Service pushing method and device, electronic equipment and computer readable storage medium

Abstract

The invention provides a service pushing method and device, electronic equipment, a computer readable storage medium and a computer program product, which can be applied to various scenes such as cloud technology, artificial intelligence, intelligent traffic, auxiliary driving and the like. The method comprises the following steps: acquiring attribution data of a push activity of a target service under a push activity dimension of a first platform and attribution data of the push activity of the target service under a push object dimension of a second platform; according to attribution data under the pushing object dimension of the second platform, respectively determining a prediction effect index of the index prediction table scheme and a prediction effect index of the index prediction model scheme so as to select a target prediction scheme; and performing prediction processing on the attribution data under the pushing activity dimension of the first platform according to the target prediction scheme to obtain a first pushing effect index of the pushing activity on the first platform. According to the method and the device, the pushing effect of the pushing activity can be accurately measured, so that the pushing mode is optimized, and resource waste in the pushing process is avoided.

Description

Technical Field

[0001] The present application relates to computer technology, and in particular to a service push method, device, electronic device, computer-readable storage medium, and computer program product. Background Art

[0002] During the operation of a service, push activities are an important part. By pushing relevant information of the service to a specific platform (such as the Android platform), the purpose of increasing the service popularity can be achieved. For example, for a game service, game advertisements can be pushed on a specific platform to attract users to play.

[0003] In order to facilitate the service provider to measure the push effect of push activities, attribution data is usually obtained after the push. However, there may be rule restrictions on some platforms, and it is impossible to obtain attribution data in the push object dimension (such as the user dimension) from these platforms, resulting in the inability to locate the behavior of the push object, and thus unable to accurately measure the push effect brought by the push activity. If push activities with poor actual push effects are continued to be pushed, resource waste will occur. Summary of the Invention

[0004] The present application provides a service push method, device, electronic device, computer-readable storage medium, and computer program product, which can accurately measure the push effect of push activities, so as to optimize the push method and avoid resource waste during the push process.

[0005] The technical solution of the present application is implemented as follows:

[0006] The present application provides a service push method, including:

[0007] Obtaining attribution data of a push activity of a target service in the push activity dimension of a first platform, and attribution data of the push activity of the target service in the push object dimension of a second platform;

[0008] According to the attribution data in the push object dimension of the second platform, respectively determining a prediction effect index of an index prediction table scheme and a prediction effect index of an index prediction model scheme, and selecting a target prediction scheme with the largest prediction effect index from the index prediction table scheme and the index prediction model scheme; wherein, the index prediction table scheme predicts a push effect index based on log data of the target service; the index prediction model scheme predicts a push effect index based on an index prediction model;

[0009] Performing prediction processing on the attribution data in the push activity dimension of the first platform according to the target prediction scheme to obtain a first push effect index of the push activity on the first platform.

[0010] The present application provides a service push device, including:

[0011] An acquisition module, configured to acquire attribution data of a push activity of a target service under a push activity dimension of a first platform and attribution data of the push activity of the target service under a push object dimension of a second platform;

[0012] A selection module, configured to respectively determine a prediction effect index of an index prediction table scheme and a prediction effect index of an index prediction model scheme according to the attribution data under the push object dimension of the second platform, and select a target prediction scheme with the largest prediction effect index from the index prediction table scheme and the index prediction model scheme; wherein, the index prediction table scheme predicts a push effect index based on log data of the target service; the index prediction model scheme predicts a push effect index based on an index prediction model;

[0013] A prediction module, configured to perform a prediction process on the attribution data under the push activity dimension of the first platform according to the target prediction scheme, and obtain a first push effect index of the push activity on the first platform.

[0014] This application provides an electronic device, including:

[0015] A memory, configured to store executable instructions;

[0016] A processor, configured to implement the service push method provided by this application when executing the executable instructions stored in the memory.

[0017] This application provides a computer-readable storage medium, storing executable instructions, which are used to cause a processor to implement the service push method provided by this application when executed.

[0018] This application provides a computer program product, which includes executable instructions, and is used to cause a processor to implement the service push method provided by this application when executed.

[0019] This application has the following beneficial effects:

[0020] For the push activity of the target service in this application, attribution data under the push activity dimension of the first platform and attribution data under the push object dimension of the second platform are obtained. According to the attribution data under the push object dimension of the second platform, the prediction effect indicators of the index prediction table scheme and the prediction effect indicators of the index prediction model scheme are respectively determined. In this way, when the first platform is restricted by rules while the second platform is not, the second platform is referred to measure which of the index prediction table scheme and the index prediction model scheme has a better prediction effect, so as to select the target prediction scheme with the largest prediction effect indicator from these two schemes. Then, the attribution data under the push activity dimension of the first platform is predicted according to the selected target prediction scheme, and the first push effect indicator of the push activity on the first platform is obtained. Since the target prediction scheme has been verified to have a good prediction effect by referring to the second platform, after the target prediction scheme is used for the first platform, the accuracy of the obtained first push effect indicator can be improved, that is, the push effect of the push activity on the first platform can be accurately measured, which is convenient for optimizing the subsequent push method and avoiding resource waste caused by continuing to push according to the push activity with poor push effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of this application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of this application. For those skilled in the art, without creative efforts, other drawings can be obtained according to these drawings.

[0022] Figure 1 is a schematic architecture diagram of the service push system provided by the embodiment of this application;

[0023] Figure 2 is a schematic structural diagram of the server provided by the embodiment of this application;

[0024] Figure 3 is a schematic flowchart of the service push method provided by the embodiment of this application;

[0025] Figure 4 is a schematic flowchart of the index prediction table scheme provided by the embodiment of this application;

[0026] Figure 5 is a schematic flowchart of the index prediction model scheme provided by the embodiment of this application;

[0027] Figure 6 is another schematic flowchart of the service push method provided by the embodiment of this application;

[0028] Figure 7It is another process schematic diagram of the service push method provided by the embodiments of the present application;

[0029] Figure 8 It is a schematic diagram of the index prediction table provided by the embodiments of the present application;

[0030] Figure 9 It is another process schematic diagram of the service push method provided by the embodiments of the present application;

[0031] Figure 10 It is a schematic diagram of the time correspondence relationship provided by the embodiments of the present application. Detailed implementation manners

[0032] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be construed as limiting the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present application.

[0033] In the following description, "some embodiments" are described, which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. In the following description, the term "a plurality" refers to at least two.

[0034] In the following description, the terms "first / second / third..." are only used to distinguish similar objects and do not represent a specific order for the objects. It can be understood that "first / second / third..." can be interchanged with a specific order or sequence when permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0036] Before further elaborating on the embodiments of the present application, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are applicable to the following explanations.

[0037] 1) Target service: Generally refers to services in various industries. By pushing information related to the target service to a specific platform, purposes such as attracting users and enhancing service popularity can be achieved. For example, the target service can be a service related to an application program, such as a game application program, an instant messaging application program, a video application program, an electronic map application program, etc.

[0038] 2) Push Campaign: Also known as an advertising campaign (Campaign), it refers to the strategy adopted for the target business. For example, a push campaign can include a series of advertising materials. For the business side of the target business (such as an advertiser or an advertising platform), the push campaign is used to reach the push target and encourage the push target to pay attention to the target business.

[0039] 3) Push Target: It refers to the object that the push campaign aims to reach. For example, it can be a specific electronic device or a user account, etc. According to the different platforms to which the push target belongs, the first platform and the second platform can be divided. Among them, there are rule restrictions on the first platform, and it is impossible to obtain attribution data in the dimension of the push target for all or part of the push targets in the first platform; while there are no rule restrictions on the second platform, and it can obtain attribution data in the dimension of the push target. The division rules of the first platform and the second platform in the embodiments of the present application are not limited, and can be divided by the operating system used by the push target. For example, the first platform is the iOS platform, and the second platform is the Android platform; or, it can also be divided by the device type or account type of the push target.

[0040] 4) Attribution Data: It refers to the data for positioning and analyzing the push effect of the push campaign. For example, it can include the behaviors (conversion events) made by the push target under the influence of the push campaign. Attribution data can be used to measure the impact of the push campaign on the push target.

[0041] In the embodiments of the present application, the attribution data includes attribution data in the dimension of the push campaign and attribution data in the dimension of the push target.

[0042] The attribution data in the dimension of the push campaign refers to the attribution data statistically calculated with the push campaign as the statistical caliber (due to the rule restrictions of the first platform, the first platform cannot use the push target as the statistical caliber); or it can be understood as the attribution data statistically calculated on the premise of ignoring the push target itself. In the statistical process of the attribution data in the dimension of the push campaign, only the behaviors made by the push target under the influence of the push campaign are concerned, and it is not concerned about which push target specifically makes the behavior, so the information of the push target itself (such as the identifier of the push target) cannot be statistically obtained. For example, in the attribution data in the dimension of the push campaign, at least the identifier of the push campaign and the behaviors made by the push target under the influence of the push campaign are included, but the identifier of the push target is not included, resulting in the inability to position the behavior of each push target from the perspective of each push target.

[0043] The attribution data under the push object dimension refers to the attribution data statistically calculated with the push object as the statistical caliber; or it can be understood as the attribution data statistically calculated on the premise of considering the push object itself. During the statistical process of the attribution data under the push object dimension, it not only focuses on the behaviors of the push object under the influence of the push activity, but also on which specific push object makes the behavior. Therefore, information about the push object itself (such as the identifier of the push object) can be statistically obtained. For example, the attribution data under the push object dimension includes at least the identifier of the push activity, the identifier of the push object, and the behaviors of the push object under the influence of the push activity, that is, it can locate the behaviors of each push object and obtain accurate push effect indicators.

[0044] 5) Push effect indicator: An indicator used to measure the push effect, such as retention, payment, etc., which can be set according to the actual application scenario. In the embodiments of this application, one push effect indicator can be predicted, or multiple push effect indicators can be predicted simultaneously.

[0045] 6) Conversion value: It represents the number of conversion events executed by the push object after being affected by the push activity (such as clicking on an advertisement). Among them, conversion events such as registration, purchase, upgrade, etc. need to be pre-configured, and the conversion events need to be associated with the push activity.

[0046] 7) Indicator prediction table solution: A solution defined in the embodiments of this application for predicting push effect indicators based on an indicator prediction table, where the indicator prediction table is constructed based on the log data of the target business.

[0047] 8) Indicator prediction model solution: A solution defined in the embodiments of this application for predicting recommendation effect indicators based on an indicator prediction model, where the indicator prediction model can be constructed based on machine learning techniques in artificial intelligence.

[0048] Artificial Intelligence (AI) uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, and is a theory, method, technology, and application system that perceives the environment, acquires knowledge, and uses knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science. It attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a way similar to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines to enable the machines to have the functions of perception, reasoning, and decision-making.

[0049] Artificial intelligence technology is an interdisciplinary subject with a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, pre-trained model technology, operation / interaction systems, mechatronics, etc. Among them, the pre-trained model, also known as the large model or the basic model, can be widely applied to downstream tasks in various directions of artificial intelligence after fine-tuning. Artificial intelligence software technology mainly includes several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0050] Machine Learning (ML) is an interdisciplinary subject that involves multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve their own performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent, and its applications cover all fields of artificial intelligence. Machine learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and rote learning. The pre-trained model is the latest development result of deep learning, which integrates the above technologies.

[0051] In the embodiments of this application, machine learning technology can be used to learn the hidden data patterns in the attribution data to obtain an indicator prediction model, so as to accurately predict the push effect indicators according to the indicator prediction model.

[0052] The embodiments of this application provide a service push method, device, electronic device, computer-readable storage medium, and computer program product, which can accurately measure the push effect of push activities, facilitate the optimization of the push method, and thus avoid resource waste during the push process. The following describes the exemplary applications of the electronic device provided in the embodiments of this application. The electronic device provided in the embodiments of this application can be implemented as various types of terminal devices or as a server.

[0053] See Figure 1 , Figure 1 is a schematic architecture diagram of the service push system 100 provided in the embodiments of this application. The terminal device 400-1 in the first platform (there may be more terminal devices in the first platform actually, Figure 1 only for example) and the terminal device 400-2 in the second platform (there may be more terminal devices in the second platform actually, Figure 1For example (only for illustration), the server 200 is connected via the network 300-1, and the terminal device 400-3 is connected to the server 200 via the network 300-2. The server 200 is connected to the database 500. Among them, the network (referring to the network 300-1 or the network 300-2) can be a wide area network, a local area network, or a combination of the two.

[0054] In some embodiments, taking the electronic device as the terminal device as an example, the service push method provided in the embodiments of the present application can be implemented by the terminal device. For example, the terminal device 400-3 can be the terminal device held by the service provider of the target service. The terminal device 400-3 obtains the attribution data of the push activity of the target service in the push activity dimension of the first platform and the attribution data of the push activity of the target service in the push object dimension of the second platform; according to the attribution data in the push object dimension of the second platform, respectively determine the prediction effect indicators of the index prediction table scheme and the prediction effect indicators of the index prediction model scheme, and select the target prediction scheme with the largest prediction effect indicator among the index prediction table scheme and the index prediction model scheme. Among them, the index prediction table scheme and the index prediction model scheme can be constructed by the server 200 and deployed to the local of the terminal device 400-3, or can be constructed locally on the terminal device 400-3; according to the target prediction scheme, perform prediction processing on the attribution data in the push activity dimension of the first platform to obtain the first push effect indicator of the push activity on the first platform. For the obtained first push effect indicator, the terminal device 400-3 can display it so that the service provider can optimize the push method according to the displayed first push effect indicator.

[0055] In some embodiments, taking the electronic device as the server as an example, the service push method provided in the embodiments of the present application can also be implemented by the server. For example, the server 200 obtains the attribution data of the push activity of the target service in the push activity dimension of the first platform and the attribution data of the push activity of the target service in the push object dimension of the second platform; according to the attribution data in the push object dimension of the second platform, respectively determine the prediction effect indicators of the index prediction table scheme and the prediction effect indicators of the index prediction model scheme, and select the target prediction scheme with the largest prediction effect indicator among the index prediction table scheme and the index prediction model scheme. Among them, the log data of the target service on which the index prediction table scheme depends can be obtained by the server 200 from the database 500; according to the target prediction scheme, perform prediction processing on the attribution data in the push activity dimension of the first platform to obtain the first push effect indicator of the push activity on the first platform.

[0056] In some embodiments, the service push method provided by the embodiments of the present application can also be implemented in cooperation with a server and a terminal device. For example, after obtaining the first push effect index, the server 200 can send the first push effect index to the terminal device 400-3 so that the terminal device 400-3 can display the first push effect index.

[0057] In some embodiments, a terminal device or a server can obtain attribution data through the transfer of a third-party attribution platform. For example, the third-party attribution platform first obtains the attribution data and then sends the attribution data to the terminal device 400-3 or the server 200. The attribution data here includes at least one of the attribution data under the push activity dimension of the first platform and the attribution data under the push object dimension of the second platform.

[0058] In some embodiments, a terminal device or a server can implement the service push method provided by the embodiments of the present application by running a computer program. For example, the computer program can be a native program or software module in an operating system; it can be a local (Native) application program (APP, Application), that is, a program that needs to be installed in the operating system to run; it can also be a small program, that is, a program that only needs to be downloaded to a browser environment to run; it can also be a small program that can be embedded in any APP. In short, the above computer program can be any form of application program, module or plug-in.

[0059] In some embodiments, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms. The terminal device can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart voice interaction device, a smart home appliance, a vehicle-mounted terminal, an aircraft, etc., but is not limited thereto. The terminal device and the server can be directly or indirectly connected through wired or wireless communication methods, which are not limited in the embodiments of the present application.

[0060] The embodiments of the present application can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, intelligent transportation, assisted driving, etc.

[0061] Taking the case where the electronic device is a server as an example, see Figure 2 , Figure 2 which is a schematic structural diagram of the server 200 provided by the embodiments of the present application, Figure 2The server 200 shown includes: at least one processor 210, a memory 250, and at least one network interface 220. Each component in the server 200 is coupled together through a bus system 240. It can be understood that the bus system 240 is used to enable connection and communication between these components. In addition to a data bus, the bus system 240 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clear illustration, in Figure 2 all kinds of buses are labeled as the bus system 240.

[0062] The processor 210 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0063] The memory 250 can be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memories, hard disk drives, optical disc drives, etc. Optionally, the memory 250 includes one or more storage devices that are physically remote from the processor 210.

[0064] The memory 250 includes volatile memory or non-volatile memory, and can also include both volatile and non-volatile memory. The non-volatile memory can be a read-only memory (ROM), and the volatile memory can be a random access memory (RAM). The memory 250 described in the embodiments of the present application is intended to include any suitable type of memory.

[0065] In some embodiments, the memory 250 is capable of storing data to support various operations. Examples of such data include programs, modules, and data structures, or subsets or supersets thereof, which are described below by way of example.

[0066] An operating system 251, including system programs for handling various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and handling hardware-based tasks;

[0067] A network communication module 252, for reaching other computing devices via one or more (wired or wireless) network interfaces 220. Exemplary network interfaces 220 include: Bluetooth, Wi-Fi (Wireless Fidelity), and Universal Serial Bus (USB), etc.;

[0068] In some embodiments, the service push device provided by the embodiments of the present application may be implemented in software. Figure 2 FIG. shows a service push device 255 stored in the memory 250, which may be software in the form of a program and a plug-in, etc., including the following software modules: an acquisition module 2551, a selection module 2552, and a prediction module 2553. These modules are logical, so they can be combined arbitrarily or further split according to the functions to be implemented. The functions of each module will be described below.

[0069] The service push method provided by the embodiments of the present application will be described in combination with the exemplary applications and implementations of the electronic devices provided by the embodiments of the present application.

[0070] See Figure 3 , Figure 3 is a flowchart of the service push method provided by the embodiments of the present application, which will be described in combination with Figure 3 the steps shown.

[0071] In step 101, obtain the attribution data of the push activity of the target service under the push activity dimension of the first platform, and the attribution data of the push activity of the target service under the push object dimension of the second platform.

[0072] Here, after the push processing of the push activity of the target service on the first platform and the second platform, obtain the attribution data of the push activity under the push activity dimension of the first platform, and the attribution data of the push activity under the push object dimension of the second platform. Among them, the push processing may refer to displaying the relevant information of the target service according to the push activity on the corresponding platform. For example, if the push activity includes several advertising materials related to the target service, then several advertising materials in the push activity may be displayed on the corresponding platform.

[0073] There are rule restrictions on the above-mentioned first platform, and it is impossible to obtain the attribution data under the push object dimension for all or part of the push objects in the first platform. For example, all the attribution data obtained from the first platform are the attribution data under the push activity dimension; or, the attribution data obtained from the first platform includes the attribution data under the push activity dimension and the attribution data under the push object dimension. Among them, the push objects corresponding to the attribution data under the push object dimension are not restricted by the rules of the first platform. For example, if the first platform is the iOS platform, then for the push objects (i.e., iOS devices) in the iOS platform that have closed the Advertising Identifier (IDFA) function, the obtained attribution data is the attribution data under the push activity dimension; for the push objects (i.e., iOS devices) in the iOS platform that have enabled the IDFA function, the obtained attribution data is the attribution data under the push object dimension.

[0074] The above-mentioned second platform has no rule restrictions. For all push objects in the second platform, attribution data at the push object dimension can be obtained.

[0075] The above-mentioned first platform and second platform both include multiple push objects, and the push objects in the first platform are different from those in the second platform. The division rules of the first platform and the second platform in the embodiments of the present application are not limited. It can be divided by the operating system used by the push object. For example, the first platform is the iOS platform and the second platform is the Android platform; alternatively, it can also be divided by the device type or account type of the push object.

[0076] It should be noted that attribution data refers to data for positioning and analyzing the push effect of a push activity. For example, it can include the behaviors (conversion events) of the push object under the influence of the push activity. Attribution data can be used to measure the influence degree of the push activity on the push object. In the embodiments of the present application, attribution data includes attribution data at the push activity dimension and attribution data at the push object dimension. In the attribution data at the push activity dimension, it at least includes the identifier of the push activity and the behaviors of the push object under the influence of the push activity (which can be reflected by the conversion value), and can also include information such as time (referring to the time when the attribution data is generated), region (referring to the region where the push object is located), and push channel, but does not include the identifier of the push object, resulting in the inability to associate the attribution data at the push activity dimension with specific push objects, and thus unable to locate the behaviors of each push object; in the attribution data at the push object dimension, it at least includes the identifier of the push activity, the identifier of the push object, and the behaviors of the push object under the influence of the push activity, that is, it can locate the behaviors of each push object and obtain accurate push effect indicators. Among them, the attribution data at the push object dimension can also include information such as time, region, and push channel. Generally speaking, the attribution data at the push object dimension can ensure accuracy and can parse accurate push effect indicators from it; while for the attribution data at the push activity dimension, it is difficult to parse accurate push effect indicators according to the solutions provided by related technologies.

[0077] In some embodiments, self-built attribution can be performed, that is, an attribution platform is independently built, and attribution data is obtained from the first platform or the second platform through this attribution platform; alternatively, attribution data forwarded by a third-party attribution platform can be obtained. Among them, data transmission with the third-party attribution platform can be realized by integrating the software development kit (SDK) of the third-party attribution platform or through the docking method of the application programming interface (API).

[0078] In step 102, according to the attribution data under the push object dimension of the second platform, the prediction effect indicators of the index prediction table scheme and the prediction effect indicators of the index prediction model scheme are respectively determined, and the target prediction scheme with the largest prediction effect indicator is selected from the index prediction table scheme and the index prediction model scheme; wherein, the index prediction table scheme predicts the push effect indicator based on the log data of the target business; the index prediction model scheme predicts the push effect indicator based on the index prediction model.

[0079] In the embodiment of the present application, two schemes for predicting the push effect indicator are constructed. The first is the index prediction table scheme. The index prediction table scheme constructs an index prediction table based on the log data of the target business, and then predicts the push effect indicator through the index prediction table. Among them, the log data of the target business includes the identifier of each object participating in the target business and the corresponding behavior. Taking the target business as a game application as an example, the log data of the game application includes the identifier of each terminal device (or user account) that installs the game and the corresponding behavior. Although it is impossible to determine whether the object recorded in the log data is converted from a push activity (i.e., whether it is a push object), and which push activity the recorded object is specifically converted from, the data law corresponding to the push effect indicator reflected in the log data can be used as a reference. Therefore, an index prediction table (including the data law corresponding to the push effect indicator) can be constructed based on the log data, and the push effect indicator corresponding to the push activity can be predicted through the index prediction table.

[0080] It should be noted that the type of the push effect indicator in the embodiment of the present application is not limited. For example, it can be an indicator related to retention and payment, and can be set according to the actual application scenario. For the index prediction table scheme, the push effect indicator to be predicted can be one or more, and the same is true for the index prediction model scheme described later.

[0081] The second is the index prediction model scheme, that is, the push effect indicator corresponding to the push activity is predicted through the index prediction model. Among them, the index prediction model can be constructed based on the principle of machine learning. For example, considering that the attribution data under the push object dimension of the second platform can ensure accuracy and the accurate push effect indicator can be parsed from it, therefore, the training samples can be determined according to the attribution data under the push object dimension of the second platform, the index prediction model is trained according to the training samples, and then the push effect indicator corresponding to the push activity is predicted according to the trained index prediction model. The type of the index prediction model in the embodiment of the present application is not limited. For example, it can be an eXtreme Gradient Boosting (XGBoost) model or a deep neural network model, etc.

[0082] For the two constructed solutions, using the attribution data under the push object dimension of the second platform, respectively determine the prediction effect indicators of the indicator prediction table solution and the prediction effect indicators of the indicator prediction model solution. Among them, the prediction effect indicators are used to measure the quality of the prediction effect of the corresponding solution. For example, the larger the prediction effect indicator, the better the prediction effect. For the convenience of understanding, the following examples will be given in this case. Of course, this does not constitute a limitation on the embodiments of the present application. After determining the prediction effect indicators corresponding to the two solutions respectively, select the solution corresponding to the largest prediction effect indicator as the target prediction solution, that is, the target prediction solution is the solution with the best prediction effect among the two solutions.

[0083] In some embodiments, the above-mentioned method of respectively determining the prediction effect indicators of the indicator prediction table solution and the prediction effect indicators of the indicator prediction model solution according to the attribution data under the push object dimension of the second platform, and selecting the target prediction solution with the largest prediction effect indicator from the indicator prediction table solution and the indicator prediction model solution can be implemented in the following way: When the running duration of the target service reaches the duration threshold, according to the attribution data under the push object dimension of the second platform, respectively determine the prediction effect indicators of the indicator prediction table solution and the prediction effect indicators of the indicator prediction model solution, and select the target prediction solution with the largest prediction effect indicator from the indicator prediction table solution and the indicator prediction model solution.

[0084] For the indicator prediction model solution, a large amount of training is required to ensure that the indicator prediction model solution has a good prediction effect, that is, sufficient training samples are needed. However, when the running duration of the target service does not reach the duration threshold, due to the insufficient number of training samples, the training effect of the indicator prediction model is not good enough and the stability is poor. At this time, the indicator prediction table solution can be directly used as the target prediction solution; when the running duration of the target service reaches the duration threshold, the indicator prediction model has been fully trained. Therefore, according to the attribution data under the push object dimension of the second platform, respectively determine the prediction effect indicators of the indicator prediction table solution and the prediction effect indicators of the indicator prediction model solution, and select the target prediction solution with the largest prediction effect indicator from the indicator prediction table solution and the indicator prediction model solution. Among them, the duration threshold can be set according to the actual application scenario, such as set to 2 weeks. Through the above method, different methods are used to determine the target prediction solution at different stages of the operation of the target service, which can improve the prediction accuracy as a whole.

[0085] In step 103, according to the target prediction solution, perform prediction processing on the attribution data under the push activity dimension of the first platform to obtain the first push effect indicator of the push activity on the first platform.

[0086] Since the target prediction solution has a good prediction effect on the second platform, it can be inferred that the target prediction solution also has a good prediction effect on the first platform. Therefore, the attribution data under the push activity dimension of the first platform is predicted and processed according to the target prediction solution, and the first push effect index of the push activity on the first platform is obtained. Thus, the accuracy of the obtained first push effect index can be improved.

[0087] For the obtained first push effect index, it can be presented to the business party so that the business party can optimize the push method. For example, adjust the advertising materials in the push activity; or, determine whether to continue pushing according to the push activity; or, in the case of obtaining the first push effect indexes corresponding to multiple push activities respectively, select and retain several push activities with the largest first push effect indexes (that is, continue to push according to these push activities) and discard other push activities. In this way, the business party can obtain better push effects at the same cost or even lower cost, effectively avoid waste of resources in the push process, and contribute to the good operation of the target business.

[0088] In some embodiments, between any steps, the business push method further includes: obtaining the attribution data under the push object dimension of the push activity on the first platform; determining the second push effect index of the push activity on the first platform according to the attribution data under the push object dimension of the first platform; after predicting and processing the attribution data under the push activity dimension of the first platform according to the target prediction solution to obtain the first push effect index of the push activity on the first platform, the business push method further includes: performing a merging process on the first push effect index and the second push effect index to obtain the merged push effect index of the push activity on the first platform.

[0089] In addition to obtaining the attribution data under the push activity dimension of the push activity on the first platform, in the embodiments of the present application, it is also possible to obtain the attribution data under the push object dimension of the push activity on the first platform. In this case, since the attribution data under the push object dimension of the first platform includes the identifiers of the push objects, the behaviors of each push object can be accurately located, and the second push effect index of the push activity on the first platform can be obtained.

[0090] Then, the first push effect index and the second push effect index of the push activity on the first platform can be merged to obtain the merged push effect index of the push activity on the first platform. In this way, the merged push effect index of the push activity on the first platform can comprehensively measure the push effect of the push activity on the first platform, which is convenient for the business party to optimize the push method for the first platform. Among them, the merging method depends on the type of the push effect index. For example, the merging process can be an addition process.

[0091] In some embodiments, between any steps, the business push method further includes: obtaining attribution data of a push activity in the push object dimension of a first platform; determining a second push effect indicator of the push activity in the first platform according to the attribution data in the push object dimension of the first platform; determining a third push effect indicator of the push activity in a second platform according to the attribution data in the push object dimension of the second platform; after performing a prediction process on the attribution data in the push activity dimension of the first platform according to a target prediction scheme to obtain a first push effect indicator of the push activity in the first platform, the business push method further includes: performing a merging process on the first push effect indicator, the second push effect indicator, and the third push effect indicator to obtain a merged push effect indicator of the push activity in the first platform and the second platform.

[0092] In addition to determining the second push effect indicator of the push activity in the first platform as described above, it is also possible to locate the behaviors of each push object according to the attribution data in the push object dimension of the first platform to obtain a third push effect indicator of the push activity in the second platform. Then, the first push effect indicator, the second push effect indicator, and the third push effect indicator can be merged to obtain a merged push effect indicator of the push activity in the first platform and the second platform. In this way, the merged push effect indicator of the push activity in the first platform and the second platform can overall measure the push effect of the push activity on all platforms (referring to the first platform and the second platform), facilitating the business side to optimize subsequent push methods. For example, if the push effect of a certain push activity in the first platform is better, then the push according to this push activity in the second platform can be stopped, and the push frequency according to this push activity in the first platform can be increased.

[0093] Such as Figure 3As shown, in the embodiments of the present application for the push activity of the target service, attribution data under the push activity dimension of the first platform and attribution data under the push object dimension of the second platform are obtained. According to the attribution data under the push object dimension of the second platform, the prediction effect indicators of the index prediction table scheme and the prediction effect indicators of the index prediction model scheme are respectively determined. In this way, when the first platform is restricted by rules while the second platform is not restricted by rules, the second platform is referred to measure which prediction effect is better between the index prediction table scheme and the index prediction model scheme, so as to select the target prediction scheme from these two schemes. Then, according to the selected target prediction scheme, prediction processing is performed on the attribution data under the push activity dimension of the first platform to obtain the first push effect indicator of the push activity on the first platform. Since the target prediction scheme has been verified to have a good prediction effect with reference to the second platform, after the target prediction scheme is used for the first platform, the accuracy of the obtained first push effect indicator can be improved, that is, the push effect of the push activity on the first platform can be accurately measured, which is convenient for optimizing the subsequent push method and avoiding resource waste caused by continuing to push according to the push activity with poor push effect.

[0094] In some embodiments, refer to Figure 4 , Figure 4 is a schematic flowchart of an index prediction table scheme provided by an embodiment of the present application, which will be described in conjunction with Figure 4 the steps shown.

[0095] In step 201, log data of the target service is obtained; wherein, the log data and the target attribution data are from the same platform; the target attribution data is attribution data under the push activity dimension of the first platform or attribution data under the push object dimension of the second platform.

[0096] Here, an example implementation manner of the index prediction table scheme is introduced. For the sake of easy understanding, an example is given for the case of performing prediction processing on the target attribution data according to the index prediction table scheme, wherein the target attribution data is attribution data under the push activity dimension of the first platform or attribution data under the push object dimension of the second platform.

[0097] First, log data of the target service is obtained. Considering that there may be differences in the data rules corresponding to the push effect indicators in different platforms, the log data obtained here and the target attribution data are from the same platform. For example, the target service may be a game application, which has an iOS end (corresponding to the iOS platform) and an Android end (corresponding to the Android platform). If the target attribution data comes from the iOS end, the log data of the game application on the iOS end is obtained.

[0098] It should be noted that the statistical period of the log data in the embodiments of the present application is not limited. For the convenience of understanding, the log data of each day (i.e., the statistical period is one day) is taken as an example for description hereinafter.

[0099] In step 202, an index prediction table is constructed according to the log data; wherein, the index prediction table includes the ratio of the number of objects corresponding to the push effect index.

[0100] Since the obtained log data includes the identifiers of the objects and the corresponding behaviors, the behaviors of each object can be effectively located to construct an index prediction table, and the index prediction table includes the ratio of the number of objects corresponding to the push effect index.

[0101] It should be noted that the push effect index involved in the embodiments of the present application has hysteresis. The push effect index describes the situation after a preset duration from a certain time on the premise that the push has been performed according to the push activity. For example, the push effect index can be the number of retained users on day 7 (i.e., the number of retained users after 7 days, the same hereinafter), etc. Based on this, the ratio of the number of objects refers to the ratio between the number of objects after a preset duration from a certain time and the number of objects at that time. Taking the push effect index as the number of retained users on day 7 as an example, if the number of retained users obtained from the log data on January 1 is 1000, and the number of retained users on January 7 is 300, then the ratio of the number of objects (here is the retention rate) can be determined as 300 / 1000 = 0.3. The determined ratio of the number of objects reflects the data law corresponding to the push effect index.

[0102] In step 203, the number of push objects in the target attribution data is weighted according to the ratio of the number of objects in the index prediction table to obtain the push effect index of the push activity on the target platform; wherein, the target platform is the platform where the target attribution data comes from.

[0103] Here, the log data corresponds to all objects in the target platform, and the target attribution data is generated by the push objects in the target platform. Since the push objects in the target platform are a subset of all objects in the target platform, the above-mentioned ratio of the number of objects can be applied to the target attribution data to predict the push effect index of the push activity on the target platform, wherein the target platform is the platform where the target attribution data comes from.

[0104] For example, the number of push objects in the target attribution data can be weighted according to the object quantity ratio in the index prediction table. Here, the weighting process can be understood as a multiplication process, that is, the object quantity ratio in the index prediction table * the number of push objects in the target attribution data = the push effect index of the push activity on the target platform, where the symbol * represents the multiplication operation. Taking the retention number as an example of the push effect index, if the object quantity ratio in the index prediction table is 0.3 and the number of push objects in the target attribution data is 20 people, then the push effect index of the push activity on the target platform can be obtained as 0.3 * 20 = 6 people.

[0105] In some embodiments, the index prediction table includes the object quantity ratios respectively corresponding to multiple conversion value ranges; the multiple conversion value ranges are obtained by bucketing multiple conversion values; the above-mentioned weighting process of the number of push objects in the target attribution data according to the object quantity ratio in the index prediction table to obtain the push effect index of the push activity on the target platform can be achieved in such a way: according to the object quantity ratios respectively corresponding to the multiple conversion value ranges in the index prediction table, perform a weighted summation process on the number of push objects respectively corresponding to the multiple conversion value ranges in the target attribution data to obtain the push effect index of the push activity on the target platform.

[0106] Here, for different objects, the behaviors performed after participating in the target business may be different. Therefore, the behaviors performed by the objects after participating in the target business can be quantified into specific values, that is, conversion values, to facilitate reflecting the above differences. For example, it is set that the conversion value 0 represents only installation, and the conversion value 1 represents only registration. For multiple objects with the same conversion value, their performances in terms of the push effect index are the same. Thus, when constructing the index prediction table based on the log data, the object quantity ratio can be calculated separately for each type of conversion value, that is, the obtained index prediction table includes the object quantity ratios respectively corresponding to multiple conversion values.

[0107] Similarly, the target attribution data includes the number of push objects corresponding to multiple conversion values. The number of push objects corresponding to multiple conversion values in the target attribution data can be weighted and summed according to the object quantity ratios corresponding to multiple conversion values in the index prediction table, so as to obtain the push effect index of the push activity on the target platform. For example, two conversion values, 0 and 1, are preset. The object quantity ratio corresponding to the conversion value 0 in the index prediction table is 0.05, and the object quantity ratio corresponding to the conversion value 1 is 0.2. The number of push objects corresponding to the conversion value 0 in the target attribution data is 60, and the number of push objects corresponding to the conversion value 1 is 30. Then, the push effect index of the push activity on the target platform can be obtained as 0.05 * 60 + 0.2 * 30 = 9 people. The above method further classifies and analyzes multiple objects. Based on the assumption that the performances of multiple objects with the same conversion value in terms of the push effect index are the same, the accuracy of predicting the push effect index can be further improved.

[0108] On this basis, the more conversion values are set, the more likely the long-tail problem will occur, that is, the number of objects corresponding to some conversion values is extremely small, resulting in little reference value. For the long-tail problem, multiple conversion values can be bucketed to obtain multiple conversion value ranges (i.e., buckets), and each conversion value range includes at least one conversion value. Among them, the rules for bucketing are not limited and can be specifically set according to the number of objects corresponding to each conversion value in the actual application scenario. Based on this, the index prediction table includes the object quantity ratios corresponding to multiple conversion value ranges, and the target attribution data includes the number of push objects corresponding to multiple conversion value ranges. The number of push objects corresponding to multiple conversion value ranges in the target attribution data can be weighted and summed according to the object quantity ratios corresponding to multiple conversion value ranges in the index prediction table, so as to obtain the push effect index of the push activity on the target platform. Through the above method, it can be ensured that the number of objects corresponding to each conversion value range will not be too small, so that the determined object quantity ratio has a certain reference value, thereby improving the accuracy of predicting the push effect index.

[0109] In some embodiments, the above-mentioned construction of the metric prediction table based on log data can be achieved in the following way: construct a metric prediction table based on the log data at the first time and the log data at the second time; wherein, the metric prediction table includes the object quantity ratio between the quantity of the second objects corresponding to the push effect metric and the quantity of the first objects; the quantity of the first objects is determined based on the log data at the first time, and the quantity of the second objects is determined based on the log data at the second time; the second time is later than the first time; the above-mentioned weighted processing of the push object quantity in the target attribution data according to the object quantity ratio in the metric prediction table can be achieved in the following way: perform weighted processing on the push object quantity in the target attribution data at the first time according to the object quantity ratio in the metric prediction table to obtain the push effect metric of the push activity on the target platform at the second time.

[0110] Here, time may affect the performance of objects in terms of the push effect metric. Therefore, the accuracy of predicting the push effect metric can be improved by constraining the consistency between the time involved in the metric prediction table and the time involved in the prediction process.

[0111] For example, a metric prediction table can be constructed based on the log data at the first time and the log data at the second time. Among them, the metric prediction table includes the object quantity ratio between the quantity of the second objects corresponding to the push effect metric and the quantity of the first objects. The quantity of the first objects is determined based on the log data at the first time, and the quantity of the second objects is determined based on the log data at the second time. Then, perform weighted processing on the push object quantity in the target attribution data at the first time according to the object quantity ratio in the metric prediction table to obtain the push effect metric of the push activity on the target platform at the second time. It should be noted that the push effect metric at the second time here is used to describe the push effect of the push activity at the second time.

[0112] Taking the push effect metric as the number of day7 retained users as an example, if the number of retained users on January 1 (the first time) recorded in the log data on January 1 is 1000 people (the quantity of the first objects), and the number of retained users on January 7 (the second time) recorded in the log data on January 7 is 300 people (the quantity of the second objects), then the object quantity ratio can be determined as 300 / 1000 = 0.3. On this basis, if the number of push objects in the target attribution data on January 1 is 20 people, then the number of day7 retained users (i.e., the number of retained users on January 7) of the push activity on the target platform can be obtained as 0.3 * 20 = 6 people. Through the above method, the push effect of the push activity at the second time can be described more accurately. Moreover, the first time and the second time can be set according to the actual application scenario, with high flexibility, and can support predicting the push effect of the push activity on any day (it can also be other time units with finer or coarser granularity).

[0113] As Figure 4 shown, in the embodiment of the present application, an index prediction table is constructed based on log data, and the number of push objects in the target attribution data is weighted according to the object quantity ratio in the index prediction table to obtain the push effect index of the push activity on the target platform, which can apply the data law corresponding to the push effect index in the log data to the target attribution data and improve the accuracy of predicting the push effect index.

[0114] In some embodiments, referring to Figure 5 , Figure 5 is a flowchart of an index prediction model solution provided by an embodiment of the present application, and will be described in conjunction with the steps shown in Figure 5 .

[0115] In step 301, an index prediction model is trained according to the attribution data under the push object dimension of the second platform.

[0116] Here, an example implementation manner of the index prediction model solution is introduced. For the sake of easy understanding, an example is given for the case of predicting and processing the target attribution data according to the index prediction model solution, where the target attribution data is the attribution data under the push activity dimension of the first platform or the attribution data under the push object dimension of the second platform.

[0117] In the index prediction model solution, the target attribution data can be predicted and processed through the index prediction model to obtain the push effect index of the push activity on the target platform. To ensure the accuracy of predicting the push effect index, the index prediction model can be trained first. Since the attribution data under the push object dimension of the second platform includes the identifiers of the objects and the corresponding behaviors, which can effectively locate the behaviors of each object to obtain accurate push effect indexes, the index prediction model is trained according to the attribution data under the push object dimension of the second platform.

[0118] In some embodiments, the above-mentioned training of the index prediction model according to the attribution data under the push object dimension of the second platform can be implemented in such a way: the attribution data under the push object dimension of the second platform at the third time is determined as the training attribution data; label extraction processing is performed on the attribution data under the push object dimension of the second platform at the fourth time to obtain the label push effect index; where the fourth time is later than the third time; a training sample is constructed according to the training attribution data and the label push effect index; and the index prediction model is trained according to the training sample.

[0119] Since the push effect indicators involved in the embodiments of this application have hysteresis, the attribution data of the second platform in the push object dimension at the third time can be determined as the training attribution data, and label extraction processing is performed on the attribution data of the second platform in the push object dimension at the fourth time to obtain the label push effect indicator. It should be noted that the third time may be the same as or different from the above-mentioned first time; the time duration between the third time and the fourth time may be equal to the time duration between the first time and the second time.

[0120] For example, if the push effect indicator is the number of day7 retained users, the attribution data of the second platform in the push object dimension on January 1st (the third time) can be determined as the training attribution data, and the number of retained users on January 7th is determined from the attribution data of the second platform in the push object dimension on January 7th (the fourth time) as the label push effect indicator.

[0121] Construct a training sample based on the training attribution data and the label push effect indicator, and train the indicator prediction model according to the training sample until the training stop condition is met. Among them, the training can be implemented in a supervised learning manner; the training stop condition can be preset, such as reaching a preset number of iterations or the loss value being less than the loss value threshold, etc.; in order to ensure the training effect of the indicator prediction model, multiple training samples can be constructed for training.

[0122] In some embodiments, the above-mentioned training of the indicator prediction model according to the training sample can be achieved in this way: extract the attribution features in the training attribution data; predict the push effect indicator to be compared corresponding to the attribution features through the indicator prediction model; calculate the loss value according to the push effect indicator to be compared and the label push effect indicator; update the model parameters of the indicator prediction model according to the loss value.

[0123] Here, the training attribution data includes various types of data, and some of the data may have little or no association with the push effect indicator. Therefore, in order to improve the training effect of the indicator prediction model, attribution features can be extracted from the training attribution data. The attribution features at least include the number of push objects corresponding to multiple conversion values, and can also include time features and regional features, etc., which can be set according to the actual application scenario.

[0124] Among them, the time in the training attribution data can be directly used as the time feature; or, the time in the training attribution data can also be converted according to the time unit of interest to obtain the time feature. For example, in a game application, the behavior of user accounts on weekdays and weekends is quite different. Therefore, the time in the training attribution data can be converted into the week number (such as Monday) to obtain the time feature. In this way, the quality of the attribution features is improved by leveraging the prior knowledge in the actual application scenario.

[0125] Similarly, the regions in the training attribution data can be directly used as region features; alternatively, the regions in the training attribution data can be transformed according to the region units of interest to obtain region features. Here, the region units can be countries, provinces, cities, etc.

[0126] The obtained attribution features are input into the metric prediction model, and the push effect metric corresponding to the attribution features is predicted through the metric prediction model. For the sake of distinction, the push effect metric obtained here is named the push effect metric to be compared. Then, according to the push effect metric to be compared and the labeled push effect metric, the loss value is calculated. Here, the type of the loss function used to calculate the loss value is not limited. For example, it can be a cross-entropy loss function, etc. The model parameters of the metric prediction model are updated according to the obtained loss value. For example, the model parameters of the metric prediction model are updated through the Back Propagation (BP) algorithm. The above method can effectively update the metric prediction model by extracting attribution features and calculating the loss value to update the model parameters of the metric prediction model, that is, it can improve the training effect.

[0127] In some embodiments, the attribution features in the training attribution data can be extracted in the following way: the number of push objects corresponding to multiple conversion value ranges in the training attribution data is determined as the attribution features; where the multiple conversion value ranges are obtained by binning multiple conversion values.

[0128] Here, the more conversion values are set, the more likely the long-tail problem will occur. For the long-tail problem, multiple conversion values can be binned to obtain multiple conversion value ranges, and each conversion value range includes at least one conversion value. Then, the number of push objects corresponding to the multiple conversion value ranges in the training attribution data can be determined as the attribution features (or a part of the attribution features). Through the above method, the poor training effect caused by overly sparse attribution features can be avoided.

[0129] In step 302, the target attribution data is predicted through the trained metric prediction model to obtain the push effect metric of the push activity on the target platform.

[0130] After the training is completed, the target attribution data can be predicted through the trained metric prediction model to obtain the push effect metric of the push activity on the target platform. For example, the attribution features in the target attribution data can be extracted and input into the trained metric prediction model, and the push effect metric corresponding to the attribution features is predicted through the trained metric prediction model as the push effect metric of the push activity on the target platform.

[0131] In some embodiments, the target attribution data includes attribution data corresponding to multiple push channels respectively; the above-mentioned prediction processing of the target attribution data by the trained metric prediction model to obtain the push effect metrics of the push activity on the target platform can be achieved in the following way: the trained metric prediction model performs prediction processing on the attribution data corresponding to each push channel to obtain the push effect metrics of the push activity on each push channel in the target platform.

[0132] Here, for a push activity, it can be pushed through multiple push channels, and the push channel can refer to the media platform in the first platform or the second platform for displaying advertisement materials. In this case, the target attribution data can be divided into attribution data corresponding to multiple push channels respectively. Taking a certain push channel, such as push channel A, as an example, the attribution data corresponding to push channel A is generated due to the push occurring in push channel A. Then, the trained metric prediction model performs prediction processing on the attribution data corresponding to push channel A to obtain the push effect metrics of the push activity on push channel A in the target platform. The above method further refines the granularity of the push effect metrics, can measure the push effect of the push activity on each push channel, and is convenient for the business side to optimize subsequent push methods, such as stopping the push on some push channels or strengthening the push on some push channels.

[0133] In some embodiments, the target attribution data is the attribution data under the push activity dimension of the first platform; the attribution data under the push activity dimension of the first platform includes attribution data corresponding to multiple conversion value ranges respectively; the multiple conversion value ranges are obtained by bucketing multiple conversion values; the above-mentioned prediction processing of the attribution data corresponding to each push channel by the trained metric prediction model to obtain the push effect metrics of the push activity on each push channel in the target platform can be achieved in the following way: the trained metric prediction model performs prediction processing on the attribution data corresponding to each conversion value range to obtain the push effect metrics of the push activity on the push channels corresponding to each conversion value range in the target platform; where each conversion value range corresponds to a push channel.

[0134] Here, for the attribution data under the push activity dimension of the first platform, it may not include push channels or the push channels are not accurate enough due to incorrect attribution. In view of this, when the target attribution data is the attribution data under the push activity dimension of the first platform, since there are significant differences in the behaviors of the objects reached through different push channels, each conversion value can be regarded as a push channel, the target attribution data can be divided into attribution data corresponding to multiple conversion values respectively, and the trained metric prediction model performs prediction processing on the attribution data corresponding to each conversion value to obtain the push effect metrics of the push activity on the push channels corresponding to each conversion value in the target platform.

[0135] On this basis, for the possible long-tail problem, multiple conversion values can be bucketed to obtain multiple conversion value ranges, and each conversion value range is regarded as a push channel. Similarly, the target attribution data is divided into attribution data corresponding to multiple conversion value ranges respectively, and the attribution data corresponding to each conversion value range is predicted by the trained metric prediction model to obtain the push effect metrics of the push activity in the target platform for each push channel corresponding to the conversion value range.

[0136] Through the above method, the push effect corresponding to each conversion value or each conversion value range can be accurately measured, which is convenient for the business side to optimize the subsequent push method.

[0137] As Figure 5 shown, in the embodiment of the present application, the metric prediction model is trained with training samples, and the push effect metrics are predicted by the trained metric prediction model. When the number of training samples is large enough, the quality is excellent enough, and the metric prediction model is fully trained, the accuracy of predicting the push effect metrics can be greatly improved.

[0138] In some embodiments, referring to Figure 6 , Figure 6 is a schematic flowchart of a business push method provided by an embodiment of the present application. Figure 3 The step 102 shown can be implemented through steps 401 to 406, and will be described in combination with each step.

[0139] In step 401, the attribution data in the push object dimension of the second platform at the fifth time is determined as the test attribution data.

[0140] Similar to the construction of the training samples above, test samples can be constructed here according to the attribution data in the push object dimension of the second platform. First, the attribution data in the push object dimension of the second platform at the fifth time is determined as the test attribution data.

[0141] In step 402, label extraction processing is performed on the attribution data in the push object dimension of the second platform at the sixth time to obtain label push effect metrics; where the sixth time is later than the fifth time.

[0142] Here, the push effect metrics are extracted from the attribution data in the push object dimension of the second platform at the sixth time as the label push effect metrics.

[0143] In some embodiments, since the testing is carried out on the basis of the constructed index prediction table scheme and the index prediction model scheme, the fifth time can be later than the first time and the third time, and the duration between the fifth time and the sixth time, the duration between the third time and the fourth time, and the duration between the first time and the second time can be equal, thereby improving the accuracy of the obtained prediction effect index.

[0144] In step 403, test samples are constructed according to the test attribution data and the label push effect index.

[0145] Here, test samples are constructed according to the test attribution data obtained in step 401 and the label push effect index obtained in step 402.

[0146] In step 404, the index prediction table scheme is tested according to the test samples to obtain the prediction effect index of the index prediction table scheme.

[0147] For example, the test attribution data in the test samples is predicted according to the index prediction table scheme to obtain the push effect index. For the sake of distinction, the push effect index obtained here is named the push effect index to be compared. Then, according to the push effect index to be compared and the label push effect index, the prediction effect index of the index prediction table scheme is calculated. The prediction effect index is used to measure the difference between the push effect index to be compared and the label push effect index. For example, the larger the prediction effect index, the smaller the difference, that is, the better the prediction effect.

[0148] In some embodiments, to improve the test effect, multiple test samples can be constructed, and the average value of the prediction effect indexes corresponding to the multiple test samples is calculated to obtain the final prediction effect index.

[0149] In step 405, the index prediction model scheme is tested according to the test samples to obtain the prediction effect index of the index prediction model scheme.

[0150] Similarly, the prediction effect index of the index prediction model scheme can be obtained according to the test samples.

[0151] In step 406, according to the prediction effect index of the index prediction table scheme and the prediction effect index of the index prediction model scheme, the target prediction scheme with the largest prediction effect index is selected from the index prediction table scheme and the index prediction model scheme.

[0152] For example, the largest prediction effect index can be selected from the prediction effect index of the index prediction table scheme and the prediction effect index of the index prediction model scheme, and the scheme corresponding to the largest prediction effect index is used as the target prediction scheme.

[0153] Such asFigure 6 As shown in the figure, in the embodiments of the present application, by constructing test samples and calculating the prediction effect indicators of two schemes based on the test samples, the prediction effect indicators can accurately measure the prediction effect, thereby ensuring that the scheme with the best prediction effect can be selected.

[0154] Next, an exemplary application of the embodiments of the present application in an actual application scenario will be described. For ease of understanding, an example will be given in the case where the first platform is the iOS platform, the second platform is the Android platform, and the target business is the game business (game application).

[0155] In the iOS platform, advertisers are restricted by the rules of App Tracking Transparency (ATT). ATT aims to protect users from cross - app location by applications. That is, when a user opens an application, the application needs to request permission from the user so that the application can access the device's advertising identifier (IDentifier For Advertisers, IDFA) and locate the user's activities. Based on this, during the advertising placement process on the iOS platform, it is impossible to obtain the original attribution data at the user granularity (the attribution data under the push object dimension mentioned above), resulting in the inability to locate key metrics such as user retention, paid amount, cumulative number of paid users, and LifeTime Value (LTV), thus posing a problem for measuring the advertising placement effect on iOS.

[0156] The iOS platform has introduced a privacy - protected attribution scheme called SKAN (StoreKit Ad Network), which aims to provide an advertising effect measurement method for advertisers and advertising platforms while protecting user privacy. Through data transmission using SKAN, advertisers and advertising platforms can obtain attribution data without disclosing the specific personal information of users. However, the positioning ability of SKAN is limited compared to IDFA. It erases the identifiable user ID and can only count app installations and a few game events that occur within a short period of time, unable to perform fine - grained behavior analysis, which affects the advertising effect evaluation and optimization of advertisers.

[0157] In response to this, the embodiments of the present application have constructed an index prediction table scheme and an index prediction model scheme based on SKAN. Combining the advantages of these two schemes, the push effect indicators of advertising campaigns (i.e., Campaigns, corresponding to the push activities mentioned above) are accurately predicted.

[0158] The embodiments of the present application provide a Figure 7 flow schematic diagram of a business push method as shown in the figure, which will be described in combination with the Figure 7 steps shown in the figure.

[0159] 1. SKAN Original Installation Table (attribution data under the push activity dimension of the first platform mentioned above).

[0160] The SKAN original installation table refers to the original data table of SKAN from the iOS platform and forwarded by a third-party attribution platform. In the SKAN original installation table, each piece of data corresponds to a user. For example, each piece of data includes time, Campaign name, CV value, etc., but there is no identifiable user ID information.

[0161] After obtaining the SKAN original installation table, installation time processing can also be performed. Installation time processing refers to converting the time zone where the time in the SKAN original installation table is located to the target time zone for easy statistics.

[0162] After obtaining the SKAN original installation table, null value (Null value) processing can also be performed. Among them, null values are generated by the null value rules of the iOS platform. For example, when the number of installations is less than the installation number threshold, the returned CV value is a null value. In response to this, the number of null values in the SKAN original installation table can be counted, and the number of null values can be evenly distributed to each CV value.

[0163] 2. Installation User Day1 - Day360 Index Table.

[0164] Here, the index table (corresponding to the daily log data mentioned above) of each day in the past year (the time here is only an example and does not constitute a limitation) of the installation users of the game application can be obtained from the overall market data (such as all log data). The index table can include push effect indicators such as payment and retention. Of course, the types of push effect indicators are not limited to this, and fewer or more push effect indicators can be set according to the actual application scenario.

[0165] It should be noted that the index table records the push effect indicators of the overall market users (all installation users) of the game. Through the index table, it is impossible to determine which Campaign brought the user, but the users can be divided into iOS platform users and Android platform users according to the platform where the users are located. For example, if the game is launched on January 1st and today is January 4th, then today, the push effect indicators of all iOS platform users and Android platform users on January 1st, January 2nd, and January 3rd can be obtained; the push effect indicators of all iOS platform users and Android platform users on January 2nd on January 1st and January 2nd; the push effect indicators of all iOS platform users and Android platform users on January 3rd on January 1st.

[0166] 3. CV Value and bucket_id Table.

[0167] For newly installed users of the game, a CV value can be assigned to each user based on their performance within 24 hours and the SKAN configuration. The CV value can be an integer within the range of 0 to 63. However, in actual application scenarios, the distribution of the number of users corresponding to the CV values is very uneven. For example, the number of users corresponding to smaller CV values (such as CV value 0 representing only installation, CV value 1 representing only registration, etc.) (corresponding to the number of objects above) is very large; while the number of users corresponding to larger CV values (such as CV value 63 representing payment) is relatively small.

[0168] Therefore, in the embodiment of the present application, the 64 CV values can be sorted, and the sorted 64 CV values can be bucketed to obtain multiple buckets (corresponding to the conversion value ranges above), and each bucket includes at least one CV value. In this way, the long-tail problem of the CV value distribution can be alleviated, the number of users in each bucket can be guaranteed, and the statistical effect is more significant. Among them, the identifier of the bucket is bucket_id.

[0169] 4. Bucket distribution table in the Campaign dimension.

[0170] Here, the SKAN original installation table is generated according to the SKAN configuration. By combining the CV value and the bucket_id table in step 3, the CV value in the SKAN original installation table can be mapped to bucket_id to obtain the bucket distribution table in the Campaign dimension.

[0171] 5. Index prediction table for the overall market from day 1 to day 360.

[0172] Based on the index table of each day of the installed users in the past year from day 1 to day 360, an index prediction table for each day from day 1 to day 360 of the overall market can be obtained. This index prediction table includes the ratio of the number of users corresponding to each CV value (or bucket) to the push effect index (such as the retention rate, corresponding to the ratio of the number of objects above).

[0173] Here, by distinguishing between iOS platform users and Android platform users, an index prediction table for the iOS platform and an index prediction table for the Android platform can be obtained.

[0174] 6. Attribution data under the user dimension of the Android platform (corresponding to the attribution data under the push object dimension above).

[0175] Here, attribution data at the user dimension can be obtained from the Android platform. The attribution data at the user dimension is reliable and can locate information such as time, push channel, region (such as country), Campaign name, CV value, etc. based on the user ID. Therefore, the attribution data at the user dimension of the Android platform is applicable to predicting the effects of the test metric prediction table solution and the metric prediction model solution.

[0176] 7. Prediction by the metric prediction table solution / Prediction by the metric prediction model solution.

[0177] For the iOS platform, predict the push effect metrics on the iOS platform through the metric prediction table solution. The data required for the metric prediction table solution here includes the bucket distribution table in the Campaign dimension in step 4 and the metric prediction table for the iOS platform in step 5. At the same time, predict the push effect metrics on the iOS platform through the metric prediction model solution. The data required for the metric prediction model solution here includes the bucket distribution table in the Campaign dimension in step 4.

[0178] For the Android platform, predict the push effect metrics on the Android platform through the metric prediction table solution. The data required for the metric prediction table solution here includes the attribution data at the user dimension of the Android platform in step 6 (the bucket distribution table in the Campaign dimension of the Android platform can be obtained by processing in the manner of step 4) and the metric prediction table for the Android platform in step 5. At the same time, predict the push effect metrics on the Android platform through the metric prediction model solution. The data required for the metric prediction model solution here includes the attribution data at the user dimension of the Android platform in step 6 (the bucket distribution table in the Campaign dimension of the Android platform can be obtained by processing in the manner of step 4).

[0179] 8. Evaluation of the Android platform solution.

[0180] Since the attribution data at the user dimension of the Android platform is reliable, it is possible to test whether the push effect metrics predicted by the metric prediction table solution on the Android platform and the push effect metrics predicted by the metric prediction model solution on the Android platform are accurate based on the attribution data at the user dimension of the Android platform, so as to select a target prediction solution with more accurate prediction from the metric prediction table solution and the metric prediction model solution. Among them, the prediction effect can be quantified through the prediction effect metrics.

[0181] 9. Result writing.

[0182] Here, after selecting the target prediction solution, write the push effect metrics predicted by the target prediction solution on the iOS platform (corresponding to the first push effect metrics above) into the big data platform.

[0183] 10. Bi data1.0

[0184] The Bi data1.0 here includes the push effect indicators determined according to the attribution data under the user dimension of the iOS platform (which can be obtained from users who have enabled IDFA in the iOS platform), corresponding to the second push effect indicator above, and the push effect indicators determined according to the attribution data under the user dimension of the Android platform, corresponding to the third push effect indicator above. It should be noted that Bi data1.0 is not predicted but directly extracted from the attribution data under the user dimension.

[0185] 11. Bi data2.0

[0186] Bi data2.0 refers to the push effect indicators written in step 9, that is, the push effect indicators predicted according to the target prediction scheme.

[0187] Finally, Bi data1.0 and Bi data2.0 can be merged and then presented. In this way, advertisers or advertising platforms can generally know the delivery effects of the Campaign on the iOS platform and the Android platform.

[0188] It should be noted that the above steps can be implemented by two big data platforms. For example, the first big data platform can implement the above steps 1 to 6, 10 to 11, and the second big data platform can implement the above steps 7 to 9. In this way, the computing power of the second big data platform is used to improve the processing efficiency. As Figure 7 shown, the first big data platform is, for example, the TBDS big data platform, and the second big data platform is, for example, the Databricks big data platform.

[0189] It should be noted that the above steps 1, 2, 10, and 11 can be implemented on the engineering side, and the above steps 3 to 9 can be implemented on the algorithm side.

[0190] Next, the indicator prediction table scheme and the indicator prediction model scheme will be described separately.

[0191] I. Indicator prediction table scheme.

[0192] As Figure 7 shown in step 5, an indicator prediction table for the overall market from day1 to day360 can be constructed. This indicator prediction table includes the user quantity ratio (such as the retention rate) of the push effect indicators corresponding to each CV value (or bucket). The core assumption is that users with the same CV value have the same performance in terms of push effect indicators.

[0193] Taking the user quantity ratio as the retention rate as an example, the embodiments of the present application provide asFigure 8 Schematic diagram of the index prediction table 81 for a certain day (such as any day from day1 to day360). For example, a CV value of 1 indicates users who are only registered. Then, look for only registered users in the overall market data (or calculate according to the index table), and calculate the retention rate of the found users to be 0.2. At the same time, a CV value of 63 indicates users who have performed all actions. Then, look for users who have performed all actions in the overall market and calculate the retention rate of the found users to be 0.3, and so on. In addition, Figure 8 The index prediction table 81 shown also shows the number of users corresponding to each CV value on that day.

[0194] Based on the index prediction table 81, the push effect index (i.e., the number of retained users) corresponding to each Campaign can be predicted. For example, there are 100 users corresponding to Campaign1, among which 60 users have a CV value of 0, 30 users have a CV value of 1, and 20 users have a CV value of 63. Then, the number of retained users for Campaign1 can be predicted to be 60*0.05 + 30*0.2 + 20*0.3 = 15.

[0195] It should be noted that Figure 8 "#0" in represents the number of users corresponding to a CV value of 0 for the Campaign, and "#1" represents the number of users corresponding to a CV value of 1 for the Campaign.

[0196] It should be noted that Figure 8 Taking the push effect index as the number of retained users as an example, but it is not limited to this, and other types of indexes can be analogized accordingly.

[0197] In some embodiments, considering that the distribution of the number of users corresponding to the CV values may be uneven. For example, the number of users corresponding to CV values greater than 10 is very small. Therefore, the 64 CV values can be bucketed to obtain multiple buckets. In this way, what is reflected in the index prediction table is the corresponding relationship between each bucket and the ratio of the number of users. Through the above method, while ensuring the distinction of the population, it is ensured that the number of users corresponding to each bucket is large enough, thus ensuring the stability and accuracy of the prediction.

[0198] II. Index prediction model solution.

[0199] The embodiments of the present application provide a Figure 9 flow schematic diagram of the business push method shown, which will be described in the form of steps.

[0200] 1) Process the attribution data under the user dimension of the Android platform to obtain training samples and test samples.

[0201] 2) Train an indicator prediction model based on training samples.

[0202] 3) Conduct offline tests on the indicator prediction table solution and the indicator prediction model solution (using the trained indicator prediction model) based on test samples to determine whether the indicator prediction model solution is superior to the indicator prediction table solution.

[0203] 4) When the indicator prediction model solution is superior to the indicator prediction table solution, predict the push effect indicators of the iOS platform according to the indicator prediction model solution.

[0204] 5) When the indicator prediction table solution is superior to the indicator prediction model solution, predict the push effect indicators of the iOS platform according to the indicator prediction table solution.

[0205] The indicator prediction model uses attribution data under the user dimension of the Android platform to implement training and offline testing. The training and offline testing processes will be described separately below.

[0206] 1) Training.

[0207] Here, training samples are constructed based on the attribution data under the user dimension of the Android platform. The training samples include training attribution data and labeled push effect indicators. For example, if the attribution data under the user dimension of the Android platform includes attribution data from day1 to day30, then the attribution data of day1 can be used as the training attribution data, and the push effect indicators are extracted from the attribution data from day2 to day30 as the labeled push effect indicators.

[0208] The training attribution data is used as the input of the indicator prediction model, and feature extraction processing is required before input to obtain attribution features. In the embodiments of the present application, the attribution features may include:

[0209] ① Time feature. Convert the time in the training attribution data into a specific day of the week to obtain the time feature. The main reason for this processing is that there are differences in aspects such as payment between weekdays and weekends. For example, the number of online players is usually higher during weekend periods. By converting the time into a specific day of the week, it helps to more accurately analyze the impact of different times on conversion events, thereby improving the accuracy of prediction and analysis.

[0210] ② Region feature. The region feature can be the country in the training attribution data (i.e., the country where the user is located).

[0211] It should be noted that when predicting the push effect metrics for the iOS platform based on the trained metric prediction model, since there may be problems of misattribution in the country information in the SKAN original installation table forwarded by the third-party attribution platform and accurate regional features cannot be extracted, the country information corresponding to the Campaign in the historical data can be directly used to replace the country information in the SKAN original installation table forwarded by the third-party attribution platform, so as to extract regional features. Here, the historical data refers to the attribution data of the iOS platform that has not been forwarded by the third-party attribution platform in history. If the country information corresponding to the Campaign cannot be found in the historical data, the country information is determined according to the country where the Campaign is mainly launched (determined at the time of launch).

[0212] ③ Conversion value feature. The conversion value feature can include the number of users corresponding to multiple buckets respectively. For example, the number of users corresponding to the bucket with bucket_id = 0 is 5, the number of users corresponding to the bucket with bucket_id = 1 is 10, and so on. Such grouping helps to more clearly show the quantity distribution of the Campaign in different buckets.

[0213] ④ Channel feature. The channel feature is not used for training, but only for the prediction process for the iOS platform. That is, the attribution data corresponding to multiple push channels can be divided to predict the push effect metrics of each push channel. In this way, it is convenient for advertisers or advertising platforms to optimize the push channels corresponding to the Campaign.

[0214] It should be noted that since there may be problems of misattribution in the push channels in the SKAN original installation table forwarded by the third-party attribution platform and accurate channel features cannot be extracted, each CV value can be regarded as a push channel, or each bucket can be regarded as a push channel.

[0215] In the embodiments of the present application, the label push effect metrics may include:

[0216] ① User payment from day 2 to day 30.

[0217] ② User retention from day 2 to day 30.

[0218] Of course, this is only an example here. In actual application scenarios, the label push effect metrics may not be limited to this.

[0219] Based on the above method, multiple training samples can be constructed, and combined with the supervised learning method, the training of the metric prediction model can be achieved. Among them, the type of the metric prediction model is not limited. For example, it can be an XGBoost model or a deep neural network model, etc. In this way, the trained metric prediction model can be used to predict the push effect metrics for each day from the 2nd day to the 30th day (day2~day30) of the Campaign. If it is necessary to predict the push effect metrics after the 30th day, then the push effect metrics predicted from the 2nd day to the 30th day can be linearly fitted, and the push effect metrics after the 30th day can be determined according to the fitted straight line.

[0220] 2) Offline testing.

[0221] Since the iOS platform cannot provide accurate data to implement offline testing, in the embodiments of the present application, a full set of SKAN processes are simulated on the Android platform, and the prediction effect metric weighted_error (we) is calculated through the test samples on the Android platform, so as to quantify the prediction effects of the metric prediction table scheme and the metric prediction model scheme.

[0222] Suppose there are m Campaigns online on the Android platform, which are C1, C2, C3... C m , the predicted d7 payment is A1, A2, A3... A m , and the actual d7 payment (label push effect metric) is B1, B2, B3... B m . Among them, A1 represents the predicted d7 payment for C1, B1 represents the actual d7 payment for C1, and the d7 payment represents the payment amount of the user on the 7th day. Then, the prediction effect metric d7_pay_we for the push effect metric of d7 payment can be calculated, and the formula is as follows:

[0223]

[0224] The closer d7_pay_we is to 1, the more accurate the prediction is. Among them, represents the weight of the d7 payment ratio of the i-th Campaign, represents the difference between the prediction result and the actual result for the i-th Campaign.

[0225] Since installations are generated every day for each Campaign, the offline test is calculated on a daily basis. On this basis, if you want to obtain the we for a certain period of time (several days), you need to aggregate the data for a specific time period for calculation. For example, if you want to obtain the d7_pay_we for 7 consecutive days and today is the 14th, since the true d7 payment situation of users installed at the latest on the 7th can be statistically counted, so starting from the 7th and pushing back 7 days, that is, screening the attribution data of m Campaigns from the 1st to the 7th as the test attribution data. The formula is as follows:

[0226]

[0227] Similarly, for d7_pay_we s the closer it is to 1, the more accurate the prediction. Among them, s represents the total number of days in the time period to be statistically counted; B ij represents the true d7 payment of the i-th Campaign on the j-th day (the j-th day when the true d7 payment can be observed), and so on. For ease of understanding, the embodiments of the present application provide a schematic diagram of the value of j as shown in Figure 10 . In Figure 10 , today is 2023 / 08 / 14 and s is 7, then the time corresponding to j with a value of 1 is 2023 / 08 / 01, and so on.

[0228] In addition to the offline test, after deploying the metric prediction model solution (that is, predicting the push effect metrics of the iOS platform according to the metric prediction model solution), online testing can be continuously carried out to ensure that the prediction effect metrics of the metric prediction model solution meet the expectations. If not, the metric prediction model can be continued to be trained. Among them, the online test is similar to the offline test, and the difference lies in whether the metric prediction model solution has been deployed.

[0229] The embodiments of the present application can at least achieve the following technical effects:

[0230] 1) Combining the two solutions of the metric prediction table solution and the metric prediction model solution, by analyzing and statistically counting the relationships between data characteristics, to measure the advertising effect (i.e., the push effect), so as to achieve the purpose of optimizing advertising placement.

[0231] 2) By processing information such as time, country, conversion value, etc., extracting more available features can improve the model training effect, thereby improving the prediction accuracy of the advertising effect.

[0232] 3) A testing method based on the Android platform is provided, which can quickly test the prediction effects of two schemes, namely the index prediction table scheme and the index prediction model scheme, so as to select a target prediction scheme with better prediction effect for prediction on the iOS platform, thereby improving the prediction accuracy of the advertising effect.

[0233] 4) It can predict the advertising effect at a finer-grained level (such as push channels, countries, etc.), which is convenient for optimizing the advertising placement strategy at these levels.

[0234] 5) In the initial stage of the game launch, due to the lack of sufficient attribution data for model training, the index prediction table scheme can be used to predict the advertising effect. As data accumulates, for example, after 2 weeks, a target prediction scheme with better prediction effect can be selected from the index prediction table scheme and the index prediction model scheme, and the advertising effect can be predicted according to the target prediction scheme. In this way, the prediction accuracy of the advertising effect can be ensured at different stages in the game life cycle.

[0235] Next, the exemplary structure of the service push device 255 provided in the embodiment of the present application implemented as a software module will be further described. In some embodiments, as Figure 2 shown, the software module in the service push device 255 stored in the memory 250 may include: an acquisition module 2551, configured to acquire the attribution data of the push activity of the target service under the push activity dimension of the first platform and the attribution data of the push activity of the target service under the push object dimension of the second platform; a selection module 2552, configured to respectively determine the prediction effect indicators of the index prediction table scheme and the index prediction model scheme according to the attribution data under the push object dimension of the second platform, and select a target prediction scheme with the largest prediction effect indicator from the index prediction table scheme and the index prediction model scheme; wherein, the index prediction table scheme predicts the push effect indicator based on the log data of the target service; the index prediction model scheme predicts the push effect indicator based on the index prediction model; a prediction module 2553, configured to perform prediction processing on the attribution data under the push activity dimension of the first platform according to the target prediction scheme to obtain the first push effect indicator of the push activity on the first platform.

[0236] In some embodiments, the indicator prediction table solution is used to perform the following processes: obtain the log data of the target service; wherein, the log data and the target attribution data are from the same platform; the target attribution data is the attribution data under the push activity dimension of the first platform or the attribution data under the push object dimension of the second platform; construct an indicator prediction table according to the log data; wherein, the indicator prediction table includes the object quantity ratio corresponding to the push effect indicator; perform a weighting process on the push object quantity in the target attribution data according to the object quantity ratio in the indicator prediction table to obtain the push effect indicator of the push activity on the target platform; wherein, the target platform is the platform from which the target attribution data is sourced.

[0237] In some embodiments, the indicator prediction table includes the object quantity ratios respectively corresponding to multiple conversion value ranges; the multiple conversion value ranges are obtained by performing a bucketing process on multiple conversion values; the indicator prediction table solution is further used to perform the following process: perform a weighted summation process on the push object quantities respectively corresponding to the multiple conversion value ranges in the target attribution data according to the object quantity ratios respectively corresponding to the multiple conversion value ranges in the indicator prediction table to obtain the push effect indicator of the push activity on the target platform.

[0238] In some embodiments, the indicator prediction table solution is further used to perform the following process: construct an indicator prediction table according to the log data at the first time and the log data at the second time; wherein, the indicator prediction table includes the object quantity ratio between the second object quantity and the first object quantity corresponding to the push effect indicator; the first object quantity is determined according to the log data at the first time, and the second object quantity is determined according to the log data at the second time; the second time is later than the first time; perform a weighting process on the push object quantity in the target attribution data at the first time according to the object quantity ratio in the indicator prediction table to obtain the push effect indicator of the push activity on the target platform at the second time.

[0239] In some embodiments, the indicator prediction model solution is used to perform the following processes: train an indicator prediction model according to the attribution data under the push object dimension of the second platform; perform a prediction process on the target attribution data through the trained indicator prediction model to obtain the push effect indicator of the push activity on the target platform; wherein, the target attribution data is the attribution data under the push activity dimension of the first platform or the attribution data under the push object dimension of the second platform; the target platform is the platform corresponding to the target attribution data.

[0240] In some embodiments, the indicator prediction model solution is further used to perform the following processing: determining the attribution data under the push object dimension of the second platform at the third time as the training attribution data; performing label extraction processing on the attribution data under the push object dimension of the second platform at the fourth time to obtain the label push effect indicator; wherein, the fourth time is later than the third time; constructing a training sample according to the training attribution data and the label push effect indicator; and training the indicator prediction model according to the training sample.

[0241] In some embodiments, the indicator prediction model solution is further used to perform the following processing: extracting the attribution features in the training attribution data; predicting, through the indicator prediction model, the push effect indicator to be compared corresponding to the attribution features; calculating a loss value according to the push effect indicator to be compared and the label push effect indicator; and updating the model parameters of the indicator prediction model according to the loss value.

[0242] In some embodiments, the indicator prediction model solution is further used to perform the following processing: determining the number of push objects corresponding to multiple conversion value ranges in the training attribution data as the attribution features; wherein, the multiple conversion value ranges are obtained by performing bucketing processing on multiple conversion values.

[0243] In some embodiments, the target attribution data includes the attribution data corresponding to multiple push channels respectively; the indicator prediction model solution is further used to perform the following processing: performing prediction processing on the attribution data corresponding to each push channel through the trained indicator prediction model to obtain the push effect indicator of each push channel of the push activity in the target platform.

[0244] In some embodiments, the target attribution data is the attribution data under the push activity dimension of the first platform; the attribution data under the push activity dimension of the first platform includes the attribution data corresponding to multiple conversion value ranges respectively; the multiple conversion value ranges are obtained by performing bucketing processing on multiple conversion values; the indicator prediction model solution is further used to perform the following processing: performing prediction processing on the attribution data corresponding to each conversion value range through the trained indicator prediction model to obtain the push effect indicator of the push channel corresponding to each conversion value range of the push activity in the target platform; wherein, each conversion value range corresponds to one push channel.

[0245] In some embodiments, the selection module 2552 is further configured to: determine the attribution data of the second platform under the push object dimension at the fifth time as the test attribution data; perform label extraction processing on the attribution data of the second platform under the push object dimension at the sixth time to obtain the label push effect index, where the sixth time is later than the fifth time; construct a test sample according to the test attribution data and the label push effect index; perform a test process on the index prediction table scheme according to the test sample to obtain the prediction effect index of the index prediction table scheme; perform a test process on the index prediction model scheme according to the test sample to obtain the prediction effect index of the index prediction model scheme.

[0246] In some embodiments, the acquisition module 2551 is further configured to: acquire the attribution data of the push activity under the push object dimension of the first platform; determine the second push effect index of the push activity on the first platform according to the attribution data of the first platform under the push object dimension. The prediction module 2553 is further configured to: perform a merging process on the first push effect index and the second push effect index to obtain the merged push effect index of the push activity on the first platform.

[0247] In some embodiments, the acquisition module 2551 is further configured to: acquire the attribution data of the push activity under the push object dimension of the first platform; determine the second push effect index of the push activity on the first platform according to the attribution data of the first platform under the push object dimension; determine the third push effect index of the push activity on the second platform according to the attribution data of the second platform under the push object dimension. The prediction module 2553 is further configured to: perform a merging process on the first push effect index, the second push effect index, and the third push effect index to obtain the merged push effect index of the push activity on the first platform and the second platform.

[0248] An embodiment of the present application provides a computer program product or a computer program, which includes executable instructions stored in a computer-readable storage medium. The processor of the electronic device reads the executable instructions from the computer-readable storage medium, and the processor executes the executable instructions, so that the electronic device executes the business push method described above in the embodiments of the present application.

[0249] An embodiment of the present application provides a computer-readable storage medium storing executable instructions, where the executable instructions, when executed by a processor, cause the processor to execute the business push method provided in the embodiments of the present application.

[0250] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disc, or CD-ROM; or may be various devices including one or any combination of the above memories.

[0251] In some embodiments, the executable instructions may be in the form of a program, software, a software module, a script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including being deployed as a stand-alone program or being deployed as a module, a component, a subroutine, or other unit suitable for use in a computing environment.

[0252] As an example, the executable instructions may or may not correspond to a file in a file system, may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, stored in a single file dedicated to the program being discussed, or, stored in multiple cooperating files (such as files that store one or more modules, subroutines, or portions of code).

[0253] As an example, the executable instructions may be deployed to execute on one computing device, or on multiple computing devices located at one site, or, on multiple computing devices distributed across multiple sites and interconnected by a communication network.

[0254] The above are only embodiments of the present application and are not intended to limit the protection scope of the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and scope of the present application are all included within the protection scope of the present application.

Claims

1. A service push method, characterized in that, Including: Obtaining attribution data of a push activity of a target service under a push activity dimension of a first platform, and attribution data of the push activity of the target service under a push object dimension of a second platform; According to the attribution data under the push object dimension of the second platform, respectively determining a prediction effect index of an index prediction table scheme and a prediction effect index of an index prediction model scheme, and selecting a target prediction scheme with the largest prediction effect index from the index prediction table scheme and the index prediction model scheme; wherein, the index prediction table scheme predicts a push effect index based on the log data of the target service; the index prediction model scheme predicts a push effect index based on an index prediction model; Performing a prediction process on the attribution data under the push activity dimension of the first platform according to the target prediction scheme to obtain a first push effect index of the push activity on the first platform.

2. The method according to claim 1, wherein The index prediction table scheme is used to perform the following processing: Obtaining the log data of the target service; wherein, the log data and the target attribution data are from the same platform; the target attribution data is the attribution data under the push activity dimension of the first platform or the attribution data under the push object dimension of the second platform; Constructing an index prediction table according to the log data; wherein, the index prediction table includes an object quantity ratio corresponding to a push effect index; Performing a weighted process on the push object quantity in the target attribution data according to the object quantity ratio in the index prediction table to obtain a push effect index of the push activity on a target platform; wherein, the target platform is the platform from which the target attribution data is sourced.

3. The method according to claim 2, wherein The index prediction table includes object quantity ratios corresponding to multiple conversion value ranges respectively; the multiple conversion value ranges are obtained by performing a bucketing process on multiple conversion values; The performing a weighted process on the push object quantity in the target attribution data according to the object quantity ratio in the index prediction table to obtain a push effect index of the push activity on a target platform includes: Performing a weighted summation process on the push object quantities corresponding to multiple conversion value ranges respectively in the target attribution data according to the object quantity ratios corresponding to multiple conversion value ranges respectively in the index prediction table to obtain a push effect index of the push activity on a target platform.

4. The method according to claim 2, characterized in that, The constructing an index prediction table according to the log data includes: Constructing an index prediction table according to the log data at a first time and the log data at a second time; wherein, the index prediction table includes an object quantity ratio between a second object quantity and a first object quantity corresponding to a push effect index; the first object quantity is determined according to the log data at the first time, and the second object quantity is determined according to the log data at the second time; the second time is later than the first time; The performing a weighted process on the push object quantity in the target attribution data according to the object quantity ratio in the index prediction table to obtain a push effect index of the push activity on a target platform includes: Weight the number of push objects in the target attribution data at the first time according to the object quantity ratio in the index prediction table to obtain the push effect index of the push activity at the second time on the target platform.

5. The method according to claim 1, wherein The index prediction model solution is used to perform the following processing: Train an index prediction model according to the attribution data under the push object dimension of the second platform; Perform a prediction process on the target attribution data through the trained index prediction model to obtain the push effect index of the push activity on the target platform; Among them, the target attribution data is the attribution data under the push activity dimension of the first platform or the attribution data under the push object dimension of the second platform; the target platform is the platform corresponding to the target attribution data.

6. The method according to claim 5, wherein The training of the index prediction model according to the attribution data under the push object dimension of the second platform includes: Determine the attribution data under the push object dimension of the second platform at the third time as the training attribution data; Perform label extraction processing on the attribution data under the push object dimension of the second platform at the fourth time to obtain the label push effect index; where the fourth time is later than the third time; Construct a training sample according to the training attribution data and the label push effect index; Train an index prediction model according to the training sample.

7. The method according to claim 6, wherein The training of the index prediction model according to the training sample includes: Extract the attribution features in the training attribution data; Predict the push effect index to be compared corresponding to the attribution features through the index prediction model; Calculate the loss value according to the push effect index to be compared and the label push effect index; Update the model parameters of the index prediction model according to the loss value.

8. The method according to claim 7, wherein The extraction of the attribution features in the training attribution data includes: Determine the number of push objects corresponding to multiple conversion value ranges in the training attribution data as the attribution features; Among them, the multiple conversion value ranges are obtained by binning multiple conversion values.

9. The method according to claim 5, wherein The target attribution data includes the attribution data corresponding to multiple push channels; the prediction process on the target attribution data through the trained index prediction model to obtain the push effect index of the push activity on the target platform includes: Perform a prediction process on the attribution data corresponding to each push channel through the trained index prediction model to obtain the push effect index of each push channel of the push activity on the target platform.

10. The method according to claim 9, wherein The target attribution data is the attribution data under the push activity dimension of the first platform; the attribution data under the push activity dimension of the first platform includes the attribution data corresponding to multiple conversion value ranges; the multiple conversion value ranges are obtained by binning multiple conversion values; The prediction process on the attribution data corresponding to each push channel through the trained index prediction model to obtain the push effect index of each push channel of the push activity on the target platform includes: Perform a prediction process on the attribution data corresponding to each conversion value range through the trained index prediction model to obtain the push effect index of the push channel corresponding to each conversion value range of the push activity on the target platform; Among them, each conversion value range corresponds to a push channel.

11. The method according to claim 1, characterized in that, Determining the prediction effect indicators of the index prediction table solution and the prediction effect indicators of the index prediction model solution respectively according to the attribution data under the push object dimension of the second platform includes: Determining the attribution data under the push object dimension of the second platform at the fifth time as the test attribution data; Performing label extraction processing on the attribution data under the push object dimension of the second platform at the sixth time to obtain label push effect indicators; wherein, the sixth time is later than the fifth time; Constructing a test sample according to the test attribution data and the label push effect indicators; Performing a test process on the index prediction table solution according to the test sample to obtain the prediction effect indicators of the index prediction table solution; Performing a test process on the index prediction model solution according to the test sample to obtain the prediction effect indicators of the index prediction model solution.

12. The method according to any one of claims 1 to 11, characterized in that, The method further includes: Obtaining the attribution data of the push activity under the push object dimension of the first platform; Determining the second push effect indicator of the push activity on the first platform according to the attribution data under the push object dimension of the first platform; After performing a prediction process on the attribution data under the push activity dimension of the first platform according to the target prediction solution to obtain the first push effect indicator of the push activity on the first platform, the method further includes: Performing a merging process on the first push effect indicator and the second push effect indicator to obtain the merged push effect indicator of the push activity on the first platform.

13. The method according to any one of claims 1 to 11, characterized in that, The method further includes: Obtaining the attribution data of the push activity under the push object dimension of the first platform; Determining the second push effect indicator of the push activity on the first platform according to the attribution data under the push object dimension of the first platform; Determining the third push effect indicator of the push activity on the second platform according to the attribution data under the push object dimension of the second platform; After performing a prediction process on the attribution data under the push activity dimension of the first platform according to the target prediction solution to obtain the first push effect indicator of the push activity on the first platform, the method further includes: Performing a merging process on the first push effect indicator, the second push effect indicator, and the third push effect indicator to obtain the merged push effect indicator of the push activity on the first platform and the second platform.

14. A service push device, characterized in that, Including: An acquisition module, configured to acquire the attribution data of the push activity of the target service under the push activity dimension of the first platform and the attribution data of the push activity of the target service under the push object dimension of the second platform; A selection module, configured to respectively determine a prediction effect index of an index prediction table solution and a prediction effect index of an index prediction model solution according to attribution data under the push object dimension of the second platform, and select a target prediction solution with the largest prediction effect index from the index prediction table solution and the index prediction model solution; wherein, the index prediction table solution predicts a push effect index based on log data of the target service; the index prediction model solution predicts a push effect index based on an index prediction model. A prediction module, configured to perform a prediction process on attribution data under the push activity dimension of the first platform according to the target prediction solution, so as to obtain a first push effect index of the push activity on the first platform.

15. An electronic device, characterized in that, Comprising: A memory, configured to store executable instructions; A processor, configured to implement the service push method according to any one of claims 1 to 13 when executing the executable instructions stored in the memory.

16. A computer-readable storage medium, characterized in that, Stored with executable instructions, which are configured to implement the service push method according to any one of claims 1 to 13 when being executed by a processor.

17. A computer program product, characterized in that, Including executable instructions, which are configured to implement the service push method according to any one of claims 1 to 13 when being executed by a processor.

Images