Information pushing method and device, equipment, storage medium and program product

Abstract

The invention discloses an information pushing method and device, equipment, a storage medium and a program product, and belongs to the technical field of information pushing. The method comprises the steps of obtaining object attribute features and object behavior features corresponding to at least two candidate objects; determining a conversion probability difference value corresponding to each candidate object based on the object attribute characteristics and the object behavior characteristics by using a causal inference network, the conversion probability difference value is used for representing a difference value between a first conversion probability value predicted by the candidate object under the condition that the push information is sent and a second conversion probability value predicted by the candidate object under the condition that the push information is not sent; and screening out a push object from the at least two candidate objects based on the conversion probability difference, and sending push information to the push object. According to the method, the object which is most likely to be influenced by the pushed information can be accurately determined in the candidate objects, so that the activeness and the participation rate of the object are improved.

Description

Technical Field

[0001] Embodiments of the present application relate to the technical field of information push, and in particular, to an information push method, device, equipment, storage medium and program product. Background Art

[0002] With the development of computer technology, precise information push is applied to many Internet services, and its goal is to accurately push information to matching objects, thereby improving the activity of the objects.

[0003] In related technologies, through a deep information model, such as a response model, the conversion probability value of an object after receiving the sent information is predicted, and the push objects are screened according to the conversion probability value for information push. Conversion means performing a certain behavior after receiving the push information. For example, a player did not originally plan to log in to the game, but logged in to the game after receiving the push information, and this behavior can be called conversion.

[0004] However, through the above method, it is difficult to determine whether the object conversion is due to the conversion of the push information or whether the conversion will still occur without receiving the push information. Therefore, how to accurately lock the push objects for information push is an important problem that needs to be solved urgently at present. Summary of the Invention

[0005] The present application provides an information push method, device, equipment, storage medium and program product, and the technical solution is as follows:

[0006] According to one aspect of the present application, an information push method is provided, and the method includes:

[0007] Obtain the object attribute features and object behavior features corresponding to at least two candidate objects, where the object attribute features are feature vectors for characterizing the object attributes of the candidate objects, and the object behavior features are feature vectors for characterizing the behaviors of the candidate objects;

[0008] Use a causal inference network to determine the conversion probability difference corresponding to each candidate object based on the object attribute features and the object behavior features, where the conversion probability difference is used to represent the difference between the first conversion probability value predicted for the candidate object when the push information is sent and the second conversion probability value predicted for the candidate object when the push information is not sent;

[0009] Screen out the push objects from at least two candidate objects based on the conversion probability difference, and send push information to the push objects.

[0010] According to one aspect of the present application, an information push device is provided, and the device includes:

[0011] An acquisition module, configured to acquire the object attribute features and object behavior features corresponding to at least two candidate objects, where the object attribute features are feature vectors for characterizing the object attributes of the candidate objects, and the object behavior features are feature vectors for characterizing the behaviors of the candidate objects;

[0012] A calculation module, configured to use a causal inference network to determine the conversion probability difference corresponding to each of the candidate objects based on the object attribute features and the object behavior features, where the conversion probability difference is used to represent the difference between a first conversion probability value predicted when the candidate object is sent a push message and a second conversion probability value predicted when the candidate object is not sent a push message;

[0013] A sending module, configured to screen out a push object from at least two candidate objects based on the conversion probability difference, and send a push message to the push object.

[0014] According to another aspect of the present application, there is provided a computer device, which includes: a processor and a memory, where at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement the information push method described in the above aspect.

[0015] According to another aspect of the present application, there is provided a computer storage medium, where at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by the processor to implement the information push method described in the above aspect.

[0016] According to another aspect of the present application, there is provided a computer program product, where the computer program product includes a computer program, and the computer program is stored in a computer-readable storage medium; the computer program is read and executed by a processor of a computer device from the computer-readable storage medium, so that the computer device executes the information push method described in the above aspect.

[0017] The beneficial effects brought by the technical solution provided by the present application at least include:

[0018] Acquire the object attribute features and object behavior features corresponding to at least two candidate objects respectively; use a causal inference network to determine the conversion probability difference corresponding to each candidate object based on the object attribute features and object behavior features; screen out a push object from at least two candidate objects based on the conversion probability difference, and send a push message to the push object. By calculating the conversion probability difference corresponding to each candidate object, the present application can accurately determine the object most likely to be affected by the push message among the candidate objects, thereby improving the activity and participation rate of the object. Description of the Drawings

[0019] To more clearly illustrate the technical solutions in the embodiments of the present application, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0020] Figure 1 It is a schematic diagram of an information push method provided by an exemplary embodiment of the present application;

[0021] Figure 2 It is a schematic diagram of the architecture of a computer system provided by an exemplary embodiment of the present application;

[0022] Figure 3 It is a flowchart of an information push method provided by an exemplary embodiment of the present application;

[0023] Figure 4 It is a flowchart of an information push method provided by an exemplary embodiment of the present application;

[0024] Figure 5 It is a schematic diagram of an intervention subnet predicting a first conversion probability value provided by an exemplary embodiment of the present application;

[0025] Figure 6 It is a schematic diagram of a non-intervention subnet predicting a second conversion probability value provided by an exemplary embodiment of the present application;

[0026] Figure 7 It is a schematic diagram of an information push method provided by an exemplary embodiment of the present application;

[0027] Figure 8 It is a schematic diagram of a causal inference network provided by an exemplary embodiment of the present application;

[0028] Figure 9 It is a block diagram of an information push device provided by an exemplary embodiment of the present application;

[0029] Figure 10 It is a schematic diagram of the structure of a computer device provided by an exemplary embodiment of the present application. Detailed implementation manners

[0030] To make the objectives, technical solutions, and advantages of this application clearer, the following will further describe the embodiments of this application in detail with reference to the accompanying drawings. Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. On the contrary, they are merely examples of devices and methods consistent with some aspects of this application as detailed in the appended claims.

[0031] The terms used in this disclosure are for the purpose of describing specific embodiments only and are not intended to limit this disclosure. The singular forms "a", "the", and "said" used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0032] It should be understood that although the terms first, second, third, etc. may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other.

[0033] For ease of understanding, the following explains several terms related to this application.

[0034] Artificial Intelligence (AI) is a theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a way similar to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines to enable the machines to have the functions of perception, reasoning, and decision-making.

[0035] Artificial intelligence technology is an interdisciplinary subject with a wide range of fields involved, including both hardware-level technologies and software-level technologies. The basic technologies of artificial intelligence generally include, for example, sensors, dedicated 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 foundation model, can be widely applied to downstream tasks in various major directions of artificial intelligence after fine-tuning. The software technologies of artificial intelligence mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0036] Cloud technology refers to a hosting technology that unifies a series of resources such as hardware, software, and networks within a wide area network or local area network to achieve data computing, storage, processing, and sharing.

[0037] Cloud technology is a general term for network technology, information technology, integration technology, management platform technology, application technology, etc. based on the cloud computing business model. It can form a resource pool, be used on demand, and is flexible and convenient. Cloud computing technology will become an important support. The back-end services of technical network systems require a large amount of computing and storage resources, such as video websites, picture websites, and more portal websites. With the highly developed application of the Internet industry, in the future, each item may have its own identification mark and needs to be transmitted to the back-end system for logical processing. Data at different levels will be processed separately, and various industry data requires a powerful system back-end support, which can only be achieved through cloud computing.

[0038] Cloud computing is a computing model that distributes computing tasks across a resource pool composed of a large number of computing devices, enabling various application systems to obtain computing power, storage space, and information services as needed. The network that provides resources is called the "cloud". The resources in the "cloud" seem to be infinitely expandable to users, and can be obtained at any time, used on demand, expanded at any time, and paid according to usage.

[0039] As a basic capability provider of cloud computing, a cloud computing resource pool will be established, abbreviated as a cloud platform, generally called an IaaS (Infrastructure as a Service) platform. Various types of virtual resources are deployed in the resource pool for external customers to choose and use. The cloud computing resource pool mainly includes: computing devices (virtual machines, including operating systems), storage devices, and network devices.

[0040] According to the logical function division, the PaaS (Platform as a Service) layer can be deployed on the IaaS (Infrastructure as a Service) layer, and the SaaS (Software as a Service) layer can be deployed on top of the PaaS layer. Or the SaaS can also be directly deployed on the IaaS. PaaS is a platform for software operation, such as databases, Web containers, etc. SaaS is various business software, such as web portal websites, SMS mass senders, etc. Generally speaking, SaaS and PaaS are upper layers relative to IaaS.

[0041] Computer Vision (CV) is a science that studies how to enable machines to "see". Further speaking, it refers to machine vision that uses cameras and computers to replace human eyes for target recognition and measurement, and further performs graphic processing to make the computer process images that are more suitable for human eyes to observe or be transmitted to instruments for detection. As a scientific discipline, computer vision studies related theories and technologies, and attempts to establish artificial intelligence systems that can obtain information from images or multi-dimensional data. Large model technology has brought important changes to the development of computer vision technology. Pretrained models in the visual field such as swin-transformer, ViT, V-MOE, and MAE can be quickly and widely applied to downstream specific tasks after fine-tuning. Computer vision technology usually includes technologies such as image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content / behavior recognition, three-dimensional object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping, and also includes common biometric recognition technologies.

[0042] Machine Learning (ML) is an interdisciplinary subject that involves multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve their own performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent, and its applications cover all fields of artificial intelligence. Machine learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and learning from demonstration. Pretrained models are the latest development results of deep learning, integrating the above technologies.

[0043] The embodiments of this application provide a schematic diagram of an information push method, as Figure 1 shown. This method can be executed by a computer device, and the computer device can be a terminal or a server.

[0044] The core of information push lies in pushing information to the matching target population, that is, sending information to the target population that will respond to the received information, so as to improve user activity, recover potential lost users, and improve application usage rate, etc. For example, in the game scenario, the core of information push is to accurately and real-time improve the active degree of users participating in the game.

[0045] Exemplarily, the computer device obtains the object attribute features 10 and object behavior features 20 corresponding to at least two candidate objects respectively; the computer device uses a causal inference network 30 to determine the conversion probability difference 60 corresponding to each candidate object based on the object attribute features 10 and object behavior features 20; the computer device screens out a push object from at least two candidate objects based on the conversion probability difference 60, and sends a push message 70 to the push object.

[0046] The object attribute feature 10 is a feature vector used to represent the object attributes of a candidate object. Optionally, the object attributes include at least one of age, gender, game rank, game title, etc., but are not limited thereto, and the embodiments of the present application do not make specific limitations in this regard.

[0047] The object behavior feature 20 is a feature vector used to characterize the behavior of a candidate object. Optionally, the behaviors of the candidate object include at least one of historical game performance, first registration time, daily startup times, daily active status, friend active status, and most recent login time, etc., but are not limited thereto, and the embodiments of the present application do not make specific limitations in this regard.

[0048] The conversion probability difference 60 is used to represent the difference between the first conversion probability value 40 predicted when the candidate object is sent the push message 70 and the second conversion probability value 50 predicted when the candidate object is not sent the push message 70.

[0049] Taking the game scenario as an example, the computer device predicts the first conversion probability value 40 that the player will log in to the game after receiving the push message 70 and the second conversion probability value 50 that the player will log in to the game without receiving the push message 70 according to the object attribute feature 10 and object behavior feature 20 corresponding to the player; by calculating the difference between the first conversion probability value 40 and the second conversion probability value 50, the influence degree of the recommendation message 70 on the player is determined.

[0050] For example, the probability value that player A will log in to the game after receiving the push message 70 is 0.9, and the probability value that player A will log in to the game after receiving the push message 70 is 0.3. Set the conversion threshold to 0.5. Since 0.9 - 0.3 is greater than 0.5, therefore, the push message will have an impact on player A, that is, player A is an object that should receive the push message 70. For example, when player A does not receive the push message 70 "Your close friend has been inactive for a long time. You will lose the combo with this close friend. Please team up and participate in the game as soon as possible", player A will not log in to the game, but when player A receives this push message, player A will log in to the game to view. That is to say, the sending of the push message has increased the player's game participation.

[0051] In summary, the method provided in this embodiment obtains the object attribute features and object behavior features corresponding to at least two candidate objects; uses a causal inference network to determine the conversion probability difference corresponding to each candidate object based on the object attribute features and object behavior features; filters out the push object from at least two candidate objects based on the conversion probability difference, and sends a push message to the push object. By calculating the conversion probability difference corresponding to each candidate object, this application can accurately determine the object most likely to be affected by the push message among the candidate objects, thereby improving the activity and participation rate of the object.

[0052] Figure 2 FIG. shows a schematic architecture diagram of a computer system provided in an embodiment of the present application. The computer system may include: a terminal 100 and a server 200.

[0053] The terminal 100 may be an electronic device such as a mobile phone, a tablet computer, a vehicle-mounted terminal (in-vehicle computer), a wearable device, a personal computer (PC), a vehicle-mounted terminal, an aircraft, a self-service vending terminal, etc. A client of a target application program may be installed and run in the terminal 100. The target application program may be an application program for pushing reference information, or other application programs with an information push function. The present application does not make any limitations in this regard. In addition, the form of the target application program is not limited in the present application, including but not limited to an application program (App), a mini-program, etc. installed in the terminal 100, and may also be in the form of a web page.

[0054] The server 200 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server, a cloud database, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery network (Content Delivery Network, CDN), and cloud servers providing basic cloud computing services such as big data and artificial palm image recognition platforms. The server 200 may be the background server of the above-mentioned target application program, and is used to provide background services for the client of the target application program.

[0055] Among them, cloud technology refers to a hosting technology that unifies a series of resources such as hardware, software, and networks within a wide area network or a local area network to achieve data computing, storage, processing, and sharing. Cloud technology is the general term for network technology, information technology, integration technology, management platform technology, application technology, etc. based on the cloud computing business model. It can form a resource pool, be used as needed, and is flexible and convenient. Cloud computing technology will become an important support. The background services of the technical network system require a large amount of computing and storage resources, such as video websites, picture websites, and more portal websites. With the high development and application of the Internet industry, in the future, each item may have its own identification mark and needs to be transmitted to the background system for logical processing. Data at different levels will be processed separately, and various industry data requires the support of a powerful system background, which can only be achieved through cloud computing.

[0056] In some embodiments, the above server can also be implemented as a node in a blockchain system. Blockchain is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithms. Blockchain is essentially a decentralized database, a string of data blocks generated by using cryptographic methods. Each data block contains information about a batch of network transactions, which is used to verify the validity (anti-counterfeiting) of the information and generate the next block. Blockchain can include the blockchain underlying platform, the platform product service layer, and the application service layer.

[0057] Communication can be carried out between the terminal 100 and the server 200 through a network, such as a wired or wireless network.

[0058] In the information push method provided by the embodiments of the present application, the execution subject of each step can be a computer device, and the computer device refers to an electronic device with data computing, processing, and storage capabilities. Taking Figure 2 the implementation environment of the shown solution as an example, the information push method can be executed by the terminal 100 (such as the client of the target application installed and running in the terminal 100 executes the information push method), or can be executed by the server 200, or can be executed by the interaction and cooperation of the terminal 100 and the server 200. The present application does not make any limitations in this regard.

[0059] Figure 3 is a flowchart of the information push method provided by an exemplary embodiment of the present application. This method can be executed by a computer device, and the computer device can be a terminal or a server. This method includes:

[0060] Step 302: Obtain the object attribute features and object behavior features respectively corresponding to at least two candidate objects.

[0061] The object attribute feature is a feature vector representing the object attributes of a candidate object. Optionally, the object attributes include at least one of age, gender, game rank, game title, etc., but are not limited thereto. The embodiments of the present application do not make specific limitations in this regard.

[0062] The object behavior feature is a feature vector used to characterize the behavior of a candidate object. Optionally, the behaviors of the candidate object include at least one of historical game performance, time of first registration, number of daily launches, daily active status, friend active status, and time of last login, etc., but are not limited thereto. The embodiments of the present application do not make specific limitations in this regard.

[0063] Step 304: Use the causal inference network to determine the conversion probability difference corresponding to each candidate object based on the object attribute feature and the object behavior feature.

[0064] The causal inference network is used to predict the influence probability value of a certain intervention on the object state or response behavior. Or, the causal inference network is used to predict the influence probability value of the push information on the response behavior of the object.

[0065] The conversion probability difference is used to represent the difference between the first conversion probability value predicted when the candidate object is sent the push information and the second conversion probability value predicted when the candidate object is not sent the push information.

[0066] The first conversion probability value refers to the probability value that the candidate object will perform a response behavior after being sent the push information. The second conversion probability value refers to the probability value that the candidate object will perform a response behavior when not being sent the push information.

[0067] Optionally, the response behavior includes at least one of logging in to the game, logging in to the application, updating the application, purchasing goods, etc., but is not limited thereto. The embodiments of the present application do not make specific limitations in this regard.

[0068] Taking the game scenario as an example, the computer device predicts the first conversion probability value that the player will log in to the game after receiving the push information, and the second conversion probability value that the player will log in to the game when not receiving the push information 70; by calculating the difference between the first conversion probability value and the second conversion probability value, the conversion probability difference is obtained, so as to determine the influence degree of the recommended information on the player.

[0069] Taking a shopping scenario as an example, the computer device predicts a first conversion probability value that the shopper will purchase product A after receiving push information about product A, and predicts a second conversion probability value that the shopper will purchase product A after not receiving push information about product A, based on the object attribute characteristics and object behavior characteristics corresponding to the shopper; by calculating the difference between the first conversion probability value and the second conversion probability value, the conversion probability difference is obtained, thereby determining the degree of influence of the recommended information on the shopper.

[0070] It should be noted that the embodiments of this application only cite two exemplary embodiments of sending push information in game scenarios and shopping scenarios, but are not limited to these. The method is applicable to any scenario that increases the activity of users or objects by sending push information. For example, merchants can push advertisements to push objects to attract new customers to their stores; merchants can send coupons to push objects in a targeted manner via SMS.

[0071] Step 306: Filter out a push object from at least two candidate objects based on the conversion probability difference, and send push information to the push object.

[0072] A push object refers to an object that is easily affected by the push information; or, a push object refers to an object that will respond based on the push information; or, a push object refers to a target object to which the push information is sent.

[0073] Exemplarily, after acquiring the conversion probability difference, the computer device screens candidate objects according to the conversion probability difference, determines a push object to which the push information is to be sent, and sends the push information to the push object.

[0074] Optionally, the computer device may send push information to some of the objects in the push object; or, the computer device may send push information to all of the objects in the push object.

[0075] In summary, the method provided in this embodiment obtains the object attribute characteristics and object behavior characteristics corresponding to at least two candidate objects; determines the conversion probability difference corresponding to each candidate object based on the object attribute characteristics and object behavior characteristics using a causal inference network; selects a push object from at least two candidate objects based on the conversion probability difference, and sends a push message to the push object. By calculating the conversion probability difference corresponding to each candidate object, the present application can accurately determine the object most likely to be affected by the push information among the candidate objects, thereby improving the activity and participation rate of the object.

[0076] Figure 4 1 is a flowchart of an information push method provided by an exemplary embodiment of the present application. The method can be executed by a computer device, which can be a terminal or a server. The method includes:

[0077] Step 402: Obtain the object attribute features and object behavior features corresponding to at least two candidate objects respectively.

[0078] The object attribute feature is a feature vector used to represent the object attributes of the candidate object. The object behavior feature is a feature vector used to characterize the behavior of the candidate object.

[0079] For the relevant introduction of the object attribute feature and the object behavior feature, reference can be made to the description in step 302, and no specific limitation is made here.

[0080] Step 404: Fuse the object attribute feature and the object behavior feature of each candidate object to obtain the object feature vector of each candidate object.

[0081] The object feature vector is a feature vector used to describe the attributes and behaviors of the object.

[0082] Optionally, the method for fusing the object attribute feature and the object behavior feature includes at least one of the following methods, but is not limited thereto:

[0083] · Add the corresponding feature values of the object attribute feature and the object behavior feature to obtain the object feature vector of each candidate object; optionally, add the feature values of the same dimension corresponding to the object attribute feature and the object behavior feature to obtain the object feature vector.

[0084] For example, if both the object attribute feature and the object behavior feature are feature vectors with a dimension of n*m, after adding the corresponding feature values of the object attribute feature and the object behavior feature, the dimension of the obtained object feature vector is still n*m.

[0085] · Concatenate the corresponding feature values of the object attribute feature and the object behavior feature to obtain the object feature vector of each candidate object. Optionally, perform arbitrary concatenation on the corresponding feature values of the object attribute feature and the object behavior feature to obtain the object feature vector.

[0086] For example, if both the object attribute feature and the object behavior feature are feature vectors with a dimension of n*m, after concatenating the corresponding feature values of the object attribute feature and the object behavior feature, the dimension of the obtained object feature vector is 2n*m.

[0087] Step 406: Input the object feature vector of each candidate object into the intervention sub-network to predict and obtain the first conversion probability value.

[0088] The first conversion probability value refers to the probability value that the candidate object will perform a response behavior after receiving the push message.

[0089] Optionally, the response behavior includes at least one of logging in to the game, logging in to the application, updating the application, purchasing goods, etc., but is not limited thereto, and no specific limitation is made in the embodiments of the present application.

[0090] The causal inference network includes at least one layer of intervention subnetworks and at least one layer of non-intervention subnetworks.

[0091] The intervention subnetwork is used to predict the probability value of a candidate object making a response behavior after receiving a push message.

[0092] Optionally, the intervention subnetwork and the non-intervention subnetwork have the same network structure but different parameters.

[0093] Exemplarily, the step of predicting the first conversion probability value by the intervention subnetwork based on the object feature vector includes: the computer device linearly adds the feature values in the object feature vector of each candidate object after adding the intervention weight to obtain the first conversion value; the computer device maps the first conversion value to the interval from 0 to 1 through a first activation function to obtain the first conversion probability value.

[0094] The intervention weight value is used to represent the coefficient added by the intervention subnetwork to the feature value in the object feature vector.

[0095] Optionally, the first activation function includes at least one of Sigmoid, Tanh, ReLU, LReLU, PReLU, Swish, but is not limited thereto, and the embodiments of the present application do not make specific limitations in this regard.

[0096] Optionally, each layer of the intervention subnetwork includes at least one intervention calculation unit.

[0097] The step for the intervention calculation unit to obtain the first conversion value includes: the computer device linearly adds the feature values in the object feature vector of each candidate object after adding the intervention weight value through the i-th intervention calculation unit in the intervention subnetwork to obtain the i-th conversion value, where i is a positive integer; the computer device aggregates the first i conversion values calculated by the first i intervention calculation units to obtain the first conversion value.

[0098] For example, as Figure 5 shown in the schematic diagram of the intervention subnetwork predicting the first conversion probability value, the computer device obtains the object feature vector 501 of each candidate object. For example, the feature values in the object feature vector 501 of a single candidate object are x1, x2, x3. The intervention subnetwork 502 includes at least one intervention calculation unit 503. A small circle in the figure represents an intervention calculation unit 503. The computer device inputs the feature values in the object feature vector 501 of the candidate object after adding different intervention weight values into the intervention calculation unit 502. The intervention calculation unit 502 linearly adds the feature values after adding different intervention weight values to obtain the first conversion value 504. For example, the feature values in the first conversion value 504 are y1, y2, y3.

[0099] Optionally, the dimension of the first transformed value 504 may be the same as that of the object feature vector 501, or may be different from that of the object feature vector 501.

[0100] Step 408: Input the object feature vector of each candidate object into the non-intervention sub-network to predict the second transformation probability value.

[0101] The second transformation probability value refers to the probability value that the candidate object will perform a response behavior without being sent a push message.

[0102] The causal inference network includes at least one non-intervention sub-network.

[0103] The non-intervention sub-network is used to predict the probability value that the candidate object will perform a response behavior after not being sent a push message.

[0104] Exemplarily, the steps for the non-intervention sub-network to predict the second transformation probability value based on the object feature vector include: the computer device linearly adds the feature values in the object feature vector of each candidate object after adding non-intervention weights to obtain a second transformed value; the computer device maps the second transformed value to the interval from 0 to 1 through a second activation function to obtain the second transformation probability value.

[0105] The non-intervention weight value is used to represent the coefficient added by the non-intervention sub-network to the feature value in the object feature vector.

[0106] Optionally, the second activation function includes at least one of Sigmoid, Tanh, ReLU, LReLU, PReLU, Swish, but is not limited thereto, and the embodiments of the present application do not make specific limitations in this regard.

[0107] It should be noted that the first activation function and the second activation function may be the same activation function or different activation functions, and the embodiments of the present application do not make specific limitations in this regard.

[0108] Optionally, each non-intervention sub-network includes at least one non-intervention calculation unit.

[0109] The steps for the non-intervention calculation unit to obtain the second transformed value include: the computer device linearly adds the feature values in the object feature vector of each candidate object after adding non-intervention weight values through the j-th intervention calculation unit in the non-intervention sub-network to obtain the j-th transformed value, where j is a positive integer; the computer device aggregates the first j transformed values calculated by the first j intervention calculation units to obtain the second transformed value.

[0110] For example, as Figure 6Schematic diagram of the non-intervention subnetwork predicting the second conversion probability value. The computer device obtains the object feature vector 601 of each candidate object. The feature values in the object feature vector of a single candidate object are x1, x2, x3. The non-intervention subnetwork 602 includes at least one non-intervention calculation unit 603. A shaded small circle in the figure is used to represent a non-intervention calculation unit 603. The computer device inputs the feature values in the object feature vector 601 of the candidate object after adding different non-intervention weight values into the non-intervention calculation unit 603. The non-intervention calculation unit 603 linearly adds the feature values after adding different non-intervention weight values to obtain the second conversion value 604. For example, the feature values in the second conversion value 604 are z1, z2, z3.

[0111] Optionally, the dimension of the second conversion value 604 can be the same as the dimension of the object feature vector 601, or the dimension of the second conversion value 604 can be different from the dimension of the object feature vector 601.

[0112] Step 410: Subtract the first conversion probability value from the second conversion probability value to obtain a conversion probability difference.

[0113] The conversion probability difference is used to represent the difference between the first conversion probability value predicted for the candidate object when the push message is sent and the second conversion probability value predicted for the candidate object when the push message is not sent.

[0114] Step 412: Screen out the push objects from at least two candidate objects based on the conversion probability difference, and send push messages to the push objects.

[0115] A push object refers to an object that is easily affected by the push message; or, a push object refers to an object that will respond based on the push message; or, a push object refers to the target object of the push message.

[0116] Exemplarily, after obtaining the conversion probability difference, the computer device determines the candidate objects with the conversion probability difference greater than the preset conversion threshold as the push objects, and sends push messages to the push objects.

[0117] Optionally, the computer device sorts the candidate objects according to the conversion probability difference to obtain a sorting result; determines the top n candidate objects as the push objects, where n is a positive integer.

[0118] In some embodiments, the computer device obtains the push plan of the push message in the target time period. The push plan is determined by at least one of traffic, time, and conversion effect; the computer device adjusts the preset push ratio parameter according to the expected push volume in the push plan to obtain a corrected push ratio parameter; the computer device sends push messages to the push objects according to the corrected push ratio parameter.

[0119] The corrected push ratio parameter is obtained based on the actual push volume, the expected push volume, and the preset push ratio parameter.

[0120] Optionally, the target time period includes at least one of holidays, weekdays, evenings, daytime, and specified time periods, but is not limited thereto. The embodiments of the present application do not make specific limitations on this.

[0121] Optionally, the corrected push ratio parameter is obtained by modifying the preset push ratio parameter according to the difference between the actual push volume and the expected push volume.

[0122] The calculation formula of the corrected push ratio parameter can be expressed as:

[0123]

[0124] where, is used to represent the corrected push ratio parameter, P(online) is used to represent the actual push volume, P(target) is used to represent the expected push volume, Δ is used to represent the preset push ratio parameter, and K is used to represent the correction coefficient.

[0125] By dynamically adjusting the delivery volume, the delivery volume (push volume) of the push information can be well stabilized at the target delivery volume. If the information is sent to all recommended objects, the total number of actual push objects of the information will deviate from the target value, especially when the information push time period is short, the deviation degree will be greater.

[0126] In summary, the method provided in this embodiment obtains the object attribute features and object behavior features corresponding to at least two candidate objects; uses a causal inference network to determine the conversion probability difference corresponding to each candidate object based on the object attribute features and object behavior features; screens out the push objects from at least two candidate objects based on the conversion probability difference, and sends push information to the push objects. By calculating the conversion probability difference corresponding to each candidate object, the present application can accurately determine the objects most likely to be affected by the push information among the candidate objects, thereby improving the activity and participation rate of the objects.

[0127] The method provided in this embodiment calculates the first conversion probability value and the second conversion probability value corresponding to each candidate object by using an intervention subnet and a non-intervention subnet with the same network structure but different parameters, and determines the influence degree of the push information on the candidate object according to the difference between the first conversion probability value and the second conversion probability value, so as to accurately determine the push objects most likely to be affected by the push information among the candidate objects.

[0128] The method provided in this embodiment offers two ways of feature fusion. One way of feature fusion is to add the features, and the other way is to splice the features. During actual use, different ways can be selected according to the characteristics of the features to achieve feature fusion, so as to obtain the object feature vector corresponding to each candidate object.

[0129] In the method provided in this embodiment, after determining the conversion probability difference, all candidate objects with a conversion probability difference greater than the preset conversion threshold can be determined as push objects, or the candidate objects ranked in the top n and with a conversion probability difference greater than the preset conversion threshold can be determined as push objects. The push objects for sending push messages can be determined according to different requirements, improving the determination efficiency of push objects.

[0130] In the method provided in this embodiment, after determining the push objects, the corrected push ratio parameter can be obtained by adjusting the push ratio parameter, and the push objects for sending push messages can be selected according to the corrected push ratio parameter. That is, the overall push volume can be controlled by the corrected push ratio parameter, thereby improving the activity and participation rate of the objects.

[0131] Figure 7 It is a schematic diagram of an information push method provided by an exemplary embodiment of the present application. This method can be executed by a computer device, and the computer device can be a terminal or a server. The method includes:

[0132] The computer device acquires the object attribute features 701 and object behavior features 702 corresponding to at least two candidate objects respectively.

[0133] The computer device fuses the object attribute features 701 and object behavior features 702 of each candidate object to obtain the object feature vector of each candidate object.

[0134] The computer device inputs the object feature vector of each candidate object into the intervention sub-network 703 to predict and obtain the first conversion probability value 705.

[0135] The computer device inputs the object feature vector of each candidate object into the non-intervention sub-network 704 to predict and obtain the second conversion probability value 706.

[0136] The computer device subtracts the second conversion probability value 706 from the first conversion probability value 705 to obtain the conversion probability difference 707.

[0137] The computer device determines the candidate objects with a conversion probability difference 707 greater than the preset conversion threshold as push objects; and sends push messages to the push objects.

[0138] For example, the probability that player A will log in to the game after receiving the push message 70 is 0.9, and the probability that player A will log in to the game after receiving the push message 70 is 0.3. The conversion threshold is set to 0.5. Since 0.9 - 0.3 is greater than 0.5, player A is determined as the push target. The push message 708 is sent to player A. For example, the push message 708 is "Your close friend has been inactive for a long time. You will lose the combo with this close friend. Please team up and participate in the game as soon as possible." In the case where player A does not receive the push message 708, player A will not log in to the game, but after receiving this push message, player A will log in to the game to check. That is to say, the sending of the push message increases the player's game participation rate.

[0139] In summary, the method provided in this embodiment obtains the object attribute features and object behavior features corresponding to at least two candidate objects respectively; uses a causal inference network to determine the conversion probability difference corresponding to each candidate object based on the object attribute features and object behavior features; screens out the push target from at least two candidate objects based on the conversion probability difference, and sends a push message to the push target. By calculating the conversion probability difference corresponding to each candidate object, this application can accurately determine the object most likely to be affected by the push message among the candidate objects, thereby improving the activity and participation rate of the object.

[0140] Figure 8 It is a schematic diagram of a causal inference network provided by an exemplary embodiment of this application. The computer device obtains the object attribute feature 801 and object behavior feature 802 corresponding to each of at least two candidate objects. The object attribute feature 801 is a feature vector used to characterize the object attribute of the candidate object, and the object behavior feature 802 is a feature vector used to characterize the behavior of the candidate object. The computer device fuses the object attribute feature 801 and object behavior feature 802 of each candidate object through the feature encoding network 803, and adjusts the feature dimension through at least one fully connected layer 804, that is, condenses the features through at least one fully connected layer 804, to obtain the object feature vector of each candidate object.

[0141] The computer device inputs the object feature vector of each candidate object into at least one intervention sub-network 805 for prediction, and maps the first conversion value obtained by the prediction to the interval from 0 to 1 through the first activation function 806 to obtain the first conversion probability value 807. Among them, the first conversion probability value 807 can be expressed as Q(1, x), where x is used to represent the xth candidate object, and 1 is used to represent sending a push message.

[0142] The computer device inputs the object feature vector of each candidate object into at least one non-intervention subnetwork 808 for prediction, and maps the predicted second conversion value to the interval of 0 to 1 through a second activation function 809 to obtain a second conversion probability value 810, wherein the second conversion probability value 810 can be expressed as Q(0, x), wherein x is used to represent the xth candidate object and 0 is used to represent that no push information has been sent.

[0143] The computer device directly maps the object feature vector of each candidate object to the interval of 0 to 1 through the third activation function 811 to obtain a bias score 812. The bias score 812 can be used to determine whether the object feature vector is biased to be input into the intervention sub-network 805 or the non-intervention sub-network 808.

[0144] For example, intelligent marketing and push methods are becoming more and more popular. Merchants can reach consumers through various channels. For example, merchants can target target groups for advertising push to attract new customers for their stores, or distribute coupons through SMS or Wangwang channels. The goal of push is to maximize the total output of push with limited costs. The most critical point here is whether we can accurately find users who can really be attracted by the information. We call them sensitive groups.

[0145] From the perspective of the population, we can make statistics on the average causal effect. Suppose we have two groups of homogeneous users, both from first-tier cities / aged 25-35 / male. We can send push messages to one group of users and not intervene in the other group. Then we can count the difference in conversion rate between the two groups. This difference can be roughly regarded as the possible average causal effect of people with the same characteristics. Therefore, the causal inference network is essentially the average causal effect of the conditions learned from the training samples. At the same time, because the model network has a certain generalization ability, it can also make predictions for samples that have not been seen.

[0146] It should be noted that the causal inference network has relatively high requirements for samples, requiring that the object attribute feature 801 and the variable of the pushed information are independent of each other. The two groups of samples obtained by splitting the traffic through A / B Test have the same distribution of features, that is, the object attribute feature 801 and the variable of the pushed information are independent of each other.

[0147] In the figure, the network model mainly uses the object attribute feature 801 and the object behavior feature 802 to represent the object feature vector of each candidate object through feature processing and the deep network hidden layer; based on the object feature vector of each candidate object, it predicts T (the response behavior of the object) and Y (whether to send push information). By calculating the difference between the first conversion probability value 807 and the second conversion probability value 810, the conversion probability difference is obtained. The computer device determines the delivery threshold according to the conversion probability difference and the actual number of objects to be delivered, and adjusts it in real time online.

[0148] The scheme for adjusting the delivery volume of push information in real time online is as follows:

[0149] The influence of each coefficient in the PID algorithm on the system is very important. Therefore, before the algorithm goes online, it is necessary to determine relatively reasonable coefficients in advance to avoid unstable algorithm performance after going online, resulting in large fluctuations in the delivery volume of push information and causing unnecessary pushes. In addition, due to the uncertain factors of the game login population, even when setting the delivery volume with the same delivery threshold all the time, different delivery volumes will still be presented.

[0150] When initially setting parameters, only the linear relationship between the delivery volume of push information and the actual delivery volume can be considered, and different parameters K are used for experiments to observe the influence of different parameters on the system performance, and initially obtain the interval of relatively reasonable parameters for each coefficient. It can be found from the experimental results that when the proportional coefficient is larger, the oscillation amplitude of the system is larger; when the integral coefficient is larger, the time for the system to reach a steady state is longer; when the derivative coefficient is larger, the system will continue to oscillate and it is difficult to reach a stable state. Usually, we will introduce a random term for fine-tuning to obtain a more ideal PID controller.

[0151] Using the PID control algorithm to dynamically adjust the delivery volume of push information can well stabilize the delivery volume of push information (push volume) at the target delivery volume. However, using a fixed information push strategy will cause the actual total number of pushed players of information push to deviate from the target value, especially when the information push time period is short, the deviation degree will be greater. In the actual delivery scenario, there are certain constraints on the actual information push time and whether it is a holiday, and it is necessary to process the output of the PID algorithm according to the business scenario to make the algorithm output stable information to the target population (that is, the push object).

[0152] Figure 9 The structural schematic diagram of the information push device provided by an exemplary embodiment of the present application is shown. This device can be implemented as all or part of a computer device through software, hardware, or a combination of both. This device includes:

[0153] An acquisition module 801, configured to acquire object attribute features and object behavior features corresponding to at least two candidate objects, where the object attribute features are feature vectors for characterizing the object attributes of the candidate objects, and the object behavior features are feature vectors for characterizing the behaviors of the candidate objects;

[0154] A calculation module 802, configured to use a causal inference network to determine a conversion probability difference corresponding to each of the candidate objects based on the object attribute features and the object behavior features, where the conversion probability difference is used to represent the difference between a first conversion probability value predicted when the candidate object is sent a push message and a second conversion probability value predicted when the candidate object is not sent a push message;

[0155] A sending module 803, configured to screen out a push object from at least two candidate objects based on the conversion probability difference, and send a push message to the push object.

[0156] The causal inference network includes at least one intervention subnet and at least one non-intervention subnet.

[0157] In some embodiments, the calculation module 802 is further configured to fuse the object attribute features and the object behavior features of each candidate object to obtain an object feature vector for each candidate object; input the object feature vector for each candidate object into the intervention subnet to predict the first conversion probability value; and input the object feature vector for each candidate object into the non-intervention subnet to predict the second conversion probability value; subtract the second conversion probability value from the first conversion probability value to obtain the conversion probability difference; where the network structures of the intervention subnet and the non-intervention subnet are the same but the parameters are different.

[0158] In some embodiments, the calculation module 802 is further configured to linearly add the feature values in the object feature vector for each candidate object after adding an intervention weight value, where the intervention weight value is used to represent the coefficient added by the intervention subnet to the feature values in the object feature vector, to obtain a first conversion value, and map the first conversion value to the interval from 0 to 1 through a first activation function to obtain the first conversion probability value.

[0159] In some embodiments, the calculation module 802 is further configured to linearly add the feature values in the object feature vector for each candidate object after adding the intervention weight value through the i-th intervention calculation unit in the intervention subnet to obtain an i-th conversion value, where i is a positive integer; summarize the first i conversion values calculated by the first i intervention calculation units to obtain the first conversion value.

[0160] In some embodiments, the computing module 802 is further configured to linearly add the eigenvalues in the object feature vectors of each of the candidate objects after adding the non-intervention weight values, to obtain a second conversion value, where the non-intervention weight value is used to represent the coefficient added by the non-intervention sub-network to the eigenvalues in the object feature vector; and map the second conversion value to the interval from 0 to 1 through a second activation function, to obtain the second conversion probability value.

[0161] In some embodiments, the computing module 802 is further configured to linearly add the eigenvalues in the object feature vectors of each of the candidate objects after adding the non-intervention weight values through the j-th non-intervention calculation unit in the non-intervention sub-network, to obtain the j-th conversion value, where j is a positive integer; and summarize the first j conversion values calculated by the first j non-intervention calculation units, to obtain the second conversion value.

[0162] In some embodiments, the computing module 802 is further configured to add the eigenvalues corresponding to the object attribute feature and the object behavior feature respectively, to obtain the object feature vector of each of the candidate objects.

[0163] In some embodiments, the computing module 802 is further configured to splice the eigenvalues corresponding to the object attribute feature and the object behavior feature respectively, to obtain the object feature vector of each of the candidate objects.

[0164] In some embodiments, the sending module 803 is further configured to determine the candidate objects with the conversion probability difference greater than a preset conversion threshold as the push objects; and send the push information to the push objects.

[0165] In some embodiments, the sending module 803 is further configured to sort the candidate objects according to the conversion probability difference, to obtain a sorting result; and determine the candidate objects ranked in the top n as the push objects, where n is a positive integer.

[0166] In some embodiments, the obtaining module 801 is further configured to obtain a push plan of the push information in a target time period, where the push plan is determined by at least one of traffic, time, and conversion effect.

[0167] In some embodiments, the computing module 802 is further configured to adjust a preset push ratio parameter according to the expected push volume in the push plan, to obtain a corrected push ratio parameter, where the corrected push ratio parameter is obtained according to the actual push volume, the expected push volume, and the preset push ratio parameter.

[0168] In some embodiments, the sending module 803 is further configured to send the push information to the push objects according to the corrected push ratio parameter.

[0169] Figure 10 FIG. Figure 10 shows a block diagram of a computer device 900 shown in an exemplary embodiment of the present application. The computer device may be implemented as the server in the above solution of the present application. The computer device 900 includes a central processing unit (CPU) 901, a system memory 904 including a random access memory (RAM) 902 and a read-only memory (ROM) 903, and a system bus 905 connecting the system memory 904 and the central processing unit 901. The computer device 900 further includes a mass storage device 906 for storing an operating system 909, application programs 910, and other program modules 911.

[0170]

[0170] The mass storage device 906 is connected to the central processing unit 901 through a mass storage controller (not shown) connected to the system bus 905. The mass storage device 906 and its associated computer-readable medium provide non-volatile storage for the computer device 900. That is to say, the mass storage device 906 may include a computer-readable medium (not shown) such as a hard disk or a compact disc read-only memory (CD-ROM) drive.

[0171]

[0171] Without loss of generality, the computer-readable medium may include computer storage media and communication media. Computer storage media includes volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes RAM, erasable programmable read-only memory (EPROM), electrically-erasable programmable read-only memory (EEPROM), flash memory or other solid-state storage technologies, CD-ROM, digital versatile disc (DVD) or other optical storage, magnetic tape cartridges, tapes, disk storage or other magnetic storage devices. Of course, those skilled in the art will appreciate that the computer storage media is not limited to the above several. The above system memory 904 and mass storage device 906 may be collectively referred to as memory.

[0172] According to various embodiments of the present disclosure, the computer device 900 may also run on a remote computer on the network through a network such as the Internet. That is, the computer device 900 may be connected to the network 908 through the network interface unit 907 connected to the system bus 905. Or rather, the network interface unit 907 may also be used to connect to other types of networks or remote computer systems (not shown).

[0173] The memory further includes at least one segment of computer program, which is stored in the memory, and the central processing unit 901 implements all or part of the steps in the information push method shown in the above various embodiments by executing the at least one segment of program.

[0174] An embodiment of the present application further provides a computer device, which includes a processor and a memory. At least one program is stored in the memory, and the at least one program is loaded and executed by the processor to implement the information push method provided in the above method embodiments.

[0175] An embodiment of the present application further provides a computer-readable storage medium, in which at least one computer program is stored, and the at least one computer program is loaded and executed by the processor to implement the information push method provided in the above method embodiments.

[0176] An embodiment of the present application further provides a computer program product, which includes a computer program stored in a computer-readable storage medium; the computer program is read and executed by a processor of a computer device, so that the computer device executes to implement the information push method provided in the above method embodiments.

[0177] It can be understood that in the specific implementation of the present application, for data, historical data, portraits and other user data processing related to user identity or characteristics, when the above embodiments of the present application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0178] It should be noted that unless otherwise clearly defined herein, all terms used in the claims are interpreted according to their ordinary meanings in the technical field. Unless otherwise clearly stated, all references to "an element, device, component, equipment, step, etc." will be interpreted openly as referring to at least one instance of the element, device, component, equipment, step, etc. Unless clearly stated, the steps of any method disclosed herein are not necessarily to be executed in the exact order disclosed.

[0179] It should be understood that the "plurality" mentioned herein refers to two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships may exist. For example, A and / or B may represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0180] Those of ordinary skill in the art can understand that all or part of the steps for implementing the above embodiments can be completed by hardware or can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium. The storage media mentioned above can be a read-only memory, a magnetic disk, an optical disk, or the like.

[0181] The foregoing are only optional embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. An information push method, characterized in that, The method includes: Obtaining the object attribute features and object behavior features corresponding to at least two candidate objects, where the object attribute features are feature vectors for characterizing the object attributes of the candidate objects, and the object behavior features are feature vectors for characterizing the behaviors of the candidate objects; Using a causal inference network to determine the conversion probability difference corresponding to each of the candidate objects based on the object attribute features and the object behavior features, where the conversion probability difference is used to represent the difference between the first conversion probability value predicted when the candidate object is sent a push message and the second conversion probability value predicted when the candidate object is not sent a push message; Screening out a push object from at least two of the candidate objects based on the conversion probability difference, and sending a push message to the push object.

2. The method according to claim 1, wherein The causal inference network includes at least one intervention sub-network and at least one non-intervention sub-network; The using a causal inference network to determine the conversion probability difference corresponding to each of the candidate objects based on the object attribute features and the object behavior features includes: Fusing the object attribute features and the object behavior features of each of the candidate objects to obtain the object feature vector of each of the candidate objects; Inputting the object feature vector of each of the candidate objects into the intervention sub-network to predict the first conversion probability value; and inputting the object feature vector of each of the candidate objects into the non-intervention sub-network to predict the second conversion probability value; Subtracting the second conversion probability value from the first conversion probability value to obtain the conversion probability difference; Wherein, the network structures of the intervention sub-network and the non-intervention sub-network are the same but the parameters are different.

3. The method according to claim 2, wherein The inputting the object feature vector of each of the candidate objects into the intervention sub-network to predict the first conversion probability value includes: Adding intervention weight values to the feature values in the object feature vector of each of the candidate objects and linearly adding them to obtain a first conversion value, where the intervention weight value is used to represent the coefficient added by the intervention sub-network to the feature value in the object feature vector; Mapping the first conversion value to the interval from 0 to 1 through a first activation function to obtain the first conversion probability value.

4. The method according to claim 3, characterized in that Each layer of the intervention sub-network includes at least one intervention calculation unit; The adding intervention weight values to the feature values in the object feature vector of each of the candidate objects and linearly adding them to obtain a first conversion value includes: Linearly adding the intervention weight value to the feature value in the object feature vector of each of the candidate objects through the i-th intervention calculation unit in the intervention sub-network to obtain the i-th conversion value, where i is a positive integer; Summarizing the first i conversion values calculated by the first i intervention calculation units to obtain the first conversion value.

5. The method according to claim 2, wherein The inputting the object feature vector of each of the candidate objects into the non-intervention sub-network to predict the second conversion probability value includes: Add the eigenvalue in the object feature vector of each candidate object with an intervention-free weight value and linearly sum them to obtain a second conversion value, where the intervention-free weight value is used to represent the coefficient added by the intervention-free sub-network to the eigenvalue in the object feature vector; Map the second conversion value to the interval of 0 to 1 through a second activation function to obtain the second conversion probability value.

6. The method according to claim 5, characterized in that Each layer of the intervention-free sub-network includes at least one intervention-free calculation unit; The step of adding the eigenvalue in the object feature vector of each candidate object with an intervention-free weight value and linearly summing them to obtain a second conversion value includes: Linearly sum the eigenvalue in the object feature vector of each candidate object with the intervention-free weight value through the j-th intervention-free calculation unit in the intervention-free sub-network to obtain the j-th conversion value, where j is a positive integer; Aggregate the first j conversion values calculated by the first j intervention-free calculation units to obtain the second conversion value.

7. The method according to claim 2, wherein The step of fusing the object attribute feature and the object behavior feature of each candidate object to obtain the object feature vector of each candidate object includes: Add the corresponding eigenvalues of the object attribute feature and the object behavior feature to obtain the object feature vector of each candidate object.

8. The method according to claim 2, wherein The step of fusing the object attribute feature and the object behavior feature of each candidate object to obtain the object feature vector of each candidate object includes: Concatenate the corresponding eigenvalues of the object attribute feature and the object behavior feature to obtain the object feature vector of each candidate object.

9. The method according to any one of claims 1 to 8, characterized in that, The step of screening out the push object from at least two candidate objects based on the conversion probability difference and sending a push message to the push object includes: Determine the candidate object with the conversion probability difference greater than the preset conversion threshold as the push object; Send the push message to the push object.

10. The method according to claim 9, wherein The step of determining the candidate object with the conversion probability difference greater than the preset conversion threshold as the push object includes: Sort the candidate objects according to the conversion probability difference to obtain a sorting result; Determine the candidate objects ranked in the top n and with the conversion probability difference greater than the preset conversion threshold as the push objects, where n is a positive integer.

11. According to the method described in any one of claims 1 to 8, characterized in that, The method further includes: Obtain the push plan of the push message within a target time period, where the push plan is determined by at least one of traffic, time, and conversion effect; Adjust the preset push ratio parameter according to the expected push volume in the push plan to obtain a corrected push ratio parameter, where the corrected push ratio parameter is obtained based on the actual push volume, the expected push volume, and the preset push ratio parameter; Send the push message to the push object according to the corrected push ratio parameter.

12. An information push device, characterized in that, The device includes: An acquisition module, configured to acquire the object attribute feature and the object behavior feature respectively corresponding to at least two candidate objects, where the object attribute feature is a feature vector used to characterize the object attribute of the candidate object, and the object behavior feature is a feature vector used to characterize the behavior of the candidate object; A calculation module, configured to determine a conversion probability difference corresponding to each of the candidate objects based on the object attribute features and the object behavior features by using a causal inference network, where the conversion probability difference is used to represent the difference between a first conversion probability value predicted when the candidate object is sent a push message and a second conversion probability value predicted when the candidate object is not sent a push message; A sending module, configured to screen out a push object from at least two candidate objects based on the conversion probability difference, and send a push message to the push object.

13. A computer device, characterized in that, The computer device includes: a processor and a memory, where at least one computer program is stored in the memory, and at least one computer program is loaded and executed by the processor to implement the information push method according to any one of claims 1 to 11.

14. A computer storage medium, characterized in that, At least one computer program is stored in the computer-readable storage medium, and at least one computer program is loaded and executed by a processor to implement the information push method according to any one of claims 1 to 11.

15. A computer program product, characterized in that, The computer program product includes a computer program, and the computer program is stored in a computer-readable storage medium; the computer program is read and executed by a processor of a computer device from the computer-readable storage medium, so that the computer device executes the information push method according to any one of claims 1 to 11.

Images