Information push method, device and storage medium

Abstract

Embodiments of the present invention provide an information push method, device, and storage medium. The method comprises: obtaining positive samples and negative samples, and performing model training based on the positive and negative samples to obtain an information push model, wherein the positive samples include data of users to whom pushes were successfully made, and the negative samples include data of users to be pushed obtained from a set to be pushed, wherein the set to be pushed includes data of all users to be pushed; and executing information push according to the information push model. This invention solves the problem of inaccurate information push model construction caused by insufficient data volume and data range in related technologies, improves the accuracy of the information push model, and increases the success rate of information push.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of computers, and in particular, to an information push method, device, and storage medium. Background Art

[0002] In recent years, intelligent information push methods that combine high-tech to portray user portraits (for example, smart marketing) have been widely praised. Simply put, it is to use big data analysis, machine learning and other methods to more deeply explore user preferences, analyze users' potential needs, and then transform the effect of information push.

[0003] According to the intelligent information push model of relevant technologies, each institution can only explore user needs based on the user portraits in its own system, because the user portrait information between institutions (for example, portrait information related to user assets, consumption behavior, etc.) is separated from each other. This data not only involves the user's personal privacy and has a certain degree of sensitivity, but is also one of the core data of the institution. It has high value and is difficult to share publicly in plain text.

[0004] It can be seen that the process of constructing an information push model often involves the problem of inaccurate construction of the information push model due to insufficient data volume and data range. Summary of the Invention

[0005] The embodiments of the present invention provide an information push method, device, and storage medium to at least solve the problem in the related art of inaccurate construction of an information push model due to insufficient data volume and data range.

[0006] According to one embodiment of the present invention, there is provided an information push method, comprising:

[0007] Obtaining positive samples and negative samples, and performing model training based on the positive samples and the negative samples to obtain an information push model, wherein the positive samples include data of users to whom information has been successfully pushed, and the negative samples include data of users to be pushed obtained from a set to be pushed, wherein the set to be pushed includes data of all users to be pushed;

[0008] Information push is performed according to the information push model.

[0009] In at least one exemplary embodiment, obtaining positive samples and negative samples, and performing model training based on the positive samples and the negative samples to obtain an information push model includes:

[0010] Sample acquisition step: acquiring the positive sample and acquiring the negative sample from the set to be pushed;

[0011] Training step: performing model training based on the positive samples and the negative samples to obtain the information push model;

[0012] Adjustment step: predicting the push success rate of all users to be pushed in the set to be pushed according to the information push model, determining potential push users from the set to be pushed according to the push success rate, adding the data of the potential push users to the positive samples, and deleting the data of the potential push users from the set to be pushed;

[0013] Control step: A control loop iteratively executes the sample acquisition step, the training step, and the adjustment step until the information push model meets a predetermined condition.

[0014] In at least one exemplary embodiment, determining potential push users from the set to be pushed according to the push success rate includes one of the following:

[0015] Determine the users in the set to be pushed whose push success rate is higher than the potential push threshold as the potential push users;

[0016] A predetermined number of users with the highest push success rates in the set to be pushed are determined as the potential push users.

[0017] In at least one exemplary embodiment, the predetermined condition includes one of the following:

[0018] The information push model meets business requirements;

[0019] The information push model reaches a convergence condition.

[0020] In at least one exemplary embodiment, the number of positive samples is a first value, and the number of negative samples is a second value, wherein a difference between the second value and the first value is less than a preset deviation value.

[0021] In at least one exemplary embodiment, the negative sample includes data of users to be pushed that is randomly selected from the set to be pushed.

[0022] In at least one exemplary embodiment, performing model training based on the positive samples and the negative samples to obtain an information push model includes:

[0023] According to a federated algorithm, model training is performed based on the positive samples and the negative samples to obtain the information push model. The federated algorithm includes one of the following: a federated logistic regression algorithm and a federated tree algorithm.

[0024] In at least one exemplary embodiment, according to a federated algorithm, performing model training based on the positive samples and the negative samples to obtain the information push model includes:

[0025] Multiple participants provide feature data of users included in the positive samples and the negative samples from different dimensions respectively, and the coordinator aggregates the model gradients and updates the model parameters, and obtains the information push model through model training.

[0026] According to another embodiment of the present invention, there is provided an information push device, comprising:

[0027] A model training module, configured to obtain positive samples and negative samples, and perform model training based on the positive samples and the negative samples to obtain an information push model, wherein the positive samples include data of users to whom information has been successfully pushed, and the negative samples include data of users to be pushed obtained from a set to be pushed, wherein the set to be pushed includes data of all users to be pushed;

[0028] The information push module is used to execute information push according to the information push model.

[0029] According to yet another embodiment of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, wherein the computer program is configured to execute the steps of any one of the above method embodiments when run.

[0030] According to another embodiment of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any one of the above method embodiments.

[0031] Through the present invention, since the data of users in the to-be-pushed set who have not yet been successfully pushed are introduced as samples, the quantity and type of sample data are greatly enriched. Therefore, it can solve the problem of inaccurate construction of information push models caused by insufficient data volume and data range in related technologies, improve the accuracy of the information push model, and increase the success rate of information push. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a hardware structure block diagram of a computer terminal for an information push method according to an embodiment of the present invention;

[0033] Figure 2 is a flowchart of an information push method according to an embodiment of the present invention;

[0034] Figure 3 is a structural block diagram of an information push device according to an embodiment of the present invention;

[0035] Figure 4 is a schematic diagram of the training process of the intelligent marketing model according to an embodiment of the present invention;

[0036] Figure 5is a flowchart of a model training sub-protocol for federated logistic regression according to an embodiment of the present invention;

[0037] Figure 6 is a flowchart of a model training sub-protocol 1 of a federated tree according to an embodiment of the present invention;

[0038] Figure 7 4 is a flowchart of the second sub-protocol for model training of the federated tree according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In recent years, intelligent information push methods that combine high-tech to portray user portraits (for example, smart marketing) have been widely praised. Simply put, it is to use big data analysis, machine learning and other methods to more deeply explore user preferences, analyze users' potential needs, and then transform the effect of information push.

[0040] Currently, intelligent information push models are often constructed using logistic regression. First, historical information push data (e.g., historical marketing data) is labeled. Users with successful information pushes (e.g., users who chose to receive and display the pushed information; or marketing success, where users chose to receive and display the pushed information and subsequently placed an order based on the information; the specific conditions for successful information push can be determined based on the current business context) are labeled as positive samples, while users with failed information pushes (e.g., users who chose to decline receiving and displaying the pushed information, users who did not choose to receive and display the pushed information, or marketing failures, where users did not place an order based on the pushed information; the specific conditions for failed information pushes can be determined based on the current business context) are labeled as negative samples. A logistic regression model is then constructed using the user's existing features to predict the information push success rate for the target user. However, the above intelligent information push model is based on the premise of data centralization. This structure assumes that all user feature data comes from the same organization, which is inconsistent with the current data silo situation and therefore lacks practical application value.

[0041] As mentioned above, ideally, only by integrating multi-dimensional data from multiple institutions can a more three-dimensional user profile be constructed, achieving the marketing goals of complementary resource advantages, expanding market breadth, and tapping into market depth. However, the current situation is that user profile information between institutions (for example, profile information related to user assets, consumer behavior, etc.) is fragmented, resulting in significant deficiencies in both data volume and data scope, which can lead to inaccurate information push model construction. On the other hand, the establishment of previous information push models faces the problem that labeled data is extremely difficult to collect, and labeling is time-consuming and labor-intensive. This results in an insufficient number of labeled data samples, which also leads to inaccurate information push model construction.

[0042] To this end, in embodiments of the present invention, federated learning enables all participating parties (e.g., organizations requiring information push) to confidentially share their user data to calculate information push models. Information push strategies are then formulated based on the modeling results, achieving a win-win, or even multi-win, joint information push solution. Furthermore, through semi-supervised learning, embodiments of the present invention can train intelligent information push models based solely on a small number of labeled samples and a large number of unlabeled samples.

[0043] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings and in combination with embodiments.

[0044] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0045] The method embodiments provided in the embodiments of the present application can be executed in a computer terminal or similar computing device. Taking running on a computer terminal as an example, Figure 1 FIG. 1 is a hardware structure diagram of a computer terminal of an information push method according to an embodiment of the present invention. Figure 1 As shown, the computer terminal may include one or more ( Figure 1 Only one is shown) a processor 102 (the processor 102 may include but is not limited to a microprocessor MCU or a programmable logic device FPGA and other processing devices) and a memory 104 for storing data. The computer terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above-mentioned computer terminal. For example, the computer terminal may also include Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0046] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the information push method in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, implementing the above-mentioned information push method. The memory 104 may include a high-speed random access memory and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely located relative to the processor 102, and these remote memories may be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0047] The transmission device 106 is used to receive or send data via a network. A specific example of the aforementioned network may include a wireless network provided by a computer terminal's communications provider. In one embodiment, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0048] In this embodiment, a method for pushing information running on a computer terminal or similar computing device is provided. Figure 2 is a flow chart of an information push method according to an embodiment of the present invention. Figure 2 As shown, the process includes the following steps:

[0049] Step S202: obtain positive samples and negative samples, and perform model training based on the positive samples and the negative samples to obtain an information push model, wherein the positive samples include data of users who have successfully pushed, and the negative samples include data of users to be pushed obtained from a set to be pushed, and the set to be pushed includes data of all users to be pushed.

[0050] In at least one exemplary embodiment, in order to iteratively and gradually obtain a more accurate information push model, obtaining positive samples and negative samples in step S202 and performing model training based on the positive samples and the negative samples to obtain the information push model can be implemented in the following manner:

[0051] Step S202-1, sample acquisition step: acquiring the positive sample, and acquiring the negative sample from the set to be pushed.

[0052] In at least one exemplary embodiment, the negative sample includes data of users to be pushed that is randomly selected from the set to be pushed.

[0053] In at least one exemplary embodiment, the number of positive samples is a first value, and the number of negative samples is a second value, wherein the number of negative samples is recommended to be approximately the same as the number of positive samples, that is, the difference between the second value and the first value is less than a preset deviation value.

[0054] Step S202-2, training step: performing model training based on the positive samples and the negative samples to obtain the information push model.

[0055] In at least one exemplary embodiment, performing model training based on the positive samples and the negative samples to obtain an information push model includes:

[0056] According to a federated algorithm, model training is performed based on the positive samples and the negative samples to obtain the information push model. The federated algorithm includes one of the following: a federated logistic regression algorithm and a federated tree algorithm.

[0057] In at least one exemplary embodiment, according to a federated algorithm, performing model training based on the positive samples and the negative samples to obtain the information push model includes:

[0058] Multiple participants provide feature data of users included in the positive samples and the negative samples from different dimensions respectively, and the coordinator aggregates the model gradients and updates the model parameters, and obtains the information push model through model training.

[0059] Step S202-3, adjustment step: predict the push success rate of all users to be pushed in the set to be pushed based on the information push model, determine potential push users from the set to be pushed based on the push success rate, add the data of the potential push users to the positive sample, and delete the data of the potential push users from the set to be pushed.

[0060] In at least one exemplary embodiment, in step S202-3, determining potential push users from the set to be pushed based on the push success rate includes one of the following:

[0061] Determine the users in the set to be pushed whose push success rate is higher than the potential push threshold as the potential push users;

[0062] A predetermined number of users with the highest push success rates in the set to be pushed are determined as the potential push users.

[0063] Step S202-4, control step: the control loop iteratively executes the sample acquisition step S202-1, the training step S202-2 and the adjustment step S202-3 until the information push model meets the predetermined conditions.

[0064] In at least one exemplary embodiment, the predetermined condition may include one of the following:

[0065] The information push model meets business requirements;

[0066] The information push model reaches a convergence condition.

[0067] Step S204: executing information push according to the information push model.

[0068] The execution subject of the above steps may be a computer terminal or a similar computing device, but is not limited thereto.

[0069] Through the above steps S202 and S204, the data of users in the to-be-pushed set that have not yet been successfully pushed are introduced as samples, which greatly enriches the quantity and type of sample data. Therefore, it can solve the problem of inaccurate construction of the information push model due to insufficient data volume and data range in related technologies, improve the accuracy of the information push model, and increase the success rate of information push.

[0070] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0071] In this embodiment, an information push device is also provided, which is used to implement the above-mentioned embodiments and preferred embodiments. The details that have been described will not be repeated here. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0072] Figure 3 is a structural block diagram of an information push device according to an embodiment of the present invention. Figure 3 As shown, the device includes:

[0073] A model training module 32 is configured to obtain positive samples and negative samples, and perform model training based on the positive samples and the negative samples to obtain an information push model, wherein the positive samples include data of users to whom information has been successfully pushed, and the negative samples include data of users to be pushed obtained from a set to be pushed, wherein the set to be pushed includes data of all users to be pushed;

[0074] The information push module 34 is coupled to the model training module 32 and is used to execute information push according to the information push model.

[0075] It should be noted that the above modules can be implemented through software or hardware. For the latter, it can be implemented in the following ways, but not limited to: the above modules are all located in the same processor; or the above modules are located in different processors in any combination.

[0076] An embodiment of the present invention further provides a computer-readable storage medium, in which a computer program is stored. The computer program is configured to execute the steps of any one of the above method embodiments when running.

[0077] In this embodiment, the computer-readable storage medium may be configured to store a computer program for performing the following steps:

[0078] S1. Obtain positive samples and negative samples, and perform model training based on the positive samples and the negative samples to obtain an information push model, wherein the positive samples include data of users who have successfully pushed, and the negative samples include data of users to be pushed obtained from a set to be pushed, and the set to be pushed includes data of all users to be pushed.

[0079] S2: Execute information push according to the information push model.

[0080] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disk.

[0081] An embodiment of the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0082] In an exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0083] In an exemplary embodiment, the processor may be configured to execute the following steps via a computer program:

[0084] S1. Obtain positive samples and negative samples, and perform model training based on the positive samples and the negative samples to obtain an information push model, wherein the positive samples include data of users who have successfully pushed, and the negative samples include data of users to be pushed obtained from a set to be pushed, and the set to be pushed includes data of all users to be pushed.

[0085] S2: Execute information push according to the information push model.

[0086] For specific examples in this embodiment, reference may be made to the examples described in the above embodiments and exemplary implementation modes, and this embodiment will not be described in detail here.

[0087] The following describes in detail the information push solution of the embodiment of the present invention by taking the intelligent marketing model as an example.

[0088] Federated learning is a machine learning framework that can effectively help multiple institutions use data and conduct machine learning modeling while meeting the requirements of user privacy protection, data security and government regulations.

[0089] Semi-supervised learning is to train labeled samples with the help of unlabeled samples to obtain a model with better performance than the model trained with labeled samples alone, thus making up for the defect of insufficient number of labeled samples.

[0090] The embodiment of the present invention proposes an intelligent marketing model under federated learning for the first time, and incorporates the advantages of semi-supervised learning into it. Figure 4 FIG. 1 is a schematic diagram of the training process of the intelligent marketing model according to an embodiment of the present invention. Figure 4 As shown, the following steps are included:

[0091] Step 1: Add the users who have been marketed to the training set as positive samples. Randomly select a user group with a number approximately equal to the number of users who have been marketed from the users to be marketed and add them to the training set as negative samples.

[0092] Step 2: Select a suitable federated algorithm and train the model based on positive and negative samples;

[0093] Step 3: Predict all users to be marketed and select the user group with the highest potential marketing success rate (potential marketing users) as new positive samples to be added to the marketed users, thus achieving the goal of expanding the labeled dataset.

[0094] Step 4: Subtract these potential marketing users from the users to be marketed;

[0095] Repeat steps 1-4 multiple times until the model effect meets business requirements.

[0096] The marketing model structure offers two federated algorithm options: federated logistic regression and federated tree. The choice of which method to use depends on specific business needs. This section describes the model training process for each algorithm.

[0097] (1) Federated Logistic Regression

[0098] Figure 5 is a flow chart of a model training sub-protocol of federated logistic regression according to an embodiment of the present invention. Figure 5 As shown, the process includes the following steps:

[0099] Step 1: The federation coordinator uses the homomorphic encryption algorithm to generate homomorphic public and private keys, and sends the public key to each participant;

[0100] Step 2: The Host (data provider) calculates and uses the public key to encrypt its own model parameters and feature dot product wx, and sends it to the Guest (data user).

[0101] Step 3: The guest sends the encrypted dot product to the host and sums it with its own wx, calculates the encrypted residual [d] based on the label, and sends it to the participating host;

[0102] Step 4: Each participant uses the encrypted residual and dot product to calculate the encrypted gradient [g] and sends the encrypted gradient to the coordinator;

[0103] In step 5, the coordinator decrypts each encrypted gradient and sends it to all participants, who then update the model parameters accordingly.

[0104] (2) Federal Tree

[0105] The federated tree model training consists of two sub-protocols:

[0106] Sub-protocol 1 is that before the algorithm starts, the federated coordinator distributes the homomorphic encryption public key to the participating guest (i.e., the data application party). The guest encrypts the first-order and second-order derivatives of the sample and transmits the ciphertext to the participating host (i.e., the data provider). Figure 6 : is a flow chart of the model training sub-protocol 1 of the federated tree according to an embodiment of the present invention. Figure 6 As shown, the process includes the following steps:

[0107] Step 1: The federation coordinator generates homomorphic public and private keys and sends them to the participating guest.

[0108] Step 2: The guest uses the local tag to calculate the first-order derivative and the second-order derivative, and encrypts the first-order and second-order derivatives using the homomorphic encryption public key.

[0109] Step 3: The guest sends its own gradient ciphertext to the host.

[0110] Sub-protocol 2 is executed at each node of the tree. The federal coordinator obtains and decrypts the candidate split point information of the participating parties Guest and Host, outputs whether the node is split and the optimal split point information, and determines the left and right subtrees or leaf nodes that each participating party can generate based on this information. The participating parties update the tree model according to the coordinator's results. Figure 7 : is a flowchart of the model training sub-protocol 2 of the federated tree according to an embodiment of the present invention. Figure 7 As shown, the process includes the following steps:

[0111] Step 1: Each participant uses the local label, first-order derivative, and second-order derivative ciphertext to calculate the feature split points, shuffle the order of all split points, and number them. Then, the split point ciphertext information is sent to the federal coordinator.

[0112] Step 2: The federal coordinator decrypts the split point ciphertext, compares the split point benefits, and obtains the optimal split point;

[0113] In step 3, the federation coordinator informs the participating party of the optimal split point number, and the participating party uses the split point to update the local model and synchronize the tree structure and sample division with other participating parties.

[0114] In summary, the present invention proposes an intelligent marketing model based on federated learning and uses semi-supervised learning to address the problem of insufficient labeled samples. Compared to related technologies, the present invention implements a federated intelligent marketing model, whereby participating organizations provide user feature data from different dimensions, with the coordinator responsible for aggregating model gradients and updating model parameters. Furthermore, based on semi-supervised learning, the present invention addresses the problem of insufficient labeled samples in real-world scenarios, preventing model training.

[0115] Obviously, those skilled in the art will appreciate that the various modules or steps of the present invention described above can be implemented using a general-purpose computing device, can be centralized on a single computing device, or can be distributed across a network of multiple computing devices. They can be implemented using program code executable by the computing device, and thus, can be stored in a storage device and executed by the computing device. In some cases, the steps shown or described herein can be performed in a different order than that shown, or can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0116] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. An information push method, characterized in that: include: Obtaining positive samples and negative samples, and performing model training based on the positive samples and the negative samples to obtain an information push model, wherein the positive samples include data of users to whom information has been successfully pushed, and the negative samples include data of users to be pushed obtained from a set to be pushed, wherein the set to be pushed includes data of all users to be pushed; Execute information push according to the information push model; Acquiring positive samples and negative samples, and performing model training based on the positive samples and the negative samples to obtain an information push model includes: Sample acquisition step: acquiring the positive sample and acquiring the negative sample from the set to be pushed; Training step: performing model training based on the positive samples and the negative samples to obtain the information push model; Adjustment step: predicting the push success rate of all users to be pushed in the set to be pushed according to the information push model, determining potential push users from the set to be pushed according to the push success rate, adding the data of the potential push users to the positive samples, and deleting the data of the potential push users from the set to be pushed; Control step: A control loop iteratively executes the sample acquisition step, the training step, and the adjustment step until the information push model meets a predetermined condition.

2. The method according to claim 1, characterized in that Determining potential push users from the set to be pushed according to the push success rate includes one of the following: Determine the users in the set to be pushed whose push success rate is higher than the potential push threshold as the potential push users; A predetermined number of users with the highest push success rates in the set to be pushed are determined as the potential push users.

3. The method according to claim 1, characterized in that The predetermined condition includes one of the following: The information push model meets business requirements; The information push model reaches a convergence condition.

4. The method according to claim 1, wherein The number of the positive samples is a first value, and the number of the negative samples is a second value, wherein a difference between the second value and the first value is less than a preset deviation value.

5. The method according to claim 1, characterized in that The negative samples include data of users to be pushed that are randomly selected from the set to be pushed.

6. The method according to any one of claims 1 to 5, characterized in that The information push model obtained by performing model training based on the positive samples and the negative samples includes: According to a federated algorithm, model training is performed based on the positive samples and the negative samples to obtain the information push model. The federated algorithm includes one of the following: a federated logistic regression algorithm and a federated tree algorithm.

7. The method according to claim 6, characterized in that According to the federation algorithm, performing model training based on the positive samples and the negative samples to obtain the information push model includes: Multiple participants provide feature data of users included in the positive samples and the negative samples from different dimensions respectively, and the coordinator aggregates the model gradients and updates the model parameters, and obtains the information push model through model training.

8. An information push device, characterized in that: include: A model training module, configured to obtain positive samples and negative samples, and perform model training based on the positive samples and the negative samples to obtain an information push model, wherein the positive samples include data of users to whom information has been successfully pushed, and the negative samples include data of users to be pushed obtained from a set to be pushed, wherein the set to be pushed includes data of all users to be pushed; An information push module, configured to execute information push according to the information push model; The model training module is also used for the sample acquisition step: obtaining the positive sample and obtaining the negative sample from the set to be pushed; the training step: performing model training based on the positive sample and the negative sample to obtain the information push model; the adjustment step: predicting the push success rate of all users to be pushed in the set to be pushed according to the information push model, determining potential push users from the set to be pushed according to the push success rate, adding the data of the potential push users to the positive sample, and deleting the data of the potential push users from the set to be pushed; the control step: controlling the loop to iteratively execute the sample acquisition step, the training step and the adjustment step until the information push model meets the predetermined conditions.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 7 when executed.

Images