then the target predicted data is added to the noisy interaction data. This avoids adding all the determined predicted data to the noisy interaction data, which would result in the promotion platform facing significant data processing pressure when adjusting its advertising strategy for the target software based on the noisy interaction data, thereby affecting the efficiency of the strategy adjustment. At the same time, because the setting of the correlation threshold can ensure a greater correlation between the predicted data added to the noisy interaction data and the strategy adjustment, the effectiveness of the strategy adjustment can be improved.

[0128] In the aforementioned description of "determining predicted data for the target object in a future time period based on the target object's account characteristics in the target software," one possible implementation method can be to use a prediction model to determine the predicted data for the target object in the future time period. Specifically, the determination method includes: using a prediction model to determine predicted data for the target object in the future time period based on the target object's account characteristics in the target software.

[0129] The specific prediction model can be trained by the following method:

[0130] B1: Obtain training samples.

[0131] The training sample includes the account features of the sample account of the target software in the first historical specified period, and the sample label of the training sample is the sample first data of the sample account in the second historical specified period. The end time of the first historical specified period is earlier than the start time of the second historical specified period.

[0132] Specifically, the account characteristics of the sample account may include the account's online time, number of logins, level, location, and transaction flow. The training sample includes the account characteristics of the sample account of the target software in the first historical specified period. Based on the training sample, the initial prediction model needs to be used to predict the sample first data of the corresponding sample account in the second historical specified period. The end time of the first historical specified period is earlier than the start time of the second historical specified period. In other words, the first historical specified period is a more historical period than the second historical specified period. For example, assuming that the first historical specified period is from March 1 to March 5, then the second historical specified period is from March 6 to March 10.

[0133] It should be noted that since both the first and second designated historical time periods are historical time periods, the sample first data corresponding to the target account's sample account within these two designated historical time periods are known. The sample label of the training sample is the sample first data of the sample account in the second historical time period. This sample label can be understood as the reference data for the sample prediction data obtained by the initial prediction model.

[0134] B2: Inputting the training sample into an initial prediction model, training the initial prediction model based on the difference between the sample prediction data obtained by the initial prediction model and the sample label, and obtaining the prediction model.

[0135] When a training sample (account characteristics of a sample account of the target software during a first specified historical period) is input into the initial prediction model, the initial prediction model obtains sample prediction data for the sample account during a second specified historical period based on the account characteristics of the sample account in the training sample during the first specified historical period. The initial prediction model is trained to obtain a prediction model based on the difference between the sample first data of the sample account during the second specified historical period indicated in the sample label of the training sample and the sample prediction data of the sample account during the second specified historical period obtained by the initial prediction model.

[0136] It should be noted that the model structure of the initial prediction model and the prediction model can include three parts: input and embedding layer, feature interaction layer and output layer. The following is a detailed introduction to each part:

[0137] (1) Input and embedding layer

[0138] In the initial prediction model and the prediction model, all account features must first pass through the embedding layer after being input. The function of the embedding layer is to map each account feature i to a low-dimensional dense vector (the vector dimension is emb_size, and the emb_size of the account feature is 32 or 8). For account features involving categories (i.e., categorical account features), each value of this type of account feature needs to be mapped to an embedding vector. For example, the two values ​​of the "gender" feature, male and female, correspond to two embedding vectors respectively. For numerical account features, each account feature is mapped to an embedding vector. Assuming that the value of the account feature is value, the final embedding vector is equal to the embedding vector of the account feature multiplied by value.

[0139] (2) Feature interaction layer

[0140] The embedding vectors obtained by the embedding layer are concatenated and fed into the feature interaction layer to learn the complex interactions between features. For categorical and numerical account features, logistic regression (LR), deep neural networks (DNN), and factorization machines (FM) are used to learn first-order and second-order interactions between features, respectively. Finally, the output vectors of the feature interaction layer are concatenated and fed into the first-order missingness perception layer.

[0141] (3) Output layer

[0142] The output layer obtains the sample prediction data of the training sample based on the first-order and second-order missing perception expressions of the training sample. The output result of the output layer is expressed as three numerical values ​​(p, μ, σ). The typical structure of the output layer is a multi-layer fully connected network. For example, the first layer of the network is a fully connected layer of 64 neural network nodes, the second layer is a fully connected layer of 32 neural network nodes, and the third layer is a fully connected layer of 3 neural network nodes. The outputs of the first two fully connected layers are processed using the linear rectified function (RectifiedLinearUnit, ReLU) activation function. The last layer uses the Sigmoid activation function to obtain the numerical value p, and obtains μ through the identity activation function and σ through the softplus activation function. The sample prediction data predicted by the final model (expressed in LTV) is:

[0143]

[0144] The loss function involved in determining the predicted data is Ziln loss, and the formula is as follows:

[0145]

[0146] The first term is the cross entropy loss function for the probability p of the first data, and the second term is the lognormal loss function. The formula is as follows:

[0147]

[0148] Where μ and σ refer to the predicted distribution mean and standard deviation, and y is the true sample label value.

[0149] The Adam optimizer was used during the training of the initial prediction model. The optimizer uses the Ziln loss function to optimize the model and continuously updates the model parameters of the initial prediction model until the difference between the sample prediction data and the sample label no longer decreases. This results in a fully trained prediction model.

[0150] The above-mentioned method for determining prediction data, which determines prediction data based on a prediction model, can improve the efficiency of determining prediction data. During the prediction model training process, data from different designated historical time periods is used as training samples and reference for sample prediction data. This can, to a certain extent, ensure the effectiveness of the initial prediction model training, thereby improving the accuracy of the predicted data for the target object in the future time period.

[0151] Figure 5 A schematic diagram of another scenario of policy adjustment provided in an embodiment of the present application is shown in FIG. Figure 5 As shown, the target software requires targeted advertising by the promotion platform. Through advertising, the promotion platform can achieve exposure to the target software and thus achieve user-level conversion. In the target software, the converted target object will perform an operation (i.e., pay) through the target software to obtain first data. The target object will be subjected to privacy protection processing. The specific privacy protection processing involves first data processing and prediction data determination. The specific first data processing process is: adding noise to the first data using differential privacy to obtain noisy interaction data. In addition to the need for first data processing to protect the privacy data of the target object, in order to further improve the effectiveness of the noisy interaction data sent to the promotion platform, prediction data determination can also be performed. The specific determination process is: obtaining the account characteristics of the target object, using a prediction model to predict based on the account characteristics to obtain prediction data, and adding the prediction data to the noisy interaction data. Finally, the noisy interaction data is sent to the promotion platform so that the promotion platform can adjust the advertising strategy of the target software based on the noisy interaction data.

[0152] Table 2 shows a comparison of the advertising effects of the data processing method provided by this application and the related technology (i.e., directly sending real data to the promotion platform), as shown in Table 2:

[0153] Table 2. Comparison of advertising effects of this application and related technologies

[0154]

[0155] CPA (Cost Per Action), ROI1 (Return on Investment), and LTV1 (Customer Lifetime Value) are used to measure the revenue the target software generates from the target account. The table above shows that the method for adjusting the advertising strategy for the target software provided in this application does not compromise the effectiveness of advertising compared to related technologies.

[0156] In the aforementioned Figure 1-5 Based on the corresponding embodiments, Figure 6 A schematic diagram of a data processing device provided in an embodiment of the present application is shown, wherein the data processing device 600 includes: an acquisition module 601, an adding module 602, and a sending module 603;

[0157] The acquisition module 601 is used to acquire first data of a target object in a target software, where the first data is numerical data used to identify an operation performed by the target object through the target software;

[0158] The adding module 602 is configured to add noise to the first data in a differential privacy manner to obtain noisy interaction data, wherein a consistency difference between a second numerical distribution of values ​​in the noisy interaction data and a first numerical distribution of values ​​in the first data meets a consistency condition;

[0159] The sending module 603 is used to send the noise-added interaction data to the promotion platform, and the noise-added interaction data is used to adjust the promotion platform's advertising strategy for the target software.

[0160] In a possible implementation, the device is specifically configured to:

[0161] According to the numerical range of the values ​​in the first data, the numerical range is divided into N different numerical intervals, where N>1;

[0162] Dividing the first data into N value sets based on the N value intervals, wherein the value intervals and the value sets correspond one to one, and the value sets include the first data in the same value interval;

[0163] The adding module 602 is specifically used for:

[0164] Noise is added to each of the N value sets in a differential privacy manner to obtain the noisy interaction data, wherein different mechanisms are used to add noise to the first data in different value sets.

[0165] In a possible implementation, the device is specifically configured to:

[0166] According to the numerical range of the numerical values ​​in the first data and the first numerical distribution, the numerical range is divided into N different numerical intervals.

[0167] In a possible implementation, the adding module 602 is specifically configured to:

[0168] Determining the numerical span of the numerical interval corresponding to the numerical set;

[0169] Determine a sampling range for adding noise according to the numerical span, wherein the larger the numerical span, the larger the corresponding sampling range;

[0170] By using a differential privacy method, noise is added to the sampling range corresponding to each of the N value sets to obtain the noisy interaction data.

[0171] In a possible implementation, the device is specifically configured to:

[0172] When it is determined that the value of the first data is greater than a value threshold, the value threshold is used as the value of the first data, and a numerical range of the values ​​in the first data is determined according to the numerical threshold, where the data threshold is used to identify the maximum value in the numerical range involved in the numerical interval;

[0173] According to the determined numerical range of the numerical values ​​in the first data, the numerical range is divided into N different numerical intervals.

[0174] In a possible implementation, the adding module 602 is specifically configured to:

[0175] Adding noise to the first data using a differential privacy method to obtain noisy data to be processed;

[0176] In response to the noisy data to be processed including the target noisy data that does not meet the reasonable numerical condition, the target noisy data is removed from the noisy data to be processed to obtain the noisy interaction data.

[0177] In a possible implementation, the device is specifically configured to:

[0178] determining, based on the account characteristics of the target object in the target software, predicted data of the target object in a future time period, the predicted data being numerical data used to identify operations performed by the target object through the target software in the future time period;

[0179] adding the predicted data to the noisy interaction data;

[0180] The noisy interaction data to which the prediction data is added is sent to a promotion platform.

[0181] In a possible implementation, the device is specifically configured to:

[0182] Selecting prediction data with a value greater than a correlation threshold from the prediction data as target prediction data;

[0183] The target prediction data is added to the noisy interaction data.

[0184] In a possible implementation, the device is specifically configured to:

[0185] Determining predicted data of the target object in a future time period using a prediction model based on the account characteristics of the target object in the target software;

[0186] The prediction model is trained in the following way:

[0187] Obtaining a training sample, the training sample including account features of a sample account of the target software during a first specified historical period, the sample label of the training sample being first sample data of the sample account during a second specified historical period, where the end time of the first specified historical period is earlier than the start time of the second specified historical period;

[0188] The training samples are input into an initial prediction model, and the initial prediction model is trained based on the difference between the sample prediction data obtained by the initial prediction model and the sample labels to obtain the prediction model.

[0189] In one possible implementation, the target object of the device is a new account of the target software, and the new account includes at least one of a target object whose registration duration of the target software is less than a first duration threshold, or a target object whose re-login duration is less than a second duration threshold after the target software has been offline for more than a preset number of days.

[0190] In a possible implementation, the numerical identifier of the first data of the device is at least one of a feature value of feature value transfer performed through target software and the number of interactions performed.

[0191] Through the data processing device provided above, when advertising is delivered to the target software through the promotion platform, the first data obtained through the target software needs to be returned to the promotion platform to guide the promotion platform to adjust the delivery strategy. In order to prevent the real values ​​involved in the first data from being leaked to the promotion platform, while maintaining the overall distribution of the values ​​in the first data to play a guiding role, before returning, noise can be added to the first data in a differential privacy manner to obtain noisy interaction data that is relatively consistent with the first data in terms of numerical distribution. Since the values ​​in the first data are blurred by noise, the account privacy is prevented from being known by the promotion platform. At the same time, since the noisy interaction data is similar to the first data in terms of overall numerical distribution, it can also reflect the operational characteristics of the target object in the target software, so that the promotion platform can also determine the target object of the target software based on the noisy interaction data. The target information is closer to the real interaction information, which can play a correct guidance effect on the advertising delivery strategy.

[0192] The embodiment of the present application further provides a computer device, including a terminal device or a server, in which the aforementioned data processing device can be configured. The computer device is described below with reference to the accompanying drawings.

[0193] If the computer device is a terminal device, see Figure 7 As shown, the embodiment of the present application provides a terminal device, taking a mobile phone as an example:

[0194] Figure 7 The block diagram shows a partial structure of the mobile phone provided by the embodiment of the present application. Figure 7 The mobile phone includes components such as a radio frequency (RF) circuit 1410, a memory 1420, an input unit 1430, a display unit 1440, a sensor 1450, an audio circuit 1460, a wireless fidelity (WiFi) module 1470, a processor 1480, and a power supply 1490. Those skilled in the art will understand that Figure 7 The mobile phone structure shown in the figure does not constitute a limitation to the mobile phone, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0195] The following combination Figure 7 A detailed introduction to the various components of a mobile phone:

[0196] The RF circuit 1410 may be used for receiving and sending signals during information transmission or calls. In particular, after receiving downlink information from the base station, it is sent to the processor 1480 for processing. In addition, the designed uplink data is sent to the base station.

[0197] Memory 1420 can be used to store software programs and modules. Processor 1480 executes the various functional applications and data processing of the mobile phone by running the software programs and modules stored in memory 1420. Memory 1420 may mainly include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as a sound playback function, an image playback function, etc.); the data storage area may store data created based on the use of the mobile phone (such as audio data, a phone book, etc.). In addition, memory 1420 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0198] The input unit 1430 may be configured to receive input digital or character information and generate key signal input related to user settings and function control of the mobile phone. Specifically, the input unit 1430 may include a touch panel 1431 and other input devices 1432 .

[0199] The display unit 1440 may be configured to display information input by the user or information provided to the user, as well as various menus of the mobile phone. The display unit 1440 may include a display panel 1441 .

[0200] The mobile phone may also include at least one sensor 1450, such as a light sensor, a motion sensor, and other sensors.

[0201] The audio circuit 1460 , the speaker 1461 , and the microphone 1462 can provide an audio interface between the user and the mobile phone.

[0202] WiFi is a short-range wireless transmission technology. The mobile phone can help users send and receive emails, browse web pages, and access streaming media through the WiFi module 1470, providing users with wireless broadband Internet access.

[0203] The processor 1480 is the control center of the mobile phone. It uses various interfaces and lines to connect various parts of the entire mobile phone. It executes various functions of the mobile phone and processes data by running or executing software programs and / or modules stored in the memory 1420 and calling data stored in the memory 1420.

[0204] The mobile phone also includes a power supply 1490 (such as a battery) for supplying power to various components.

[0205] In this embodiment, the processor 1480 included in the terminal device is also used to execute the steps in the methods of each embodiment of the present application.

[0206] If the computer device is a server, this embodiment of the application also provides a server, see Figure 8 As shown, Figure 8 The structural diagram of the server 1500 provided in the embodiment of the present application, the server 1500 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPUs) 1522 (for example, one or more processors) and a memory 1532, and one or more storage media 1530 (for example, one or more mass storage devices) for storing application programs 1542 or data 1544. Among them, the memory 1532 and the storage medium 1530 can be temporary storage or permanent storage. The program stored in the storage medium 1530 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Furthermore, the central processing unit 1522 can be configured to communicate with the storage medium 1530 to execute a series of instruction operations in the storage medium 1530 on the server 1500.

[0207] The server 1500 may also include one or more power supplies 1526, one or more wired or wireless network interfaces 1550, one or more input and output interfaces 1558, and / or one or more operating systems 1541, such as Windows Server 2003. TM , Mac OS X TM , Unix TM ,Linux TM , FreeBSD TM etc.

[0208] The steps performed by the server in the above embodiment can be based on Figure 8 The server structure shown.

[0209] In addition, an embodiment of the present application further provides a storage medium, which is used to store a computer program, and the computer program is used to execute the method provided by the above embodiment.

[0210] An embodiment of the present application further provides a computer program product including a computer program, which, when executed on a computer device, enables the computer device to execute the method provided in the above embodiment.

[0211] Those skilled in the art will understand that all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to program instructions, and the above-mentioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the above-mentioned storage medium can be at least one of the following media: read-only memory (English: Read-only Memory, abbreviated: ROM), RAM, magnetic disk or optical disk, etc., various media that can store computer programs.

[0212] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program that has a predetermined function and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories) or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0213] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0214] The above is only one specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the scope of protection of the present application. Moreover, based on the implementation methods provided in the above aspects, the present application can also be further combined to provide more implementation methods. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A data processing method, characterized in that: The method comprises: Acquire first data of a target object in target software, where the first data is numerical data used to identify an operation performed by the target object through the target software; adding noise to the first data in a differential privacy manner to obtain noisy interaction data, wherein a consistency difference between a second numerical distribution of values ​​in the noisy interaction data and a first numerical distribution of values ​​in the first data meets a consistency condition; The noise-added interaction data is sent to a promotion platform, where the noise-added interaction data is used to adjust a strategy for the promotion platform to place advertisements for the target software.

2. The method according to claim 1, characterized in that The method further comprises: According to the numerical range of the values ​​in the first data, the numerical range is divided into N different numerical intervals, where N>1; Dividing the first data into N value sets based on the N value intervals, wherein the value intervals and the value sets correspond one to one, and the value sets include the first data in the same value interval; The adding noise to the first data in a differential privacy manner to obtain noisy interaction data includes: Noise is added to each of the N value sets in a differential privacy manner to obtain the noisy interaction data, wherein different mechanisms are used to add noise to the first data in different value sets.

3. The method according to claim 2, characterized in that The method of dividing the numerical range into N different numerical intervals according to the numerical range of the numerical value in the first data includes: According to the numerical range of the numerical values ​​in the first data and the first numerical distribution, the numerical range is divided into N different numerical intervals.

4. The method according to claim 2, characterized in that The step of adding noise to each of the N value sets in a differential privacy manner to obtain the noisy interaction data includes: Determining the numerical span of the numerical interval corresponding to the numerical set; Determine a sampling range for adding noise according to the numerical span, wherein the larger the numerical span, the larger the corresponding sampling range; By using a differential privacy method, noise is added to the sampling range corresponding to each of the N value sets to obtain the noisy interaction data.

5. The method according to claim 2, characterized in that The method of dividing the numerical range into N different numerical intervals according to the numerical range of the numerical value in the first data includes: When it is determined that the value of the first data is greater than a value threshold, the value threshold is used as the value of the first data, and a numerical range of the values ​​in the first data is determined according to the numerical threshold, where the data threshold is used to identify the maximum value in the numerical range involved in the numerical interval; According to the determined numerical range of the numerical values ​​in the first data, the numerical range is divided into N different numerical intervals.

6. The method according to claim 1, characterized in that The adding noise to the first data in a differential privacy manner to obtain noisy interaction data includes: Adding noise to the first data using a differential privacy method to obtain noisy data to be processed; In response to the noisy data to be processed including the target noisy data that does not meet the reasonable numerical condition, the target noisy data is removed from the noisy data to be processed to obtain the noisy interaction data.

7. The method according to claim 1, characterized in that The method further comprises: determining, based on the account characteristics of the target object in the target software, predicted data of the target object in a future time period, the predicted data being numerical data used to identify operations performed by the target object through the target software in the future time period; adding the predicted data to the noisy interaction data; The noisy interaction data to which the prediction data is added is sent to a promotion platform.

8. The method according to claim 7, characterized in that The adding the predicted data to the noisy interaction data comprises: Selecting prediction data with a value greater than a correlation threshold from the prediction data as target prediction data; The target prediction data is added to the noisy interaction data.

9. The method according to claim 7, characterized in that The determining, based on the account characteristics of the target object in the target software, predicted data of the target object in a future time period includes: Determining predicted data of the target object in a future time period using a prediction model based on the account characteristics of the target object in the target software; The prediction model is trained in the following way: Obtaining a training sample, the training sample including account features of a sample account of the target software during a first specified historical period, the sample label of the training sample being first sample data of the sample account during a second specified historical period, where the end time of the first specified historical period is earlier than the start time of the second specified historical period; The training samples are input into an initial prediction model, and the initial prediction model is trained based on the difference between the sample prediction data obtained by the initial prediction model and the sample labels to obtain the prediction model.

10. The method according to any one of claims 1 to 9, characterized in that: The target object is a new account of the target software, and the new account includes at least one of a target object whose registration duration of the target software is less than a first duration threshold, or a target object whose re-login duration is less than a second duration threshold after the target software has been offline for more than a preset number of days.

11. The method according to any one of claims 1 to 9, characterized in that: The numerical identifier of the first data is at least one of a feature value of feature value transfer performed through target software and the number of interactions performed.

12. A data processing device, characterized in that: The device includes: an acquisition module, an adding module and a sending module; The acquisition module is configured to acquire first data of a target object in the target software, wherein the first data is numerical data used to identify an operation performed by the target object through the target software; The adding module is configured to add noise to the first data in a differential privacy manner to obtain noisy interaction data, wherein a consistency difference between a second numerical distribution of values ​​in the noisy interaction data and a first numerical distribution of values ​​in the first data meets a consistency condition; The sending module is used to send the noise-added interaction data to the promotion platform, and the noise-added interaction data is used to adjust the promotion platform's advertising strategy for the target software.

13. A computer device, characterized in that: The computer device includes a processor and a memory: The memory is used to store computer programs; The processor is configured to execute the method according to any one of claims 1 to 11 according to the computer program.

14. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store a computer program, and when the computer program is executed by a computer device, the computer program implements the method according to any one of claims 1 to 11.

15. A computer program product comprising a computer program, which, when run on a computer device, causes the computer device to perform the method according to any one of claims 1 to 11.

Images