Information processing method and apparatus, storage medium, and device

Abstract

An information processing method, device, storage medium and equipment, the method comprises: in response to the trigger operation of sending a target resource for a target object, obtaining common information required to be used when sending the target resource for the target object in M scenes; obtaining the reference importance of each sub-information in the common information corresponding to sending the target resource for the target object in the i-th scene, and performing conversion processing on the common information according to the reference importance of each sub-information in the i-th scene to obtain the reference common information corresponding to the i-th scene; obtaining unique information required to be used in the i-th scene; performing splicing processing on the reference common information corresponding to the i-th scene and the corresponding unique information to obtain the sending indication information of the i-th scene, and predicting the sending probability of the target resource to the target object in the i-th scene according to the sending indication information. Through the method of the present application, the resource processing efficiency can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to an information processing method and device, a storage medium and equipment. BACKGROUND

[0002] With the rapid development of network technology, the resources that can be provided by network platforms are growing explosively, which makes it difficult to obtain the actual demand resources from the massive data. As an effective solution to information explosion, information sending systems have been widely used in the field of information sending. The existing information sending method is mostly based on the relevant data of users to send the interested resources to the corresponding users. However, research shows that the existing information processing method has the problem of low resource processing efficiency. SUMMARY

[0003] The embodiments of the present application provide an information processing method, device, storage medium and equipment, which can effectively improve the resource processing efficiency.

[0004] In one aspect, the embodiments of the present application provide an information processing method, which comprises:

[0005] In response to a trigger operation of sending a target resource to a target object, obtaining common information required by each scene when sending the target resource to the target object; M is an integer greater than 1, and the common information includes a plurality of sub-information;

[0006] Obtaining the reference importance of each sub-information in the i-th scene when sending the target resource to the target object, and converting the common information according to the reference importance of each sub-information in the i-th scene to obtain the reference common information corresponding to the i-th scene; i is a positive integer less than or equal to M;

[0007] Obtaining unique information required by the i-th scene when sending the target resource;

[0008] Splicing the reference common information corresponding to the i-th scene and the corresponding unique information to obtain the sending indication information of the i-th scene, and predicting the sending probability of the target resource to the target object in the i-th scene according to the sending indication information.

[0009] In one aspect, the embodiments of the present application provide an information processing device, which comprises:

[0010] The obtaining unit is configured to, in response to a trigger operation of sending a target resource to a target object, obtain common information required by each scene when sending the target resource to the target object; M is an integer greater than 1, and the common information includes a plurality of sub-information;

[0011] The acquisition unit is further configured to acquire a reference importance corresponding to sending of the target resource to the target object by each sub-information in the ith scene;

[0012] The processing unit is configured to perform conversion processing on the common information according to the reference importance of each sub-information in the ith scene, to obtain reference common information corresponding to the ith scene; i is a positive integer less than or equal to M;

[0013] The acquisition unit is further configured to acquire unique information required by the ith scene to use when sending the target resource in the ith scene.

[0014] The processing unit is further configured to perform splicing processing on the reference common information corresponding to the ith scene and the corresponding unique information, to obtain sending indication information of the ith scene, and to predict a sending probability of sending the target resource to the target object in the ith scene according to the sending indication information.

[0015] In an aspect, an embodiment of the present application provides a computer device, which comprises a processor, a communication interface and a memory, the processor, the communication interface and the memory are connected with each other, wherein the memory stores a computer program, and the processor is configured to invoke the computer program to execute the information processing method in any possible implementation manner.

[0016] In an aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the information processing method in any possible implementation manner.

[0017] In an aspect, an embodiment of the present application further provides a computer program product, the computer program product comprises a computer program or computer instructions, and the computer program or computer instructions are executed by a processor to implement the steps of the information processing method provided by the embodiment of the present application.

[0018] In an aspect, an embodiment of the present application further provides a computer program, the computer program comprises computer instructions, the computer instructions are stored in a computer readable storage medium, a processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the information processing method provided by the embodiment of the present application.

[0019] In the embodiments of the present application, the common information used in each scene and the unique information used in the i-th scene are determined, and the important information (i.e., the reference common information) applicable to the i-th scene can be adaptively selected from the common information by the reference importance of the sub-information in the common information in the i-th scene, which can solve the problem that the common information has the same semantics but different importance in each scene; then the reference common information of the i-th scene and the corresponding unique information are spliced to obtain the complete information representation (i.e., the sending indication information) of the i-th scene, and the sending probability of the target resource to the target object in the i-th scene can be determined based on the complete information representation of the i-th scene. Since the complete information representation retains the commonality between scenes and the uniqueness of the i-th scene, the prediction accuracy of the resource sending probability can be effectively improved, and thus the resource processing efficiency is improved in the subsequent resource sending process. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical method of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0021] Figure 1 A model structure diagram of a STAR model provided by the embodiments of the present application;

[0022] Figure 2 A system architecture diagram of an information processing system provided by the embodiments of the present application;

[0023] Figure 3 A flow diagram of an information processing method provided by the embodiments of the present application;

[0024] Figure 4 A flow diagram of another information processing method provided by the embodiments of the present application;

[0025] Figure 5 A structure diagram of an information gating sub-module provided by the embodiments of the present application;

[0026] Figure 6 A structure diagram of a star topology module provided by the embodiments of the present application;

[0027] Figure 7 A structure diagram of a prediction module provided by the embodiments of the present application;

[0028] Figure 8 A structure diagram of a personalized processing module provided by the embodiments of the present application;

[0029] Figure 9 A flowchart of a target prediction model acquisition method provided by an embodiment of the present application is shown in FIG. 1.

[0030] Figure 10 A model structure diagram of an initial target prediction model provided by an embodiment of the present application is shown in FIG. 2.

[0031] Figure 11 A structure diagram of an information processing apparatus provided by an embodiment of the present application is shown in FIG. 3.

[0032] Figure 12 A structure diagram of a computer device provided by an embodiment of the present application is shown in FIG. 4. DETAILED DESCRIPTION

[0033] The technical methods in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0034] First, the related terms and concepts involved in the present application are introduced and explained:

[0035] (1) Artificial Intelligence (AI): using a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, perceiving the environment, acquiring knowledge and using the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which aims to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that the machine has the functions of perception, reasoning and decision-making.

[0036] (2) Machine Learning (ML): a multi-disciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, etc. It is a subject that studies how computers simulate or implement human learning to acquire new knowledge or skills, and reorganize existing knowledge structure to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent, and its applications are widespread in various fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and example-based learning techniques. For example, the machine learning technology is used to learn the process of resource sending in the embodiments of the present application, so that the computer obtains the skill of resource sending.

[0037] (3) Post-view clickthrough & conversion rate (CTCVR): the product of post-view Click-Through Rate (CTR) and post-click Conversion Rate (CVR). CTR is the ratio of click times to display times. CVR is the ratio of conversion times to click times.

[0038] (4) Multilayer Perceptron (MLP): a kind of artificial neural network (feedforward neural network) with forward structure, which maps a set of input information to a set of output information.

[0039] (5) M-STAR (Multi-gate Star Topology Adaptive Recommender) model: a kind of multi-domain information sending model based on multi-gate star topology structure (i.e., the target prediction model described below) proposed by the present application.

[0040] (6) STAR (Star Topology Adaptive Recommender): a kind of star topology multi-domain information sending model proposed by the present application for comparison with the M-STAR model. The M-STAR model can be regarded as a further improvement of the STAR model. The STAR model is briefly introduced as follows.

[0041] The input of the STAR model includes the description information of the object, which can include the description information related to the object, the description information related to the resource, and the scene identifier (domain id, used to uniquely identify the scene). As shown in FIG. 1, the STAR model generally includes three main modules: Figure 1

[0042] (6.1) Scene-partitioned batch normalization: Partitioned Normalization (PN). The network parameters in the PN are different in different scenes, and its expression is shown in equation (1).

[0043]

[0044] wherein γ, γ p , β, β p are all learnable parameters, γ and β are common parameters of the scene, γ p and β p are unique parameters of the pth scene, and γ and β are obtained by training all samples, γ p ​and β p The learning is obtained using the unique sample of the pth scene, z' is the output of PN, z is the input sample, σ is the mean of all input samples, and ∈ is the variance of all input samples.

[0045] (6.2) Star topology network: star topology fully-connected neural network (StarTopology FCN). The core idea of the star topology network is to divide the neural network parameters into two parts, which are the shared network parameters of all scenes, parameters W and b, and the unique network parameters of each scene, which are W p and b p The parameter definition of StarTopology FCN is shown in formula (2).

[0046]

[0047] wherein, represents element-wise multiplication, i.e. bitwise multiplication, in p represents the current layer input of the neural network of the pth scene, and φ represents the activation function of the neural network, such as the relu function or the sigmoid function.

[0048] (6.3) Auxiliary network: auxiliary network. The auxiliary network is composed of two layers of feedforward neural network, and the input is the relevant information of each scene (such as domain id). As shown in formula (3) below, the output of the auxiliary network is s a , the output of the star topology network is s m , and the final CTR prediction result is obtained by adding them and passing through the sigmoid activation function.

[0049]

[0050] In addition, the shared embedding layer (Share Embedding Layer) in the STAR model is used to convert each data in the description information of the object, and the pooling & concatenation (Pooling & Concatention) is used to pool the converted information and splice the pooled information.

[0051] In the process of STAR model implementing resource transmission, it is found that the STAR model has the following problems:

[0052] Problem 1: The STAR model does not consider the importance of the same information in different scenarios. For example, for an old object (an object that has operated on the resources in the scene), the operation data generated in the scene is more important, but for a new object (an object that has not operated on the resources in the scene), the operation data generated in the scene is less important.

[0053] Problem 2: All neural network parameters of the star topology network are split into shared parameter networks and unique network parameters of each scene, and the final neural network parameters of each scene are obtained in the form of multiplication. Although it can theoretically utilize all scene information and retain the uniqueness of each scene, it objectively increases the difficulty of model learning, reduces the stability and convergence speed of the model (two small weights multiplied become smaller).

[0054] To solve the above problems, an improved information processing scheme (which can be represented as M-STAR model) is proposed in the embodiments of the present application, which includes:

[0055] (1) The input information is divided into common information used by each scene and unique information used by the i-th scene. The common information is information common to all scenes, such as object basic data (used to describe the object, which can be information inherent to the object), resource basic data (used to describe the resource, such as the color, price, and category of the resource), and the current time. The unique information is information unique to the i-th scene, such as the scene identifier of the i-th scene, the operation data of the object in the i-th scene, and the statistical data of the resources in the i-th scene, such as the sales of the resources and the number of times the object accesses the i-th scene.

[0056] (2) Determine the reference importance of each sub-information in the common information when sending a target resource to a target object in the i-th scene, so as to adaptively select important information in the common information suitable for the i-th scene (i.e., reference common information), solving the problem that the common information has different importance in different scenes.

[0057] (3) The reference common information of the i-th scene and the corresponding unique information are spliced to obtain the complete information representation of the i-th scene (i.e., the sending instruction information). The reference common information in the complete information representation can to some extent retain the commonality between scenes and the uniqueness of the i-th scene, while the unique information in the complete information representation can further strengthen the uniqueness of the i-th scene. Finally, based on the complete information representation of the i-th scene, the sending probability of the target resource to the target object in the i-th scene can be accurately determined. The uniqueness can make the resource sending in the i-th scene more targeted, and the commonality, as a property common to all scenes, can to some extent serve as a kind of supplementary information to assist the resource sending in the i-th scene.

[0058] The application can better solve the problems of the STAR model through a carefully designed information processing scheme: by fusing expert experience information (i.e., experience obtained by experts in the information sending field through practice), the input information is split into common information used in each scene and unique information used in the i-th scene, and by the reference importance of the sub-information in the common information in the i-th scene, the reference common information (which can indicate important information in the common information) of the i-th scene is adaptively selected from the common information, thereby solving problem 1 of the STAR model; the complete information representation obtained by splicing the important common information of the i-th scene and the corresponding unique information can retain all scene information and the unique information of the i-th scene, thereby solving problem 2 of the STAR model. Therefore, the application is beneficial to improving the prediction accuracy of the resource sending probability.

[0059] Please refer to Figure 2 , Figure 2 is a structural schematic diagram of an information processing system provided by an embodiment of the application. As Figure 2 indicated, the information processing system includes a computer device 200 and a database 201, and the computer device 200 and the database 201 can be communicatively connected through a wired or wireless manner.

[0060] The database 201 can be a local database of the computer device 200 or a cloud database accessible by the computer device 200, and the database 201 can be used to store the description information of the target object.

[0061] The computer device 200 can be a terminal device or a server. In an embodiment, the terminal device can be a smart phone, a tablet computer, a smart wearable device, a smart voice interaction device, a smart home appliance, a personal computer, a vehicle-mounted terminal, and the like, without limitation. The server can be a standalone physical server, a server cluster or a distributed system composed of multiple physical servers, a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content distribution networks (CDN), and basic cloud computing services such as big data and artificial intelligence platforms, without limitation.

[0062] The interaction process between the computer device 200 and the database 201 is as follows:

[0063] The computer device 200 obtains description information of the target object from the database 201 in response to the trigger operation of sending the target resource to the target object, the description information including one or more object description information of the target object and one or more resource description information of the target resource to be sent to the target object, obtains description information that is required to be used in all of the M scenes when the target resource is sent to the target object, and obtains common information required to be used in each of the M scenes when the target resource is sent to the target object by using the obtained description information.

[0064] The computer device 200 obtains the reference importance of each sub information in the obtained common information in the i th scene when the target resource is sent to the target object, and performs conversion processing on the common information according to the reference importance of each sub information in the i th (i is a positive integer less than or equal to M) scene to obtain reference common information corresponding to the i th scene.

[0065] The computer device 200 obtains object description information related to the i th scene of the M scenes and resource description information related to the i th scene from the description information of the target object, and obtains unique information required to be used in the i th scene when the target resource is sent in the i th scene by using the obtained object description information and resource description information.

[0066] The computer device 200 performs splicing processing on the reference common information corresponding to the i th scene and the corresponding unique information to obtain sending indication information of the i th scene, and predicts a sending probability of sending the target resource to the target object in the i th scene according to the sending indication information.

[0067] It can be seen that the information processing system provided in the embodiments of the present application can obtain common information required to be used in each scene and unique information required to be used in the i th scene by using the description information, can adaptively select reference common information of the i th scene from the common information by using the reference importance of the sub information in the i th scene in the common information, and can solve the problem that the common information has the same semantics but different importance in each scene. Further, the complete information representation (i.e., the sending indication information) of the i th scene can be obtained by splicing the unique information of the i th scene and the corresponding reference common information, so that the sending probability of sending the target resource to the target object in each scene can be accurately determined based on the complete information representation of each scene, the prediction accuracy of the resource sending probability is effectively improved, and the resource processing efficiency is improved in the subsequent resource sending process.

[0068] The specific implementation of the information processing method provided in the embodiments of the present application will be described in detail below with reference to the accompanying drawings. Please refer to Figure 3 , Figure 3 is a flowchart of an information processing method provided in the embodiments of the present application. The method can be performed by a computer device (for example, the computer device 200 shown in FIG. 1)Figure 2 The information processing method is executed by the computer device 200 shown in FIG. 1. The information processing method includes the following steps.

[0069] S301, in response to a trigger operation of sending a target resource to a target object, obtaining common information required by each of M scenes when sending the target resource to the target object; M is an integer greater than 1, and the common information includes multiple sub-information.

[0070] The target resource refers to a resource to be sent to the target object, which can include one or more of video, goods, topics, information, articles, books, movies, and games. The trigger operation of sending the target resource to the target object can be used to instruct the computer device to perform an information (resource) sending task. The trigger operation of sending the target resource to the target object can be initiated by the target object through a client used by the target object. The client can be installed in the computer device (which can be a terminal device) and directly detected by the computer device. The trigger operation of sending the target resource to the target object can also be detected by the client and communicated to the computer device (which can be a server) to perform subsequent prediction processing of sending probability by the computer device. The client can be an e-commerce application, a live broadcast application, a financial application, a music application, a game application, etc. In an embodiment, when the target object opens an information display page of the client or refreshes the information display page in the client, it can be determined that the client initiates the trigger operation of sending the target resource to the target object, and the information display page is used to display resources.

[0071] The M scenes can be various activity scenes (or fields) in the client. For example, when the client is an e-commerce application, the M scenes involved can include a mall activity area, a home page advertisement area, etc. When the client is a financial application, the M scenes involved can include a new object information display home page, an old object information display home page, a fund tab page (fund management page), a stable tab page (stable management page), a holding page, etc. When the client is a music application, the M scenes involved can include a daily sending area and a home page advertisement area. When the client is a news application, the M scenes involved can include a home page information stream and a search box drop-down.

[0072] It should be noted that the M scenes have certain commonalities, but the data scale and business form often have certain differences. For example, the same resource can be sent in the M scenes. In addition, the M scenes can also include activity scenes in different clients.

[0073] When the trigger operation of sending the target resource to the target object occurs, common information used by each scene when sending the target resource to the target object in M scenes can be obtained. The common information is information commonly owned by each scene, which is suitable for any scene and can help the prediction process of the sending probability in each scene. For example, the information obtained by converting the basic data of the target object, the basic data of the target resource, and the current time can be summarized as common information. It can be seen that the common information is universal in different scenes and does not change because of the difference between scenes. The basic data of the target object can be data irrelevant to the operation of the target object in each scene, which is mainly used to describe the target object, such as the inherent information of the height of the target object. The basic data of the target resource can be data irrelevant to the operation of the target resource, which is mainly used to describe the target resource object, such as the inherent information of the color, size, and category of the target resource. The common information includes multiple sub-information, and one sub-information is information converted from data describing one dimension, for example, the color of the target resource can be one dimension, and the color of the target resource being red is data describing one dimension.

[0074] In S302, the reference importance of each sub-information in the i-th scene when sending the target resource to the target object is obtained, and the common information is converted and processed according to the reference importance of each sub-information in the i-th scene to obtain the reference common information corresponding to the i-th scene; i is a positive integer less than or equal to M.

[0075] The same sub-information has different importance in different scenes, and the importance can reflect the influence degree of the sub-information on the final prediction result (for example, the sending probability described below) when sending the target resource to the target object in the corresponding scene. For example, the greater the absolute value of the importance, the greater the influence degree. The reference importance of each sub-information in the i-th scene when sending the target resource to the target object can be obtained, that is, the reference importance is used to reflect the influence degree of the sub-information on the final prediction result (for example, the sending probability described below) in the i-th scene when sending the target resource to the target object. By using the reference importance of each sub-information in the i-th scene to convert and process the common information, the reference common information corresponding to the i-th scene can be obtained. If the absolute value of the reference importance of one sub-information in the i-th scene is greater, the conversion processing will tend to make the absolute value of the converted sub-information greater, and correspondingly, if the absolute value of the reference importance of one sub-information in the i-th scene is smaller (for example, a decimal less than 1), the conversion processing will tend to make the absolute value of the converted sub-information smaller.

[0076] Different sub-information in the common information has different importance in different scenes, and the conversion processing can strengthen the decision-making ability (influence degree) of the sub-information with higher reference importance in the i-th scene, so that when predicting the sending probability in the i-th scene, more attention can be paid to the sub-information with higher reference importance, and therefore the reference common information in the i-th scene has higher applicability to the i-th scene than the common information, which is beneficial to improving the prediction accuracy of the sending probability.

[0077] S303, obtaining unique information used by the i-th scene when sending the target resource in the i-th scene.

[0078] When the target resource is to be sent in the i-th scene, unique information used by the i-th scene needs to be obtained. The unique information refers to information exclusively used in the i-th scene, which can be converted and generated according to operation data of the target object in the i-th scene and statistical data of the target resource in the i-th scene. The operation data of the target object in the i-th scene is mainly obtained according to the operation of the target object in the i-th scene. For example, which resources are purchased by the target object in the i-th scene, which category of resources the target object tends to purchase, and the number of times the i-th scene is used. The statistical data of the target resource in the i-th scene is obtained by counting the operation of the target resource in the i-th scene. For example, the number of clicks, sales ranking, and search times of the target resource in the i-th scene. The unique information can also include converted information of the scene identifier of the i-th scene and converted information of the environment information of the i-th scene. The environment information of the i-th scene can include the size and position of the interface region corresponding to the i-th scene.

[0079] The present application relates to operation data of the target object in each scene, basic data of the target object, and other data related to the target object. When the embodiments of the present application are applied to specific products or technologies, the data has been permitted or agreed by the target object, and the extraction, use and processing of the related data comply with local laws and regulations. Specifically, before obtaining the basic data of the target object, an authorization agreement for obtaining the basic data of the target object can be sent to the terminal device of the target object. When the target object agrees to the authorization agreement, the computer device can obtain the basic data of the target object, otherwise, the computer device cannot obtain the basic data of the target object.

[0080] S304, splicing the reference common information corresponding to the i-th scene and the corresponding unique information to obtain sending indication information of the i-th scene, and predicting a sending probability of sending the target resource to the target object in the i-th scene according to the sending indication information.

[0081] The sending indication information of the i-th scene can be obtained by splicing the reference common information corresponding to the i-th scene and the corresponding unique information. The splicing can be performed by splicing the reference common information corresponding to the i-th scene after the corresponding unique information, or by splicing the unique information corresponding to the i-th scene after the corresponding reference common information. The splicing obtained sending indication information is the complete information representation of the i-th scene. The reference common information in the sending indication information can strengthen the decision-making ability of the important sub-information of the i-th scene, and the unique information can strengthen the uniqueness of the i-th scene, thereby avoiding interference between scenes, so that the shared part between scenes and the unique part of the i-th scene can be used to improve the prediction effect of the i-th scene when predicting the sending probability of the target resource to the target object in the i-th scene. The sending probability is used to represent the probability of sending the target resource to the target object. The greater the sending probability, the greater the possibility of sending the target resource. When determining the sending probability, the sending indication information can be used as the input of a classification network model (for example, the personalized processing module described below), and the output of the classification network model can be used as the sending probability. The model parameters of the classification network model can be generated through a learning process of multi-scene information transmission.

[0082] In an embodiment, if the sending probability of the target resource to the target object in the i-th scene is greater than the probability threshold, the target resource is sent to the target object in the i-th scene, and the resource processing efficiency can be improved by sending the resource in the i-th scene.

[0083] By using the method of the present application, the common information used in each scene and the unique information used in the i-th scene are determined. The reference importance of the sub-information in the common information in the i-th scene can be used to adaptively select the reference common information suitable for the i-th scene from the common information, which can solve the problem that the common information has the same semantics but different importance in each scene. The reference common information of the i-th scene and the corresponding unique information are spliced to obtain the complete information representation (i.e., the sending indication information) of the i-th scene. The complete information representation not only retains the commonality between scenes but also retains the uniqueness of the i-th scene, so that the sending probability of the target resource to the target object in the i-th scene can be accurately determined based on the complete information representation of the i-th scene. It can be seen that the present application can effectively improve the prediction accuracy of the resource sending probability, and thus improve the resource processing efficiency in the subsequent resource sending process.

[0084] Please refer to Figure 4 , Figure 4 is a flowchart of another information processing method provided by an embodiment of the present application. The method can be executed by a computer device (for example, the computer device 200 shown in Figure 2 . The information processing method includes:

[0085] S401, in response to the trigger operation of sending the target resource to the target object, obtaining description information of the target object, the description information comprising one or more object description information of the target object and one or more resource description information of the target resource to be sent to the target object.

[0086] The description information of the target object comprises one or more object description information of the target object, and the object description information can comprise basic data of the target object and operation data of the target object. The basic data of the target object is used to describe the target object, and can comprise inherent data such as height of the target object and statistical data of the target object which does not change with the scene. The operation data of the target object is mainly obtained based on the operation of the target object in each scene, for example, which resources are purchased by the target object in each scene, which category of resources the target object tends to purchase, and the number of times of using a certain scene. One object description information is data for describing one dimension of the target object, for example, the height of the target object is one dimension. It should be noted that the description information of the target object is obtained and used after the permission or consent of the target object is obtained.

[0087] The description information of the target object further comprises one or more resource description information of the target resource to be sent to the target object. The resource description information can comprise basic data of the target resource and statistical data of the target resource. The basic data of the target resource is used to describe inherent information of the target resource, for example, the category, color, size, price and the like of the target resource. The statistical data of the target resource can be obtained according to the operation of the target resource in each scene, for example, the number of clicks, sales ranking, search times of the target resource in each scene and the like. One resource description information is data for describing one dimension of the target resource, for example, the number of clicks of the target resource is one dimension.

[0088] The present application relates to the description information of the target object and the like data, and when the embodiments of the present application are applied to specific products or technologies, the permission or consent of the target object is obtained, and the extraction, use and processing of the related data comply with the local laws and regulations. For example, before obtaining the description information of the target object, an authorization agreement for obtaining the description information of the target object can be sent to the terminal device of the target object. When the target object agrees to the authorization agreement, the computer device can obtain the description information of the target object, otherwise, the computer device cannot obtain the description information of the target object.

[0089] S402, obtain, from the description information, description information that is required to be used when the target resource is sent to the target object in M scenes, and convert the obtained description information into information, which is used as common information; one description information obtained from the description information is converted into one sub-information in the common information, M is an integer greater than 1, and the common information includes a plurality of sub-informations.

[0090] The description information obtained from the description information is required to be used when the target resource is sent to the target object in M scenes, for example, data such as basic data of the target object and the target resource, which does not change with the scene, that is, the description information required to be used in each scene. Further, the information converted from the description information obtained from the description information is used as common information, and one description information obtained from the description information is converted into one sub-information in the common information.

[0091] In an embodiment, an embedding layer can be used to convert one description information obtained from the description information into one d-dimensional fixed-length information, assuming that there are D c sub-informations, the common information EMD c is represented as:

[0092]

[0093] Among them, represents each description information obtained from the description information. E(inp i ) is a simplified representation of the embedding module of the neural network, which is used to map sparse information into fixed-length information with a dimension of d, inp i represents the input information. Assuming that the input information ,

[0094] The embedding module is obtained after the initial target prediction model is trained by the multi-scene information sending learning, and the detailed learning process can be referred to the embodiments Figure 9 involved.

[0095] S403, obtain the reference importance of each sub-information when the target resource is sent to the target object in the i-th scene, and convert the common information according to the reference importance of each sub-information in the i-th scene to obtain the reference common information corresponding to the i-th scene; i is a positive integer less than or equal to M.

[0096] The importance of the same sub-information in shared information varies across different scenarios. Therefore, it is necessary to determine the reference importance of each sub-information in scenario i. In one embodiment, the reference importance of each sub-information in scenario i and the reference shared information in scenario i are both obtained by calling an information gating module. The information gating module has M information selection parameter groups, and each information selection parameter group corresponds to a scenario. Obtaining the reference importance of each sub-information in scenario i when sending target resources to the target object includes: obtaining the information selection parameter group corresponding to scenario i from the M information selection parameter groups in the information gating module. Each information selection parameter group contains multiple information selection parameters, and each information selection parameter corresponds to an information dimension. The information selection parameter in the obtained information selection parameter group that corresponds to the same information dimension as any sub-information is used as the reference importance of any sub-information in scenario i when sending target resources to the target object.

[0097] Furthermore, each sub-information is weighted according to its reference importance to obtain corresponding weighted sub-information. The combined information obtained by combining the weighted sub-information is used as the reference common information corresponding to the i-th scenario.

[0098] In one embodiment, the process of determining the reference common information corresponding to the i-th scene can be achieved through, as follows: Figure 5 The information gating submodule implementation shown can include M information gating submodules (feature gates). Each information gating submodule corresponds to a scene and can be used to determine a set of information selection parameters. Specifically, the information gating submodule corresponding to the i-th scene can be determined, and common information is used as input information inp. i Input a specific information gating submodule. Within this submodule: ① Common information EMD c After reducing the number of parameters using the mean pooling layer, the common information after dimensionality reduction is obtained: pooling(inp i This is a simplified representation of the mean pooling layer in a neural network. If the input information is inp i Includes D c If each piece of information is d-dimensional, then the input information is inp. i Represented as: After mean pooling, we will get The principle behind this is to calculate the mean across all dimensions for each d-dimensional piece of information, thus obtaining a floating-point average result. ② (The last part, "inp," appears to be a typo and can be left as is.) ipool After passing through the first layer of a feedforward neural network with an output dimension of K (using the ReLU function as the activation function), the output is obtained as follows: ③ Pass o1 through the second layer to output dimension D c A 3D feedforward neural network (using the sigmoid function as the activation function) is used to obtain the information selection parameter set corresponding to the i-th scene: ④ Using the obtained information, select the parameters in the parameter group, adjust the corresponding sub-information in the common information, and obtain the reference common information corresponding to the i-th scenario: weight includes D c Select parameters for each information, weight*EMD c This will cause the information selection parameters in weight to be related to EMD. c The sub-information at corresponding positions (i.e., corresponding to the same information dimension) is weighted (i.e., multiplied) to obtain weighted sub-information. It can be seen that the information dimensions of both the reference shared information and the shared information are... The information dimensions have not changed, but each sub-information has been adjusted according to the characteristics of the i-th scenario.

[0099] In one feasible implementation, the information dimension (i.e., the length or data quantity of the unique information after dimensionality reduction) corresponding to the unique information of the i-th scenario can be obtained. Dimensionality reduction is then performed on each weighted sub-information according to its information dimension, ensuring that the information dimension of the weighted sub-information is the same as that of the unique information after dimensionality reduction. The combined information obtained by combining the weighted sub-information after dimensionality reduction is then used as the reference common information corresponding to the i-th scenario. For example, an average pooling layer can be used to process the EMD output by the information gating submodule. ci Dimensionality reduction is performed to obtain common reference information: Dimensionality reduction can reduce the number of parameters and facilitate subsequent splicing with the unique information obtained after dimensionality reduction.

[0100] The information gating module is derived from the initial target prediction model after learning through multiple scenario information transmissions. For a detailed learning process, please refer to [link to documentation]. Figure 9 The embodiments involved.

[0101] In feasible embodiments, information cross-processing can also be performed on the various sub-information in the shared information, transforming the various processing of the shared information into processing of the shared information after information cross-processing.

[0102] S404. From one or more object description information in the description information, obtain the object description information related to the i-th scene of M scenes, and from one or more resource description information in the description information, obtain the resource description information related to the i-th scene. Use the transformed information of the object description information related to the i-th scene and the transformed information of the resource description information related to the i-th scene as the unique information of the i-th scene.

[0103] From the one or more object description information, the object description information related to the i-th scene of the M scenes is obtained, such as the operation data of the target object in the i-th scene. From the one or more resource description information, the resource description information related to the i-th scene is obtained, such as the statistical data of the target resource in the i-th scene. Further, the converted information of the resource description information related to the i-th scene and the converted information of the object description information related to the i-th scene are taken as the unique information of the i-th scene. In addition, the unique information can also include the converted information of the scene identification (domainid, used for uniquely identifying the scene) and the converted information of the environmental information such as the size and position of the interface region where the scene is located.

[0104] In an embodiment, the resource description information related to the i-th scene and the object description information related to the i-th scene obtained from the description information can be converted into a d-dimensional fixed-length information by using an embedding layer, assuming that there are D p total data, and the unique information EMD pi of the i-th scene is represented as:

[0105]

[0106] wherein, represents the resource description information related to the i-th scene and the object description information related to the i-th scene obtained from the description information.

[0107] In addition, the converted information of the scene identification and the converted information of the environmental information such as the size and position of the interface region where the scene is located can also be obtained by using the embedding layer.

[0108] S405, the reference common information corresponding to the i-th scene is spliced with the corresponding unique information to obtain the sending indication information of the i-th scene, and the sending probability of the target resource to the target object in the i-th scene is predicted according to the sending indication information.

[0109] In a feasible implementation, the reference common information corresponding to the i-th scene is spliced with the corresponding unique information to obtain the transmission indication information of the i-th scene, including: obtaining the reference common information corresponding to each scene in the M scenes respectively, and determining the information commonality between the reference common information corresponding to different scenes, and the information difference between the reference common information corresponding to the i-th scene and other scenes. The information commonality refers to the characteristics or rules possessed by the reference common information of different scenes, and the information difference refers to the unique characteristics or rules of the reference common information of the i-th scene compared with the reference common information of other scenes. The reference common information corresponding to the i-th scene can be converted according to the information commonality and the information difference to obtain the conversion information STAR ci of the i-th scene. The conversion processing can utilize the information commonality and the information difference to make the obtained conversion information retain the information common to all scenes and the unique information of the i-th scene.

[0110] Further, the unique information of the i-th scene can be processed by dimension reduction to obtain the unique information of the i-th scene after dimension reduction, for example, using a mean pooling layer to obtain the unique information after dimension reduction: PEMD pi = pooling(EMD pi ). Further, the conversion information STAR ci of the i-th scene is spliced with the corresponding unique information after dimension reduction PEMD pi to obtain the transmission indication information input i of the i-th scene. By splicing the unique information and the conversion information, the uniqueness of the i-th scene can be further strengthened.

[0111]

[0112] In an embodiment, the conversion processing of the reference common information corresponding to the i-th scene is performed by calling a star topology module, as shown in Figure 6 . The star topology module includes a shared network and M scene networks (three scene networks are taken as an example), and one scene network is associated with one scene; the conversion processing of the reference common information corresponding to the i-th scene is performed according to the information commonality and the information difference to obtain the conversion information of the i-th scene, including: obtaining the shared network parameters W and b of the shared network corresponding to the i-th scene, and the scene network parameters W i and b i of the scene network corresponding to the i-th scene; wherein the shared network parameters are used to represent the information commonality between the reference common information corresponding to different scenes, and the scene network parameters are used to represent the information difference between the reference common information corresponding to the i-th scene and other scenes. The conversion parameters W and b are generated according to the shared network parameters and the scene network parameters. Specifically, the shared network parameters ω and b and the scene network parameters W of the i-th scene are determined i and b i The conversion parameters are determined by substituting the following formula (7) and

[0113]

[0114] Further, the reference common information corresponding to the i-th scene is converted by using the conversion parameters to obtain the conversion information STAR of the i-th scene ci Specifically, the reference common information PEMD corresponding to the i-th scene is substituted into the following formula (8) to obtain the conversion information STAR of the i-th scene ci ci。

[0115]

[0116] wherein, D in the formula (8) is the input dimension of the star topology module, i.e. c is determined according to the formula (7), and d is the output dimension of the star topology module.

[0117] The formula (8) is actually a calculation expression of a single-layer feedforward neural network. The star topology module can also include a multi-layer feedforward neural network. Each layer of the feedforward neural network is also divided into shared network parameters and scene network parameters. The input of the first layer of the feedforward neural network is the reference common information. The input of each subsequent layer of the feedforward neural network can be the output of the previous layer of the feedforward neural network. The output of the last layer of the feedforward neural network is the conversion information.

[0118] The STAR model does not distinguish between common information and unique information. Therefore, the star topology network in the STAR model uses the same information representation in all scenes and only learns the differences between scenes through the unique network parameters in the star topology network. Although the auxiliary network in the STAR model introduces domain id as an input to enable the network to perceive the data distribution in different scenes, the shallow number of layers and weak learning ability make the STAR model unable to well learn the differences between different scenes. In addition, the star topology network in the STAR model involves all the description information of the object, so the unique information of each scene inevitably interferes with the learning of the scene differences of the star topology network.

[0119] ​​Different from the star topology network in the STAR model, the same information representation is no longer used as the input of the star topology module, but each scene has corresponding reference common information, which is used as the input of the star topology module. On the one hand, the star topology module can use the important sub-information of the corresponding scene to better learn the commonality and specificity between scenes. On the other hand, since the star topology module no longer processes unique information, it can also prevent unique information from interfering with the star topology module learning the differences between scenes.

[0120] The star topology module is an initial target prediction model obtained after multi-scene information transmission learning, and the detailed learning process can be referred to Figure 9 Embodiments.

[0121] In an embodiment, the sending probability of the target resource to the target object in the i-th scene is predicted according to the sending indication information, including: obtaining the object identifier (used to uniquely identify the target object) of the target object, the resource identifier (used to uniquely identify the target resource) of the target resource and the scene identifier (used to uniquely identify the i-th scene) of the i-th scene. Wherein, the object identifier, the resource identifier and the scene identifier can be represented in the form of one-hot. For example, including three scenes, the scene identifiers of the three scenes are represented by one-hot as: [1, 0, 0], [0, 1, 0], [0, 0, 1].

[0122] Further generate the sending weight of the target resource to the target object in the i-th scene according to the object identifier, the resource identifier and the scene identifier of the i-th scene, including: respectively converting the object identifier, the resource identifier and the scene identifier of the i-th scene to obtain the identifier information u emb of the object identifier, the identifier information i emb of the resource identifier and the identifier information p emb of the scene identifier of the i-th scene. Specifically, the object identifier, the object identifier and the scene identifier of the i-th scene can be respectively converted into d-dimensional fixed-length information by an embedding module to obtain the respective identifier information. The identifier information of the object identifier, the identifier information of the resource identifier and the identifier information of the scene identifier of the i-th scene are spliced to obtain the spliced information. The prediction module is called to predict the spliced information to obtain the sending weight of the target resource to the target object in the i-th scene. As Figure 7 shown, the prediction module includes a personalized weight gate module (domain gate). The gate input of the personalized weight gate module is the spliced information obtained by using the object identifier, the resource identifier and the scene identifier: The spliced information is input into a first layer of feedforward neural network with K-dimensional output (with relu function as the activation function) to obtain the output: out1 gate After passing through a second-layer feedforward neural network with an output dimension of D (using the sigmoid function as the activation function), the transmission weights are obtained: The sending weight is D. p +D c The dimension is the same as the information dimension of the sending instruction information, because the subsequent sending weight needs to be matrix-multiplied with the sending instruction information.

[0123] Further, a transmission prediction process is performed using transmission weights combined with transmission indication information to obtain the transmission probability of sending the target resource to the target object in scenario i. For example, the transmission weights and transmission indication information are multiplied, and the result is used as the input to a classification neural network (e.g., a feedforward neural network with the sigmoid function as the activation function). The output of the classification neural network is the transmission probability of sending the target resource to the target object in scenario i.

[0124] In one specific implementation method, the following can be utilized: Figure 8 The personalized processing module shown performs transmission prediction processing. The personalized processing module includes a personalized parameter module (domain tower) and a prediction module. The prediction module includes personalized weight gating module 1 (domain gate1) and personalized weight gating module 2 (domain gate2). Domain gate1 and domain gate2 are connected to... Figure 7 The prediction modules shown have the same model structure. The concatenated information can be used as input to domain gate1 to obtain the first transmission weights: And by using the concatenated information as input to domain gate2, the second transmission weight is obtained: Will send instruction information input i Multiply by the first sending weight out1 gate1 The output is: The output of o1 is obtained by passing it through the first layer of the personalized parameter module, which is a feedforward neural network with an output dimension of D (using the ReLU function as the activation function): Multiply o2 by the second sending weight out2 gate2 Output: By passing o3 through the second layer of the personalized parameter module, which has a 1-dimensional output feedforward neural network (using the sigmoid function as the activation function), the probability of sending the target resource to the target object in the i-th scenario is obtained:

[0125] Considering that the same object and resource often have great differences in different scenarios, the application introduces a sending weight near the model output (i.e., the individual processing module). By processing the complete information representation with the sending weight of the i-th scenario, personalized scenario information (i.e., which scenario, which object, and which resource to send) can be introduced. That is, the sending weight is equivalent to an individual weight that increases personalized scenario information. The individual weight can improve the prediction accuracy of the sending probability by strengthening the uniqueness of the scenario (i.e., the uniqueness of the target object and the target resource in the i-th scenario), and can reduce the adverse effects of shared network parameters in the star-shaped topology module on the prediction accuracy of each scenario.

[0126] The individual processing module is obtained after the initial target prediction model is learned by multiple scenario information sending. The detailed learning process can be found in Figure 9 the embodiments.

[0127] In an embodiment, the above method can be used to determine the sending probability of the target resource to the target object in each of the M scenarios. If there is a sending probability greater than the probability threshold, the target resource is sent to the target object in the scenario corresponding to the sending probability, which helps to improve the prediction accuracy of multi-scenario resource sending and thus improves the multi-scenario resource processing efficiency.

[0128] In addition, other structures of information selection networks can be used for the information gating module to achieve similar effects. In addition, the unique information of the scenario can also be selected by the information gating module. The individual parameter module can also directly use a general MLP or other structures of deep neural networks, but the model parameters of each scenario are not shared.

[0129] To protect the object privacy of the corresponding object, the computer device obtains the object description information of the target object after obtaining the information authorization of the target object, and uses the object description information of the target object only for sending the target resource. In addition, the computer device sends the target resource to the target object after obtaining the corresponding sending authorization of the target object.

[0130] By adopting the method, the input information can be split into common information used in each scene and unique information used in the i-th scene by fusing expert experience information, each sub-information in the common information can be obtained in the reference importance of the i-th scene through the information gating module, the reference importance can be used to adaptively select the reference common information of the i-th scene from the common information, the problem that the common information has the same semantics but different importance in each scene is solved, the reference common information of each scene is calculated through the star topology module, each output obtained can retain the commonality between each scene and the uniqueness of the i-th scene, the unique information of each scene and the output of the star topology module are spliced to obtain complete information representation of the i-th scene, the unique information in the complete information representation can strengthen the uniqueness of the i-th scene, and the personalized parameter module (domain tower) unique to the i-th scene and the personalized weight gating module (domain gate) based on the object identifier, the resource identifier and the scene identifier are introduced to strengthen the scene uniqueness, improve the prediction accuracy of the resource sending probability, improve the resource processing efficiency, and reduce the adverse effect of the shared network parameters in the star topology module on the prediction accuracy of each scene.

[0131] Please refer to Figure 9 , Figure 9 is a flowchart of a target prediction model acquisition method provided by an embodiment of the present application. The method can be executed by a computer device (for example, the computer device 200 shown in FIG. 1). Figure 2 The target prediction model acquisition method includes the following steps.

[0132] S901, an initial target prediction model is acquired, and the initial target prediction model includes a personalized processing module, a star topology module, an information gating module and an embedding module.

[0133] If the initial target prediction model involves performing the sending task in M scenes, the initial target prediction model includes M personalized processing modules, one personalized processing module corresponding to one scene, one personalized processing module including a domain tower and a prediction module, the prediction module can include one or more domain gates or prediction modules for determining a sending weight. The information gate module in the initial target prediction model includes M information gate sub-modules, one information gate sub-module corresponding to one scene, and the information gate sub-module can be used to determine an information selection parameter group. The embedding module in the initial target prediction model is used to convert the input description information into fixed-length information of d (a positive integer) dimensions, and the input description information can be object information (such as basic data of an object), resource information (such as basic data of a resource), statistical information (such as operation data of an object in each scene, statistical data of a resource in each scene), scene identification, etc. The star topology module in the initial target prediction model includes a shared network and a scene network corresponding to each scene, which can take into account the commonality and difference between scenes.

[0134] Referring to Figure 10 , Figure 10 A model structure diagram of an initial target prediction model provided by an embodiment of the present application is shown, which exemplarily involves performing a sending task in three scenes, and it can be seen that each of the three scenes corresponds to an information gate sub-module, a scene network and a personalized processing module (including a domain tower and a domain gate).

[0135] S902, obtain a training sample set, the training sample set including one or more sample information, one sample information including description information of one object and a sending label corresponding to the description information of the object.

[0136] The training sample set can include description information of a target object. The sending label corresponding to the description information of the target object can be used to indicate whether the target resource is successfully sent to the target object in the ith scene. For example, if the target object performs a target operation (such as viewing or purchasing) on the target resource sent in the ith scene, it can be considered that the target resource is successfully sent to the target object. The sending label can be specifically represented as: y i ∈{0,1}, which indicates whether the target object performs a target operation on the target resource in the ith scene, if y i =1, it indicates that the target object performs a target operation on the target resource in the ith scene, if y i =0, it indicates that the target object does not perform a target operation on the target resource in the ith scene.

[0137] S903, utilize the initial target prediction model to process the description information included in any sample information in the training sample set, obtain the sending probability corresponding to any sample information, utilize the sending probability corresponding to any sample information and the sending label, adjust the model parameter of the initial target prediction model, and obtain the adjusted target prediction model.

[0138] If the any sample information includes the description information of the target object, the description information included in the any sample information in the training sample set is processed by utilizing the initial target prediction model, including: obtaining the description information required to be used in each scene from the description information of the target object, utilizing the embedding module in the initial target prediction model to perform conversion processing on the obtained description information, and obtaining common information EMD c . The common information EMD c is input into the information gate sub-module corresponding to the i-th scene, the information selection parameter group corresponding to the i-th scene is obtained, the weighting processing is performed on each sub-information in the common information EMD c by utilizing the information selection parameter group corresponding to the i-th scene, the reference common information corresponding to the i-th scene is obtained. The shared network parameter W and b corresponding to the shared network in the star-shaped topology module in the initial target prediction model, and the scene network parameter W i and b i of the scene network corresponding to the i-th scene are utilized to generate conversion parameters and The conversion processing is performed on the reference common information corresponding to the i-th scene by utilizing the conversion parameters and , and the conversion information STAR ci of the i-th scene is obtained.

[0139] Further, the object description information related to the i-th scene of the M scenes is obtained from one or more object description information of the description information of the target object, and the resource description information related to the i-th scene is obtained from one or more resource description information of the description information, the conversion processing is performed on the object description information related to the i-th scene and the resource description information related to the i-th scene by utilizing the embedding module in the initial target prediction model, and the unique information EMD pi of the i-th scene is obtained.

[0140] The conversion information STAR ci and the reduced-dimension unique information PEMD pi are spliced to obtain the sending indication information input i of the i-th scene, and the sending indication information input iIn the personalized processing module corresponding to the i-th scene, the sending probability corresponding to the description information of the target object is obtained, i.e., the probability of sending the target resource to the target object in the i-th scene. Further, a loss parameter can be determined based on the difference between the sending label corresponding to the description information of the target object and the sending probability, and the model parameters of the embedding module in the initial target prediction model, the personalized processing module corresponding to the i-th scene, the information gating submodule corresponding to the i-th scene, and the model parameters of the shared network in the star topology module and the scene network corresponding to the i-th scene are adjusted along the direction in which the loss parameter decreases (which can be achieved by stochastic gradient descent), and an adjusted target prediction model is obtained. When the number of times of adjusting the target prediction model is greater than a preset number of times or the loss parameter is less than a preset loss value, the embedding module, the personalized processing module, the information gating module, and the star topology module in the adjusted target prediction model can be used to predict the sending probability of sending the target resource to the target object in the i-th scene. It can be seen that the modules common to all scenes in the target prediction model can be trained using samples of all scenes to learn more complete information representation, and the modules unique to each scene should be trained using samples of the scene to avoid interference between scenes.

[0141] In addition, Figure 5 The "X" in the above formula indicates that the information gating submodule uses the truncated gradient method to update the model parameters. The truncated gradient method is mainly used to prevent the reverse propagation adjustment process of the model parameters before "X" from affecting the adjustment process of the model parameters after "X", which can prevent the learning of the information gating submodule from affecting the multi-scene sending learning of the target prediction model.

[0142] The adjustment process of the model parameters of the target prediction model is the learning process of the multi-scene information sending of the target prediction model. In the multi-scene information sending learning process, the personalized scene information can be learned by introducing scene identifiers, resource identifiers, and object identifiers in the personalized processing module. Subsequently, when predicting the sending probability using the target prediction model, the target prediction model can introduce personalized weights (i.e., sending weights) using the personalized scene information learned in the training phase. To avoid the problem that the target prediction model does not learn the personalized information about the target object and the target resource, resulting in inaccurate personalized weights, some unrepresentative resources and objects can be indicated using general identifiers in the training phase of the target prediction model. For example, the training sample set involves 1000 items, of which 800 items appear less frequently in the training sample set. These 800 items can share a resource identifier, thereby improving the generalization ability of the target prediction model. Even if the target prediction model does not learn the personalized information about the target resource and the target object in the i-th scene, it can still accurately predict the sending probability.

[0143] In an implementable embodiment, the training sample set can include a plurality of training batches, each training batch including N sample information, each sample information including description information for generating common information and description information for generating unique characteristics of each of the M scenes. The common information corresponding to each training batch can be represented as: Since there are M scenes, there are M unique information. Therefore, the unique information of each training batch in the i-th scene can be represented as:

[0144] The common information of each training batch is input into each information gate sub-module to obtain the output of each information gate sub-module: {EMD c1 , EMD c2 , …, EMD cM}. Further dimension reduction processing of {EMD c1 , EMD c2 , …, EMD cM} can obtain M reference common information: {PEMD c1 , PEMD c2 , …, PEMD cM}, PEMD ci is the reference common information corresponding to the i-th scene. Further inputting the M reference common information into the star topology module can obtain M conversion information: {STAR c1 , STAR c2 , …, STAR cM}, STAR ci is the conversion information corresponding to the i-th scene. By splicing the conversion information corresponding to the i-th scene and the unique information PEMD pi of the i-th scene after dimension reduction, the sending indication information of the i-th scene can be obtained: Inputting the sending indication information of each scene into the individualization parameter module in the corresponding individualization processing module and introducing individualization scene information by using the individualization weight gate included in the prediction module can finally output the sending probability corresponding to all sample information in the training batch: wherein, represents the prediction probability corresponding to the i-th scene of the N sample information.

[0145] wherein, the sending indication information of the i-th scene is input into the individualization parameter module in the individualization processing module corresponding to the i-th scene, and the individualization scene information is introduced by using the individualization weight gate included in the prediction module therein. Finally, the sending probability corresponding to all sample information in the i-th scene in the training batch is output, including: obtaining the splicing information corresponding to each training batch: The first layer output dimension of the feature gate 1 corresponding to the i-th scene is D p +D c The output of the feedforward neural network with a dimension of D is obtained The out1 gate The first sending weight of the feedforward neural network with a dimension of D obtained by the second layer of the feature gate 1 corresponding to the i-th scene is obtained Similarly, the splicing information corresponding to each training batch is outputted by the feature gate 2 corresponding to the i-th scene to obtain the second sending weight By Multiplying the out gate1 , the output is obtained At this time, the personalized scene information is introduced for the first time, and the o1 is further inputted into the first layer of the feedforward neural network in the personalized parameter module corresponding to the i-th scene to obtain the output The output o2 is multiplied by the out2 gate2 , the output is obtained At this time, the personalized scene information can be further strengthened, and the o3 is inputted into the second layer of the feedforward neural network in the personalized parameter module corresponding to the i-th scene to obtain the final output That is, the above

[0146] Since the training sample set involves M scenes, it is necessary to extend the calculation formula of the loss parameter loss to M scenes, as shown in the following formula (9).

[0147]

[0148] Wherein, represents the sending label of the i-th sample information in the p-th scene, represents the sending probability of the i-th sample information in the p-th scene, and N p The number of sample information in the training batch.

[0149] It should be noted that the loss parameter calculation formula shown in formula (8) is actually calculated for the loss parameter corresponding to each scene, that is, The loss parameter corresponding to each scene is used to adjust the modules shared by each scene and the modules exclusively owned by the corresponding scene in the target prediction model.

[0150] The loss parameter corresponding to each training batch can be obtained through formula (8), and the initial target prediction model is adjusted along the direction of reducing (which can be realized by stochastic gradient descent) the loss parameter corresponding to each training batch. When the adjustment times are greater than a preset number of times, or the loss parameter is less than a preset loss value, the embedding module, the individual processing module, the information gating module and the star topology module in the adjusted target prediction model can be used to predict the sending probability of the target resource to the target object in each scene. The batch processing method can accelerate the training speed of the target prediction model.

[0151] The target prediction model (i.e., M-STAR model) provided by the embodiments of the present application can be applied to a financial platform, such as a fund individualization sending scene. The financial platform has thousands of high-quality funds for investment parties to choose from, but the investment parties cannot efficiently select the funds of their interest from the thousands of funds, and therefore need to be individually and quickly matched with the funds of their interest. Since the financial product form is relatively complex, there are a new object information display homepage (the new object is an object that has not performed any operation on the funds of the financial platform, and the information display homepage is used to display the funds), an old object information display homepage (the old object is an object that has performed an operation on the funds of the financial platform), a fund tab page (a fund financial page), a stable tab page (a stable financial page), a holding page, and the like, which are multiple regions and positions. Although these positions are all fund display scenes, the product forms, candidate item pools (funds), and investment party groups are different, and are a typical multi-scene resource sending. The present application is specifically applied to the fund sending in the three scenes of the new object information display homepage (referred to as the new object homepage), the old object information display homepage (referred to as the old object homepage), and the fund tab page of the financial platform, and has achieved significant offline effects and online benefits. The offline evaluation index is AUC (area under the curve, a model evaluation index), and the online index is exposure conversion rate (i.e., CTCVR). The test results are shown in Table 1:

[0152] Table 1

[0153]

[0154] In addition, ablation experiments are also performed on the effectiveness of each part of the target prediction model structure, and the results are shown in Table 2:

[0155] Table 2

[0156]

[0157]

[0158] From the above experiments, it can be seen that the target prediction model is better than the original STAR model whether offline or online, and has achieved benefits in multiple scenarios. The ablation experiment also verifies the necessity and effectiveness of each module of the target prediction model.

[0159] By adjusting the model parameters of the target prediction model, the information gating module can automatically select the required reference common information for each scene. Meanwhile, the star topology module splits the network parameters into shared network parameters and scene-specific network parameters for each scene, so that the star topology module can learn both the commonality between the reference common information of each scene and the uniqueness of each scene. The personalized parameter module for each scene concatenates the output of the star topology module for each scene and the unique information representation of the corresponding scene as input, and then uses the personalized weight gating module to strengthen the uniqueness of each scene and improve the personalized ability, which is conducive to improving the prediction accuracy of the target prediction model for the resource sending probability and thus improving the resource processing efficiency.

[0160] It can be understood that in the specific embodiments of the present application, the description information of the target object and other related data are involved. When the above embodiments of the present application are applied to specific products or technologies, the permission or consent of the target object needs to be obtained, and the collection, use and processing of the related data need to comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0161] The above describes the method of the embodiments of the present application in detail. In order to better implement the above method of the embodiments of the present application, correspondingly, the device of the embodiments of the present application is provided below. Please refer to Figure 11 , Figure 11 is a structural schematic diagram of an information processing device provided by the embodiments of the present application. The information processing device 110 can include:

[0162] The acquisition unit 1101 is configured to, in response to a trigger operation of sending a target resource to a target object, acquire common information required by each scene when sending a target resource to the target object in M scenes; M is an integer greater than 1, and the common information includes a plurality of sub-information.

[0163] The acquisition unit 1101 is further configured to acquire a reference importance of each sub-information corresponding to the target object when sending a target resource in the i-th scene.

[0164] The processing unit 1102 is configured to convert the common information according to the reference importance of each sub-information in the i-th scene to obtain reference common information corresponding to the i-th scene; i is a positive integer less than or equal to M.

[0165] The acquisition unit 1101 is further configured to acquire unique information that needs to be used by the i-th scene when the target resource is transmitted in the i-th scene.

[0166] The processing unit 1102 is further configured to splice the reference common information corresponding to the i-th scene and the corresponding unique information to obtain transmission indication information of the i-th scene, and predict a transmission probability of the target resource to the target object in the i-th scene according to the transmission indication information.

[0167] In an embodiment, the acquisition unit 1101 is specifically configured to: acquire description information of the target object, the description information including one or more object description information of the target object and one or more resource description information of the target resource to be transmitted to the target object; and acquire, from the description information, description information that needs to be used when the target resource is transmitted to the target object in the M scenes.

[0168] The processing unit 1102 is specifically configured to: convert the description information acquired from the description information into common information; and convert one description information acquired from the description information into one sub-information in the common information.

[0169] In an embodiment, the acquisition unit 1101 is specifically configured to: acquire description information of the target object, the description information including one or more object description information of the target object and one or more resource description information of the target resource to be transmitted to the target object; acquire, from the one or more object description information of the description information, object description information related to the i-th scene of the M scenes; and acquire, from the one or more resource description information of the description information, resource description information related to the i-th scene.

[0170] The processing unit 1102 is specifically configured to: convert the object description information related to the i-th scene into information, and convert the resource description information related to the i-th scene into information, as unique information of the i-th scene.

[0171] In an embodiment, each sub-information is obtained by calling an information gating module, the information gating module is provided with M information selection parameter groups, one information selection parameter group corresponds to one scene, the acquisition unit 1101 is specifically configured to: acquire, from the M information selection parameter groups in the information gating module, an information selection parameter group corresponding to the i-th scene; the information selection parameter group includes a plurality of information selection parameters, and one information selection parameter corresponds to one information dimension.

[0172] The processing unit 1102 is specifically configured to: select, as reference importance of any sub-information, information selection parameters corresponding to the same information dimension of the any sub-information in the acquired information parameter group.

[0173] In an embodiment, the processing unit 1102 is specifically configured to: perform weighted processing on the any sub-information by using the reference importance of the any sub-information to obtain a corresponding weighted sub-information; and obtain, as reference common information corresponding to the i-th scene, combined information obtained by combining the weighted sub-information.

[0174] In an embodiment, the acquisition unit 1101 is specifically configured to: acquire an information dimension corresponding to the unique information of the i-th scene after dimension reduction processing.

[0175] The processing unit 1102 is specifically configured to: perform dimension reduction processing on each weighted sub-information according to the information dimension, so that the information dimension of the weighted sub-information after dimension reduction processing is the same as the information dimension of the unique information after dimension reduction processing; and obtain, as reference common information corresponding to the i-th scene, combined information obtained by combining the weighted sub-information after dimension reduction processing.

[0176] In an embodiment, the acquisition unit 1101 is specifically configured to: acquire reference common information corresponding to each of the M scenes respectively, and determine information commonality between the reference common information corresponding to different scenes, and information difference between the reference common information corresponding to the i-th scene and other scenes.

[0177] The processing unit 1102 is specifically configured to: perform conversion processing on the reference common information corresponding to the i-th scene according to the information commonality and the information difference, to obtain conversion information of the i-th scene; and perform splicing processing on the conversion information of the i-th scene and the unique information after dimension reduction processing corresponding thereto, to obtain sending indication information of the i-th scene.

[0178] In an embodiment, the conversion processing on the reference common information corresponding to the i-th scene is performed by calling a star topology module, the star topology module includes a shared network and M scene networks, one scene network is associated with one scene, and the acquisition unit 1101 is specifically configured to: acquire shared network parameters corresponding to the shared network, and scene network parameters of the scene network corresponding to the i-th scene; wherein the shared network parameters are used to represent the information commonality between the reference common information corresponding to different scenes, and the scene network parameters are used to represent the information difference between the reference common information corresponding to the i-th scene and other scenes.

[0179] The processing unit 1102 is specifically configured to generate a conversion parameter according to the shared network parameter and the scene network parameter, and perform conversion processing on reference common information corresponding to the i th scene by using the conversion parameter, to obtain conversion information of the i th scene.

[0180] In an embodiment, the acquisition unit 1101 is specifically configured to acquire an object identifier of the target object, a resource identifier of the target resource, and a scene identifier of the i th scene.

[0181] The processing unit 1102 is specifically configured to generate a sending weight of sending the target resource to the target object in the i th scene according to the object identifier, the resource identifier, and the scene identifier of the i th scene, and perform sending prediction processing by using the sending weight in combination with the sending indication information, to obtain a sending probability of sending the target resource to the target object in the i th scene.

[0182] In an embodiment, the processing unit 1102 is specifically configured to perform conversion processing on the object identifier, the resource identifier, and the scene identifier of the i th scene respectively, to obtain identifier information of the object identifier, identifier information of the resource identifier, and identifier information of the scene identifier of the i th scene, perform splicing processing on the identifier information of the object identifier, the identifier information of the resource identifier, and the identifier information of the scene identifier of the i th scene, to obtain splicing information, and call a prediction module to predict the splicing information, to obtain the sending weight of sending the target resource to the target object in the i th scene.

[0183] In an embodiment, the sending probability of sending the target resource to the target object is obtained by calling a target prediction model, and the common information is obtained from description information of the target object. When the description information of the target object is sample information, the acquisition unit 1101 is specifically configured to acquire a sending label corresponding to the description information of the target object. The sending label is used to indicate whether the target resource is successfully sent to the target object in the i th scene.

[0184] The processing unit 1102 is specifically configured to adjust model parameters of an initial target prediction model based on the sending label and the sending probability, to obtain an adjusted target prediction model.

[0185] It can be understood that the functions of each functional unit of the information processing apparatus described in the embodiments of the present application can be specifically implemented according to the methods in the above method embodiments, and the specific implementation process can refer to the related description of the above method embodiments, which will not be described here.

[0186] In the embodiments of the present application, the common information used in each scene and the unique information used in the i-th scene are determined, and the reference common information suitable for the i-th scene can be adaptively selected from the common information according to the reference importance of the sub-information in the common information in the i-th scene, which can solve the problem that the common information has the same semantics but different importance in each scene; then the reference common information of the i-th scene and the corresponding unique information are spliced to obtain the complete information representation (i.e., the sending indication information) of the i-th scene, and the sending probability of the target resource to the target object in the i-th scene can be determined based on the complete information representation of the i-th scene. Since the complete information representation retains the commonality between scenes and the uniqueness of the i-th scene, the prediction accuracy of the resource sending probability can be effectively improved, and thus the resource processing efficiency is improved in the subsequent resource sending process.

[0187] As shown in Figure 12 , Figure 12 is a structural schematic diagram of a computer device provided by the embodiments of the present application, and the internal structure of the computer device 120 is shown in Figure 12 , which includes one or more processors 1201, a memory 1202, and a communication interface 1203. The above-mentioned processor 1201, memory 1202, and communication interface 1203 can be connected through a bus 1204 or other means, and the embodiments of the present application take the connection through the bus 1204 as an example.

[0188] The processor 1201 (or CPU (Central Processing Unit)) is the computing core and control core of the computer device 120, which can parse various instructions in the computer device 120 and process various data of the computer device 120. For example, the CPU can be used to parse the power-on / off instruction sent to the computer device 120 and control the computer device 120 to perform power-on / off operation; for another example, the CPU can transmit various interactive data between the internal structures of the computer device 120, and the like. The communication interface 1203 can optionally include a standard wired interface, a wireless interface (such as Wi-Fi, a mobile communication interface, etc.), and is controlled by the processor 1201 to receive and transmit data. The memory 1202 is a memory device in the computer device 120, used to store computer programs and data. It can be understood that the memory 1202 herein can include the built-in memory of the computer device 120, and of course can also include the expansion memory supported by the computer device 120. The memory 1202 provides a storage space that stores the operating system of the computer device 120, which can include but is not limited to: Windows system, Linux system, Android system, iOS system, etc., and the present application does not make any limitation thereto. The processor 1201 performs the following operations by running the computer program stored in the memory 1202:

[0189] In response to the trigger operation of sending the target resource to the target object, obtain common information required by each scene when sending the target resource to the target object, M is an integer greater than 1, and the common information includes a plurality of sub-information;

[0190] Obtain the reference importance of each sub-information in the ith scene when sending the target resource to the target object, and perform conversion processing on the common information according to the reference importance of each sub-information in the ith scene to obtain the reference common information corresponding to the ith scene; i is a positive integer less than or equal to M;

[0191] Obtain the unique information required by the ith scene when sending the target resource;

[0192] Splice the reference common information corresponding to the ith scene and the corresponding unique information to obtain the sending instruction information of the ith scene, and predict the sending probability of the target resource to the target object in the ith scene according to the sending instruction information.

[0193] In an embodiment, the processor 1201 is specifically configured to: obtain description information of the target object, the description information comprising one or more object description information of the target object and one or more resource description information of a target resource to be sent for the target object; obtain, from the description information, description information that is required to be used when the target resource is sent for the target object in M scenes; and convert the obtained description information into converted information as common information; wherein one description information obtained from the description information is converted into one sub-information in the common information.

[0194] In an embodiment, the processor 1201 is specifically configured to: obtain description information of the target object, the description information comprising one or more object description information of the target object and one or more resource description information of a target resource to be sent for the target object; obtain, from the one or more object description information of the description information, object description information related to an i-th scene of the M scenes, and obtain, from the one or more resource description information of the description information, resource description information related to the i-th scene; and convert the object description information related to the i-th scene and the resource description information related to the i-th scene into converted information as unique information of the i-th scene.

[0195] In an embodiment, the reference importance degree of the i-th scene and the reference common information of the i-th scene are obtained by calling an information gating module, the information gating module is provided with M information selection parameter groups, one information selection parameter group corresponds to one scene, the processor 1201 is specifically configured to: obtain, from the M information selection parameter groups in the information gating module, an information selection parameter group corresponding to the i-th scene; the information selection parameter group comprises a plurality of information selection parameters, and one information selection parameter corresponds to one information dimension; and obtain, as a reference importance degree of any sub-information, an information selection parameter corresponding to the same information dimension of the any sub-information in the obtained information selection parameter group.

[0196] In an embodiment, the processor 1201 is specifically configured to: perform weighted processing on any sub-information by using the reference importance degree of the any sub-information to obtain a corresponding weighted sub-information; and obtain, as the reference common information corresponding to the i-th scene, combined information obtained by combining the weighted sub-information.

[0197] In an embodiment, the processor 1201 is specifically configured to: obtain an information dimension corresponding to the unique information of the i-th scene after dimension reduction processing; perform dimension reduction processing on each weighted sub-information according to the information dimension, so that the information dimension of the weighted sub-information after dimension reduction processing is the same as the information dimension of the unique information after dimension reduction processing; and combine the combined information obtained by combining the weighted sub-information after dimension reduction processing as the reference common information corresponding to the i-th scene.

[0198] In an embodiment, the processor 1201 is specifically configured to: obtain the reference common information corresponding to each of the M scenes respectively, and determine the information commonality between the reference common information corresponding to different scenes and the information difference between the reference common information corresponding to the i-th scene and other scenes; perform conversion processing on the reference common information corresponding to the i-th scene according to the information commonality and the information difference, to obtain conversion information of the i-th scene; and perform splicing processing on the conversion information of the i-th scene and the unique information after dimension reduction processing corresponding thereto, to obtain sending indication information of the i-th scene.

[0199] In an embodiment, the conversion processing on the reference common information corresponding to the i-th scene is performed by calling a star topology module, and the star topology module includes a shared network and M scene networks, one scene network being associated with one scene; the processor 1201 is specifically configured to: obtain a shared network parameter corresponding to the shared network and a scene network parameter of the scene network corresponding to the i-th scene; wherein the shared network parameter is used to represent the information commonality between the reference common information corresponding to different scenes, and the scene network parameter is used to represent the information difference between the reference common information corresponding to the i-th scene and other scenes; generate a conversion parameter according to the shared network parameter and the scene network parameter, and perform conversion processing on the reference common information corresponding to the i-th scene by using the conversion parameter, to obtain the conversion information of the i-th scene.

[0200] In an embodiment, the processor 1201 is specifically configured to: obtain an object identifier of the target object, a resource identifier of the target resource, and a scene identifier of the i-th scene; generate a sending weight of sending the target resource to the target object in the i-th scene according to the object identifier, the resource identifier, and the scene identifier of the i-th scene; and perform sending prediction processing by using the sending weight in combination with the sending indication information, to obtain a sending probability of sending the target resource to the target object in the i-th scene.

[0201] In an embodiment, the processor 1201 is specifically configured to: perform conversion processing on the object identifier, the resource identifier, and the scene identifier of the i th scene respectively to obtain identification information of the object identifier, identification information of the resource identifier, and identification information of the scene identifier of the i th scene; perform splicing processing on the identification information of the object identifier, the identification information of the resource identifier, and the identification information of the scene identifier of the i th scene to obtain splicing information; and call a prediction model to perform prediction on the splicing information to obtain a sending weight of sending the target resource to the target object in the i th scene.

[0202] In an embodiment, the sending probability of sending the target resource to the target object is obtained by calling a target prediction model, and the common information is obtained from the description information of the target object. When the description information of the target object is sample information, the processor 1201 is specifically configured to: obtain a sending label corresponding to the description information of the target object, the sending label being used to indicate whether the target resource is successfully sent to the target object in the i th scene; and adjust model parameters of an initial target prediction model based on the sending label and the sending probability to obtain an adjusted target prediction model.

[0203] In a specific implementation, the processor 1201, the memory 1202, and the communication interface 1203 described in the embodiments of the present application can perform the implementation manner described in the information processing method provided by the embodiments of the present application, and can also perform the implementation manner described in the information processing apparatus provided by the embodiments of the present application, which will not be described here.

[0204] By adjusting the model parameters of the target prediction model, the information gating module can automatically select the required important common information (i.e., reference common information) for each scene. Meanwhile, the star topology module can learn the commonality between the reference common information of each scene and the uniqueness of each scene by splitting the network parameters into shared network parameters and scene network parameters unique to each scene. In addition, the personalized parameter module of each scene splices the output of the star topology module of each scene and the unique information of the corresponding scene as input, and then strengthens the uniqueness of each scene and improves the personalized ability through the personalized weight gating module, which is beneficial to improving the prediction accuracy of the target prediction model for the resource sending probability and thus improving the resource processing efficiency.

[0205] The embodiments of the present application also provide a computer-readable storage medium having a computer program stored therein, which, when running on a computer device, causes the computer device to execute the information processing method of any possible implementation manner described above. The specific implementation manners can refer to the foregoing description, which will not be described here.

[0206] The embodiment of the present application further provides a computer program product, which comprises a computer program or computer instructions, and the computer program or computer instructions realize the steps of the information processing method provided by the embodiment of the present application when executed by a processor. The specific implementation manner can refer to the foregoing description, and will not be described here.

[0207] The embodiment of the present application further provides a computer program, which comprises computer instructions, the computer instructions are stored in a computer readable storage medium, a processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the information processing method provided by the embodiment of the present application. The specific implementation manner can refer to the foregoing description, and will not be described here.

[0208] It should be noted that, for each of the foregoing method embodiments, in order to simply describe, each is expressed as a combination of a series of actions, but those skilled in the art should know that the present application is not limited to the action order described, because according to the present application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions and modules involved are not necessarily necessary for the present application.

[0209] Those of ordinary skill in the art can understand that all or part of the steps of the various methods in the above embodiments can be completed by a program instructing relevant hardware, and the program can be stored in a computer readable storage medium, and the storage medium can include a flash disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0210] The above only describes some embodiments of the present application, and of course cannot limit the scope of the rights of the present application, so equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. An information processing method characterized by comprising: The method comprises: In response to sending a target resource to a target object, obtaining common information required by each scene when sending the target resource to the target object in M scenes; M is an integer greater than 1, the common information comprises a plurality of sub-information; the common information is information commonly owned by each scene; the reference importance degree of each sub-information in the ith scene and the reference common information of the ith scene are obtained by calling an information gating module, and the information gating module is provided with M information selection parameter groups, and one information selection parameter group corresponds to one scene; From the M information selection parameter groups in the information gating module, obtain the information selection parameter group corresponding to the ith scene; the information selection parameter group comprises a plurality of information selection parameters, and one information selection parameter corresponds to one information dimension; The information selection parameters corresponding to the same information dimension of any sub-information in the obtained information selection parameter group are taken as the reference importance degree of the any sub-information, and the common information is converted according to the reference importance degree of each sub-information in the ith scene to obtain the reference common information corresponding to the ith scene; i is a positive integer less than or equal to M; the reference importance degree is used to reflect the influence degree of the corresponding sub-information on the sending probability of the ith scene when the target resource is sent to the target object in the ith scene; Obtain the unique information required by the ith scene when sending the target resource; the unique information is information exclusively owned by the ith scene; Obtain the reference common information corresponding to each scene in the M scenes, and determine the information commonality between the reference common information corresponding to different scenes and the information difference between the reference common information corresponding to the ith scene and other scenes; Obtain the shared network parameter corresponding to the shared network in the star topology module and the scene network parameter of the scene network corresponding to the ith scene in the star topology module; wherein the star topology module comprises a shared network and M scene networks, one scene network is associated with one scene, the shared network parameter is used to represent the information commonality between the reference common information corresponding to different scenes, and the scene network parameter is used to represent the information difference between the reference common information corresponding to the ith scene and other scenes; generate a conversion parameter according to the shared network parameter and the scene network parameter, and convert the reference common information corresponding to the ith scene by using the conversion parameter to obtain the conversion information of the ith scene; the conversion information comprises the information commonality and the information difference; The conversion information of the ith scene and the unique information after dimension reduction processing are spliced to obtain the sending indication information of the ith scene, and the sending probability of the target resource sent to the target object in the ith scene is predicted according to the sending indication information.

2. The method of claim 1, wherein, The method comprises: Obtaining description information of the target object, the description information comprising one or more object description information of the target object and one or more resource description information of a target resource to be sent to the target object; Obtaining description information that is required to be used in sending the target resource to the target object in the M scenes from the description information; Converting the obtained description information from the description information into common information, wherein one description information obtained from the description information is converted into one sub-information in the common information.

3. The method of claim 1, wherein, The obtained unique information of the i-th scene in sending the target resource comprises: Obtaining description information of the target object, the description information comprising one or more object description information of the target object and one or more resource description information of a target resource to be sent to the target object; Obtaining object description information related to the i-th scene of the M scenes from the one or more object description information of the description information, and obtaining resource description information related to the i-th scene from the one or more resource description information of the description information; Converting the object description information related to the i-th scene and the resource description information related to the i-th scene into unique information of the i-th scene.

4. The method of claim 1, wherein, The conversion processing of the common information according to the reference importance of each sub-information in the i-th scene comprises: Performing weighted processing on any sub-information according to the reference importance of the any sub-information to obtain a corresponding weighted sub-information; Combining the obtained weighted sub-information to obtain combination information as the reference common information corresponding to the i-th scene.

5. The method of claim 4, wherein, The combination of the obtained weighted sub-information to obtain combination information as the reference common information corresponding to the i-th scene comprises: Obtaining an information dimension corresponding to the unique information of the i-th scene after dimension reduction processing; Performing dimension reduction processing on each weighted sub-information according to the information dimension, so that the information dimension of the weighted sub-information after dimension reduction processing is the same as the information dimension of the unique information after dimension reduction processing; Combining the weighted sub-information after dimension reduction processing to obtain combination information as the reference common information corresponding to the i-th scene.

6. The method of claim 1, wherein, The prediction of a sending probability of sending the target resource to the target object in the i-th scene according to the sending indication information comprises: Obtaining an object identifier of the target object, a resource identifier of the target resource and a scene identifier of the i-th scene; Generating a sending weight of sending the target resource to the target object in the i-th scene according to the object identifier, the resource identifier and the scene identifier of the i-th scene; Performing sending prediction processing by combining the sending weight with the sending indication information to obtain the sending probability of sending the target resource to the target object in the i-th scene.

7. The method of claim 6, wherein, The generating, according to the object identifier, the resource identifier and the scene identifier of the i-th scene, of a sending weight of the target resource sent by the i-th scene to the target object comprises: respectively converting the object identifier, the resource identifier and the scene identifier of the i-th scene to obtain identifier information of the object identifier, identifier information of the resource identifier and identifier information of the scene identifier of the i-th scene; splicing the identifier information of the object identifier, the identifier information of the resource identifier and the identifier information of the scene identifier of the i-th scene to obtain spliced information; calling a prediction module to predict the spliced information to obtain the sending weight of the target resource sent by the i-th scene to the target object.

8. The method of claim 1, wherein, The sending probability of the target resource sent to the target object is obtained by calling a target prediction model, and the common information is obtained from the description information of the target object. When the description information of the target object is sample information, the method further comprises: obtaining a sending label corresponding to the description information of the target object; the sending label is used to indicate whether the target resource is successfully sent to the target object in the i-th scene; adjusting model parameters of an initial target prediction model based on the sending label and the sending probability to obtain an adjusted target prediction model.

9. An information processing apparatus, comprising: The device comprises: an obtaining unit configured to, in response to a trigger operation of sending a target resource to a target object, obtain common information required by each scene when the target resource is sent to the target object in M scenes; M is an integer greater than 1, the common information comprises a plurality of sub-information, the common information is information common to each scene, and the reference importance degree of each sub-information in the i-th scene and the reference common information of the i-th scene are obtained by calling an information gating module, the information gating module is provided with M information selection parameter groups, and one information selection parameter group corresponds to one scene; the obtaining unit is further configured to obtain an information selection parameter group corresponding to the i-th scene from the M information selection parameter groups in the information gating module; the information selection parameter group comprises a plurality of information selection parameters, and one information selection parameter corresponds to one information dimension; and the information selection parameter corresponding to the same information dimension of any sub-information in the obtained information selection parameter group is used as the reference importance degree of the any sub-information; a processing unit configured to convert the common information according to the reference importance degree of each sub-information in the i-th scene to obtain reference common information corresponding to the i-th scene; i is a positive integer less than or equal to M; and the reference importance degree is used to reflect the influence degree of the corresponding sub-information on the sending probability of the i-th scene when the target resource is sent to the target object in the i-th scene; the obtaining unit is further configured to obtain unique information required by the i-th scene when the target resource is sent in the i-th scene; and the unique information is information exclusively owned by the i-th scene. The processing unit is further configured to obtain reference common information corresponding to each of the M scenes respectively, and determine information commonality between the reference common information corresponding to different scenes, and information difference between the reference common information corresponding to the i-th scene and other scenes; The processing unit is further configured to obtain a shared network parameter corresponding to a shared network in a star topology module, and a scene network parameter of a scene network corresponding to the i-th scene in the star topology module; the star topology module includes the shared network and M scene networks, one scene network is associated with one scene, the shared network parameter is used to represent the information commonality between the reference common information corresponding to different scenes, and the scene network parameter is used to represent the information difference between the reference common information corresponding to the i-th scene and other scenes; a conversion parameter is generated according to the shared network parameter and the scene network parameter, and the conversion parameter is used to perform conversion processing on the reference common information corresponding to the i-th scene to obtain conversion information of the i-th scene; the conversion information includes the information commonality and the information difference; The processing unit is further configured to perform splicing processing on the conversion information of the i-th scene and corresponding unique information after dimension reduction processing to obtain sending indication information of the i-th scene, and predict a sending probability of the target resource to the target object in the i-th scene according to the sending indication information.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the information processing method in any one of claims 1-8.

11. A computer device, comprising: The computer device includes a memory, a communication interface and a processor, and the memory, the communication interface and the processor are connected with each other; the memory stores a computer program, and the processor invokes the computer program stored in the memory to implement the information processing method in any one of claims 1-8.

Images