Resource distribution strategy generation method and device, medium and equipment

Abstract

The invention provides a resource distribution strategy generation method and device, a medium and equipment, and relates to the technical field of computers.The method comprises the steps that a sample set representing the relation between a resource distribution result and user behaviors can be obtained, and the behavior occurrence probability and probability gain of a user under all resource types can be predicted based on the sample set; based on each behavior occurrence probability set, each probability gain set, a resource distribution decision variable, a user set, a resource type set, a resource value and a resource budget, a problem model used for representing resource distribution gain maximization can be constructed, the problem model can face each user in the user set, and user grouping does not need to be carried out like related technologies. The problem model can be relaxed into a Lagrangian dual problem model, a target Lagrangian multiplier which is more accurate than that obtained by directly solving the problem model can be obtained by solving the problem model, and the target Lagrangian multiplier can guide generation of a resource distribution strategy for achieving the purpose of accurate resource distribution.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more specifically, to a method for generating a resource allocation strategy, a device for generating a resource allocation strategy, a computer-readable storage medium, and an electronic device. Background Technology

[0002] With the continuous advancement of computer technology, users can enjoy increasingly diverse online services, such as online shopping, online top-up, and online financial services. To accelerate the promotion of these online services, various resources are typically pushed to users to encourage their use. For example, coupons can be sent to encourage users to place orders based on online shopping services; cashback offers can be sent to encourage users to recommend online shopping services to friends; and high-interest rate alerts can be sent to encourage users to use online financial services.

[0003] Generally, users differ in personal habits, social customs, economic conditions, and other factors. Therefore, it's understandable that users with such diverse needs may require different resources. Even for the same online service's specific function, relevant personnel can design a variety of resources (e.g., coupons for purchases over 200 yuan, coupons for purchases over 200 yuan, coupons for purchases over 200 yuan, coupons for purchases over 200 yuan, etc.) to meet users' diverse needs.

[0004] To achieve precise resource allocation and avoid both inappropriate resource distribution in one area leading to poor user engagement and inappropriate distribution in another causing budget overruns, relevant technologies typically group massive amounts of users based on similarity data. This grouping guides subsequent resource allocation. However, even within the same group, users can have differences, meaning that resource allocation based on this method still suffers from insufficient precision.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this application is to provide a method, apparatus, computer-readable storage medium, and electronic device for generating resource allocation strategies. The method first acquires a sample set representing the relationship between resource allocation results and user behavior. Based on this sample set, the probability of user behavior and probability gain under each resource type can be predicted. Then, based on the sets of behavior probabilities, probability gain sets, resource allocation decision variables, user set, resource type set, resource value, and resource budget, a problem model representing the maximization of resource allocation gain can be constructed. This problem model can be applied to each user in the user set, providing each user with a model to maximize resource allocation gain without requiring user grouping as in related technologies. Furthermore, this problem model can be relaxed to a Lagrange dual problem model. Solving the Lagrange dual problem model yields a more accurate target Lagrange multiplier than directly solving the problem model. This target Lagrange multiplier can guide the generation of a resource allocation strategy, providing personalized resource allocation schemes for each user and maximizing resource allocation gain. Therefore, it is understood that this application can achieve precise resource allocation.

[0007] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0008] According to one aspect of this application, a method for generating a resource allocation strategy is provided, the method comprising:

[0009] Based on a sample set representing the relationship between resource allocation results and user behavior, the probability of user behavior and probability gain for each resource type are predicted, resulting in a set of probability of behavior and a set of probability gain for each resource type.

[0010] Based on the set of probabilities of each behavior, the set of probabilities of gain, the resource allocation decision variables, the user set, the set of resource types, the resource value, and the resource budget, construct a problem model to characterize the maximization of resource allocation gain.

[0011] Lagrange relaxation is applied to the problem model to obtain the Lagrange dual problem model, and the Lagrange dual problem model is solved iteratively to determine the target Lagrange multiplier;

[0012] A resource allocation strategy for maximizing resource allocation gain is generated based on the objective Lagrange multiplier.

[0013] In one exemplary embodiment of this application, it further includes:

[0014] Obtain behavioral datasets and resource allocation records;

[0015] Perform a full database join operation on the behavior dataset and resource allocation records to obtain an intermediate sample set; the intermediate sample set includes a first class of samples belonging to the resource-assisted behavior type, a second class of samples belonging to the no-resource-assisted behavior type, and a third class of samples belonging to the no-behavior type of allocated resources.

[0016] Four types of samples belonging to the "no resource, no behavior" category are generated based on the intermediate sample set;

[0017] Determine the sample set containing Class I, Class II, Class III, and Class IV samples.

[0018] In one exemplary embodiment of this application, it further includes:

[0019] If the proportion of a certain type of sample in the intermediate sample set is lower than the preset proportion, then sample augmentation is performed on that type of sample based on behavioral cost and resource acquisition threshold.

[0020] In one exemplary embodiment of this application, four types of samples belonging to the "no resource, no behavior" type are generated based on an intermediate sample set, including:

[0021] Copy the intermediate sample set to obtain a duplicate sample set;

[0022] Based on the sampling rules, sample users in the replica sample set are sampled to obtain the sample user set;

[0023] Based on the first and second type samples in the replica sample set, and the frequency of behavior corresponding to each user identifier in the sample user set, four types of samples belonging to the no-resource, no-behavior type are constructed in the replica sample set.

[0024] In one exemplary embodiment of this application, determining a sample set comprising four types of samples (a first type, a second type, a third type, and a fourth type) includes:

[0025] Find the union of the duplicate sample set and the intermediate sample set to obtain a sample set containing samples of class 1, class 2, class 3, and class 4.

[0026] In one exemplary embodiment of this application, it further includes:

[0027] If the number of samples of class 1, class 2, class 3, and class 4 does not conform to the sample balance rule, then the sample set will be balanced proportionally.

[0028] In one exemplary embodiment of this application, based on a sample set representing the relationship between resource allocation results and user behavior, the probability of user behavior and probability gain for each resource type are predicted, resulting in a set of behavior occurrence probabilities and a set of probability gains corresponding to each resource type, including:

[0029] A prediction model is constructed based on a sample set representing the relationship between resource allocation results and user behavior;

[0030] Based on the prediction model, predict the probability of a user's behavior and the probability of no behavior under each resource type in the user set, and obtain the set of probability of behavior and the set of probability of no behavior corresponding to each resource type.

[0031] Based on the set of probability of occurrence of behavior and the set of probability of no behavior occurring for each resource type, determine the probability gain set corresponding to each resource type.

[0032] In one exemplary embodiment of this application, a Lagrange relaxation is performed on the problem model to obtain a Lagrange dual problem model, and the Lagrange dual problem model is iteratively solved to determine the target Lagrange multiplier, including:

[0033] By applying Lagrange relaxation to the problem model, a Lagrange dual problem model is obtained.

[0034] The Lagrange dual problem model is simplified into the objective problem model;

[0035] The target problem model is solved iteratively to determine the target Lagrange multiplier.

[0036] In one exemplary embodiment of this application, a resource allocation strategy for maximizing resource allocation gain is generated based on a target Lagrange multiplier, including:

[0037] Use the Lagrange heuristic algorithm to solve for the resource allocation cost containing the target Lagrange multiplier;

[0038] Based on a greedy strategy and resource allocation costs, a resource allocation strategy is generated to maximize the resource allocation gain; wherein, the resource allocation strategy includes resource allocation schemes for each user in the user set.

[0039] According to one aspect of this application, a resource allocation strategy generation apparatus is provided, comprising:

[0040] The probability prediction unit is used to predict the probability of a user set’s behavior and the probability gain under each resource type based on a sample set representing the relationship between resource allocation results and user behavior, thus obtaining the set of probability of behavior and the set of probability gain corresponding to each resource type.

[0041] The problem model construction unit is used to construct a problem model that represents the maximization of resource allocation gain based on the set of probability of occurrence of each behavior, the set of probability gains, resource allocation decision variables, user set, resource type set, resource value, and resource budget.

[0042] The Lagrange relaxation unit is used to perform Lagrange relaxation on the problem model to obtain the Lagrange dual problem model, and to iteratively solve the Lagrange dual problem model to determine the target Lagrange multiplier;

[0043] The resource allocation strategy generation unit is used to generate a resource allocation strategy based on the target Lagrange multiplier to maximize the resource allocation gain.

[0044] In one exemplary embodiment of this application, it further includes:

[0045] The data acquisition unit is used to acquire behavioral datasets and resource allocation records;

[0046] The fully connected unit is used to perform a full database connection operation on the behavior dataset and resource allocation records to obtain an intermediate sample set. The intermediate sample set includes a first class of samples belonging to the resource-assisted behavior type, a second class of samples belonging to the no-resource-assisted behavior type, and a third class of samples belonging to the no-behavior type of allocated resources.

[0047] The sample generation unit is used to generate four types of samples belonging to the no-resource and no-behavior type based on the intermediate sample set.

[0048] The sample set determination unit is used to determine the sample set containing Class I, Class II, Class III, and Class IV samples.

[0049] In one exemplary embodiment of this application, it further includes:

[0050] The sample augmentation unit is used to augment a sample class if the proportion of a sample class in the intermediate sample set is lower than a preset proportion, based on the behavioral cost and resource acquisition threshold.

[0051] In one exemplary embodiment of this application, the sample generation unit generates four types of samples belonging to the "no resource, no behavior" type based on an intermediate sample set, including:

[0052] Copy the intermediate sample set to obtain a duplicate sample set;

[0053] Based on the sampling rules, sample users in the replica sample set are sampled to obtain the sample user set;

[0054] Based on the first and second type samples in the replica sample set, and the frequency of behavior corresponding to each user identifier in the sample user set, four types of samples belonging to the no-resource, no-behavior type are constructed in the replica sample set.

[0055] In one exemplary embodiment of this application, the sample set determination unit determines a sample set containing four types of samples: a first type, a second type, a third type, and a fourth type, including:

[0056] Find the union of the duplicate sample set and the intermediate sample set to obtain a sample set containing samples of class 1, class 2, class 3, and class 4.

[0057] In one exemplary embodiment of this application, it further includes:

[0058] The sample balancing unit is used to balance the proportions of the sample set if the quantities of samples of class 1, class 2, class 3, and class 4 do not conform to the sample balancing rules.

[0059] In one exemplary embodiment of this application, the probability prediction unit predicts the probability of a user set's behavior and the probability gain for each resource type based on a sample set characterizing the relationship between resource allocation results and user behavior, thereby obtaining a set of behavior occurrence probabilities and a set of probability gains corresponding to each resource type, including:

[0060] A prediction model is constructed based on a sample set representing the relationship between resource allocation results and user behavior;

[0061] Based on the prediction model, predict the probability of a user's behavior and the probability of no behavior under each resource type, and obtain the set of probability of behavior and the set of probability of no behavior corresponding to each resource type.

[0062] Based on the set of probability of occurrence of behavior and the set of probability of no behavior occurring for each resource type, determine the probability gain set corresponding to each resource type.

[0063] In one exemplary embodiment of this application, a Lagrange relaxation unit performs Lagrange relaxation on the problem model to obtain a Lagrange dual problem model, and iteratively solves the Lagrange dual problem model to determine the target Lagrange multiplier, including:

[0064] By performing Lagrange relaxation on the problem model, a Lagrange dual problem model is obtained;

[0065] The Lagrange dual problem model is simplified into the objective problem model;

[0066] The target problem model is solved iteratively to determine the target Lagrange multiplier.

[0067] In one exemplary embodiment of this application, the resource allocation strategy generation unit generates a resource allocation strategy based on a target Lagrange multiplier to maximize the resource allocation gain, including:

[0068] Use the Lagrange heuristic algorithm to solve for the resource allocation cost containing the target Lagrange multiplier;

[0069] Based on a greedy strategy and resource allocation costs, a resource allocation strategy is generated to maximize the resource allocation gain; wherein, the resource allocation strategy includes resource allocation schemes for each user in the user set.

[0070] According to one aspect of this application, a computer-readable storage medium is provided, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the method of any one of the above.

[0071] According to one aspect of this application, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the method of any of the above by executing the executable instructions.

[0072] The exemplary embodiments of this application may have some or all of the following beneficial effects:

[0073] In a resource allocation strategy generation method provided in an example embodiment of this application, a sample set representing the relationship between resource allocation results and user behavior can be obtained first. Based on this sample set, the probability of user behavior and probability gain under each resource type can be predicted. Then, based on the set of probability occurrences of each behavior, the set of probability gains of each behavior, resource allocation decision variables, user set, resource type set, resource value, and resource budget, a problem model representing the maximization of resource allocation gain can be constructed. This problem model can be applied to each user in the user set, providing each user with a model that maximizes resource allocation gain, without the need for user grouping as in related technologies. Furthermore, this problem model can be relaxed to a Lagrange dual problem model. Solving the Lagrange dual problem model yields a more accurate target Lagrange multiplier than directly solving the problem model. The target Lagrange multiplier can guide the generation of a resource allocation strategy, which can provide personalized resource allocation schemes for each user and maximize resource allocation gain. Therefore, it is understood that this application can achieve precise resource allocation. Furthermore, since this application employs Lagrange relaxation, a high-precision target Lagrange multiplier can be obtained. The resource allocation strategy generated based on this target Lagrange multiplier can avoid the impact of data scale on the accuracy of resource allocation.

[0074] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0075] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0076] Figure 1 A flowchart illustrating a resource allocation strategy generation method according to an embodiment of this application is shown schematically.

[0077] Figure 2 This illustration schematically shows a problem-scale diagram to be solved according to one embodiment of the present application;

[0078] Figure 3 A flowchart illustrating a resource allocation strategy generation method according to another embodiment of this application is shown schematically;

[0079] Figure 4 This illustration schematically shows a system architecture diagram for implementing a resource allocation strategy generation method according to an embodiment of this application;

[0080] Figure 5 This schematic diagram illustrates a structural block diagram of a resource allocation strategy generation apparatus according to one embodiment of the present application;

[0081] Figure 6 The schematic diagram illustrates the structure of a computer system suitable for implementing the electronic devices of the present application. Detailed Implementation

[0082] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of the embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced with one or more of the specific details omitted, or other methods, components, apparatus, steps, etc., can be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of this application.

[0083] Furthermore, the accompanying drawings are merely illustrative of this application and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0084] Please see Figure 1 , Figure 1 A flowchart illustrating a resource allocation strategy generation method according to an embodiment of this application is shown schematically. Figure 1 As shown, the resource allocation strategy generation method may include steps S110 to S140.

[0085] Step S110: Based on the sample set representing the relationship between resource allocation results and user behavior, predict the probability of user behavior and probability gain under each resource type, and obtain the set of probability of behavior and the set of probability gain corresponding to each resource type.

[0086] Step S120: Based on the set of probabilities of each behavior, the set of probability gains, the resource allocation decision variables, the user set, the set of resource types, the resource value, and the resource budget, construct a problem model to characterize the maximization of resource allocation gains.

[0087] Step S130: Perform Lagrange relaxation on the problem model to obtain the Lagrange dual problem model, and iteratively solve the Lagrange dual problem model to determine the target Lagrange multiplier.

[0088] Step S140: Generate a resource allocation strategy based on the target Lagrange multiplier to maximize the resource allocation gain.

[0089] Implementation Figure 1The method described above first obtains a sample set representing the relationship between resource allocation results and user behavior. Based on this sample set, the probability of user behavior and probability gain under each resource type can be predicted. Then, based on the sets of probability occurrences, probability gain sets, resource allocation decision variables, user set, resource type set, resource value, and resource budget, a problem model representing the maximization of resource allocation gain can be constructed. This problem model can be applied to each user in the user set, providing each user with a model to maximize resource allocation gain without requiring user grouping as in related technologies. Furthermore, this problem model can be relaxed to a Lagrange dual problem model. Solving the Lagrange dual problem model yields a more accurate target Lagrange multiplier than directly solving the problem model. This target Lagrange multiplier can guide the generation of a resource allocation strategy, providing personalized resource allocation schemes for each user, and maximizing resource allocation gain. Therefore, it is understood that this application can achieve precise resource allocation. Furthermore, since this application employs Lagrange relaxation, a high-precision target Lagrange multiplier can be obtained. The resource allocation strategy generated based on this target Lagrange multiplier can avoid the impact of data scale on the accuracy of resource allocation.

[0090] The steps described above in this example implementation will now be explained in more detail.

[0091] In step S110, based on the sample set representing the relationship between resource allocation results and user behavior, the probability of user behavior and probability gain under each resource type are predicted, resulting in the set of probability of behavior and the set of probability gain corresponding to each resource type.

[0092] Specifically, the relationship between resource distribution results and user behavior is used to characterize whether user behavior is performed with the assistance of resource distribution results. Resource distribution results are used to characterize which users received what type of resources (e.g., a coupon for 100 off a purchase of 200) under the "spend 200 and get a discount" category, while user behavior refers to user-triggered actions (e.g., placing an order, making an investment).

[0093] There are usually multiple resource types. For example, in the online shopping field, resource types can include: discounts for purchases over 200, discounts for purchases over 500, discounts for purchases over 1000, etc. The probability of a behavior occurring refers to the probability that users within a user group will trigger a user action under a particular resource type. For example, user A has an 80% probability of placing an order under the "discount for purchases over 200" type. In other words, if user A is given a 100 RMB discount coupon for purchases over 200, user A has an 80% probability of triggering an order based on the 100 RMB discount coupon.

[0094] Understandably, due to the diversity of resource types and the varying sensitivities of different users to different resource types, precise resource allocation is beneficial for maximizing the returns from resource distribution. For example, if user A is given a coupon for 100 off a purchase of 200, user A has the highest probability of placing an order; if given a coupon for 10 off a purchase of 200, user A has the lowest probability of placing an order. Therefore, provided that the resource budget is not exceeded and resources are also reasonably allocated to other users, giving user A a coupon for 100 off a purchase of 200 is a better decision. User B is not sensitive to any form of coupon, and there is no necessary correlation between their user behavior and the resource distribution result. Therefore, it is understandable that user B's user behavior is more likely to depend on the user themselves, rather than on the type of resource given to them.

[0095] Furthermore, probability gain refers to the difference between the probability of a behavior occurring and the probability of a behavior not occurring. Under each resource type, the probability of a behavior occurring and the probability gain for each user can be calculated. Based on this, a set of behavior occurrence probabilities and a set of probability gains corresponding to each resource type can be obtained. For example, in the set of behavior occurrence probabilities A and the set of probability gains A corresponding to resource type A, the set of behavior occurrence probabilities A contains the probability of a behavior occurring for each user, and the set of probability gains A contains the probability gains for each user. Moreover, the data in the set of behavior occurrence probabilities A and the set of probability gains A are in one-to-one correspondence; a set of corresponding data corresponds to the same user.

[0096] As an optional embodiment, before step S110, the method further includes:

[0097] Step S010: Obtain the behavior dataset and resource allocation records;

[0098] Step S020: Perform a full database join operation on the behavior dataset and resource allocation records to obtain an intermediate sample set; wherein, the intermediate sample set includes a first class of samples belonging to the resource auxiliary behavior type, a second class of samples belonging to the no resource auxiliary behavior type, and a third class of samples belonging to the allocated resource no behavior type;

[0099] Step S030: Generate four types of samples belonging to the no-resource, no-behavior type based on the intermediate sample set;

[0100] Step S040: Determine the sample set containing Class I, Class II, Class III, and Class IV samples.

[0101] As can be seen, implementing this optional embodiment can obtain a sample set with more comprehensive sample dimensions. Based on this sample set, it is beneficial to predict a more accurate set of probability of occurrence and probability gain set of behavior, which is conducive to generating a resource allocation strategy that can achieve the purpose of accurate resource allocation.

[0102] Specifically, the behavior dataset includes a data volume determined by the actual situation, and the resource allocation records are similarly structured. In the behavior dataset, any behavior data point may include at least: user identifier (e.g., user A), behavior trigger time (e.g., 2022-02-02 00:00:00), behavior target (e.g., men's trench coat), and behavior assistance method (e.g., based on a 200-100 coupon). In the resource allocation record, any resource allocation record may include at least: user identifier (e.g., user A), resource type (e.g., 200-100 discount type), resource name (e.g., 200-100 coupon), and allocation time (e.g., 2022-02-01 00:00:00).

[0103] In most cases, to comply with database design specifications, data is stored in different tables within a database, but this storage method is not conducive to efficient data retrieval. Joins are typically used to achieve efficient data retrieval. Join methods include inner join, left join, right join, and full join. The result set of a full join combines the result sets of left and right joins. In this application, to ensure that all data related to the resource allocation records and the behavior dataset are retrieved, a full join can be used to merge the behavior dataset and resource allocation records to obtain an intermediate sample set. The fields in the intermediate sample set can come from both the behavior dataset and the resource allocation records.

[0104] In addition, the resource-assisted behavior type refers to the type of user behavior executed based on the allocated resources; the no-resource-assisted behavior type refers to the type of user behavior executed without the allocated resources; the resource-allocated no-behavior type refers to the type of user behavior not triggered after resources are allocated to the user; and the no-resource no-behavior type refers to the type of user behavior not triggered either, without resources being allocated to the user.

[0105] As an optional embodiment, it further includes: if the proportion of a certain type of sample in the intermediate sample set is lower than a preset proportion (e.g., 5%), then sample augmentation is performed on the certain type of sample based on behavioral cost and resource acquisition threshold.

[0106] As can be seen, implementing this optional embodiment can solve the problem of sparse sample quantity of a class of samples through sample augmentation. Augmenting sparse samples is beneficial to improving prediction accuracy when predicting the probability of occurrence of behavior and probability gain based on the sample set.

[0107] Specifically, the preset ratio is used to limit the minimum sample ratio. When the sample ratio is lower than the preset ratio, the samples corresponding to the sample ratio are likely to affect the accuracy of subsequent steps, so sample augmentation is required.

[0108] In the above embodiments, if the proportion of a certain type of sample in the intermediate sample set is lower than a preset proportion, sample augmentation is performed on the certain type of sample based on the behavioral cost value and resource acquisition threshold. This includes: if the proportion of a certain type of sample in the intermediate sample set is lower than the preset proportion, substituting the behavioral cost value and resource acquisition threshold into a specified expression (e.g., (behavioral cost value + resource acquisition threshold) / 2) to calculate the value of the expression, constructing a new type of sample based on the value and the information recorded in the certain type of sample (e.g., username, behavioral cost resource (e.g., product name)), and supplementing the new type of sample into the intermediate sample set to achieve sample augmentation for the intermediate sample set.

[0109] The behavioral value refers to the cost (e.g., 30 yuan) that a user needs to pay for their behavior. This cost can be the price of a product or the amount of investment. The specific meaning of the cost depends on the field in which this application is applied, and this application does not limit it. In addition, the resource acquisition threshold is used to limit the usage threshold of resources (e.g., physical goods, financial products, etc.). This usage threshold is used to limit the minimum behavioral cost. When the resource acquisition threshold (e.g., 200 yuan) is less than or equal to the behavioral cost (e.g., 300 yuan), the user can perform user behavior based on the resources provided. In one example, the user can place an order based on a 100 yuan discount coupon for orders over 200 yuan, thereby reducing the behavioral cost of 300 yuan to 200 yuan.

[0110] In addition, optionally, it may include: if the proportion of Class II samples in the intermediate sample set is lower than a preset proportion, then perform sample augmentation on Class II samples based on behavioral cost and resource acquisition threshold; and / or, if the proportion of Class III samples in the intermediate sample set is lower than a preset proportion, then perform sample augmentation on Class III samples based on behavioral cost and resource acquisition threshold.

[0111] As an optional embodiment of step S030, four types of samples belonging to the no-resource, no-behavior type are generated based on the intermediate sample set, including:

[0112] Step S0301: Copy the intermediate sample set to obtain a duplicate sample set;

[0113] Step S0302: Sample users in the replica sample set based on the sampling rules to obtain the sample user set;

[0114] Step S0303: Based on the first and second type samples in the replica sample set, and the frequency of behavior corresponding to each user identifier in the sample user set, construct four types of samples belonging to the no-resource, no-behavior type in the replica sample set.

[0115] As can be seen, implementing this optional embodiment can achieve the construction of four types of samples without affecting the intermediate sample set, which helps to ensure the accuracy of the intermediate sample set.

[0116] Specifically, the replica sample set and the intermediate sample set are consistent. Based on the sampling rules used to limit the sampling method (e.g., sampling at an 80% ratio), sample users in the replica sample set can be sampled to obtain the sample user set. The number of sample users in the sample user set is less than the number of users in the replica sample set. Based on the first-class and second-class samples and the frequency of user behavior (e.g., user behavior is triggered once every 7 days), the time when user behavior should have occurred but did not occur can be determined (e.g., in the two weeks from January 1 to January 15, user behavior should have occurred once every 7 days, but in fact, no user behavior occurred from January 1 to January 15. Therefore, a random time point can be determined from the two time periods of January 1 to January 7 and January 8 to January 15 as the construction time of the four types of samples). Based on this time, four types of samples belonging to the no-resource, no-behavior type are constructed.

[0117] As an optional embodiment of step S040, determining a sample set containing Class I, Class II, Class III, and Class IV samples includes: taking the union of the duplicate sample set and the intermediate sample set to obtain a sample set containing Class I, Class II, Class III, and Class IV samples.

[0118] As can be seen, implementing this optional embodiment can yield a sample set with high accuracy and availability based on the union of datasets, which is beneficial for generating resource allocation strategies with higher accuracy.

[0119] Specifically, since the replica sample set contains four types of samples that are not in the intermediate sample set, the union of the replica sample set and the intermediate sample set can yield a sample set with more comprehensive sample dimensions.

[0120] As an optional embodiment, it also includes:

[0121] Step S050: If the number of samples of class 1, class 2, class 3, and class 4 does not conform to the sample balance rule, then the sample set is subjected to proportional balance processing.

[0122] As can be seen, by implementing this optional embodiment, sample balance can be achieved within the sample set through sample balancing processing, thereby improving the availability of the sample set.

[0123] Specifically, the sample balance rule is used to limit the ratio between Class I, Class II, Class III, and Class IV samples, such as 1:1:1:1. If the number of Class I, Class II, Class III, and Class IV samples does not conform to the sample balance rule, the sample set can be balanced. After the balance is achieved, the ratio between Class I, Class II, Class III, and Class IV samples will conform to the sample balance rule.

[0124] As an optional embodiment of step S110, based on a sample set representing the relationship between resource allocation results and user behavior, the probability of user behavior and probability gain for each resource type are predicted, resulting in a set of behavior occurrence probabilities and a set of probability gains corresponding to each resource type, including:

[0125] Step S1101: Construct a prediction model based on a sample set representing the relationship between resource allocation results and user behavior;

[0126] Step S1102: Based on the prediction model, predict the probability of behavior and the probability of no behavior for users in the user set under each resource type, and obtain the set of probability of behavior and the set of probability of no behavior corresponding to each resource type;

[0127] Step S1103: Determine the probability gain set corresponding to each resource type based on the set of probability of occurrence of behavior and the set of probability of no behavior occurrence for each resource type.

[0128] As can be seen, by implementing this optional embodiment, it is possible to predict the probability gain set and the probability set of behavior occurrence that are beneficial to the construction of the problem model, which can help to obtain a more accurate resource allocation strategy.

[0129] Optionally, the sample set representing the relationship between resource distribution results (e.g., coupon distribution results) and user behavior (e.g., order placement behavior) can be represented as shown in the following table:

[0130]

[0131] A prediction model, label(Y), can be trained based on a sample set representing the relationship between resource allocation results and user behavior. Y represents a user, label(Y) = 1 indicates that an order has been placed, and label(Y) = 0 indicates that no order has been placed. The prediction model can be trained based on the LightGBM model.

[0132] Furthermore, the prediction model can be used to predict the probability of a user's behavior occurring and the probability of no behavior occurring under each resource type, thus obtaining the corresponding set of probability of behavior occurring and the set of probability of no behavior occurring for each resource type. The set of probability of behavior occurring (e.g., the probability of placing an order with coupons) and the set of probability of no behavior occurring (e.g., the probability of placing an order without coupons) are specifically represented in the following table:

[0133]

[0134]

[0135] Taking user A as an example, the probability of placing an order with a coupon is represented by P(Y|X,T=1), and the probability of placing an order without a coupon is represented by P(Y|X,T=0); where X represents the feature of the prediction model, T=1 indicates that a coupon is issued, and T=0 indicates that a coupon is not issued.

[0136] Furthermore, for each user i and each coupon type j, the order probability gain uplift(Δp) can be calculated. ij The probability gain set for each resource type can be determined based on the probability set of occurrence of behavior and the probability set of no behavior occurrence for each resource type. The probability gain set is specifically represented by the following table:

[0137]

[0138] It is evident that user B, who has the highest probability of placing an order with a coupon, does not have a high probability gain, while user A has a high probability gain. Therefore, issuing coupons to user A would be more profitable.

[0139] Please see Figure 2 , Figure 2 This illustration schematically depicts the scale of the problem to be solved according to one embodiment of the present application. Figure 2 As shown, in real-world business scenarios, the number of users is large. Assuming there are k users, m types of thresholds, and n denominations, m*n coupon types can be constructed. When accurate to each user, k*m*n results can be calculated. In this situation, achieving precise coupon distribution through related technologies is difficult. However, the method defined in this application's embodiments can generate resource distribution strategies with high accuracy and low cost.

[0140] In step S120, a problem model is constructed to characterize the maximization of resource allocation gain based on the set of probability of occurrence of each behavior, the set of probability gains, resource allocation decision variables, user set, resource type set, resource value, and resource budget.

[0141] Specifically, the probability of an action occurring in the set of probabilities of each action can be represented as p. ij p ij This represents the probability that user i will use resource j for auxiliary behavior. The probability gain in each probability gain set can be expressed as Δp. ij Δp ij This refers to the probability gain resulting from allocating resource j to user i. The resource allocation decision variable can be represented as x. ij x ij This indicates whether resource j has been allocated to user i. If it has, then x represents whether resource j has been allocated to user i. ij =1, if not issued, x ij =0. The user set can be represented as I, where the number of users depends on the actual situation. The resource type set can be represented as J. For example, a resource type can be represented as 200 minus 10, and the number of resource types in the resource type set depends on the actual situation. Resource value can be represented as a. j a j This refers to the face value of resources, such as 10. The resource budget can be represented as b, where b represents the upper limit of the sum of all resource face values.

[0142] According to the above p ij Δp ij x ij 、I、J、a j b, can be used to construct a problem model to characterize the maximization of resource allocation gain. Optionally, the problem model to characterize the maximization of resource allocation gain can be represented by the following expressions (1) to (4):

[0143] Optionally, the problem model used to characterize the maximization of resource allocation gain can be represented by the following expressions (1) to (4):

[0144] max∑ i∈I ∑ j∈J Δp ij x ij (1)

[0145]

[0146] ∑ i∈I ∑ j∈J p ij a j x ij ≤b (3)

[0147] x ij ∈{0,1} (4)

[0148] Specifically, expression (1) can be understood as the objective function, which represents the technical objective of maximizing the order probability gain brought about by resource allocation; expression (2) represents the knapsack grouping constraint, which means that each user is allocated one resource, where optional, not allocating resources is represented as a resource type as "full 0-0"; expression (3) represents the knapsack capacity constraint, which means that the total amount of resource allocation does not exceed the resource budget; expression (4) represents the range of values ​​for the resource decision variable.

[0149] In step S130, the problem model is subjected to Lagrange relaxation to obtain the Lagrange dual problem model, and the Lagrange dual problem model is solved iteratively to determine the target Lagrange multiplier.

[0150] Lagrangian relaxation refers to a technique for handling constraints in constrained optimization problems. Constraints are categorized as simple or difficult. A typical Lagrangian relaxation problem includes the primal problem, the relaxation problem, and the dual problem. Lagrangian relaxation involves applying a Lagrange multiplier to the objective function, creating a difficult constraint. Solving the function containing the Lagrange multiplier yields its value, which provides the answer to the constrained problem. The relaxation problem represents a lower bound on the primal problem, while the dual problem represents the largest lower bound on the primal problem. Therefore, we can relax the problem model in step S120 into a dual problem and solve for the Lagrange multipliers to obtain the desired final strategy.

[0151] Specifically, after performing Lagrange relaxation on the problem model defined by expressions (1) to (4), the resulting Lagrange dual problem model can be represented by expression (5); where Lagrange relaxation refers to using the Lagrange multipliers λ and μ i The problem model is relaxed to a Lagrange dual problem model.

[0152]

[0153] As an optional embodiment of step S130, the problem model is subjected to Lagrange relaxation to obtain a Lagrange dual problem model, and the Lagrange dual problem model is solved iteratively to determine the target Lagrange multiplier, including:

[0154] Step S1301: Perform Lagrange relaxation on the problem model to obtain the Lagrange dual problem model;

[0155] Step S1302: Simplify the Lagrange dual problem model into the target problem model;

[0156] Step S1303: Iteratively solve the target problem model to determine the target Lagrange multiplier.

[0157] As can be seen, implementing this optional embodiment can calculate higher-quality target Lagrange multipliers based on Lagrange relaxation, and improve the solution efficiency of the problem model, which is conducive to the efficient generation of resource allocation strategies.

[0158] Specifically, in order to efficiently solve for the target Lagrange multiplier, expression (5) can be simplified to the target problem model, which can be represented by expression (6).

[0159]

[0160] Based on this, optionally, the target problem model is iteratively solved to determine the target Lagrange multipliers, including:

[0161] The Lagrange multipliers λ and μ in the target problem model i The initial value is configured as a first value (e.g., 0); then the iteration step size α and the Lagrange multiplier μ corresponding to the Lagrange multiplier λ are determined. i The corresponding iteration step size β i , α and β i The initial value is configured as a second value (e.g., 0.01); then, based on the configured Lagrange multipliers λ and μ i and α and β i Solve the expression (7) to be iterated corresponding to expression (6);

[0162]

[0163] Based on this, the rc is based on distributing resources j to user i. ij =Δp ij -μ i -λ,rc ij This refers to the cost of resource allocation. If rc ij If x > 0, then x ij =1, otherwise x ij =0;

[0164] Furthermore, based on x ij The value of can be used to solve the expression (8) corresponding to expression (5);

[0165]

[0166] Among them, the curve corresponding to the expression (8) to be solved is represented as a convex curve. According to the convex optimization theory, the negative gradient direction is the steepest descent direction. Therefore, the gradient of the Lagrange dual problem model can be represented by expressions (9) to (11):

[0167] g λ =b-∑i∈I ∑ j∈J a j x ij (9)

[0168]

[0169]

[0170] Based on this, λ and μ i The iterative formula can be expressed as λ=λ-αg λ and

[0171] If the iteration termination condition is met (e.g., the maximum number of iterations is reached, the objective value of the dual problem converges, etc.), then step S140 is executed; otherwise, the iteration continues from solving the expression to be iterated (7).

[0172] In step S140, a resource allocation strategy is generated based on the target Lagrange multiplier to maximize the resource allocation gain.

[0173] Specifically, the resource allocation strategy can support modification operations, custom operations, etc., which are not limited in the embodiments of this application.

[0174] As an optional embodiment of step S140, a resource allocation strategy for maximizing resource allocation gain is generated based on the target Lagrange multiplier, including:

[0175] Step S1401: Call the Lagrange heuristic algorithm to solve for the resource allocation cost containing the target Lagrange multiplier;

[0176] Step S1402: Based on the greedy strategy and resource allocation cost, generate a resource allocation strategy to maximize the resource allocation gain; wherein, the resource allocation strategy includes a resource allocation scheme for each user in the user set.

[0177] As can be seen, implementing this optional embodiment can efficiently and cost-effectively generate resource allocation schemes for each user in a user set based on high-precision target Lagrange multipliers.

[0178] Specifically, resource allocation cost rc ij =Δp ij -μ i -λ, calling the Lagrange heuristic algorithm can solve for the objective Lagrange multipliers λ and μ. i The resource distribution cost of issuing coupon j to user i.

[0179] The greedy strategy can be represented as follows:

[0180]

[0181] Based on a greedy strategy and resource allocation costs, a resource allocation plan corresponding to each user can be generated, achieving efficient resource allocation down to the individual level without grouping users.

[0182] Please see Figure 3 , Figure 3 A flowchart illustrating a resource allocation strategy generation method according to another embodiment of this application is shown schematically. Figure 3 As shown, the resource allocation strategy generation method includes steps S310 to S380.

[0183] Step S310: Obtain the behavior dataset and resource allocation records.

[0184] Step S320: Perform a full database join operation on the behavior dataset and resource allocation records to obtain an intermediate sample set. The intermediate sample set includes three types of samples: one type of samples belonging to resource-assisted behavior, two types of samples belonging to no-resource-assisted behavior, and three types of samples belonging to no-behavior type with allocated resources. If the proportion of a type of sample in the intermediate sample set is lower than a preset proportion, then sample augmentation is performed on the type of sample based on the behavior cost and resource acquisition threshold.

[0185] Step S330: Copy the intermediate sample set to obtain the replica sample set. Based on the sampling rules, sample users in the replica sample set are sampled to obtain the sample user set. Based on the first and second type samples in the replica sample set, and the frequency of behavior corresponding to each user identifier in the sample user set, four types of samples belonging to the no-resource and no-behavior type are constructed in the replica sample set.

[0186] Step S340: Calculate the union of the duplicate sample set and the intermediate sample set to obtain a sample set containing samples of class 1, class 2, class 3, and class 4. If the number of samples of class 1, class 2, class 3, and class 4 does not conform to the sample balance rule, then the sample set is subjected to proportional balance processing.

[0187] Step S350: Based on the sample set representing the relationship between resource allocation results and user behavior, construct a prediction model, and predict the probability of behavior and the probability of no behavior for users in the user set under each resource type according to the prediction model, to obtain the set of probability of behavior and the set of probability of no behavior corresponding to each resource type. Then, based on the set of probability of behavior and the set of probability of no behavior for each resource type, determine the set of probability gain corresponding to each resource type.

[0188] Step S360: Based on the set of probabilities of each behavior, the set of probabilities of gain, the resource allocation decision variables, the user set, the set of resource types, the resource value, and the resource budget, construct a problem model to characterize the maximization of resource allocation gain.

[0189] Step S370: Perform Lagrange relaxation on the problem model to obtain the Lagrange dual problem model, and simplify the Lagrange dual problem model into the objective problem model. Iterate through the objective problem model to determine the objective Lagrange multiplier.

[0190] Step S380: Solve the resource allocation cost containing the objective Lagrange multiplier using the Lagrange heuristic algorithm, and generate a resource allocation strategy to maximize the resource allocation gain based on the greedy strategy and the resource allocation cost; wherein, the resource allocation strategy includes a resource allocation scheme for each user in the user set.

[0191] It should be noted that steps S310 to S380 are the same as... Figure 1 For the specific implementation methods of steps S310 to S380, please refer to the examples shown. Figure 1 The steps and their embodiments shown are not described in detail here.

[0192] It is evident that implementation Figure 3 The method described above first obtains a sample set representing the relationship between resource allocation results and user behavior. Based on this sample set, the probability of user behavior and probability gain under each resource type can be predicted. Then, based on the sets of probability occurrences, probability gain sets, resource allocation decision variables, user set, resource type set, resource value, and resource budget, a problem model representing the maximization of resource allocation gain can be constructed. This problem model can be applied to each user in the user set, providing each user with a model to maximize resource allocation gain without requiring user grouping as in related technologies. Furthermore, this problem model can be relaxed to a Lagrange dual problem model. Solving the Lagrange dual problem model yields a more accurate target Lagrange multiplier than directly solving the problem model. This target Lagrange multiplier can guide the generation of a resource allocation strategy, providing personalized resource allocation schemes for each user, and maximizing resource allocation gain. Therefore, it is understood that this application can achieve precise resource allocation. Furthermore, since this application employs Lagrange relaxation, a high-precision target Lagrange multiplier can be obtained. The resource allocation strategy generated based on this target Lagrange multiplier can avoid the impact of data scale on the accuracy of resource allocation.

[0193] Please see Figure 4 , Figure 4The diagram illustrates a system architecture schematic for implementing a resource allocation policy generation method according to an embodiment of this application. Figure 4 As shown, the system architecture includes at least: a data input module 410, a sample augmentation module 420, an intelligent coupon issuance algorithm module 430, an output module 440, and a parameter setting module 450; wherein, the intelligent coupon issuance algorithm module 430 includes at least a prediction algorithm module 431, a causal algorithm module 432, and an operations research algorithm module 433.

[0194] The data input module 410 is used to acquire behavior datasets and resource allocation records; perform a full database connection operation on the behavior datasets and resource allocation records to obtain an intermediate sample set; the intermediate sample set includes a first-class sample belonging to the resource-assisted behavior type, a second-class sample belonging to the no-resource-assisted behavior type, and a third-class sample belonging to the allocated resource no-behavior type; a fourth-class sample belonging to the no-resource no-behavior type is generated based on the intermediate sample set; and a sample set containing the first-class, second-class, third-class, and fourth-class samples is determined.

[0195] The sample enhancement module 420 is used to enhance the sample set of a certain type of sample if the proportion of that type of sample in the intermediate sample set is lower than a preset proportion, based on the behavioral cost and resource acquisition threshold. It also includes: replicating the intermediate sample set to obtain a duplicate sample set; sampling the sample users in the duplicate sample set according to sampling rules to obtain a sample user set; and constructing four types of samples belonging to the no-resource, no-behavior type in the duplicate sample set based on the first and second type samples in the duplicate sample set and the frequency of behavior corresponding to each user identifier in the sample user set. Furthermore, if the quantities of the first, second, third, and fourth type samples do not conform to the sample balance rules, the sample set is subjected to proportional balancing processing.

[0196] The prediction algorithm module 431 is used to construct a prediction model based on a sample set representing the relationship between resource allocation results and user behavior; and to predict the probability of behavior and the probability of no behavior for users in the user set under each resource type according to the prediction model, thereby obtaining the set of probability of behavior and the set of probability of no behavior corresponding to each resource type.

[0197] The causal algorithm module 432 is used to determine the probability gain set corresponding to each resource type based on the set of probability of occurrence of behavior and the set of probability of no behavior occurrence for each resource type.

[0198] The operations research algorithm module 433 is used to construct a problem model representing the maximization of resource allocation gain based on the set of occurrence probabilities of each behavior, the set of probability gains of each behavior, resource allocation decision variables, user set, resource type set, resource value, and resource budget; to perform Lagrange relaxation on the problem model to obtain the Lagrange dual problem model, and to iteratively solve the Lagrange dual problem model to determine the target Lagrange multiplier; to generate a resource allocation strategy to maximize the resource allocation gain based on the target Lagrange multiplier.

[0199] Output module 440 is used to output resource allocation strategies.

[0200] The parameter setting module 450 is used to configure the parameters of at least one of the prediction algorithm module 431, the causal algorithm module 432, and the operations research algorithm module 433 in response to the configuration operation.

[0201] It is evident that implementation Figure 4 The system shown can first acquire a sample set representing the relationship between resource allocation results and user behavior. Based on this sample set, the probability of user behavior and probability gain under each resource type can be predicted. Then, based on the sets of probability occurrences, probability gain sets, resource allocation decision variables, user set, resource type set, resource value, and resource budget, a problem model representing the maximization of resource allocation gain can be constructed. This problem model can be applied to each user in the user set, providing each user with a model to maximize resource allocation gain without requiring user grouping as in related technologies. Furthermore, this problem model can be relaxed to a Lagrange dual problem model. Solving the Lagrange dual problem model yields a more accurate target Lagrange multiplier than directly solving the problem model. This target Lagrange multiplier can guide the generation of a resource allocation strategy, which can provide personalized resource allocation schemes for each user and maximize resource allocation gain. Therefore, it is understood that this application can achieve precise resource allocation. Furthermore, since this application employs Lagrange relaxation, a high-precision target Lagrange multiplier can be obtained. The resource allocation strategy generated based on this target Lagrange multiplier can avoid the impact of data scale on the accuracy of resource allocation.

[0202] Please see Figure 5 , Figure 5 This schematically illustrates a structural block diagram of a resource allocation policy generation apparatus according to one embodiment of the present application. The resource allocation policy generation apparatus 500 and... Figure 1 The methods shown correspond to, for example Figure 5As shown, the resource allocation strategy generation device 500 includes:

[0203] The probability prediction unit 501 is used to predict the probability of user set behavior and probability gain under each resource type based on a sample set representing the relationship between resource allocation results and user behavior, so as to obtain the set of probability of behavior and the set of probability gain corresponding to each resource type.

[0204] Problem model construction unit 502 is used to construct a problem model that represents the maximization of resource allocation gain based on the set of probability of occurrence of each behavior, the set of probability gains, resource allocation decision variables, user set, resource type set, resource value, and resource budget.

[0205] The Lagrange relaxation unit 503 is used to perform Lagrange relaxation on the problem model to obtain the Lagrange dual problem model, and to iteratively solve the Lagrange dual problem model to determine the target Lagrange multiplier;

[0206] The resource allocation strategy generation unit 504 is used to generate a resource allocation strategy based on the target Lagrange multiplier to maximize the resource allocation gain.

[0207] It is evident that implementation Figure 5 The apparatus shown can first acquire a sample set representing the relationship between resource allocation results and user behavior. Based on this sample set, the probability of user behavior and probability gain under each resource type can be predicted. Then, based on the sets of probability occurrences, probability gain sets, resource allocation decision variables, user set, resource type set, resource value, and resource budget, a problem model representing the maximization of resource allocation gain can be constructed. This problem model can be applied to each user in the user set, providing each user with a model to maximize resource allocation gain without requiring user grouping as in related technologies. Furthermore, this problem model can be relaxed to a Lagrange dual problem model. Solving the Lagrange dual problem model yields a more accurate target Lagrange multiplier than directly solving the problem model. This target Lagrange multiplier can guide the generation of a resource allocation strategy, which can provide personalized resource allocation schemes for each user and maximize resource allocation gain. Therefore, it is understood that this application can achieve precise resource allocation. Furthermore, since this application employs Lagrange relaxation, a high-precision target Lagrange multiplier can be obtained. The resource allocation strategy generated based on this target Lagrange multiplier can avoid the impact of data scale on the accuracy of resource allocation.

[0208] In one exemplary embodiment of this application, it further includes:

[0209] The data acquisition unit is used to acquire behavioral datasets and resource allocation records;

[0210] The fully connected unit is used to perform a full database connection operation on the behavior dataset and resource allocation records to obtain an intermediate sample set. The intermediate sample set includes a first class of samples belonging to the resource-assisted behavior type, a second class of samples belonging to the no-resource-assisted behavior type, and a third class of samples belonging to the no-behavior type of allocated resources.

[0211] The sample generation unit is used to generate four types of samples belonging to the no-resource and no-behavior type based on the intermediate sample set.

[0212] The sample set determination unit is used to determine the sample set containing Class I, Class II, Class III, and Class IV samples.

[0213] As can be seen, implementing this optional embodiment can obtain a sample set with more comprehensive sample dimensions. Based on this sample set, it is beneficial to predict a more accurate set of probability of occurrence and probability gain set of behavior, which is conducive to generating a resource allocation strategy that can achieve the purpose of accurate resource allocation.

[0214] In one exemplary embodiment of this application, it further includes:

[0215] The sample augmentation unit is used to augment a sample class if the proportion of a sample class in the intermediate sample set is lower than a preset proportion, based on the behavioral cost and resource acquisition threshold.

[0216] As can be seen, implementing this optional embodiment can solve the problem of sparse sample quantity of a class of samples through sample augmentation. Augmenting sparse samples is beneficial to improving prediction accuracy when predicting the probability of occurrence of behavior and probability gain based on the sample set.

[0217] In one exemplary embodiment of this application, the sample generation unit generates four types of samples belonging to the "no resource, no behavior" type based on an intermediate sample set, including:

[0218] Copy the intermediate sample set to obtain a duplicate sample set;

[0219] Based on the sampling rules, sample users in the replica sample set are sampled to obtain the sample user set;

[0220] Based on the first and second type samples in the replica sample set, and the frequency of behavior corresponding to each user identifier in the sample user set, four types of samples belonging to the no-resource, no-behavior type are constructed in the replica sample set.

[0221] As can be seen, implementing this optional embodiment can achieve the construction of four types of samples without affecting the intermediate sample set, which helps to ensure the accuracy of the intermediate sample set.

[0222] In one exemplary embodiment of this application, the sample set determination unit determines a sample set containing four types of samples: a first type, a second type, a third type, and a fourth type, including:

[0223] Find the union of the duplicate sample set and the intermediate sample set to obtain a sample set containing samples of class 1, class 2, class 3, and class 4.

[0224] As can be seen, implementing this optional embodiment can yield a sample set with high accuracy and availability based on the union of datasets, which is beneficial for generating resource allocation strategies with higher accuracy.

[0225] In one exemplary embodiment of this application, it further includes:

[0226] The sample balancing unit is used to balance the proportions of the sample set if the quantities of samples of class 1, class 2, class 3, and class 4 do not conform to the sample balancing rules.

[0227] As can be seen, by implementing this optional embodiment, sample balance can be achieved within the sample set through sample balancing processing, thereby improving the availability of the sample set.

[0228] In one exemplary embodiment of this application, the probability prediction unit 501 predicts the probability of a user set's behavior and the probability gain for each resource type based on a sample set representing the relationship between resource allocation results and user behavior, thereby obtaining a set of behavior occurrence probabilities and a set of probability gains corresponding to each resource type, including:

[0229] A prediction model is constructed based on a sample set representing the relationship between resource allocation results and user behavior;

[0230] Based on the prediction model, predict the probability of a user's behavior and the probability of no behavior under each resource type in the user set, and obtain the set of probability of behavior and the set of probability of no behavior corresponding to each resource type.

[0231] Based on the set of probability of occurrence of behavior and the set of probability of no behavior occurring for each resource type, determine the probability gain set corresponding to each resource type.

[0232] As can be seen, by implementing this optional embodiment, it is possible to predict the probability gain set and the probability set of behavior occurrence that are beneficial to the construction of the problem model, which can help to obtain a more accurate resource allocation strategy.

[0233] In one exemplary embodiment of this application, the Lagrange relaxation unit 503 performs Lagrange relaxation on the problem model to obtain a Lagrange dual problem model, and iteratively solves the Lagrange dual problem model to determine the target Lagrange multiplier, including:

[0234] By applying Lagrange relaxation to the problem model, a Lagrange dual problem model is obtained.

[0235] The Lagrange dual problem model is simplified into the objective problem model;

[0236] The target problem model is solved iteratively to determine the target Lagrange multiplier.

[0237] As can be seen, implementing this optional embodiment can calculate higher-quality target Lagrange multipliers based on Lagrange relaxation, and improve the solution efficiency of the problem model, which is conducive to the efficient generation of resource allocation strategies.

[0238] In one exemplary embodiment of this application, the resource allocation strategy generation unit 504 generates a resource allocation strategy based on a target Lagrange multiplier to maximize the resource allocation gain, including:

[0239] Use the Lagrange heuristic algorithm to solve for the resource allocation cost containing the target Lagrange multiplier;

[0240] Based on a greedy strategy and resource allocation costs, a resource allocation strategy is generated to maximize the resource allocation gain; wherein, the resource allocation strategy includes resource allocation schemes for each user in the user set.

[0241] As can be seen, implementing this optional embodiment can efficiently and cost-effectively generate resource allocation schemes for each user in a user set based on high-precision target Lagrange multipliers.

[0242] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0243] Since the functional modules of the resource allocation strategy generation apparatus in the example embodiments of this application correspond to the steps of the example embodiments of the resource allocation strategy generation method described above, for details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the resource allocation strategy generation method described above.

[0244] Please see Figure 6 , Figure 6 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown.

[0245] It should be noted that, Figure 6The computer system 600 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0246] like Figure 6 As shown, the computer system 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 602 or programs loaded from storage section 608 into random access memory (RAM) 603. The RAM 603 also stores various programs and data required for system operation. The CPU 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0247] The following components are connected to I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to I / O interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 610 as needed so that computer programs read from it can be installed into storage section 608 as needed.

[0248] Specifically, according to embodiments of this application, the processes described in the above-described flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 609, and / or installed from removable medium 611. When the computer program is executed by central processing unit (CPU) 601, it performs the various functions defined in the methods and apparatus of this application.

[0249] In another aspect, this application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the methods described in the above embodiments.

[0250] It should be noted that the computer-readable medium shown in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0251] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0252] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0253] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

Claims

1. A method for generating a resource allocation strategy, characterized in that, include: Based on a sample set representing the relationship between resource allocation results and user behavior, the probability of user behavior and probability gain for each resource type are predicted, resulting in a set of probability of behavior and a set of probability gain for each resource type. Based on the set of probabilities of each behavior, the set of probability gains, the resource allocation decision variables, the user set, the set of resource types, the resource value, and the resource budget, a problem model is constructed to characterize the maximization of resource allocation gains. Lagrange relaxation is applied to the problem model to obtain the Lagrange dual problem model, and the Lagrange dual problem model is solved iteratively to determine the target Lagrange multiplier; Based on the target Lagrange multiplier, a resource allocation strategy is generated to maximize the resource allocation gain.

2. The method according to claim 1, characterized in that, Also includes: Obtain behavioral datasets and resource allocation records; Perform a full database join operation on the behavior dataset and the resource allocation record to obtain an intermediate sample set; wherein, the intermediate sample set includes a first type of sample belonging to the resource-assisted behavior type, a second type of sample belonging to the no-resource-assisted behavior type, and a third type of sample belonging to the allocated resource no-behavior type. Based on the intermediate sample set, four types of samples belonging to the no-resource, no-behavior type are generated; Determine the sample set that includes the first type of sample, the second type of sample, the third type of sample, and the fourth type of sample.

3. The method according to claim 2, characterized in that, Also includes: If the proportion of a certain type of sample in the intermediate sample set is lower than a preset proportion, then sample augmentation is performed on the certain type of sample based on behavioral cost and resource acquisition threshold.

4. The method according to claim 2, characterized in that, Based on the intermediate sample set, four types of samples belonging to the "no resource, no behavior" type are generated, including: Copy the intermediate sample set to obtain a duplicate sample set; Based on the sampling rules, sample users in the replica sample set are sampled to obtain the sample user set; Based on the first and second type samples in the replica sample set, and the frequency of behavior corresponding to each user identifier in the sample user set, four types of samples belonging to the no-resource, no-behavior type are constructed in the replica sample set.

5. The method according to claim 4, characterized in that, Determining the sample set comprising the first type of samples, the second type of samples, the third type of samples, and the fourth type of samples includes: The union of the replica sample set and the intermediate sample set is calculated to obtain the sample set containing the first type of sample, the second type of sample, the third type of sample, and the fourth type of sample.

6. The method according to claim 2, characterized in that, Also includes: If the number of samples of type I, type II, type III, and type IV does not conform to the sample balance rule, then the sample set is subjected to proportional balance processing.

7. The method according to claim 1, characterized in that, Based on a sample set representing the relationship between resource allocation results and user behavior, the probability of user behavior and probability gain are predicted for each resource type, resulting in a set of behavior probability and a set of probability gains corresponding to each resource type, including: A prediction model is constructed based on a sample set representing the relationship between resource allocation results and user behavior; Based on the prediction model, predict the probability of a user's behavior and the probability of no behavior under each resource type, and obtain the set of probability of behavior and the set of probability of no behavior corresponding to each resource type. Based on the set of probability of occurrence of behavior and the set of probability of no behavior occurring for each resource type, determine the probability gain set corresponding to each resource type.

8. The method according to claim 1, characterized in that, The problem model is subjected to Lagrange relaxation to obtain the Lagrange dual problem model, and the Lagrange dual problem model is solved iteratively to determine the target Lagrange multiplier, including: Lagrange relaxation is applied to the problem model to obtain the Lagrange dual problem model; The Lagrange dual problem model is simplified into the target problem model; The target problem model is solved iteratively to determine the target Lagrange multiplier.

9. The method according to claim 1, characterized in that, Based on the target Lagrange multiplier, a resource allocation strategy is generated to maximize the resource allocation gain, including: The resource allocation cost containing the target Lagrange multiplier is solved by invoking the Lagrange heuristic algorithm. Based on a greedy strategy and resource allocation costs, a resource allocation strategy is generated to maximize the resource allocation gain; wherein, the resource allocation strategy includes a resource allocation scheme for each user in the user set.

10. A resource allocation strategy generation device, characterized in that, include: The probability prediction unit is used to predict the probability of a user set’s behavior and the probability gain under each resource type based on a sample set representing the relationship between resource allocation results and user behavior, thereby obtaining a set of probability of behavior and a set of probability gain corresponding to each resource type. The problem model construction unit is used to construct a problem model that represents the maximization of resource allocation gain based on the set of probability of occurrence of each behavior, the set of probability gains of each behavior, the resource allocation decision variables, the user set, the set of resource types, the resource value, and the resource budget. A Lagrange relaxation unit is used to perform Lagrange relaxation on the problem model to obtain a Lagrange dual problem model, and to iteratively solve the Lagrange dual problem model to determine the target Lagrange multiplier; The resource allocation strategy generation unit is used to generate a resource allocation strategy based on the target Lagrange multiplier to maximize the resource allocation gain.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-9.

12. An electronic device, characterized in that, include: processor; as well as Memory for storing the executable instructions of the processor; The processor is configured to execute the method of any one of claims 1-9 by executing the executable instructions.

Images