Resource allocation optimization method and device, equipment, storage medium and product

Abstract

The invention discloses a resource allocation optimization method and device, equipment, a storage medium and a product, relates to the technical field of computers, and discloses a resource allocation optimization method which comprises the following steps: acquiring target user characteristics of a user to be subjected to resource allocation; on the basis of the target user features, resource allocation is carried out through a preset optimal allocation model, ranking scores of all the users to be subjected to resource allocation are output, the ranking scores reflect contribution of all the users to be subjected to resource allocation to an optimization target, and the optimal allocation model is composed of a neural network containing a loss convex function; and generating a corresponding optimal allocation scheme based on the ranking score. According to the method, the optimal allocation model formed by the neural network containing the loss convex function is provided, and the loss convex function has the global minimum, so that the neural network can converge to the global minimum in the training process, unbiased estimation of the ranking score is realized, and the accuracy of resource allocation optimization is improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to resource allocation optimization methods, devices, equipment, storage media and products. Background Art

[0002] With the rapid development of data mining and neural network technology, intelligent decision-making has become increasingly important in many fields. Especially when resources are limited, how to maximize benefits through intelligent decision-making has become a key issue. In clinical medicine, operation and maintenance, advertising industry, recommendation and marketing fields, the rational allocation of resources is crucial to improving benefits and revenue.

[0003] In related technologies, the direct ranking model DRM (Direct Ranking Model) is usually used to predict the difference between revenue (Conditional Average Treatment Effect on Revenue) and cost (Conditional Average Treatment Effect on Cost), that is, the ranking score, and determine the optimal resource allocation plan based on the ranking score to achieve resource allocation optimization. However, the loss function in the DRM model cannot converge to a stable extreme point, so when the loss converges, the correct ranking score cannot be obtained, resulting in low accuracy in resource allocation optimization.

[0004] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of this application is to provide a resource allocation optimization method, aiming to solve the technical problem of low accuracy of resource allocation optimization.

[0006] To achieve the above objectives, the present application proposes a resource allocation optimization method, which includes:

[0007] Obtain target user characteristics of users to be allocated resources;

[0008] Based on the target user characteristics, resources are allocated through a preset optimization allocation model, and a ranking score of each user of resources to be allocated is output, wherein the ranking score reflects the contribution of each user of resources to be allocated to the optimization target, and the optimization allocation model is composed of a neural network including a loss convex function;

[0009] Based on the ranking scores, a corresponding optimal allocation solution is generated.

[0010] Optionally, the formula of the loss convex function is as follows:

[0011]

[0012] Among them, N1 represents the sample size of the experimental group, N0 represents the sample size of the control group, i represents the i-th user, and q i represents the ranking score of the i-th user, and tanh is The hyperbolic tangent transform, h i is the intermediate output of any neural network, λ is represented as a hyperparameter, y r Represents the income result value, y c Represents the cost result value.

[0013] Optionally, the step of performing resource allocation based on the target user characteristics through a preset optimization allocation model and outputting a ranking score of each user to be allocated resources comprises:

[0014] Based on the target user characteristics, resources are allocated through a preset optimization allocation model, and the first condition average processing impact value of income and the second condition average processing impact value of cost corresponding to each user of the resources to be allocated are predicted;

[0015] The difference between the first condition average processing impact value and the second condition average processing impact value is calculated to obtain a ranking score of each user to be allocated resources.

[0016] Optionally, the step of performing resource allocation based on the target user characteristics through a preset optimization allocation model to predict the first conditional average processing impact value of income and the second conditional average processing impact value of cost corresponding to each user of the resources to be allocated includes:

[0017] Based on the preset user behavior level division information and the target user characteristics, determining the first user behavior level characteristics of each user of the to-be-allocated resources regarding income and the second user behavior level characteristics regarding cost;

[0018] Based on the first user behavior level characteristics and the second user behavior level characteristics, resources are allocated through a preset optimization allocation model to predict the first conditional average processing impact value of income and the second conditional average processing impact value of cost for each user of the resources to be allocated.

[0019] Optionally, before the step of obtaining the allocation scheme to be optimized, the method includes:

[0020] Obtaining user feature samples and ranking score labels of the user feature samples;

[0021] Based on the user feature samples and the ranking score labels, the preset model to be trained is iteratively trained to obtain an optimized allocation model, wherein the optimized allocation model is a dual causal learning model, and the optimized allocation model is composed of a neural network including a loss convex function.

[0022] Optionally, the step of iteratively training a preset model to be trained based on the user feature sample and the ranking score label to obtain an optimized allocation model includes:

[0023] Based on the user feature samples, resources are allocated through a preset model to be trained, and a predicted ranking score of each user feature sample is output;

[0024] Based on a preset loss convex function, calculating an error result between the predicted ranking score and the ranking score label;

[0025] Based on the error result, determining whether the error result meets an error standard indicated by a preset error threshold range;

[0026] If the error result does not meet the error standard indicated by the preset error threshold range, then return to the step of allocating resources based on the user feature sample through the preset model to be trained, and output the predicted ranking score of each user feature sample, until the error result meets the error standard indicated by the preset error threshold range, and stop training to obtain the optimized allocation model.

[0027] In addition, to achieve the above purpose, the present application also proposes a resource allocation optimization device, the resource allocation optimization device comprising:

[0028] An acquisition module, used to acquire target user characteristics of users to be allocated resources;

[0029] An allocation module, configured to allocate resources based on the target user characteristics through a preset optimization allocation model, and output a ranking score of each user of resources to be allocated, wherein the ranking score reflects the contribution of each user of resources to be allocated to the optimization target, and the optimization allocation model is composed of a neural network including a loss convex function;

[0030] A generation module is used to generate a corresponding optimal allocation plan based on the ranking score.

[0031] In addition, to achieve the above-mentioned purpose, the present application also proposes a resource allocation optimization device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the resource allocation optimization method described above.

[0032] In addition, to achieve the above objectives, the present application also proposes a storage medium, which is a computer-readable storage medium, and stores a computer program on the storage medium. When the computer program is executed by a processor, the steps of the resource allocation optimization method described above are implemented.

[0033] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the resource allocation optimization method described above are implemented.

[0034] One or more technical solutions proposed in this application have at least the following technical effects:

[0035] Compared with the related art, the direct ranking model DRM is usually used to predict the difference between income CATE and cost CATE, that is, the ranking score, and the optimal resource allocation scheme is determined according to the ranking score, so as to achieve the optimization of resource allocation. However, the loss function in the DRM model cannot converge to a stable extreme point. Therefore, when the loss converges, the correct ranking score cannot be obtained, resulting in low accuracy of resource allocation optimization. In comparison, the present application proposes to obtain the target user characteristics of the resource users to be allocated; based on the target user characteristics, resource allocation is performed through a preset optimization allocation model, and the ranking score of each resource user to be allocated is output, wherein the ranking score reflects the contribution of each resource user to be allocated to the optimization target, and the optimization allocation model is composed of a neural network containing a loss convex function; based on the ranking score, the corresponding optimal allocation scheme is generated. It can be understood that the present application proposes an optimization allocation model composed of a neural network containing a loss convex function, which realizes the accurate prediction of the ranking score of each resource user to be allocated. Since the loss convex function has a global minimum, the neural network can converge to the global minimum during the training process, thereby realizing an unbiased estimation of the ranking score, thereby improving the accuracy of resource allocation optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0037] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0038] Figure 1 A schematic diagram of a flow chart provided for the first embodiment of the resource allocation optimization method of the present application;

[0039] Figure 2 A schematic diagram of user hierarchical definition of the resource allocation optimization method of this application;

[0040] Figure 3 A flowchart of the second embodiment of the resource allocation optimization method of the present application is provided;

[0041] Figure 4 A schematic diagram of the neural network structure of the resource allocation optimization method of this application;

[0042] Figure 5 This is a schematic diagram of the module structure of the resource allocation optimization device according to an embodiment of the present application;

[0043] Figure 6 Schematic diagram of the device structure of the hardware operating environment involved in the resource allocation optimization method in the embodiment of the present application.

[0044] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0045] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0046] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0047] The main solution of the embodiment of the present application is: obtaining the target user characteristics of the user to be allocated resources; based on the target user characteristics, performing resource allocation through a preset optimization allocation model, and outputting the ranking score of each user to be allocated resources, wherein the ranking score reflects the contribution of each user to be allocated resources to the optimization target, and the optimization allocation model is composed of a neural network including a loss convex function; based on the ranking score, generating a corresponding optimal allocation plan.

[0048] In this embodiment, the resource allocation optimization device is used as the execution body. For the convenience of description, it will be described as "device" hereinafter.

[0049] In the related art, the direct ranking model DRM is usually used to predict the difference between revenue CATE and cost CATE, that is, the ranking score, and the optimal resource allocation scheme is determined based on the ranking score to achieve resource allocation optimization. However, the loss function in the DRM model cannot converge to a stable extreme point, so when the loss converges, the correct ranking score cannot be obtained, resulting in low accuracy of resource allocation optimization.

[0050] The present application provides a solution to achieve unbiased estimation of ranking scores of users of resources to be allocated, thereby improving the accuracy of resource allocation optimization.

[0051] From the above embodiments, it can be seen that the present application proposes to obtain the target user characteristics of the resource users to be allocated; based on the target user characteristics, perform resource allocation through a preset optimization allocation model, and output the ranking score of each resource user to be allocated, wherein the ranking score reflects the contribution of each resource user to be allocated to the optimization target, and the optimization allocation model is composed of a neural network containing a loss convex function; based on the ranking score, generate the corresponding optimal allocation plan. It can be understood that the present application proposes an optimization allocation model composed of a neural network containing a loss convex function, which realizes the accurate prediction of the ranking score of each resource user to be allocated. Since the loss convex function has a global minimum, the neural network can converge to the global minimum during the training process, thereby realizing an unbiased estimate of the ranking score, thereby improving the accuracy of resource allocation optimization.

[0052] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a terminal system, etc. The following takes the resource allocation optimization device as an example to illustrate this embodiment and the following embodiments.

[0053] Based on this, the embodiment of the present application provides a resource allocation optimization method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the resource allocation optimization method of the present application.

[0054] In this embodiment, the resource allocation optimization method includes steps S100 to S300:

[0055] Step S100, obtaining target user characteristics of users to be allocated resources;

[0056] It should be noted that the target user characteristics refer to the specific characteristics of the user groups that are about to receive resources that need to be identified, that is, before the device allocates resources, it needs to identify and understand the specific characteristics of the user groups that are about to receive resources. These characteristics help to determine how to allocate resources most effectively to achieve specific goals or optimize specific results.

[0057] In a specific implementation, the target user characteristics include but are not limited to the user's age, gender, past purchase history, browsing history, income level, etc. The device may obtain the target user characteristics of the user to be allocated resources by extracting them from a database containing target user characteristics, or by receiving target user characteristics uploaded by an administrator or user, which is not specifically limited here.

[0058] Step S200, based on the target user characteristics, resources are allocated through a preset optimization allocation model, and a ranking score of each user to be allocated resources is output, wherein the ranking score reflects the contribution of each user to be allocated resources to the optimization target, and the optimization allocation model is composed of a neural network including a loss convex function;

[0059] In the specific implementation, the device uses a preset optimization model to decide how to allocate resources based on these target user characteristics. This model aims to maximize certain business indicators, such as user satisfaction, revenue, or efficiency. Using this optimization model, the device allocates resources. The model determines how much resources each user should get based on user characteristics and preset goals.

[0060] It should be noted that the ranking score output by the model is a numerical indicator for each user of the resources to be allocated. This score reflects the user's potential contribution to the optimization goal (such as maximizing benefits, minimizing costs, etc.).

[0061] Furthermore, the optimization allocation model is composed of a neural network with a convex loss function. This means that the loss function of the model is a convex function, which helps to ensure that the optimization problem during model training is easy to solve because any local minimum of the convex function is also a global minimum.

[0062] In the specific implementation, by training a neural network with a convex loss function, the model can learn how to allocate resources based on user characteristics to maximize the optimization goal. The ranking score is a result of this optimization, which indicates the contribution of each user to achieving the optimization goal.

[0063] Specifically, the above process describes how to use a neural network-based optimization model to allocate resources by considering user characteristics and preset business goals, and generate a ranking score for each user, which represents the user's contribution to the optimization goal. This method can ensure that resources are allocated to users who can best improve business goals.

[0064] In the specific implementation, this application proposes a new model, named Duality Causal Learner (DCL), which includes a neural network Its loss function is as follows:

[0065]

[0066] Among them, N1 represents the sample size of the experimental group, N0 represents the sample size of the control group, i represents the i-th user, and q i represents the ranking score of the i-th user, and tanh is The hyperbolic tangent transform, h i is the intermediate output of any neural network, λ is represented as a hyperparameter, y r Represents the income result value, y c Represents the cost result value.

[0067] Furthermore, the proof of the unbiased estimate of the loss convex function includes the following Theorem 1 and Theorem 2:

[0068] In the specific implementation, Theorem 1, consider a dataset of size N obtained from a randomized controlled trial (RCT), where the i-th data point is represented as The number of samples that received treatment is N1, and the number of samples that did not receive treatment is N0. In this case, the loss function L(s) is about s i Convex function of .

[0069] Specifically, first, define f(x i )=q i , simplifying the complex expression in the loss function L(s) into i Based on the form. Using the properties of RCT: Based on the properties of randomized controlled trials (RCT), the confirmation expression can be simplified to the form of expected value, thereby simplifying the loss function. Calculate the first-order derivative: Calculate the loss function L(s) for each parameter s i The first-order derivative shows that when the first-order derivative is zero, it reaches the extreme point. Calculate the second-order derivative: Calculate the loss function L(s) for each parameter s i The second-order derivative of shows that the second-order derivative is positive, thus proving that the loss function is about s i Convex function of .

[0070] Conclusion: Since the second-order derivative is positive, the loss function L(s) is a convex function, which means that it has a unique global minimum.

[0071] Furthermore, Theorem 2, given the loss function L(s), when the loss function converges to a stable point, the output of the tanh function is f(x i ) is an unbiased estimate of .

[0072] In the specific implementation, convexity ensures the existence of a minimum: Theorem 1 has proved that the loss function L(s) is a convex function, which means that there is a unique global minimum. The first-order derivative is equal to zero: When the loss function converges to the global minimum, the derivative of the loss function will be equal to zero. This shows that at the optimal point, the output value of tanh f(x i ) and the target value τ r (x i )-λτ c (x i ). Unbiased estimator: Since the output of tanh is equal to the target value, this is an unbiased estimator. This means that when the model training converges, the output value of the tanh function is the target causal effect f(x i ) accurately estimated.

[0073] In a specific implementation, the device allocates resources based on the target user characteristics through a preset optimization allocation model, and outputs the ranking score of each user to be allocated resources, including:

[0074] Based on the target user characteristics, resources are allocated through a preset optimization allocation model to predict the first conditional average processing impact value of income and the second conditional average processing impact value of cost for each user to be allocated resources; the difference between the first conditional average processing impact value and the second conditional average processing impact value is calculated to obtain the ranking score of each user to be allocated resources.

[0075] It should be noted that the first conditional average treatment effect on income is also known as income CATE (Conditional Average Treatment Effect on Revenue): Income CATE represents the average increase in expected income after a user receives treatment (e.g., seeing an ad) given feature X, which measures the causal effect of treatment on income results; Cost CATE (Conditional Average Treatment Effect on Cost): The second conditional average treatment effect on cost represents the average increase in expected cost after a user receives treatment (e.g., seeing an ad) given feature X, which measures the causal effect of treatment on cost results. Treatment indicates whether a specific ad is shown to the user, T = 1 indicates that the user has seen the ad, and T = 0 indicates that the user has not seen the ad; Outcome indicates whether the user has purchased the product promoted in the ad, and Y(1) represents the purchase result when the user has seen the ad, and Y(0) represents the purchase result when the user has not seen the ad.

[0076] In the specific implementation, the difference between income CATE (i.e., the average treatment impact value of the first condition) and cost CATE (i.e., the average treatment impact value of the second condition) is an important indicator used in the DCL model to determine the contribution of samples to the objective function. Specifically, this difference reflects the relative impact of each sample on income and cost after receiving a certain "treatment" (e.g., advertising exposure).

[0077] Further, in the DCL model, this difference is usually defined as the difference or ratio between revenue CATE and cost CATE. The size of this difference indicates whether a particular user (or sample) is worth investing resources in under a given budget constraint. If the difference is positive, it means that for the sample, the expected increase in revenue is greater than the increase in cost, so the sample is a favorable target and may be preferred; if the difference is negative or close to zero, it means that the increase in cost has not brought about a corresponding increase in revenue, and it may not be suitable for priority selection.

[0078] In the specific implementation, this difference is the basis used to calculate the ranking score. In the DCL model, the larger the difference, the more samples contribute to the goal (such as maximizing revenue or ROI), so they should be ranked higher in resource allocation.

[0079] In a specific implementation, the device allocates resources based on the target user characteristics through a preset optimization allocation model, and predicts the steps of obtaining the first conditional average processing impact value of income and the second conditional average processing impact value of cost corresponding to each user of the resources to be allocated, including:

[0080] Based on the preset user behavior level division information and the target user characteristics, the first user behavior level characteristics regarding income and the second user behavior level characteristics regarding cost of each user of resources to be allocated are determined; based on the first user behavior level characteristics and the second user behavior level characteristics, resources are allocated through a preset optimization allocation model, and the first conditional average processing impact value regarding income and the second conditional average processing impact value regarding cost of each user of resources to be allocated are predicted.

[0081] In a specific implementation, first, based on the preset user characteristic information, the device divides users into different levels according to their behavioral characteristics. These levels include divisions based on the user's historical behavior, purchasing habits, history of responding to marketing activities, etc. Secondly, the device combines the target user characteristics, which may include the user's demographic information, consumption ability, preferences, etc., to further refine the behavioral hierarchical characteristics of each user.

[0082] Further, the device determines the first user behavior level characteristics (regarding income), that is, the device determines the behavior level characteristics of each user of the resources to be allocated in terms of income. This may involve the potential of the user to contribute to income, such as their purchasing power, purchase frequency, etc. Secondly, the device also determines the second user behavior level characteristics (regarding cost), that is, the behavior level characteristics of each user of the resources to be allocated in terms of cost. This includes user acquisition cost, maintenance cost, service cost, etc.

[0083] In a specific implementation, the device uses a preset optimization allocation model to determine the allocation of resources based on the above-mentioned first and second user behavior level characteristics. This model aims to maximize revenue, minimize costs or achieve other business goals. The model predicts the first condition average processing impact value on revenue and the second condition average processing impact value on cost for each user of the resources to be allocated. The CATE here refers to the average impact of a certain process (such as providing resources or services) on user revenue and cost under specific conditions. Ultimately, through the prediction of the model, the potential impact of each user in terms of revenue and cost can be obtained, which helps to decide how to allocate resources to optimize business goals.

[0084] In the specific implementation, refer to Figure 2 In this application, all users related to income are divided into four levels, defined as "Lost Causes", "Persuadables", "Sleeping Dogs" and "Sure Things", which are user behaviors that participate but do not convert after resource delivery, user behaviors that will generate conversion if resources are delivered but will not generate conversion if resources are not delivered, user behaviors that will generate conversion if resources are not delivered but will not generate conversion if resources are delivered, and user behaviors that participate after resource delivery and will generate conversion; all users related to cost are divided into two levels, defined as "Never Takers" and "Coupon Takers", which are user behaviors that will not make purchases regardless of whether coupons are provided and user behaviors that will make purchases only when coupons are provided.

[0085] In a specific implementation, the present application considers all hierarchical combinations of income CATE and cost CATE when defining ranking scores, so that the DCL model can comprehensively consider the different reactions of all users when evaluating and allocating resources, avoiding the strict restrictions of the prior art on specific user stratification. Specifically, the present application considers all hierarchical combinations of income CATE and cost CATE when defining ranking scores, so that the DCL model can comprehensively consider the different reactions of all users when evaluating and allocating resources, process and identify all user stratification combinations, avoiding the strict restrictions of the prior art on specific user stratification, that is, its DCL method can process and identify all user stratification combinations, not just specific categories, thereby enhancing the applicability and robustness of the model.

[0086] It should be noted that in this application, the device adopts the potential outcome framework in causal inference. It is assumed that there is a fully controlled randomized controlled trial (RCT) sample, and the "Stable Unit Treatment Value Assumption" (SUTVA) is maintained in this proposal. Due to the random treatment allocation mechanism in RCT, the positivity hypothesis and the non-confounding hypothesis will also hold.

[0087] Furthermore, the treatment allocation mechanism, the positivity hypothesis, and the non-confounding hypothesis are explained below through the scenario of advertising marketing.

[0088] 1. Treatment Assignment Mechanism

[0089] Meaning: In advertising marketing, the treatment allocation mechanism refers to how ads are delivered to users. Treatment in this context can be understood as "whether the user sees the ad." In a randomized experiment (such as an A / B test), ads are randomly assigned to ensure that each user has the same chance of seeing or not seeing the ad.

[0090] For example: suppose an e-commerce platform wants to test the effectiveness of a new advertisement. The platform randomly selects a portion of users as the experimental group and shows them the advertisement; another portion of users as the control group does not show the advertisement. Since users are randomly assigned to the experimental group and the control group, the treatment allocation mechanism is random, which means that any user has an equal chance of seeing the advertisement or not seeing the advertisement. Doing so can effectively eliminate the bias caused by the user's own characteristics (such as purchase history, age, gender, etc.), thereby more accurately evaluating the impact of advertising on user behavior.

[0091] 2. Positivity Assumption

[0092] Meaning: In advertising marketing, the positivity assumption requires that every user has a non-zero chance of seeing an ad or not seeing an ad. In other words, the platform should not design an ad delivery strategy that makes it impossible or necessary for users with certain characteristics to see an ad forever.

[0093] For example: suppose that when this e-commerce platform is running ads, it only shows ads to users who have purchased products in the past month, and never shows ads to users who have not purchased products. This violates the positivity assumption because users who have not purchased products have no chance to see the ads. Since these users have never been assigned to the "see ads" treatment group, the platform cannot assess the potential impact of the ads on them. Therefore, in order to ensure that the positivity assumption holds, ads should be displayed to all types of users, not just users with certain specific characteristics.

[0094] 3. No Unmeasured Confounders Assumption

[0095] Meaning: In advertising marketing, the unconfounded hypothesis requires that when analyzing the effectiveness of advertising, all factors that may affect whether users see the advertisement and their purchasing behavior have been measured and controlled. In other words, there cannot be other unobserved factors that affect whether users see the advertisement and their purchasing behavior at the same time, otherwise it will affect the correct estimation of the advertising effect.

[0096] For example: suppose that when this e-commerce platform places advertisements, it decides whether to show the advertisements to users based on their purchase history, browsing history, age and other characteristics. If the analysis only considers purchase history and age, but ignores browsing history, which actually affects both whether the user sees the advertisement and the user's purchase decision (for example, users who browse more related products are more likely to see the advertisement and purchase the products), then this violates the assumption of no confounding. Unmeasured browsing history as a confounding factor will lead to biased estimates of advertising effectiveness. Therefore, in order to meet the assumption of no confounding, the platform should try to measure and control all factors that may affect advertising and purchasing behavior when analyzing advertising effectiveness, such as users' browsing history, social interactions, ad click history, etc.

[0097] In general, the treatment allocation mechanism refers to the mechanism that determines which users will see the advertisement in advertising marketing (for example, by eliminating bias through random assignment); the positivity assumption refers to the requirement that the advertising delivery strategy gives all users the opportunity to see the advertisement to ensure that the impact of the advertisement on all users can be evaluated; the non-confounding assumption refers to the requirement that when analyzing the effectiveness of advertising, all confounding factors that may simultaneously affect whether users see the advertisement and their purchasing behavior must be controlled to avoid bias in the estimation of causal effects.

[0098] Step S300: generating a corresponding optimal allocation plan based on the ranking scores.

[0099] In a specific implementation, the optimal allocation scheme refers to a resource allocation plan that determines which users should obtain resources and the amount of resources allocated based on the ranking scores. This plan aims to maximize overall benefits, such as improving resource utilization efficiency, increasing total revenue, or achieving other business goals. This process involves sorting users according to ranking scores and determining resource allocation based on the sorting results. Users with higher ranking scores will be given priority to allocate more resources because they contribute more to the optimization goal.

[0100] Furthermore, the optimal allocation plan is generated to achieve specific business goals. These goals may include improving user satisfaction, increasing sales, reducing costs, or increasing market share. This plan provides decision makers with a resource allocation suggestion based on data and models, helping them make more scientific and reasonable decisions. It is understandable that the optimal allocation plan is not static and can be dynamically adjusted according to changes in the business environment, market feedback, or further optimization of the model.

[0101] In the specific implementation, the present application adopts a single neural network structure to predict the ranking score instead of training the CATE model of income and cost separately, which not only reduces the computational complexity but also avoids the difficulty of capturing the complex interdependencies between variables.

[0102] Compared with the related art, the direct ranking model DRM is usually used to predict the difference between income CATE and cost CATE, that is, the ranking score, and the optimal resource allocation scheme is determined according to the ranking score, so as to achieve the optimization of resource allocation. However, the loss function in the DRM model cannot converge to a stable extreme point. Therefore, when the loss converges, the correct ranking score cannot be obtained, resulting in low accuracy of resource allocation optimization. In comparison, the present application proposes to obtain the target user characteristics of the resource users to be allocated; based on the target user characteristics, resource allocation is performed through a preset optimization allocation model, and the ranking score of each resource user to be allocated is output, wherein the ranking score reflects the contribution of each resource user to be allocated to the optimization target, and the optimization allocation model is composed of a neural network containing a loss convex function; based on the ranking score, the corresponding optimal allocation scheme is generated. It can be understood that the present application proposes an optimization allocation model composed of a neural network containing a loss convex function, which realizes the accurate prediction of the ranking score of each resource user to be allocated. Since the loss convex function has a global minimum, the neural network can converge to the global minimum during the training process, thereby realizing an unbiased estimation of the ranking score, thereby improving the accuracy of resource allocation optimization.

[0103] Based on the above first embodiment, the present application also proposes another embodiment, referring to Figure 3 , the resource allocation optimization method comprises:

[0104] In a specific implementation, before the step of the device acquiring the allocation scheme to be optimized, the method includes:

[0105] Step A100, obtaining user feature samples and ranking score labels of the user feature samples;

[0106] It should be noted that user feature samples refer to data collected from user groups, which contain various characteristics of users, such as demographic information, behavioral data, historical transaction records, etc. Ranking score labels refer to labels corresponding to user feature samples, which indicate the ranking or priority of users in a specific sorting task. These labels are determined based on the contribution of users to business goals (such as increasing revenue, reducing costs, etc.).

[0107] Step A200, based on the user feature samples and the ranking score labels, iteratively train the preset model to be trained to obtain an optimized allocation model, wherein the optimized allocation model is a dual causal learning model, and the optimized allocation model is composed of a neural network including a loss convex function.

[0108] It should be noted that the preset model to be trained is an initial model, which will be used to learn how to predict ranking score labels based on user characteristics. Through multiple iterations, the model parameters are continuously adjusted to minimize the difference between the predicted ranking score and the actual ranking score. The model obtained after iterative training can predict the ranking score of each user based on user characteristics, thereby guiding the optimal allocation of resources.

[0109] In specific implementation, the Dual Causal Learning (DCL) model is a specific machine learning model that optimizes resource allocation by combining causal inference and machine learning techniques. The DCL model is able to handle the heterogeneity of treatment effects, that is, different users may respond differently to the same treatment (such as coupons, advertisements, etc.).

[0110] Furthermore, the DCL model in this application uses a neural network as its basic architecture, and its loss function is a convex function. Using a convex loss function helps ensure that the optimization problem in the model training process is easy to solve, because any local minimum of a convex function is also a global minimum, which helps the model converge to the optimal solution faster.

[0111] In the specific implementation, refer to Figure 4The DCL proposed in this application uses a neural network with one hidden layer. The hidden layer contains eight neurons (neurons in the neural network play the role of processing input information, extracting features and calculating outputs), and uses the ReLU activation function, and the L2 regularization parameter is 2.5e-5. In the output layer, the model has one neuron, that is, the DCL uses the Tanh activation function, and the L2 regularization parameter is 2.5e-5. During the training process of the neural network, the model uses an adaptive momentum estimation (Adam) optimizer.

[0112] The above process describes that the device collects user feature samples and corresponding ranking score labels, and uses these data to train a dual causal learning model, which is a neural network with a convex loss function. This model can predict the ranking score of each user based on user features, thereby helping to achieve optimal resource allocation. This method can improve resource utilization efficiency and optimize business results.

[0113] In a specific implementation, the device iteratively trains a preset model to be trained based on the user feature sample and the ranking score label to obtain the step of optimizing the allocation model, including:

[0114] Based on the user feature samples, resources are allocated through the preset model to be trained, and the predicted ranking score of each user feature sample is output; based on the preset loss convex function, the error result between the predicted ranking score and the ranking score label is calculated; based on the error result, it is determined whether the error result meets the error standard indicated by the preset error threshold range; if the error result does not meet the error standard indicated by the preset error threshold range, then return to the step of allocating resources based on the user feature samples through the preset model to be trained, and outputting the predicted ranking score of each user feature sample, until the error result meets the error standard indicated by the preset error threshold range, and the training is stopped to obtain the optimized allocation model.

[0115] In a specific implementation, the device uses a preset model to be trained, allocates resources according to user feature samples, and outputs a predicted ranking score for each user feature sample, which is obtained based on the model's analysis of the user features.

[0116] Furthermore, the device calculates the error result between the ranking score predicted by the model and the actual ranking score label based on a preset loss convex function, wherein the loss convex function is a mathematical function used to quantify the difference between the model prediction and the actual result.

[0117] In a specific implementation, the device compares the calculated error result with a preset error threshold range, wherein the preset error threshold range is a standard set before model training to determine whether the model performance reaches an acceptable level.

[0118] Furthermore, if the calculated error result exceeds the preset error threshold range, that is, the model performance does not meet expectations, then it is necessary to return to the resource allocation step. Continue resource allocation, output a new predicted ranking score, and recalculate the error result. When the error result meets the preset error threshold range, it is considered that the model has been trained to a sufficient degree. At this time, training is stopped. The obtained model is called the optimized allocation model, which can accurately predict the ranking score based on the user feature sample and is used for actual resource allocation.

[0119] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the resource allocation optimization method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0120] This application also provides a resource allocation optimization device, please refer to Figure 5 , the resource allocation optimization device comprises:

[0121] An acquisition module 10 is used to acquire target user characteristics of users to be allocated resources;

[0122] The allocation module 20 is used to allocate resources based on the target user characteristics through a preset optimization allocation model, and output the ranking score of each user to be allocated resources, wherein the ranking score reflects the contribution of each user to be allocated resources to the optimization target, and the optimization allocation model is composed of a neural network including a loss convex function;

[0123] The generating module 30 is used to generate a corresponding optimal allocation scheme based on the ranking score.

[0124] Optionally, the allocation module 20 includes:

[0125] A resource allocation module, configured to allocate resources based on the target user characteristics through a preset optimization allocation model, and predict the average processing impact value of the first condition on the income and the average processing impact value of the second condition on the cost of each resource user to be allocated;

[0126] The calculation module is used to calculate the difference between the average processing impact value of the first condition and the average processing impact value of the second condition to obtain the ranking score of each user to be allocated resources.

[0127] Optionally, the resource allocation module includes:

[0128] An extraction module, used to determine the first user behavior level characteristics of each user of the to-be-allocated resources regarding income and the second user behavior level characteristics regarding cost based on the preset user behavior level division information and the target user characteristics;

[0129] A prediction module is used to perform resource allocation based on the first user behavior level characteristics and the second user behavior level characteristics through a preset optimization allocation model, and predict the first conditional average processing impact value of income and the second conditional average processing impact value of cost corresponding to each user of the resources to be allocated.

[0130] Optionally, the resource allocation optimization device further includes:

[0131] A sample acquisition module, used to acquire user feature samples and ranking score labels of the user feature samples;

[0132] A training module is used to iteratively train a preset model to be trained based on user feature samples and the ranking score labels to obtain an optimized allocation model, wherein the optimized allocation model is a dual causal learning model, and the optimized allocation model is composed of a neural network containing a loss convex function.

[0133] Optionally, the training module includes:

[0134] A resource allocation module, configured to allocate resources based on the user feature samples through a preset model to be trained, and output a predicted ranking score of each user feature sample;

[0135] An error calculation module, used to calculate the error result between the predicted ranking score and the ranking score label based on a preset loss convex function;

[0136] A judgment module, configured to judge, based on the error result, whether the error result satisfies an error standard indicated by a preset error threshold range;

[0137] The iterative training module is used to return to the step of allocating resources based on the user feature samples through a preset model to be trained, and outputting the predicted ranking score of each user feature sample if the error result does not meet the error standard indicated by the preset error threshold range, and stop training after the error result meets the error standard indicated by the preset error threshold range to obtain an optimized allocation model.

[0138] The resource allocation optimization device provided by the present application adopts the resource allocation optimization method in the above embodiment to solve the technical problem of resource allocation optimization. Compared with the prior art, the beneficial effects of the resource allocation optimization device provided by the present application are the same as the beneficial effects of the resource allocation optimization method provided by the above embodiment, and the other technical features in the resource allocation optimization device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0139] The present application provides a resource allocation optimization device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the resource allocation optimization method in the above-mentioned embodiment one.

[0140] Reference below Figure 6 , which shows a schematic diagram of the structure of a resource allocation optimization device suitable for implementing the embodiment of the present application. The resource allocation optimization device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 6 The resource allocation optimization device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0141] like Figure 6As shown, the resource allocation optimization device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the resource allocation optimization device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the resource allocation optimization device to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a resource allocation optimization device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or provided alternatively.

[0142] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0143] The resource allocation optimization device provided by the present application adopts the resource allocation optimization method in the above embodiment to solve the technical problem of resource allocation optimization. Compared with the prior art, the beneficial effects of the resource allocation optimization device provided by the present application are the same as the beneficial effects of the resource allocation optimization method provided by the above embodiment, and the other technical features in the resource allocation optimization device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0144] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0145] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0146] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the resource allocation optimization method in the above-mentioned embodiment.

[0147] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0148] The computer-readable storage medium may be included in the resource allocation optimization device; or may exist independently without being assembled into the resource allocation optimization device.

[0149] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the resource allocation optimization device, the resource allocation optimization device performs resource allocation optimization.

[0150] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0151] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0152] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0153] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned resource allocation optimization method, and can solve the technical problem of resource allocation optimization. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the resource allocation optimization method provided in the above-mentioned embodiment, and will not be repeated here.

[0154] The present application also provides a computer program product, including a computer program, which implements the steps of the resource allocation optimization method as described above when executed by a processor.

[0155] The computer program product provided by this application can solve the technical problem of resource allocation optimization. Compared with the prior art, the beneficial effects of the computer program product provided by this application are the same as the beneficial effects of the resource allocation optimization method provided by the above embodiment, which will not be repeated here.

[0156] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A resource allocation optimization method, characterized in that: The method for optimizing resource allocation comprises: Obtain target user characteristics of users to be allocated resources; Based on the target user characteristics, resources are allocated through a preset optimization allocation model, and a ranking score of each user of resources to be allocated is output, wherein the ranking score reflects the contribution of each user of resources to be allocated to the optimization target, and the optimization allocation model is composed of a neural network including a loss convex function; Based on the ranking scores, a corresponding optimal allocation solution is generated.

2. The resource allocation optimization method according to claim 1, characterized in that: The formula of the loss convex function is as follows: Among them, N1 represents the sample size of the experimental group, N0 represents the sample size of the control group, i represents the i-th user, and q i represents the ranking score of the i-th user, and tanh is The hyperbolic tangent transform, h i is the intermediate output of any neural network, λ is represented as a hyperparameter, y r Represents the income result value, y c Represents the cost result value.

3. The resource allocation optimization method according to claim 1, characterized in that: The step of allocating resources based on the target user characteristics through a preset optimization allocation model and outputting the ranking score of each user to be allocated resources includes: Based on the target user characteristics, resources are allocated through a preset optimization allocation model, and the first condition average processing impact value of income and the second condition average processing impact value of cost corresponding to each user of the resources to be allocated are predicted; The difference between the first condition average processing impact value and the second condition average processing impact value is calculated to obtain a ranking score of each user to be allocated resources.

4. The resource allocation optimization method according to claim 3, characterized in that: The step of performing resource allocation based on the target user characteristics through a preset optimization allocation model to predict the first conditional average processing impact value of income and the second conditional average processing impact value of cost corresponding to each resource user to be allocated includes: Based on the preset user behavior level division information and the target user characteristics, determining the first user behavior level characteristics of each user of the to-be-allocated resources regarding income and the second user behavior level characteristics regarding cost; Based on the first user behavior level characteristics and the second user behavior level characteristics, resources are allocated through a preset optimization allocation model to predict the first conditional average processing impact value of income and the second conditional average processing impact value of cost for each user of the resources to be allocated.

5. The resource allocation optimization method according to claim 1, characterized in that: Before the step of obtaining the allocation scheme to be optimized, the method includes: Obtaining user feature samples and ranking score labels of the user feature samples; Based on the user feature samples and the ranking score labels, the preset model to be trained is iteratively trained to obtain an optimized allocation model, wherein the optimized allocation model is a dual causal learning model, and the optimized allocation model is composed of a neural network including a loss convex function.

6. The resource allocation optimization method according to claim 5, characterized in that: The step of iteratively training the preset model to be trained based on the user feature sample and the ranking score label to obtain the optimized allocation model includes: Based on the user feature samples, resources are allocated through a preset model to be trained, and a predicted ranking score of each user feature sample is output; Based on a preset loss convex function, calculating an error result between the predicted ranking score and the ranking score label; Based on the error result, determining whether the error result meets an error standard indicated by a preset error threshold range; If the error result does not meet the error standard indicated by the preset error threshold range, then return to the step of allocating resources based on the user feature sample through the preset model to be trained, and output the predicted ranking score of each user feature sample, until the error result meets the error standard indicated by the preset error threshold range, and stop training to obtain the optimized allocation model.

7. A resource allocation optimization device, characterized in that: The device comprises: An acquisition module, used to acquire target user characteristics of users to be allocated resources; An allocation module, configured to allocate resources based on the target user characteristics through a preset optimization allocation model, and output a ranking score of each user of resources to be allocated, wherein the ranking score reflects the contribution of each user of resources to be allocated to the optimization target, and the optimization allocation model is composed of a neural network including a loss convex function; A generation module is used to generate a corresponding optimal allocation plan based on the ranking score.

8. A resource allocation optimization device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the resource allocation optimization method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the resource allocation optimization method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the resource allocation optimization method according to any one of claims 1 to 6 are implemented.

Images