Resource processing method, electronic device and computer program product

Abstract

The invention provides a resource processing method, electronic equipment and a computer program product, and the method comprises the steps: for each resource in a plurality of resources, predicting a first probability that a user object executes a resource transfer event, and a second probability that the user object executes the resource transfer event to consume resources; determining first behavior data of executing the resource transfer event by the user object; correcting the first probability based on the first behavior data to obtain a third probability, and correcting the second probability based on the first behavior data to obtain a fourth probability; and based on the third probability and the fourth probability corresponding to each resource, selecting a resource matched with the user object from the plurality of resources. In this way, the resource most matched with the user object can be selected from the multiple resources.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a resource processing method, electronic device, and computer program product. Background Art

[0002] In the coupon promotion scenario, different coupons have different degrees of influence on users' decision-making results. Therefore, determining the coupon that best matches the user and distributing the corresponding coupon to the user can help business personnel attract customers more accurately. Summary of the Invention

[0003] Embodiments of the present application provide a resource processing method, apparatus, electronic device, and computer-readable storage medium, which can select a resource that best matches a user object from multiple resources.

[0004] The technical solution of the embodiment of the present application is implemented as follows:

[0005] An embodiment of the present application provides a method for processing resources, the method comprising:

[0006] For each resource among the plurality of resources, predicting a first probability that a user object executes a resource transfer event and a second probability that the user object consumes the resource by executing the resource transfer event;

[0007] Determine first behavior data of the user object executing the resource transfer event;

[0008] Correcting the first probability based on the first behavior data to obtain a third probability, and correcting the second probability based on the first behavior data to obtain a fourth probability;

[0009] Based on the third probability and the fourth probability corresponding to each resource, a resource matching the user object is selected from the multiple resources.

[0010] An embodiment of the present application provides a resource processing device, the device comprising:

[0011] a prediction module, configured to predict, for each of a plurality of resources, a first probability that a user object executes a resource transfer event and a second probability that the user object consumes the resource by executing the resource transfer event;

[0012] A determination module, configured to determine first behavior data of the user object executing the resource transfer event;

[0013] a correction module, configured to correct the first probability based on the first behavior data to obtain a third probability, and correct the second probability based on the first behavior data to obtain a fourth probability;

[0014] The selection module is configured to select a resource matching the user object from the multiple resources based on the third probability and the fourth probability corresponding to each resource.

[0015] An embodiment of the present application provides an electronic device, including:

[0016] Memory for storing computer-executable instructions or computer programs;

[0017] The processor is used to implement the resource processing method provided in the embodiment of the present application when executing the computer executable instructions or computer program stored in the memory.

[0018] An embodiment of the present application provides a computer-readable storage medium, which stores computer-executable instructions or a computer program. When the computer-executable instructions or the computer program are executed by a processor, the processor will execute the resource processing method provided by the embodiment of the present application.

[0019] The embodiments of the present application provide a computer program product, which includes a computer program or computer-executable instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer-executable instructions or computer program from the computer-readable storage medium and executes the computer-executable instructions or computer program, causing the electronic device to perform the resource processing method provided in the embodiments of the present application.

[0020] The embodiments of the present application have the following beneficial effects:

[0021] For each of the multiple resources, a first probability of the user object executing a resource transfer event and a second probability of the user object consuming resources when executing the resource transfer event are predicted. Then, based on the first behavior data of the user object executing the resource transfer event, the first probability is corrected to obtain a third probability, and based on the first behavior data, the second probability is corrected to obtain a fourth probability. Therefore, based on the third probability and the fourth probability corresponding to each resource, a resource matching the user object is selected from the multiple resources. In this way, compared to the solution of selecting the resource with the highest probability of executing the resource transfer event as the resource that best matches the user object, the present application combines the probability of the user object executing the resource transfer event for the resource and the probability of the user object consuming the corresponding resource to select the resource that best matches the user object. This can ensure the correlation between the behavior of the user object executing the resource transfer event and the resource, which can not only improve the utilization rate of the resources, but also ensure that the selected resource has the best incentive effect on the user object to execute the resource transfer event, that is, the selected resource is the resource that best matches the user object; in addition, based on the first behavior data, the first probability and the second probability are corrected respectively, so that based on the corrected two probabilities, the resource that matches the user object is selected, which improves the accuracy of the selected resource, that is, further improves the degree of matching between the selected resource and the user object. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 1 is a schematic diagram of the architecture of a resource processing system 100 provided in an embodiment of the present application;

[0023] Figure 2 is a structural diagram of an electronic device provided in an embodiment of the present application;

[0024] Figure 3 This is a flowchart of a method for processing resources provided in an embodiment of the present application;

[0025] Figure 4 is a schematic diagram of a process for predicting a first probability provided by an embodiment of the present application;

[0026] Figure 5 is a schematic diagram of a flow chart for determining the fourth feature provided in an embodiment of the present application;

[0027] Figure 6 Schematic diagram of the model structure of the first probability prediction model provided in an embodiment of the present application;

[0028] Figure 7 is a schematic diagram of a process for training a first probability prediction model provided in an embodiment of the present application;

[0029] Figure 8 is a schematic diagram of a flow chart for correcting a first probability provided in an embodiment of the present application;

[0030] Figure 9 This is a schematic diagram of establishing a mapping relationship provided in an embodiment of the present application;

[0031] Figure 10 This is a schematic diagram of the process of determining the number of resource transfers provided by an embodiment of the present application;

[0032] Figure 11 This is a technical architecture diagram of the resource processing method provided in the embodiment of the present application;

[0033] Figure 12 It is a structural diagram of the multi-expert network prediction model provided in the embodiment of the present application. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0035] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0036] In the following description, the terms "first\second\third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0037] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0038] Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations.

[0039] 1) Artificial Intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. AI technology is an interdisciplinary discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Foundational AI technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics.

[0040] 2) Client, also known as the user end, refers to the program corresponding to the server that provides local services to users. Except for some applications that can only run locally, it is generally installed on an ordinary client and needs to cooperate with the server to run. That is, there must be corresponding servers and service programs in the network to provide corresponding services. In this way, a specific communication connection needs to be established between the client and the server to ensure the normal operation of the application.

[0041] During the research process, the inventors found that in the coupon promotion scenario, a model is generally used to identify users with the highest probability of changing their decision results due to the coupon activity, helping business personnel to attract customers more accurately. Specifically, for coupons of different amounts, the user's conversion probability for the promotion activity is predicted separately, so as to select the coupon with the highest probability as the coupon to be allocated to the user; however, some users are less sensitive to coupons. For example, after being allocated a certain coupon, the user may not use the coupon for various reasons, but the corresponding user also performs conversion behavior. In this way, the utilization rate of the coupon is low, and the correlation between the coupon and the user's conversion behavior is also low, that is, the matching degree between the corresponding coupon and the user is low; or, the user will consume the corresponding coupon and perform conversion behavior under coupons of different denominations, but the denomination of the coupon selected by the prediction is the largest. At this time, the incentive effect of the predicted selected coupon on the user's conversion behavior is relatively general, that is, the predicted selected coupon is also less matching with the user. Based on this, the matching degree between the coupon selected by the coupon allocation method in the related art and the user may be low.

[0042] Based on this, an embodiment of the present application provides a resource processing method, which combines the probability of a user object executing a resource transfer event for a resource and the probability of a user object consuming the corresponding resource to select the resource that best matches the user object. This can ensure the correlation between the behavior of the user object executing the resource transfer event and the resource, which can not only improve the resource utilization, but also ensure that the selected resource has the best incentive effect on the user object executing the resource transfer event, that is, the selected resource is the resource that best matches the user object; in addition, based on the first behavior data, the first probability and the second probability are corrected respectively, so that based on the corrected two probabilities, the resource that matches the user object is selected, thereby improving the accuracy of the selected resource, that is, further improving the degree of matching between the selected resource and the user object.

[0043] See also Figure 1 , Figure 1 This is an architectural diagram of a resource processing system 100 provided in an embodiment of the present application. To implement an application scenario of resource processing, a terminal (terminal 400 is shown as an example) is connected to a server 200 via a network 300. The network 300 may be a wide area network or a local area network, or a combination of the two. The terminal 400 is used for users to use a client 401, which is displayed on a display interface (display interface 401-1 is shown as an example). The terminal 400 and the server 200 are connected to each other via a wired or wireless network.

[0044] The terminal 400 is used to send multiple resources to the server 200;

[0045] Server 200 is used to receive multiple resources; for each of the multiple resources, predict a first probability that the user object executes a resource transfer event, and a second probability that the user object consumes resources when executing the resource transfer event; determine first behavior data of the user object executing the resource transfer event; based on the first behavior data, correct the first probability to obtain a third probability, and based on the first behavior data, correct the second probability to obtain a fourth probability; based on the third probability corresponding to each resource and the fourth probability, select a resource that matches the user object from the multiple resources.

[0046] In some embodiments, the server 200 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks (CDN, Content Deliver Network), and big data and artificial intelligence platforms. The terminal 400 can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a set-top box, an intelligent voice interaction device, a smart home appliance, a vehicle-mounted terminal, an aircraft, and a mobile device (for example, a mobile phone, a portable music player, a personal digital assistant, a dedicated messaging device, a portable gaming device, a smart speaker, and a smart watch), etc., but is not limited thereto. The terminal device and the server can be directly or indirectly connected via wired or wireless communication, which is not limited in the embodiments of the present application.

[0047] See also Figure 2 , Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. In practical applications, the electronic device can be Figure 1 The server 200 or terminal 400 shown, see Figure 2 , Figure 2 The electronic device shown includes: at least one processor 410, a memory 450, at least one network interface 420 and a user interface 430. The various components in the terminal 400 are coupled together via a bus system 440. It is understood that the bus system 440 is used to achieve connection and communication between these components. In addition to including a data bus, the bus system 440 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 440 is not shown in FIG. Figure 2 Various buses are labeled as bus system 440 .

[0048] The processor 410 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0049] The user interface 430 includes one or more output devices 431 that enable presentation of media content, including one or more speakers and / or one or more visual display screens. The user interface 430 also includes one or more input devices 432, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0050] The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard drives, optical drives, etc. The memory 450 may optionally include one or more storage devices that are physically remote from the processor 410.

[0051] The memory 450 includes volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 450 described in the embodiments of the present application is intended to include any suitable type of memory.

[0052] In some embodiments, the memory 450 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplified below.

[0053] Operating system 451, including system programs for processing various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, and driver layer, which are used to implement various basic services and process hardware-based tasks;

[0054] A network communication module 452 is used to reach other electronic devices via one or more (wired or wireless) network interfaces 420. Exemplary network interfaces 420 include Bluetooth, Wi-Fi, and Universal Serial Bus (USB).

[0055] a presentation module 453 for enabling presentation of information via one or more output devices 431 (e.g., a display screen, a speaker, etc.) associated with the user interface 430 (e.g., a user interface for operating peripheral devices and displaying content and information);

[0056] The input processing module 454 is configured to detect one or more user inputs or interactions from one of the one or more input devices 432 and to translate the detected inputs or interactions.

[0057] In some embodiments, the apparatus provided in the embodiments of the present application may be implemented in software. Figure 2 The processing device 455 for resources stored in the memory 450 is shown. This device can be software in the form of a program or plug-in, and includes the following software modules: a prediction module 4551, a determination module 4552, a correction module 4553, and a selection module 4554. These modules are logical and can be arbitrarily combined or further separated according to the functions they implement. The functions of each module will be described below.

[0058] In other embodiments, the apparatus provided in the embodiments of the present application may be implemented in hardware. As an example, the resource processing apparatus provided in the embodiments of the present application may be a processor in the form of a hardware decoding processor, which is programmed to execute the resource processing method provided in the embodiments of the present application. For example, the processor in the form of a hardware decoding processor may be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0059] In some embodiments, the terminal or server can implement the resource processing method provided in the embodiment of the present application by running a computer program. For example, the computer program can be a native program or software module in the operating system; it can be a local (Native) application (APP, Application), that is, a program that needs to be installed in the operating system to run, such as an instant messaging APP, a web browser APP; it can also be a small program, that is, a program that can be run only by downloading it into a browser environment; it can also be a small program that can be embedded in any APP. In short, the above-mentioned computer program can be any form of application, module or plug-in.

[0060] Based on the above description of the resource processing system and electronic device provided by the embodiment of the present application, the resource processing method provided by the embodiment of the present application is described below. In actual implementation, the resource processing method provided by the embodiment of the present application can be implemented by the terminal or the server alone, or by the terminal and the server in collaboration, so that Figure 1 The server 200 in the embodiment of the present application alone executes the resource processing method provided by the embodiment of the present application as an example for explanation. Figure 3 , Figure 3 This is a flow chart of the resource processing method provided by the embodiment of the present application. Figure 3 The steps shown will be described.

[0061] Step 101: The server predicts, for each of a plurality of resources, a first probability that a user object executes a resource transfer event and a second probability that the user object consumes resources by executing the resource transfer event.

[0062] It should be noted that the resource transfer event here is used to indicate a consumption event executed by a user object, that is, an event that the user object needs to complete by paying assets, that is, an event completed by transferring its own assets to another object; for example, it can be a purchase event for a product, or a download event for an application, etc., which is not limited in this embodiment of the present application; and resources are used to indicate vouchers that can reduce the assets that the user needs to pay when executing a resource transfer event, such as coupons, discount coupons, vouchers, etc., which are not limited in this embodiment of the present application;

[0063] At the same time, different resources can be of different types, for example, resource 1 can be a coupon, resource 2 can be a discount coupon, etc.; if multiple resources are of the same type, the denominations of different resources are different, for example, if the resources are discount coupons, resource 1 can be a 10% discount coupon, resource 2 can be an 20% discount coupon, etc.; in addition, multiple resources can include both resources of different types and resources of different denominations. For example, multiple resources correspond to multiple types, and the number of resources under the same type is also multiple. Here, the denominations of resources under the same type are different.

[0064] It should be noted that, for each of the multiple resources, predicting the first probability of the user object executing the resource transfer event means, assuming that the user object is allocated the corresponding resource, determining the probability of the user object executing the resource transfer event under the influence of the corresponding resource, for example, the probability of purchasing a certain product or downloading a certain application; and the second probability of the user object consuming resources when executing the resource transfer event means, the probability of using the corresponding resources when the user object executes the resource transfer event, for example, the probability of the user using a coupon when executing a consumption event.

[0065] In actual implementation, see Figure 4 , Figure 4 This is a flow chart of predicting the first probability provided by the embodiment of the present application, based on Figure 4 The process of predicting the first probability of a user object executing a resource transfer event for each of a plurality of resources is implemented by the following steps.

[0066] Step 1011: extract features from the user object to obtain a first feature, and extract features based on the resource to obtain a second feature.

[0067] In actual implementation, the process of extracting features from user objects to obtain the first feature can be: determining multiple feature types corresponding to the user object, and extracting features from the user object for each of the multiple feature types to obtain a first intermediate feature corresponding to the feature type; determining the weight corresponding to each feature type based on resources, and multiplying each weight by the corresponding first intermediate feature to obtain a second intermediate feature; and fusing multiple second intermediate features to obtain the first feature.

[0068] It should be noted that the multiple feature types corresponding to the user object include at least user attributes, interactive behaviors, activity and consumption willingness, among which user attributes refer to the basic attributes of the user, such as age, gender, permanent residence, occupational information, etc. Interactive behaviors include the user's historical usage statistics of coupons, activity includes the user's login behavior statistics in the target application, and consumption willingness includes the user's historical transaction history statistics, such as the total amount of internal consumption, etc. The embodiments of this application do not limit this.

[0069] In actual implementation, the process of determining the weight corresponding to each feature type based on resources is to extract features based on resources to obtain the second feature, and then extract the weight corresponding to each feature type based on the second feature, that is:

[0070] g(x)=S(w g x t )...Formula (1);

[0071] Among them, S indicates the weight list, w g is a hyperparameter, x t Indicates the second feature, and g(x) indicates the weight corresponding to the feature type.

[0072] It should be noted that in the process of extracting features based on resources to obtain the second feature, the second feature refers to the attribute characteristics of the resource itself, such as resource type, denomination size, historical consumption rate, etc., which can characterize the characteristics of the resource, as well as some characteristics that may affect resource allocation decisions, such as the model classification of pre-services. This embodiment of the present application does not limit this.

[0073] Then, each weight is multiplied by the corresponding first intermediate feature to obtain the second intermediate feature, and multiple second intermediate features are fused to obtain the first feature.

[0074] It should be noted that fusing multiple second intermediate features here may be fusing and summing the multiple second intermediate features, or concatenating the multiple second intermediate features, which is not limited in the embodiments of the present application; wherein, when fusing multiple second intermediate features means summing the multiple second intermediate features, the multiple second intermediate features are fused to obtain the first feature, that is:

[0075]

[0076] Among them, F(x) is the first feature, n is the number of feature types, f i (x i ) is the first intermediate feature corresponding to the i-th feature type, g(x t ) i is the weight corresponding to the i-th feature type, g(x t ) i f i (x i ) is the second intermediate feature corresponding to the i-th feature type.

[0077] In this way, user features are extracted in various dimensions, and the extracted features are weighted, and then the weighted features are fused to obtain user features. In this way, user features can be extracted in an all-round and multi-dimensional manner, ensuring the comprehensiveness of the extracted user features. At the same time, through weighted processing, the features are adjusted according to the importance of different types of features, which improves the accuracy of the extracted user features. In the subsequent prediction process based on user features, the accuracy of the prediction results can also be improved, and it helps to reduce the risk of overfitting.

[0078] Step 1012: extract features based on the user object's unconsumed behavior of resources to obtain a third feature.

[0079] It should be noted that the unconsumed behavior of resources refers to the situation where users have allocated resources but have not used them. In this case, it may be that the user is unwilling to use the resources, or it may be that the resource usage standards are not met, resulting in non-use rather than the user's true unwillingness to use the resources. Therefore, by extracting corresponding features for prediction, it can play a certain role in debiasing the prediction results.

[0080] Step 1013: determine first allocation data of the plurality of resources, and determine a fourth feature for eliminating causal deviation based on the first allocation data.

[0081] It should be noted that the first allocation data of multiple resources refers to the allocation data of resources within the target period, for example, the various coupons allocated within the target period and how each coupon is allocated; wherein the target period can be pre-set, such as the past month.

[0082] Causal bias refers to the phenomenon in causal inference or machine learning where model predictions or conclusions deviate due to failure to correctly model or consider the causal relationship between variables. For example, when the resource is a coupon, under normal circumstances, that is, under common sense, assuming the coupon type is the same, the larger the denomination of the coupon, the higher the consumption rate. Here, causal bias can be, for example, the counterintuitive phenomenon that the consumption rate of low-denomination coupons is greater than that of high-denomination coupons due to selection bias caused by user selection strategies.

[0083] In actual implementation, for the process of determining the fourth feature for eliminating causal bias based on the first distribution data, specifically, see Figure 5 , Figure 5 This is a flow chart of determining the fourth feature provided by the embodiment of the present application, based on Figure 5 , the process of determining the fourth feature for eliminating causal bias can be achieved through the following steps.

[0084] Step 10131: Determine a first probability distribution of multiple resources of different types based on the first allocation data.

[0085] In actual implementation, as mentioned above, the first allocation data is the allocation data of resources within the target period. Therefore, first, based on the first allocation data, the total number of resources issued within the target period is counted, and the number of each type of resources is obtained; for each type of resource, the ratio of the number of resources of that type to the total number of resources is obtained, and then based on multiple ratios, a first probability distribution of different types of resources is generated.

[0086] It should be noted that the first probability distribution refers to the probability distribution of resources released in the past target period, and N refers to the type of resources, p N It refers to the ratio corresponding to the Nth type of resources. The types here include the types and denominations mentioned above. That is, different types of resources include resources of different types, as well as resources of different denominations under the same type. The total number of resources can be, for example, the total number of coupons.

[0087] Step 10132: Group the resources based on their types to obtain multiple groups of resources, and extract features from each of the multiple groups of resources to obtain a first group feature of the resources.

[0088] It should be noted that resources are grouped based on their types. As mentioned above, if there are N types of resources, N groups will be obtained. Then, for each group of resources, feature extraction is performed on the group of resources to obtain the corresponding resource features. Then, the resource features corresponding to each group of resources are spliced ​​to obtain the first group feature of the resource, i.e., i = [i1,…,ij ,…,i N ], where N refers to the type of resource.

[0089] For example, if there are three groups of resources, namely resource group 1, resource group 2 and resource group 3, then the process of performing feature extraction based on each group of resources in the multiple groups of resources to obtain the first grouping feature of the resources is as follows: for resource group 1, feature extraction is performed on the resources in resource group 1 to obtain i1; for resource group 2, feature extraction is performed on the resources in resource group 2 to obtain i2; for resource group 3, feature extraction is performed on the resources in resource group 3 to obtain i3; then, i1, i2 and i3 are spliced ​​to obtain the first grouping feature, i.e., i = [i1, i2, i3].

[0090] Step 10133: cluster the user objects to obtain multiple clusters of user objects, and extract features based on each cluster of user objects in the multiple clusters to obtain second grouping features of the user objects.

[0091] In actual implementation, there are multiple user objects. Then, according to the age, occupation and other information of multiple users, the multiple user objects are clustered to obtain multiple clusters of user objects, such as K clusters of user objects. Then, for each cluster of user objects, the features of the cluster user objects are extracted to obtain the corresponding user object features. Then, the user object features corresponding to each cluster of user objects are spliced ​​to obtain the second grouping feature of the user object, i.e., u=[u1,…,u i ,…,u K ], where K refers to the number of clusters of user objects obtained by clustering.

[0092] For example, if there are three clusters of user objects, namely user object cluster 1, user object cluster 2 and user object cluster 3, then feature extraction is performed based on each cluster of user objects in the multiple clusters of user objects to obtain the second grouping feature of the user objects. The process is as follows: for user object cluster 1, feature extraction is performed on the user objects in user object cluster 1 to obtain u1; for user object cluster 2, feature extraction is performed on the user objects in user object cluster 2 to obtain u2; for user object cluster 3, feature extraction is performed on the user objects in user object cluster 3 to obtain u3; then, u1, u2 and u3 are spliced ​​to obtain the second grouping feature, namely u = [u1, u2, u3].

[0093] Step 10134: fuse the first probability distribution, the first grouping feature, and the second grouping feature to obtain a fourth feature.

[0094] In actual implementation, the fusion of the first probability distribution, the first grouping feature, and the second grouping feature can be performed by at least one of feature concatenation, multiplication, or addition, etc., and this is not limited in the embodiments of the present application. In this way, after grouping users and resources, the first probability distribution of the resources, the user grouping feature, and the resource grouping feature are fused to obtain features for eliminating causal bias. In this way, causal bias can be effectively eliminated, which not only enhances the representation ability of the features, but also improves the generalization ability of the model, thereby improving the interpretability of the subsequent probability-based prediction process.

[0095] In some embodiments, the process of fusing the first probability distribution, the first grouping feature, and the second grouping feature to obtain the fourth feature may specifically be: multiplying the first probability distribution by the first grouping feature to obtain the third grouping feature; concatenating the third grouping feature with the second grouping feature to obtain the second concatenated feature; and performing feature cross-processing based on the second concatenated feature to obtain the fourth feature.

[0096] It should be noted that the third grouping feature is obtained by multiplying the first probability distribution by the first grouping feature. Since the first probability distribution refers to the allocation ratio of different types of resources, multiplying the first probability distribution by the first grouping feature is equivalent to weighting the first grouping feature, thereby obtaining the weighted first grouping feature, which is also the third grouping feature. The third grouping feature is then concatenated with the second grouping feature to obtain the second concatenated feature, and feature cross-processing is performed based on the second concatenated feature to obtain the fourth feature.

[0097] In this way, after the first grouping feature is weighted based on the first probability distribution to obtain the third grouping feature, the third grouping feature and the second grouping feature are spliced ​​together, and the second spliced ​​feature is subjected to feature cross processing to obtain the fourth feature. In this way, not only the accuracy of the feature is improved through the weighting process, but also, since feature cross refers to generating new composite features by combining original features, the interactive relationship between features can be captured through feature cross processing, thereby improving the expression ability and prediction performance of the model.

[0098] In actual implementation, the process of performing feature cross-processing based on the second splicing feature to obtain the fourth feature can be: performing feature decomposition on the second splicing feature to obtain multiple sub-features, and combining the multiple sub-features in pairs to obtain a feature combination; for each feature combination, based on the correlation relationship between the two sub-features in the feature combination, determining the associated feature corresponding to the feature combination; and fusing the sub-features and the associated features to obtain the fourth feature.

[0099] It should be noted that feature decomposition maps high-dimensional sparse features to low-dimensional dense latent vector space, which reduces the computational complexity and can capture the potential structure of features. Specifically, assuming that the second concatenated feature is x = [x1,…,x n ], thereby decomposing the second concatenated feature to obtain n sub-features x, where for the i-th sub-feature x i , and at the same time be able to determine the latent vector v corresponding to the sub-feature i ;

[0100] Then, multiple sub-features are combined in pairs to obtain feature combinations. At the same time, for each feature combination, the latent vectors corresponding to the two sub-features in the feature combination are determined, and the dot product of the two latent vectors is determined. The dot product and the two sub-features are then fused, such as multiplied, to obtain the associated features corresponding to the feature combination;

[0101] Finally, the sub-features and related features are fused, that is, multiple sub-features and multiple related features are fused, such as adding them together, to obtain the fourth feature, namely:

[0102]

[0103] Among them, z is the fourth feature, w0 is the global bias term, and w i is the weight corresponding to the i-th sub-feature, x i is the i-th sub-feature, x j is the jth sub-feature, v i is the latent vector corresponding to the i-th sub-feature, v j is the latent vector corresponding to the j-th sub-feature, <v i ,v j > is the dot product of two latent vectors, <v i ,v i >x i *x j is the associated feature corresponding to the feature combination formed by the i-th sub-feature and the j-th sub-feature, and n is the number of sub-features; at the same time, is a linear term, representing the independent contribution of each sub-feature, It is a feature combination term, which represents the interactive contribution between features.

[0104] In this way, through feature decomposition, each feature can be mapped to a low-dimensional latent vector space to obtain a latent vector. Then, through feature combination, the interaction relationship between features can be modeled based on the dot product of the latent vector, avoiding the direct calculation of high-dimensional feature interactions. In this way, the interaction relationship between features can be efficiently captured, which not only improves the efficiency of feature extraction, but also improves the accuracy of the extracted features.

[0105] Step 1014 , performing feature splicing on the first feature, the second feature, the third feature, and the fourth feature to obtain a first spliced ​​feature.

[0106] It should be noted that after determining the first feature, the second feature, the third feature, and the fourth feature, the first feature, the second feature, the third feature, and the fourth feature can be spliced ​​to obtain a first spliced ​​feature, that is:

[0107] f=f(x)‖x bias ‖x t ‖x q ...Formula (4);

[0108] Among them, f is the first splicing feature, ‖ is used to indicate the splicing process, and x bias is the third feature, x t is the second feature, x q is the fourth feature, and f(x) is the first feature.

[0109] Step 1015 : Based on the first splicing feature, the probability of the user object executing the resource transfer event is predicted to obtain a first probability.

[0110] It should be noted that after determining the first splicing feature, the probability of the user object executing the resource transfer event can be predicted based on the first splicing feature to obtain the first probability, that is,

[0111] y=h(f(x)‖x bias ‖x t ‖x q )...Formula (5);

[0112] Among them, y is the first probability, ‖ is used to indicate the splicing process, and x bias is the third feature, x t is the second feature, x q is the fourth feature, and f(x) is the first feature.

[0113] In this way, after the features extracted from different dimensions are spliced ​​to obtain the first spliced ​​feature, the probability of the user object executing the resource transfer event is predicted based on the first spliced ​​feature to obtain the first probability. In this way, the probability of the user object executing the resource transfer event can be accurately determined, which helps to select resources that match the user object based on the corresponding probability in the subsequent process.

[0114] It should be noted that the process of determining the second probability that a user object executes a resource transfer event and consumes resources is similar to the process of determining the first probability that a user object executes a resource transfer event, except that at the end, the probability of the user object executing a resource transfer event based on the first splicing feature is predicted to obtain the first probability, which is replaced by the probability of the user object executing a resource transfer event and consuming resources based on the first splicing feature to obtain the second probability; this is not elaborated in the embodiments of the present application.

[0115] In actual implementation, the process of predicting the first probability of the user object executing the resource transfer event can be implemented by a first probability prediction model, wherein, see Figure 6 , Figure 6 This is a schematic diagram of the model structure of the first probability prediction model provided in the embodiment of the present application, based on Figure 6 The first probability prediction model includes a feature extraction layer, a feature debiasing layer, a feature splicing layer and a prediction layer, so as to extract features from user objects to obtain the first feature, and extract features based on resources to obtain the second feature. The process can be, through the feature extraction layer, extracting features from user objects to obtain the first feature, and extracting features based on resources to obtain the second feature; and the process of extracting features based on the non-consumption behavior of user objects for resources to obtain the third feature can be, through the feature extraction layer, extracting features based on the non-consumption behavior of user objects for resources to obtain the third feature; and based on the first allocation data, determining the method for eliminating causal bias The process of obtaining the fourth feature of the resource transfer event may be, through a feature debiasing layer, based on the first allocation data, determining the fourth feature for eliminating the causal bias; the process of performing feature splicing on the first feature, the second feature, the third feature and the fourth feature to obtain the first splicing feature may be, through a feature splicing layer, performing feature splicing on the first feature, the second feature, the third feature and the fourth feature to obtain the first splicing feature; and the process of predicting the probability of the user object executing the resource transfer event based on the first splicing feature to obtain the first probability may be, through a prediction layer, based on the first splicing feature, predicting the probability of the user object executing the resource transfer event to obtain the first probability.

[0116] In some embodiments, before predicting the first probability of the user object executing the resource transfer event through the first probability prediction model, the first probability prediction model needs to be trained. Specifically, see Figure 7 , Figure 7 This is a flow chart of training the first probability prediction model provided by the embodiment of the present application, based on Figure 7 The process of training the first probability prediction model is achieved through the following steps.

[0117] In step 201 , the server obtains an untrained first probability prediction model, sample user objects carrying labels, and sample resources.

[0118] The label here is used to indicate the actual result of whether the sample user object executes the resource transfer event. For example, when the sample object executes the resource transfer event, the label is 0; when the sample object does not execute the resource transfer event, the label is 1.

[0119] Step 202: Perform feature extraction on the sample user object through the feature extraction layer to obtain the sample first feature, and perform feature extraction based on the sample resource to obtain the sample second feature.

[0120] Step 203: extracting features of the sample user object's unconsumed behavior of the sample resources through the feature extraction layer to obtain a third feature of the sample;

[0121] Step 204 : determining second allocation data of the plurality of sample resources, and determining a fourth feature of the sample for eliminating causal bias based on the second allocation data through a feature debiasing layer.

[0122] As described above with respect to the first allocation data, the second allocation data is allocation data of sample resources in the past target period.

[0123] Step 205 , performing feature splicing on the first feature of the sample, the second feature of the sample, the third feature of the sample, and the fourth feature of the sample through a feature splicing layer to obtain a first spliced ​​feature of the sample.

[0124] Step 206 : The prediction layer predicts the probability of the sample user object executing the resource transfer event based on the sample first splicing feature to obtain the sample first probability.

[0125] Step 207: Obtain the difference between the first probability of the sample and the label, and update the model parameters of the untrained first probability prediction model based on the difference.

[0126] In actual implementation, the difference between the first probability of the sample and the label is obtained, and the model parameters of the untrained first probability prediction model are updated based on the difference. That is, a pre-set loss function is obtained, and then the target loss corresponding to the first probability of the sample is determined based on the first probability of the sample and the label, as well as the loss function. The model parameters of the untrained first probability prediction model are updated based on the target loss, that is:

[0127]

[0128] Among them, L(w) is the target loss, Indicates the predicted value corresponding to the i-th sample user object, that is, the corresponding sample first probability, is the true value corresponding to the i-th sample user object, that is, the corresponding label, w t ,w s are the parameters of the model.

[0129] It should be noted that the process of predicting the second probability of a user object consuming resources when executing a resource transfer event can be implemented through a second probability prediction model, wherein the model results of the second probability model and the first probability model are the same, but there are differences between the first probability model and the second probability model during the training process. For example, for the first probability model, the label of the sample user object is used to indicate the actual result of whether the sample user object executes the resource transfer event, while for the second probability prediction model, the label of the sample user object is used to indicate the actual result of whether the sample user object consumes resources when executing the resource transfer event; based on this, the training process of the second probability prediction model is also similar to the training process of the first probability prediction model, and this will not be elaborated in the embodiments of the present application.

[0130] Step 102: Determine the first behavior data of the user object executing the resource transfer event.

[0131] It should be noted that the first behavior data of the user object executing the resource transfer event includes whether the user object executes the resource transfer event in the past target period, and whether resources are consumed when the user object executes the resource transfer event.

[0132] Step 103: Based on the first behavior data, correct the first probability to obtain a third probability, and based on the first behavior data, correct the second probability to obtain a fourth probability.

[0133] In actual implementation, for the process of correcting the first probability based on the first behavior data to obtain the third probability, see Figure 8 , Figure 8 This is a flow chart of correcting the first probability provided by the embodiment of the present application, based on Figure 8 Based on the first behavior data, the first probability is corrected to obtain the third probability through the following steps.

[0134] Step 1031: construct a mapping relationship for correcting the first probability based on the first behavior data.

[0135] It should be noted that the mapping relationship here is used to indicate the relationship between the true first probability and the predicted first probability (that is, the predicted first probability); wherein, the true first probability here is determined based on the first behavior data, and is the actual first probability of the user object executing the resource transfer event within the past target period, and the predicted first probability is the first probability of the user object executing the resource transfer event within the past target period, according to the method described above.

[0136] In actual implementation, there are multiple user objects, so the process of constructing a mapping relationship for correcting the first probability based on the first behavior data can be: based on the first behavior data, predicting a reference first probability of the user object executing a resource transfer event; based on the reference first probability, bucketing the user objects to obtain multiple buckets, each bucket corresponding to a reference first probability; for each bucket, performing the following processing to obtain the true first probability of the user object executing the resource transfer event in the bucket: determining a first number of user objects included in the bucket, and based on the first behavior data, determining a second number of user objects executing the resource transfer event in the bucket; determining the ratio of the second number to the first number as the true first probability corresponding to the bucket; based on the reference first probability and the true first probability corresponding to each bucket, constructing a linear relationship between the predicted first probability and the true first probability, and determining the linear relationship as a mapping relationship.

[0137] It should be noted that, based on the first behavior data, the reference first probability of the user object executing the resource transfer event is predicted, that is, within the past target period corresponding to the first behavior data, according to the method described above, the first probability of the user object executing the resource transfer event is predicted, that is, the reference first probability;

[0138] Then, based on the reference first probability, the user objects are bucketed to obtain multiple buckets. For example, if the number of reference first probabilities of multiple user objects is 100, the user objects corresponding to the same reference first probability are divided into one bucket, so that one bucket includes at least one user object, and one bucket corresponds to one reference first probability. At the same time, if the number of reference first probabilities here is greater than the quantity threshold, that is, greater than the preset number of buckets, when bucketing, the reference first probability of the user object can be rounded to reduce the number of reference first probabilities, thereby performing bucketing based on the rounded reference first probability.

[0139] Then, for each bucket, a ratio of the second number of user objects that have executed a resource transfer event in the bucket to the first number of user objects included in the bucket is determined as the true first probability corresponding to the bucket, so that for each bucket, there is a true first probability and a reference first probability;

[0140] Next, based on the reference first probability and the true first probability corresponding to each bucket, a linear relationship between the predicted first probability and the true first probability is constructed. Specifically, see Figure 9 , Figure 9 This is a schematic diagram of a mapping relationship provided by an embodiment of the present application, based on Figure 9 After bucketing multiple user objects, two adjacent buckets can constitute the two endpoints of a linear function, thereby constructing a linear function. Similarly, N buckets can construct N-1 linear functions. These N-1 linear functions are also the linear relationship between the predicted first probability and the actual first probability mentioned above.

[0141] In this way, based on the first probability predicted in the past target time period and the actual first probability in the same time period, a mapping relationship for correcting the first probability is constructed, which can ensure the accuracy of the constructed mapping relationship, so that when the first probability is subsequently corrected based on the mapping relationship, the first probability can be accurately corrected.

[0142] Step 1032: Based on the mapping relationship, correct the first probability to obtain a third probability.

[0143] It should be noted that, as mentioned above, the mapping relationship here is a multi-segment linear function. Therefore, based on the mapping relationship, the first probability is corrected to obtain the third probability. That is, the first probability is substituted as the independent variable into the mapping relationship, that is, the multi-segment linear function, and the dependent variable obtained is also the third probability.

[0144] In this way, a mapping relationship between the predicted first probability and the actual first probability is constructed based on the first behavior data, and the first probability is corrected based on the mapping relationship. In this way, the first probability can be accurately corrected.

[0145] It should be noted that the process of correcting the second probability based on the first behavior data to obtain the fourth probability is similar to the process of correcting the first probability based on the first behavior data to obtain the third probability. This embodiment of the present application will not be elaborated on this.

[0146] In some embodiments, the corrected second probability may be further corrected or adjusted. Specifically, after the second probability is corrected based on the first behavior data to obtain a fourth probability, an influence coefficient of the resource on the user object may be determined for each resource; the influence coefficient is used to indicate the influence of the resource on the user object's execution of the resource transfer event. Based on each influence coefficient, the third probability is adjusted to obtain a fifth probability, namely:

[0147] uplift_adjust u =uplift u *wi ...Formula (7);

[0148] Among them, uplift_adjust u For the fifth probability, uplift u is the third probability, w i is the influence coefficient.

[0149] Thus, the process of selecting a resource that matches the user object from multiple resources based on the third probability and the fourth probability corresponding to each resource may be selecting a resource that matches the user object from multiple resources based on the fifth probability and the fourth probability corresponding to each resource.

[0150] It should be noted that some user objects have a high probability of executing a resource transfer event before and after resource allocation, while some user objects have a low probability of executing a resource transfer event before and after resource allocation. This is equivalent to the user objects having a sensitivity to resources. If the user object has a high sensitivity to resources, then whether or not to allocate resources or what type of resources to allocate will have a significant impact on the probability of the user object executing a resource transfer event. If the user object has a low sensitivity to resources, then whether or not to allocate resources or what type of resources to allocate will have a minimal impact on the probability of the user object executing a resource transfer event.

[0151] Based on this, for user objects with low sensitivity to resources, even if the first probability of the user object is high, it has little correlation with the corresponding resource, that is, the corresponding resource has a lower impact on the user object. For user objects with high sensitivity to resources, the corresponding resource has a greater impact on the user object; therefore, for each resource, the impact coefficient of the resource on the user object can also be determined, and the third probability can be adjusted based on the impact coefficient.

[0152] In this way, based on the impact of resources on the user object's execution of the resource transfer event, the probability of the user object executing the resource transfer event is adjusted again. This can further improve the accuracy of the determined probability of the user object executing the resource transfer event, thereby helping to select resources that match the user object based on the corresponding probability in the subsequent process.

[0153] In actual implementation, the process of determining the influence coefficient of resources on user objects can be, when the user object is not allocated multiple resources, predicting the sixth probability of the user object executing a resource transfer event for each resource, and determining the third probability of the user object for the corresponding resource; based on the third probability and the sixth probability, determining the influence coefficient of each resource on the user object.

[0154] It should be noted that, when the user object is not allocated multiple resources, the sixth probability of predicting the user object to execute the resource transfer event for each resource refers to the probability of predicting the user to execute the resource transfer event if no resources are allocated. As for the process of determining the third probability of the user object for the corresponding resource, as mentioned above, a corresponding third probability has been determined for each of the multiple resources. Therefore, there is a corresponding third probability for each resource; thus, by comparing the difference between the sixth probability and the third probability, we can know the impact of the corresponding resource on the user object, that is, the impact coefficient of the corresponding resource on the user object.

[0155] In this way, the impact of resources on user objects is determined based on the probability of a user object executing a resource transfer event when no resources are allocated and the probability of a user executing a resource transfer event when resources are allocated. This allows for an accurate assessment of the causal effect of resources and an optimization of resource allocation.

[0156] In some embodiments, the process of determining the influence coefficient of each resource on the user object based on the third probability and the sixth probability may be to subtract the third probability from the sixth probability to obtain a probability difference; and determine the target probability interval to which the probability difference belongs from multiple probability intervals, wherein different probability intervals correspond to different influence coefficients, thereby determining the influence coefficient corresponding to the target probability interval as the influence coefficient of the corresponding resource on the user object.

[0157] In other embodiments, the probability of a user object executing a resource transfer event corresponds to N probability intervals, different probability intervals correspond to different probability types, the N probability types constitute N*N first probability type pairs, different first probability type pairs correspond to different impact coefficients, and N is an integer greater than 1; thus, the process of determining the impact coefficient of each resource on the user object based on the third probability and the sixth probability can be, from the N probability intervals, determining the first probability interval to which the third probability belongs and the second probability interval to which the sixth probability belongs; determining the first probability type corresponding to the user object based on the first probability interval, and determining the second probability type corresponding to the user object based on the second probability interval; constructing a second probability type pair based on the first probability type and the second probability type; selecting a first probability type pair consistent with the second probability type pair from the N*N first probability type pairs, and determining the impact coefficient corresponding to the selected first probability type pair as the impact coefficient of the corresponding resource on the user object; wherein the corresponding resource here is the resource corresponding to the third probability.

[0158] It should be noted that, assuming N is 5, the 5 probability intervals can be [0%, 20%), [20%, 40%), [40%, 60%), [60%, 80%), [80%, 100%), and the probability types corresponding to these five probability intervals are ultra-low, low, normal, high, and ultra-high respectively; then based on the 5 probability types, 25 first probability type pairs are constructed, namely ultra-low and ultra-low, ultra-low and low, ultra-low and normal, ultra-low and high, ultra-low and ultra-high, low and ultra-low, low and low, low and normal, low and high, low and ultra-high, normal and ultra-low, normal and low, normal and normal, normal and high, normal and ultra-high, high and ultra-low, high and low, high and normal, high and high, high and ultra-high, ultra-high and ultra-low, ultra-high and low, ultra-high and normal, ultra-high and high, ultra-high and ultra-high, and then different influence coefficients are pre-set for different first probability type pairs;

[0159] Next, based on the first probability type corresponding to the first probability interval to which the third probability belongs and the second probability type corresponding to the second probability interval to which the sixth probability belongs, a second probability type pair is constructed. For example, if the first probability type corresponding to the first probability interval to which the third probability belongs is ultra-low and the second probability type corresponding to the second probability interval to which the sixth probability belongs is ultra-high, then the second probability type pair is ultra-low and ultra-high.

[0160] Finally, a first probability type pair that is consistent with the second probability type pair is selected from the 25 first probability type pairs, and the influence coefficient corresponding to the selected first probability type pair is determined as the influence coefficient of the corresponding resource on the user object.

[0161] In this way, according to the above method, the probability of the user object before and after resource allocation is used to determine which situation the user object hits, thereby determining the influence coefficient of the resource on the user object. This not only improves the accuracy of the determined influence coefficient, but also improves the interpretability of the influence coefficient.

[0162] Step 104 : Select a resource that matches the user object from a plurality of resources based on the third probability and the fourth probability corresponding to each resource.

[0163] In actual implementation, before selecting a resource that matches the user object from multiple resources based on the third probability and the fourth probability corresponding to each resource, when the user object has not been assigned multiple resources, predicting the sixth probability of the user object executing a resource transfer event, thereby selecting a resource that matches the user object from multiple resources based on the third probability and the fourth probability corresponding to each resource, can be a process of selecting a resource that matches the user object from multiple resources based on the third probability and the fourth probability corresponding to each resource, with at least two preset constraints as constraints, and based on the third probability and the fourth probability corresponding to each resource, selecting a resource from multiple resources whose probability difference between the corresponding third probability and the sixth probability is greater than or equal to a difference threshold, as the resource that matches the user object.

[0164] It should be noted that the resources for which the probability difference between the corresponding third probability and the sixth probability is greater than or equal to the difference threshold, that is, the probability difference between the corresponding third probability and the sixth probability is greater than or equal to the difference threshold, refer to maximizing the probability difference between the corresponding third probability and the sixth probability, that is, the difference threshold is the maximum value of the probability difference between the corresponding third probability and the sixth probability;

[0165] As for the at least two constraints preset here, illustratively, four constraints can be included, namely: Constraint 1: ROI constraint, that is, in the process of resource allocation, the promoter sets the minimum requirement for the return on investment (ROI, Return on Investment). Here, the purpose of the ROI constraint is to ensure that the input-output ratio of resource investment reaches a certain standard, thereby controlling the promotion cost and optimizing the promotion effect; Constraint 2: The consumption cost of resources does not exceed the target budget; Constraint 3: Each user object can only send one resource at most; Constraint 4: The allocation ratio of various resources must be within the corresponding range, that is, for various resources, the allocation ratio of the resource is between the pre-set maximum and minimum proportions.

[0166] In actual implementation, based on the third probability and the fourth probability corresponding to each resource, a resource whose probability difference between the corresponding third probability and the sixth probability is greater than or equal to the difference threshold is selected from multiple resources as the resource matching the user object, that is:

[0167]

[0168] Δp u,i =P(Y|t=i)-P(Y|t=0)...Formula (9);

[0169] Where Δp u,i It refers to the probability difference between the third probability and the sixth probability, U is the total number of user objects, I is the number of resource types, x u,i Refers to whether to send resource i to user u, where: P(Y|t=i) is the third probability, P(Y|t=0) is the sixth probability, and t refers to the type of resource, where the type here includes the type and denomination mentioned above, that is, different types of resources include resources of different types, and also include resources of different denominations under the same type.

[0170] It should be noted that, based on formula (8) and formula (9), the above four constraints can also be expressed by the formula:

[0171]

[0172] Among them, for formula (10)-formula (13), q u,i It refers to the probability that user object consumes resource i when resource i is allocated to user object u. For example, the fourth probability, A u,i It refers to the amount of resource transfer of user object u when resource i is allocated to user object u, m i refers to the type of resource i, such as denomination, B refers to the target budget, p u,i It refers to the probability of a user object executing a resource transfer event when resource i is allocated to user object u, such as the third probability and the fifth probability; Lr i It refers to the minimum proportion of resource i allocated, Ur i Refers to the highest percentage of coupon i allocated.

[0173] It should be noted that formula (10) corresponds to constraint condition 1, formula (11) corresponds to constraint condition 2, formula (12) corresponds to constraint condition 3, and formula (13) corresponds to constraint condition 4.

[0174] In this way, by presetting at least two constraints, and then, when the constraints are met, selecting the resource that maximizes the difference in the user's conversion probability before and after resource allocation as the resource that matches the user object, it can be ensured that the selected resource is the resource that best matches the user object.

[0175] It should be noted that A in the above process u,i That is, when resource i is allocated to user object u, the resource transfer quantity of user object u is a historical statistical value. Specifically, the process of determining this value involves obtaining multiple user objects and bucketing them in descending order according to the total resource transfer amount of each user object in the past target period to obtain multiple buckets. Each bucket contains user objects that are not allocated resources and user objects that are allocated resources. For each bucket, the user objects that are allocated resources in the bucket are divided into multiple first user object groups according to the type of resources, and the user objects that are not allocated resources are grouped as a second user object group. For each user object group, the average resource transfer quantity per user object in the user object group is determined. For each bucket, the average resource transfer quantity per user object corresponding to each first user object group is subtracted from the average resource transfer quantity per user object group in the second user object group to obtain the resource transfer quantity corresponding to different types of resources in the bucket.

[0176] Based on this, when allocating resource i to user object u, the resource transfer quantity of user object u is first determined by determining the bucket to which user object u belongs, and then determining the resource transfer quantity corresponding to resource i in the bucket, that is, A u,i .

[0177] For example, see Figure 10 , Figure 10 This is a schematic diagram of the process of determining the number of resource transfers provided by the embodiment of the present application, based on Figure 10 Taking coupons as an example, there are M types of coupons. Therefore, after bucketing multiple user objects into N buckets, each bucket contains M coupon groups (i.e., the first user object group) and 1 blank group (i.e., the second user object group). Therefore, the per capita consumption amount corresponding to each coupon group is subtracted from the per capita consumption amount of the blank group to obtain the consumption amount corresponding to different types of coupons in the bucket.

[0178] It should be noted that after selecting a resource that matches the user object from multiple resources based on the third probability and the fourth probability corresponding to each resource, the selected resource can be allocated to the user object so that the user object can execute a resource transfer event under the incentive of the selected resource.

[0179] By applying the above-mentioned embodiments of the present application, for each of a plurality of resources, a first probability of a user object executing a resource transfer event and a second probability of the user object consuming resources when executing the resource transfer event are predicted. Then, based on the first behavior data of the user object executing the resource transfer event, the first probability is corrected to obtain a third probability, and based on the first behavior data, the second probability is corrected to obtain a fourth probability. Thus, based on the third probability and the fourth probability corresponding to each resource, a resource matching the user object is selected from a plurality of resources. In this way, compared to the solution of selecting the resource with the highest probability of executing the resource transfer event as the resource that best matches the user object, the present application combines the probability of the user object executing the resource transfer event for the resource and the probability of the user object consuming the corresponding resource to select the resource that best matches the user object. This can ensure the correlation between the behavior of the user object executing the resource transfer event and the resource, which can not only improve the utilization rate of the resources, but also ensure that the selected resource has the best incentive effect on the user object to execute the resource transfer event, that is, the selected resource is the resource that best matches the user object; in addition, based on the first behavior data, the first probability and the second probability are corrected respectively, so that based on the corrected two probabilities, the resource that matches the user object is selected, which improves the accuracy of the selected resource, that is, further improves the degree of matching between the selected resource and the user object.

[0180] The following describes an exemplary application of the embodiments of the present application in a practical application scenario.

[0181] During the research process, the inventors found that in the coupon promotion scenario, a model is generally used to identify users with the highest probability of changing their decision results due to the coupon activity, helping business personnel to attract customers more accurately. Specifically, for coupons of different amounts, the user's conversion probability for the promotion activity is predicted separately, so as to select the coupon with the highest probability as the coupon to be allocated to the user; however, some users are less sensitive to coupons. For example, after being allocated a certain coupon, the user may not use the coupon for various reasons, but the corresponding user also performs conversion behavior. In this way, the utilization rate of the coupon is low, and the correlation between the coupon and the user's conversion behavior is also low, that is, the matching degree between the corresponding coupon and the user is low; or, the user will consume the corresponding coupon and perform conversion behavior under coupons of different denominations, but the denomination of the selected coupon is predicted to be the largest. At this time, the incentive effect of the selected coupon on the user's conversion behavior is relatively general, that is, the matching degree between the selected coupon and the user is also low. Based on this, the coupon selected by the coupon allocation method in the related art has a low matching degree with the user.

[0182] Based on this, an embodiment of the present application provides a resource processing method, which combines the probability of a user object executing a resource transfer event for a resource and the probability of a user object consuming the corresponding resource to select the resource that best matches the user object. This can ensure the correlation between the behavior of the user object executing the resource transfer event and the resource, which can not only improve the resource utilization, but also ensure that the selected resource has the best incentive effect on the user object executing the resource transfer event, that is, the selected resource is the resource that best matches the user object; in addition, based on the first behavior data, the first probability and the second probability are corrected respectively, so that based on the corrected two probabilities, the resource that matches the user object is selected, thereby improving the accuracy of the selected resource, that is, further improving the degree of matching between the selected resource and the user object.

[0183] In actual implementation, see Figure 11 , Figure 11 This is a technical architecture diagram of the resource processing method provided in the embodiment of the present application, based on Figure 11 The technical solution provided by the embodiment of this application includes four processes: data preparation and processing, multi-objective model offline training, model reasoning and coupon optimal matching.

[0184] Among them, data preparation and processing include feature data processing and grouped withdrawal amount uplift historical statistical value processing. Among them, feature data mainly includes: Bias feature, also known as deviation feature (third feature). The bias feature here refers to the situation where coupons are not used. This type of bias feature is added in the modeling stage to play a certain debiasing role in the model from the feature level; Treatment feature, also known as intervention feature (second feature), refers to the attribute characteristics of the coupon (resource) itself, such as coupon type, denomination, historical consumption rate, etc., which can characterize the characteristics of the coupon, as well as some characteristics that may affect the decision to issue coupons, such as the model classification of pre-service; User characteristics (first feature) including interaction features, user attribute features, consumption willingness features and activity features. Among them, interaction features mainly include historical interaction data between users and coupons, such as usage statistics of various coupons; User attribute features mainly include basic attribute features of users, such as age, gender, permanent residence, occupation information, etc.; Consumption willingness features mainly include historical external and internal transaction history statistics of users, such as the total amount of internal consumption; Activity features mainly include active behavior statistics of users such as logins on the site over a period of time.

[0185] For the group consumption amount uplift (A u,i ), specifically, Figure 10 As shown in the figure, by using propensity scores to group homogeneous users, historical statistical values ​​are used instead of model predictions to reduce the loss caused by the regression model: the external model is called to obtain the propensity scores (the user's total past consumption amount) of users in the historical statistical window and bucket them in descending order. Each group contains blank users (users without coupons) and users of each coupon strategy group. The per capita consumption amount of coupon group m in the nth bucket is subtracted from the per capita consumption amount of the blank group in the bucket to obtain the consumption amount gain of coupon m in bucket n.

[0186] In actual implementation, for the offline training process of the model, such as Figure 12 As shown, Figure 12 This is a schematic diagram of the structure of the multi-expert network prediction model provided in the embodiment of the present application, based on Figure 12 ,The multi-expert network prediction model includes input layer, representation layer, ,feature extraction layer, feature splicing layer and multi-target prediction layer.

[0187] It should be noted that the multi-expert network model here has the same model structure as the first probability prediction model and the second probability prediction model in the previous article. The only difference is that when the multi-expert network model is trained with samples labeled with the real conversion probability, the first probability prediction model is obtained, and when the multi-expert network model is trained with samples labeled with the real consumption probability, the second probability prediction model is obtained.

[0188] Among them, the input layer receives bias features, treatment features, user features, interaction features, activity features and consumption willingness features, etc.; the representation layer converts discrete type features into dense embedding features; the feature extraction layer contains four expert networks, among which expert network 1 is a user representation expert, which receives various attribute features of users to make a comprehensive representation of user portraits; expert network 2 is a joint representation expert, which mainly expresses the user's preference for coupons based on various feature data of the user's historical interaction with coupons; expert network 3 is a user activity representation expert, which is mainly responsible for representing the user's historical activity information; expert network 4 is a willingness representation expert, which mainly uses the statistical characteristics of external and internal consumption scenarios to represent the user's preferences for coupons. The user's consumption willingness is represented; finally, the output of the expert network (the first feature) is spliced ​​with the deviation feature (the third feature) and intervention feature (the second feature) from the input layer, and the feature (the fourth feature) output by the backdoor adjustment debiasing module (feature debiasing layer), and sent together to the final target prediction layer. The output is shown in the above formula (5); wherein, the loss function is shown in the above formula (6). Using the loss function, the multi-expert network prediction model is trained twice to obtain a model (the first probability prediction model) for predicting the conversion probability (the first probability) of whether the user will perform consumption behavior (resource transfer behavior), and a model (the second probability prediction model) for predicting the consumption probability (the second probability) of whether the user will consume the coupon when performing consumption behavior.

[0189] It should be noted that the backdoor adjustment debiasing module is mainly used to remove the counterintuitive phenomenon that the consumption rate of low-denomination coupons is greater than that of high-denomination coupons due to the selection bias caused by the manual selection strategy during the coupon issuance process (common sense phenomenon: assuming the coupon types are the same, the larger the denomination, the higher the consumption rate). Since the frequency of issuing coupons to the same user in business scenarios is low, the user history and coupon interaction data are relatively sparse. Therefore, Group-Level user features are used in this module, and users are grouped by the Kmeans method. The embedding of grouped users represents the user embedding for overall training. Specifically, the input includes: historical coupon grouping probability distribution Coupon grouping feature i=[i1,…,ij ,…,i N ] (first grouping feature) and user grouping feature u=[u1,…,u i ,…,u K ](Second grouping feature). First, combine the coupon grouping feature with Multiply them together to get the weighted coupon grouping vector e coupon (the third grouping feature), and then the vector e of the user group user (Second group feature) splicing to obtain the input vector e of the FM module FM =[e coupon ,e user ] (the second concatenated feature) is passed through the FM module to obtain the confounding adjustment variable (the fourth feature). This variable (the fourth feature) is fed into the main network along with the user representation (the first feature), the coupon representation (the second feature), and the bias feature (the third feature) to achieve debiasing. After the above calculations, the backdoor path can be cut off, thereby eliminating selection bias and improving model performance.

[0190] In actual implementation, for the model inference process, that is, using the two debiased models (first probability prediction model and second probability prediction model) produced by the model training process to perform model inference on the data, the inference process is as follows: 1) Obtain inference samples and various types of feature data except intervention features introduced in the data preparation and processing process; 2) Assign different coupons to each sample, and extract the intervention features of each coupon; 3) Call the debiased model to predict the model score of the same user under each coupon, including conversion probability (first probability) and consumption probability (second probability); 4) Post-process the model score, that is, for conversion probability and consumption probability, the usual practice is to directly apply the model's predicted value for subsequent optimization allocation, but this method is too dependent on the accuracy of the model output, and if the model or strategy changes slightly, the distribution of the model output will change, and the subsequent results will have unpredictable changes. Therefore, in order to solve this problem, a smooth mapping method between the model prediction value and the true statistical value is used here, such as Figure 9As shown, for the user to be predicted, obtain the statistical data of a historical period, including the real statistical value and the model prediction score, and divide the buckets in ascending order according to the model score, and obtain the real indicator statistical mean (real first probability) and the model score mean (reference first probability) in each bucket. The two adjacent buckets can constitute the two endpoints of the linear function, and the linear function is constructed. Similarly, N buckets can construct N-1 linear functions, and these N-1 linear functions constitute the piecewise linear smoothing function of the model score. For the new user to be predicted, input the model score into the piecewise linear function to obtain the corrected real indicator value; 5) For the corrected conversion probability, Specifically, after the conversion probability is corrected once under different coupons, the downstream will usually directly use the corrected conversion probability to identify sensitive groups. However, the conversion probability of some user objects is very high before and after the coupons are distributed, while the conversion probability of some user objects is very low before and after the coupons are distributed. It is meaningless to intervene in the promotion of such users. Therefore, the conversion probability needs to be corrected again, that is, the conversion rate before and after the coupons are distributed is divided into five cases, namely, super high, high, normal, low, and super low. After user intervention, there will be 25 corresponding migration cases, and an uplift scaling factor w is set for each case. i , then uplift_adjust u =uplift u *w i .

[0191] In actual implementation, for the optimal coupon allocation process, specifically, coupon promotion activities generally involve maximizing objectives, which can be ROI, conversion rate, etc. The embodiment of the present application uses maximizing conversion probability as the optimization objective function. Based on the predicted values ​​of conversion probability and coupon consumption probability generated by the model reasoning process, and some data constraints (constraints 1, 2, 3, 4) in the promotion scenario as indicated by the above formulas (10)-(13), an allocation algorithm such as the above formulas (8) and (9) is constructed, so that when the constraints are met, the coupon that maximizes the difference in the user's conversion probability before and after the intervention is selected and allocated to the corresponding user.

[0192] By applying the above-mentioned embodiments of the present application, for each of a plurality of resources, a first probability of a user object executing a resource transfer event and a second probability of the user object consuming resources when executing the resource transfer event are predicted. Then, based on the first behavior data of the user object executing the resource transfer event, the first probability is corrected to obtain a third probability, and based on the first behavior data, the second probability is corrected to obtain a fourth probability. Thus, based on the third probability and the fourth probability corresponding to each resource, a resource matching the user object is selected from a plurality of resources. In this way, compared to the solution of selecting the resource with the highest probability of executing the resource transfer event as the resource that best matches the user object, the present application combines the probability of the user object executing the resource transfer event for the resource and the probability of the user object consuming the corresponding resource to select the resource that best matches the user object. This can ensure the correlation between the behavior of the user object executing the resource transfer event and the resource, which can not only improve the utilization rate of the resources, but also ensure that the selected resource has the best incentive effect on the user object to execute the resource transfer event, that is, the selected resource is the resource that best matches the user object; in addition, based on the first behavior data, the first probability and the second probability are corrected respectively, so that based on the corrected two probabilities, the resource that matches the user object is selected, which improves the accuracy of the selected resource, that is, further improves the degree of matching between the selected resource and the user object.

[0193] The following continues to describe the exemplary structure of the resource processing device 455 provided in the embodiment of the present application as a software module. In some embodiments, such as Figure 2 As shown, the software modules stored in the resource processing device 455 of the memory 450 may include:

[0194] Prediction module 4551, configured to predict, for each of a plurality of resources, a first probability that a user object executes a resource transfer event and a second probability that the user object consumes the resource by executing the resource transfer event;

[0195] A determination module 4552 is configured to determine first behavior data of the user object executing the resource transfer event;

[0196] a correction module 4553 configured to correct the first probability based on the first behavior data to obtain a third probability, and correct the second probability based on the first behavior data to obtain a fourth probability;

[0197] The selection module 4554 is configured to select a resource matching the user object from the multiple resources based on the third probability and the fourth probability corresponding to each resource.

[0198] In some embodiments, the correction module 4553 is further used to construct a mapping relationship for correcting the first probability based on the first behavior data; and based on the mapping relationship, correct the first probability to obtain the third probability.

[0199] In some embodiments, the correction module 4553 is also used to predict a reference first probability of the user object executing the resource transfer event based on the first behavior data; based on the reference first probability, the user object is bucketed to obtain multiple buckets, each bucket corresponding to a reference first probability; for each bucket, the following processing is performed to obtain the true first probability of the user object in the bucket executing the resource transfer event: determine the first number of user objects included in the bucket, and based on the first behavior data, determine the second number of user objects executing the resource transfer event in the bucket; determine the ratio of the second number to the first number as the true first probability corresponding to the bucket; based on the reference first probability and the true first probability corresponding to each bucket, construct a linear relationship between the predicted first probability and the true first probability, and determine the linear relationship as the mapping relationship.

[0200] In some embodiments, the device also includes a second correction module, which is used to determine, for each of the resources, an impact coefficient of the resource on the user object; wherein the impact coefficient is used to indicate the impact of the resource on the user object executing a resource transfer event; based on each of the impact coefficients, the third probability is adjusted to obtain a fifth probability; the selection module 4554 is also used to select a resource that matches the user object from the multiple resources based on the fifth probability and the fourth probability corresponding to each of the resources.

[0201] In some embodiments, the second correction module is further used to predict the sixth probability of the user object executing the resource transfer event for each of the resources when the user object is not allocated the multiple resources, and determine the third probability of the user object for the corresponding resource; based on the third probability and the sixth probability, determine the influence coefficient of each of the resources on the user object.

[0202] In some embodiments, the probability of the user object executing a resource transfer event corresponds to N probability intervals, different probability intervals correspond to different probability types, and the N probability types constitute N*N first probability type pairs, different first probability type pairs correspond to different impact coefficients, and N is an integer greater than 1; the second correction module is also used to determine, from the N probability intervals, the first probability interval to which the third probability belongs, and the second probability interval to which the sixth probability belongs; based on the first probability interval, determine the first probability type corresponding to the user object, and based on the second probability interval, determine the second probability type corresponding to the user object; construct a second probability type pair based on the first probability type and the second probability type; from the N*N first probability type pairs, select the first probability type pair that is consistent with the second probability type pair, and determine the impact coefficient corresponding to the selected first probability type pair as the impact coefficient of the corresponding resource on the user object.

[0203] In some embodiments, the prediction module 4551 is also used to predict the sixth probability of the user object executing the resource transfer event when the user object is not allocated the multiple resources; the selection module 4554 is also used to, based on the third probability and the fourth probability corresponding to each of the resources, select from the multiple resources a resource whose probability difference between the corresponding third probability and the sixth probability is greater than or equal to a difference threshold, based on at least two preset constraints, as a resource that matches the user object.

[0204] In some embodiments, the prediction module 4551 is also used to extract features from the user object to obtain a first feature, and to extract features based on the resources to obtain a second feature; to extract features based on the user object's unconsumed behavior of resources to obtain a third feature; to determine first allocation data for the multiple resources, and based on the first allocation data, to determine a fourth feature for eliminating causal bias; to perform feature splicing on the first feature, the second feature, the third feature, and the fourth feature to obtain a first spliced ​​feature; and to predict the probability of the user object executing the resource transfer event based on the first spliced ​​feature to obtain the first probability.

[0205] In some embodiments, there are multiple user objects, and the prediction module 4551 is further used to determine the first probability distribution of the multiple resources of different types based on the first allocation data; group the resources based on the type of the resources to obtain multiple groups of resources, and extract features based on each group of resources in the multiple groups of resources to obtain a first grouping feature of the resources; cluster the user objects to obtain multiple clusters of user objects, and extract features based on each cluster of user objects in the multiple clusters of user objects to obtain a second grouping feature of the user objects; and fuse the first probability distribution, the first grouping feature, and the second grouping feature to obtain the fourth feature.

[0206] In some embodiments, the prediction module 4551 is further used to multiply the first probability distribution by the first grouping feature to obtain a third grouping feature; concatenate the third grouping feature with the second grouping feature to obtain a second concatenated feature; and perform feature cross-processing based on the second concatenated feature to obtain the fourth feature.

[0207] In some embodiments, the prediction module 4551 is further used to perform feature decomposition on the second splicing feature to obtain multiple sub-features, and combine the multiple sub-features in pairs to obtain a feature combination; for each of the feature combinations, based on the correlation relationship between the two sub-features in the feature combination, determine the associated feature corresponding to the feature combination; and fuse the sub-features and the associated features to obtain the fourth feature.

[0208] In some embodiments, the prediction module 4551 is also used to determine multiple feature types corresponding to the user object, and perform feature extraction on the user object for each of the multiple feature types to obtain a first intermediate feature corresponding to the feature type; based on the resources, determine the weight corresponding to each of the feature types, and multiply each weight by the corresponding first intermediate feature to obtain a second intermediate feature; and fuse multiple of the second intermediate features to obtain the first feature.

[0209] The embodiment of the present application provides a computer program product, which includes computer-executable instructions or a computer program, which is stored in a computer-readable storage medium. A processor of an electronic device reads the computer-executable instructions or the computer program from the computer-readable storage medium, and the processor executes the computer-executable instructions or the computer program, so that the electronic device executes the resource processing method described in the embodiment of the present application, for example, Figure 3 Shows the resource processing method.

[0210] The embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions or computer programs. When the computer-executable instructions or computer programs are executed by a processor, the processor will execute the resource processing method provided by the embodiment of the present application, for example, Figure 3 Shows the resource processing method.

[0211] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface storage, optical disk, or CD-ROM; or various devices including one or any combination of the above memories.

[0212] In some embodiments, computer-executable instructions may be in the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0213] As an example, computer-executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file that stores other programs or data, such as, for example, in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinating files (e.g., files storing one or more modules, subroutines, or code portions).

[0214] By way of example, computer-executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one site, or on multiple electronic devices distributed across multiple sites and interconnected by a communication network.

[0215] It should be noted that in the embodiments of the present application, when it comes to obtaining the first behavior data of the user object and other related data, when the embodiments of the present application are applied to specific products or technologies, it is necessary to obtain user permission or consent, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0216] The above description is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. Any modifications, equivalent replacements, and improvements made within the spirit and scope of the present application are included in the scope of protection of the present application.

Claims

1. A method for processing resources, characterized in that: The method comprises: For each resource among the plurality of resources, predicting a first probability that a user object executes a resource transfer event and a second probability that the user object consumes the resource by executing the resource transfer event; Determine first behavior data of the user object executing the resource transfer event; Correcting the first probability based on the first behavior data to obtain a third probability, and correcting the second probability based on the first behavior data to obtain a fourth probability; Based on the third probability and the fourth probability corresponding to each resource, a resource matching the user object is selected from the multiple resources.

2. The method according to claim 1, characterized in that Correcting the first probability based on the first behavior data to obtain a third probability includes: constructing a mapping relationship for correcting the first probability based on the first behavior data; Based on the mapping relationship, the first probability is corrected to obtain the third probability.

3. The method according to claim 2, characterized in that The constructing, based on the first behavior data, a mapping relationship for correcting the first probability includes: Predicting, based on the first behavior data, a reference first probability that the user object executes a resource transfer event; Bucketing the user object based on the reference first probability to obtain a plurality of buckets, each bucket corresponding to one of the reference first probabilities; For each bucket, perform the following process to obtain a true first probability of the user object in the bucket executing the resource transfer event: Determining a first number of user objects included in the bucket, and determining a second number of user objects in the bucket that execute the resource transfer event based on the first behavior data; and determining a ratio of the second number to the first number as a true first probability corresponding to the bucket; Based on the reference first probability and the true first probability corresponding to each bucket, a linear relationship between the predicted first probability and the true first probability is constructed, and the linear relationship is determined as the mapping relationship.

4. The method according to claim 1, wherein After correcting the second probability based on the first behavior data to obtain a fourth probability, the method further includes: For each of the resources, determining an influence coefficient of the resource on the user object; wherein the influence coefficient is used to indicate the influence of the resource on the user object executing a resource transfer event; Adjusting the third probability based on each of the influence coefficients to obtain a fifth probability; The selecting, from the plurality of resources, a resource matching the user object based on the third probability and the fourth probability corresponding to each resource includes: Based on the fifth probability and the fourth probability corresponding to each resource, a resource matching the user object is selected from the multiple resources.

5. The method according to claim 4, characterized in that Determining the influence coefficient of the resource on the user object includes: In a case where the user object is not allocated the plurality of resources, predicting a sixth probability that the user object performs the resource transfer event for each of the resources, and determining a third probability of the user object for the corresponding resource; An influence coefficient of each resource on the user object is determined based on the third probability and the sixth probability.

6. The method according to claim 1, wherein Before selecting a resource matching the user object from the plurality of resources based on the third probability and the fourth probability corresponding to each resource, the method further includes: In a case where the user object is not allocated the plurality of resources, predicting a sixth probability that the user object executes the resource transfer event; The selecting, from the plurality of resources, a resource matching the user object based on the third probability and the fourth probability corresponding to each resource includes: With at least two preset constraints as constraints, based on the third probability and the fourth probability corresponding to each of the resources, from the multiple resources, select the resource whose probability difference between the corresponding third probability and the sixth probability is greater than or equal to the difference threshold as the resource that matches the user object.

7. The method according to claim 1, characterized in that The predicting a first probability of the user object executing a resource transfer event includes: Performing feature extraction on the user object to obtain a first feature, and performing feature extraction based on the resource to obtain a second feature; Extracting features based on the user object's unconsumed behavior of resources to obtain a third feature; determining first allocation data of the plurality of resources, and determining a fourth feature for eliminating causal bias based on the first allocation data; Performing feature splicing on the first feature, the second feature, the third feature, and the fourth feature to obtain a first spliced ​​feature; Based on the first splicing feature, the probability of the user object executing the resource transfer event is predicted to obtain the first probability.

8. The method according to claim 7, characterized in that The determining, based on the first distribution data, a fourth feature for eliminating causal bias includes: determining a first probability distribution of different types of the plurality of resources based on the first allocation data; Grouping the resources based on the types of the resources to obtain multiple groups of resources, and extracting features based on each group of resources in the multiple groups of resources to obtain a first grouping feature of the resources; Clustering the user objects to obtain multiple clusters of user objects, and extracting features based on each cluster of user objects in the multiple clusters of user objects to obtain second grouping features of the user objects; The first probability distribution, the first grouping feature, and the second grouping feature are fused to obtain the fourth feature.

9. An electronic device, characterized in that: include: Memory for storing computer-executable instructions or computer programs; A processor, configured to implement the resource processing method according to any one of claims 1 to 8 when executing the computer executable instructions or computer program stored in the memory.

10. A computer program product comprising computer executable instructions or a computer program, characterized in that When the computer executable instructions or computer program are executed by a processor, the resource processing method according to any one of claims 1 to 8 is implemented.

Images