CN115049433A

CN115049433A - Model training method and device, storage medium and electronic equipment

Info

Publication number: CN115049433A
Application number: CN202210676916.3A
Authority: CN
Inventors: 唐波; 余东亮; 谢乾龙; 王兴星
Original assignee: Beijing Sankuai Online Technology Co Ltd
Current assignee: Beijing Sankuai Online Technology Co Ltd
Priority date: 2022-06-15
Filing date: 2022-06-15
Publication date: 2022-09-13

Abstract

In the model training method provided by the present specification, the obtained sample user information, sample environment information, and user resource amount are input into each first model corresponding to a service, and the resource amount to be optimized allocated to each service by the sample user is obtained; determining the rewards obtained by the sample users in the services under the condition that the resource amount distributed by the sample users to the services is the resource amount to be optimized corresponding to each service; and training the first models simultaneously by taking the maximum sum of the rewards obtained by the sample users under each business as an optimization target and the constraint that the sum of the resource amount distributed by the sample users to each business is not more than the resource amount of the users. The model trained by the model training method provided by the specification can simultaneously consider all services required by the user under the constraint condition that the resource quantity of the user is limited, and provides a resource allocation scheme which enables the comprehensive benefits of the user in all services to be the highest, thereby effectively solving the problem that the user cannot obtain the globally optimal resource allocation scheme.

Description

Model training method and device, storage medium and electronic equipment

Technical Field

The present disclosure relates to the field of machine learning, and in particular, to a model training method, apparatus, storage medium, and electronic device.

Background

Today, the internet is inseparable from people's lives, and various platforms or software can provide a variety of services to users via the internet. Such as instant distribution, information recommendation, etc. At present, the number of users on the internet is huge, and the number of services that each platform can provide for the users is limited. Therefore, users often need to pay certain resources to be able to use services, and when there are a plurality of users who wish to use the same service at the same time, the platform will generally provide services to the user who pays the highest resources preferentially.

However, the resources of the user are also limited. Therefore, it is critical for users how to reasonably allocate resources to obtain the optimal service. Currently, most users use a method for allocating resources to calculate the number of resources that can bring the highest reward to the user for each service provided by the user using a preset model, and the calculated number is used as the resource allocated to the service. And the platform provides the service for the user with the highest paid resource and deducts the corresponding resource of the user.

However, this method for allocating resources for a single service has great limitations. Generally, a user needs multiple services, and the existing method does not consider how to allocate resources on multiple services simultaneously, so that the user can achieve global profit maximization on multiple services.

Disclosure of Invention

The present specification provides a model training method, apparatus, storage medium, and electronic device to at least partially solve the above problems in the prior art.

The technical scheme adopted by the specification is as follows:

the present specification provides a model training method, comprising:

acquiring sample user information, sample environment information and user resource amount of the sample user;

for each service required by the sample user, inputting the sample user information, the sample environment information and the user resource amount into a first model to be trained corresponding to the service, so as to determine the resource amount to be optimized, which is allocated to the service by the sample user, through the first model;

when the resource amount distributed to each service by the sample user is the resource amount to be optimized corresponding to each service, determining the reward obtained by the sample user under each service;

and training each first model by taking the maximum sum of the rewards obtained by the sample user under each business as an optimization target and the constraint that the sum of the resource quantity to be optimized distributed by the sample user to each business is not more than the user resource quantity.

Optionally, for each service required by the sample user, the sample user information, the sample environment information, and the user resource amount are input into a first model to be trained corresponding to the service, so as to determine, through the first model, a resource amount to be optimized allocated to the service by the sample user, specifically including:

and inputting the sample user information, the sample environment information and the user resource amount of the sample user when the sample user uses the service at this time into a first model to be trained corresponding to the service aiming at each service required by the sample user for using each time, so as to determine the resource amount to be optimized distributed to the service when the sample user uses the service at this time through the first model.

Optionally, the obtaining of the user resource amount of the sample user specifically includes:

determining a plurality of preset periods and the total amount of period resources of the periods;

determining the period of the current time as the current period;

and inputting the sample user information, the sample environment information and the total amount of the periodic resources into a pre-trained second model, and determining the amount of the periodic resources allocated to the current period by the sample user through the second model to be used as the user resource amount of the sample user.

determining a sample user group where the sample user is located;

acquiring information of the sample user group and total resource amount of the sample user group;

inputting the sample user information, the sample user group information and the total resource amount into a pre-trained third model to determine the user resource amount of the sample user through the third model.

Optionally, the pre-training of the second model specifically includes:

acquiring sample user information and sample environment information; determining a plurality of preset periods and the total amount of period resources of the periods;

determining the period of the current time as the current period;

inputting the sample user information, the sample environment information and the total cycle resource amount into a second model to be trained, so as to determine the cycle resource amount to be optimized, which is distributed to the current cycle by the sample user, through the second model;

determining the reward of the sample user in the current period under the condition that the resource amount distributed by the sample user to the current period is the resource amount of the period to be optimized;

and training the second model by taking the maximum sum of the rewards in each period as an optimization target and taking the constraint that the sum of the resource amount of the period to be optimized corresponding to each period is not more than the total resource amount of the period.

Optionally, the training the third model in advance specifically includes:

obtaining sample user information, and determining a sample user group where the sample user is located;

inputting the sample user information, the information of the sample user group and the total resource amount into a third model to be trained so as to determine the user resource amount to be optimized of the sample user through the third model;

determining the rewards obtained by the sample users under the condition that the user resource amount of the sample users is the user resource amount to be optimized;

and training the third model by taking the maximum sum of the rewards obtained by each sample user in the sample user group as an optimization target and the constraint that the sum of the resource amount of the user to be optimized of each sample user in the sample user group is not more than the total resource amount.

The present specification provides a service providing method, including:

acquiring user information, environment information and user resource amount of the user;

acquiring each first model corresponding to each service according to the user information;

for each service required by the user, inputting the user information, the environment information and the user resource amount into a first model corresponding to the service to determine the resource amount allocated to the service by the user through the first model, wherein the first model is obtained according to a model training method provided by the specification;

and aiming at each service, determining a user for distributing the resource amount to the service, selecting a target user according to the determined resource amount distributed to the service by each user, providing the service to the target user, and deducting the resource amount distributed to the service by the target user.

The present specification provides a model training apparatus, the apparatus comprising:

the acquisition module is used for acquiring sample user information, sample environment information and user resource amount of the sample user;

the resource quantity determining module inputs the sample user information, the sample environment information and the user resource quantity into a first model to be trained corresponding to the service aiming at each service required by the sample user so as to determine the resource quantity to be optimized distributed to the service by the sample user through the first model;

the reward determining module is used for determining the reward obtained by the sample user under each business when the resource amount distributed by the sample user to each business is the resource amount to be optimized corresponding to each business;

and the first training module is used for training each first model by taking the maximum sum of the rewards obtained by the sample user under each business as an optimization target and taking the constraint that the sum of the resource quantity to be optimized distributed by the sample user to each business is not more than the resource quantity of the user.

A service providing apparatus provided in this specification, the apparatus includes:

the information acquisition module is used for acquiring user information, environment information and user resource amount of the user;

the model acquisition module is used for acquiring each first model corresponding to each service according to the user information;

the allocation module is used for inputting the user information, the environment information and the user resource quantity into a first model corresponding to each service required by the user so as to determine the resource quantity allocated to the service by the user through the first model, wherein the first model is obtained according to a model training method provided by the specification;

and the execution module determines a user for distributing the resource amount to the service aiming at each service, selects a target user according to the determined resource amount distributed to the service by each user, provides the service for the target user and deducts the resource amount distributed to the service by the target user.

The present specification provides a computer-readable storage medium storing a computer program which, when executed by a processor, implements the above-described model training method.

The present specification provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the above model training method when executing the program.

The technical scheme adopted by the specification can achieve the following beneficial effects:

in the model training method provided by the specification, the obtained sample user information, sample environment information and user resource amount are input into each first model corresponding to the service, and the resource amount to be optimized distributed to each service by the sample user is obtained; determining the rewards obtained by the sample users in the services under the condition that the resource amount distributed by the sample users to the services is the resource amount to be optimized corresponding to each service; and training the first models simultaneously by taking the maximum sum of the rewards acquired by the sample users under each business as an optimization target and the constraint that the sum of the resource amount distributed by the sample users to each business is not more than the resource amount of the users. The model trained by the model training method provided by the specification can simultaneously consider all services required by the user under the constraint condition that the resource quantity of the user is limited, and provides a resource allocation scheme which enables the comprehensive benefits of the user in all services to be the highest, thereby effectively solving the problem that the user cannot obtain the globally optimal resource allocation scheme.

Drawings

The accompanying drawings, which are included to provide a further understanding of the specification and are incorporated in and constitute a part of this specification, illustrate embodiments of the specification and together with the description serve to explain the specification and not to limit the specification in a non-limiting sense. In the drawings:

FIG. 1 is a schematic flow chart of a model training method in the present specification;

fig. 2 is a schematic flow chart of a service providing method in this specification;

FIG. 3 is a schematic diagram of a model training apparatus provided herein;

fig. 4 is a schematic diagram of a service providing apparatus provided in the present specification;

fig. 5 is a schematic diagram of an electronic device corresponding to fig. 1 provided in the present specification.

Detailed Description

At present, the development of various platforms on the internet is mature, and various services can be provided for users, and the life of the users is improved in all aspects, so that more and more users select to use the services of the internet platform. As such, the influx of a large number of users leads to a shortage of service supplies provided by the platform, and the platform cannot meet each user, so that most internet platforms introduce a scheme for providing services to users based on resource amount.

In this scheme, each user would provide several algorithmic models to the platform in advance, each model being applied to one service. When a plurality of users need to use a service at the same time, the platform determines the resource amount paid by each user for using the service this time according to the model corresponding to the service provided by the user, provides the service to the user with the highest paid resource amount, and deducts the corresponding resource amount of the user.

Most of models used by users at present determine the payment resource amount with the highest reward return rate aiming at the services which the users want to use at present. But in most cases, the user will need multiple services. Under the condition that the resource amount of a user is limited, the existing model only considers the return rate of single-use service, which cannot bring the highest global benefit to the user.

In order to solve the above problems, the present specification provides a model training method, which can train a model for performing reasonable global resource allocation on all services required by a user.

In order to make the objects, technical solutions and advantages of the present disclosure more clear, the technical solutions of the present disclosure will be clearly and completely described below with reference to the specific embodiments of the present disclosure and the accompanying drawings. It is to be understood that the embodiments described are only a few embodiments of the present disclosure, and not all embodiments. All other embodiments obtained by a person skilled in the art based on the embodiments in the present specification without any inventive step are within the scope of the present application.

The technical solutions provided by the embodiments of the present description are described in detail below with reference to the accompanying drawings.

Fig. 1 is a schematic flow chart of a model training method in this specification, which specifically includes the following steps:

s100: and acquiring sample user information, sample environment information and user resource amount of the sample user.

All steps in the model training method provided in the present specification can be implemented by any electronic device with computing function, such as a terminal, a server, and the like.

In this step, the sample user is a user selected from users related to the service, and the sample user may include a user who has allocated a resource amount to a service and successfully used the service; or may include users who do not use a service after allocating an amount of resources to the service. The user or sample user mentioned in this specification may mean not only an individual who uses the service but also a group or organization that uses the service, such as a school, a company, various stores, and the like. Therefore, the user information mentioned in the present specification may contain different contents according to the user type. For example, when the user is a person, the user information may be a representation of the user, including information about the user's gender, age, occupation, etc.; when the user is an organization, the user information includes information such as the type, age, and operation status of the organization.

It should be noted that all actions related to user information in this specification, such as operations of obtaining and using user information, are legal and are only executed after user's approval.

The service mentioned in this specification may be any available service that the platform provides to the user, such as instant delivery, online taxi appointment, information recommendation, etc. The specific content of the resource may vary from service to service, but in general, the user may pay for the same resource when using many different services, such as more general currency. The amount of user resources represents the amount of resources owned by the user. The sample environment information mentioned in this specification may be information of an environment and a state where the user is located when requesting to use the service, and may generally include information of a time, a place, and the like where the user is located, and information related to the service when aiming at different services, for example, when the sample user is a store and the service is information recommendation, the sample environment information may further include information of a consuming capability of a resident near the store and the like.

S102: and inputting the sample user information, the sample environment information and the user resource quantity into a first model to be trained corresponding to the service aiming at each service required by the sample user, so as to determine the resource quantity to be optimized distributed to the service by the sample user through the first model.

In the model training method provided by the specification, a plurality of first models to be trained exist, the specific number may be not less than the number of all services required by a user, the internal structures of the first models are the same, and the parameters may be different. And inputting the sample user information, the sample environment information and the user resource amount obtained in the step S100 into each first model corresponding to each service, and obtaining the resource amount to be optimized distributed to each service by the sample user through the first models. Wherein, the sample information input into each first model can contain different contents.

S104: and when the resource amount distributed to each service by the sample user is the resource amount to be optimized corresponding to each service, determining the reward obtained by the sample user under each service.

And taking the resource quantity to be optimized distributed to each service determined by the first model in the step S102 as the resource quantity distributed to each service by the sample user. That is, for each service required by the sample user, the amount of the resource to be optimized allocated to the service determined by the first model is used as the amount of the resource allocated to the service by the sample user, and the reward obtained by the user in the service in this case is determined.

The reward obtained by the user may vary from service to service, for example, when the service is an instant delivery service, the reward may be the time taken to deliver to the user; when the business is an information recommendation service, the reward may be a return on investment ratio. It is worth mentioning that when a user allocates resources to a service but fails to use the service, the resources allocated to the user are not deducted, and there is no incentive for the user to pay in the service.

S106: and training each first model by taking the maximum sum of the rewards obtained by the sample user under each business as an optimization target and the constraint that the sum of the resource quantity to be optimized distributed by the sample user to each business is not more than the user resource quantity.

The goal that the model training method provided by the specification is expected to achieve is that the user can obtain the benefit of the global maximization, and therefore the optimization goal in training the model is that the sum of the rewards obtained by the sample user under each business is the maximum. Meanwhile, the amount of resources that can be consumed by the user is limited, so that each first model is trained under the constraint that the sum of the amounts of resources to be optimized, which are allocated to each service by the sample user, is not greater than the amount of the user resources during training. It should be noted that in the model training method provided in this specification, there are a plurality of first models that need to be trained, and all the first models are trained simultaneously during training.

For the constrained optimization objectives in the first model, a markov decision process may be used for representation, and for each first model, the optimization objective may be specifically represented as follows:

wherein m represents the m-th model, J _m Represents the maximum reward that the m model can obtain corresponding to the business, E represents the expectation, r _m Representing the reward, s, earned by the service corresponding to the mth model _m State information representing the m-th model, i.e., user information and environment information, s represents input sample information and environment information, g _m Representing the amount of resources that can be consumed by the service corresponding to the m-th model, g representing the amount of user resources that are input, a _m Represents the action of the m-th model, i.e. the amount of resources allocated to the m-th service, and a _m Is according to s _m Obtained, pi _m The decision made for the m-th model, i.e. the parameters internal to the m-th model. The meaning of the above formula is that under the condition of the input sample user information, sample environment information and user resource amount, the reward which can make the service obtain is found to be the mostParameters of the large first model.

While for all first models, a common optimization objective can be represented by the following formula:

Q＝ΣJ _m

where Q represents the maximum total reward that can be obtained for all services required by the user.

By adopting the model training method provided by the specification, the first models corresponding to various services can be trained simultaneously in a reinforcement learning mode. And under the condition that the sum of the resource amount distributed to each service by the model does not exceed the resource amount of the user, obtaining the reward of the user under each service by taking the resource amount to be optimized distributed to each service obtained by the model as the resource amount paid out to each service by the user, and adjusting the parameters of the model according to the reward sum obtained by the user under each service until the reward sum obtained by the user under each service is maximum. By the method, the resource allocation scheme with the highest global benefit can be determined for the user according to the limited resource amount of the user, the defect that the service benefit can only be considered once when the resource is allocated by the existing model is effectively overcome, and the problem that the user cannot obtain the highest benefit under various services is solved.

Sometimes, each service is used more than once by the user, and therefore, the model needs to consider the situation that the user uses each service multiple times when allocating resources for the user. Specifically, for each service required by the sample user for using the service each time, the sample user information, the sample environment information, and the user resource amount of the sample user when using the service this time may be input into the first model to be trained corresponding to the service, so as to determine, through the first model, the resource amount to be optimized allocated to the service when the sample user uses the service this time.

In the above case, the time when the user uses each service may be different, and thus, when training each first model, the time when each first model corresponding to each service inputs sample data may be different. In order to ensure that all the first models can be trained simultaneously, when the first model corresponding to any service receives new sample data, other first models can adopt the sample data received last time to train simultaneously, so that the training of all the first models can be ensured to be synchronous.

For example, the sample user uses a network car booking service at nine points and an instant delivery service at eleven points. At eleven o 'clock, the first model corresponding to the instant delivery service may receive new sample data, and the first model corresponding to the network appointment service may reuse the sample data received at nine o' clock and train with the first model corresponding to the instant delivery service. It should be noted that each first model has initialized data for synchronous training with other models when the first model does not receive any sample data.

In fact, it can be found from historical data and experience that many users have a certain periodicity when using services provided by the platform. That is, one user uses several services provided by the platform in this period, and will repeat the services used in this period in the next period. However, although the service used by the user in each period is the same, the amount of resources the user pays in each period and the prize won may be completely different due to the change of the environment. For example, when the user is an advertiser and uses the information recommendation service to place advertisements on the platform, the amount of user resources may be the cost of placing the advertisements. Advertisers typically keep placing advertisements for a period of time, such as a week or a month, i.e., reuse the information recommendation service. The day can be used as a period, and the cost of placing the advertisement consumed each day is different because the access amount of the platform is different each day and the influence of competition of the advertisement space with other advertisers exists.

Thus, the amount of resources that the user needs to pay in each period can be periodically predicted. Specifically, the total amount of the cycle resources of a plurality of preset cycles and a plurality of cycles can be determined; determining the period of the current time as the current period; and inputting the sample user information, the sample environment information and the total amount of the periodic resources into a pre-trained second model, and determining the amount of the periodic resources allocated to the current period by the sample user through the second model to be used as the user resource amount of the sample user. Wherein, the total amount of the periodic resources is the total amount of the resources which can be consumed by the user in a plurality of periods. According to different sample environment information at different time, the determined available cycle resource amount in different cycles may be different.

The second model used in the above process is pre-trained, and the method for pre-training the second model specifically can be used for obtaining sample user information and sample environment information; determining a plurality of preset periods and the total amount of period resources of the periods; determining the period of the current time as the current period; inputting the sample user information, the sample environment information and the total cycle resource amount into a second model to be trained, so as to determine the cycle resource amount to be optimized, which is distributed to the current cycle by the sample user, through the second model; determining the reward of the sample user in the current period under the condition that the resource amount distributed by the sample user to the current period is the resource amount of the period to be optimized; and training the second model by taking the maximum sum of the rewards in each period as an optimization target and taking the constraint that the sum of the resource amount of the period to be optimized corresponding to each period is not more than the total resource amount of the period.

And in the process of training the second model, the resource amount of the cycle to be optimized, which is determined by the second model and is allocated to the current cycle for the sample user, is determined. For each period, assuming that the resource amount allocated by the sample user for the current period is the resource amount of the period to be optimized determined by the second model, and determining the reward obtained by the sample user in the current period under the assumption. At this time, for the sample user, the global optimal benefit is that the sum of the rewards obtained in each period is the maximum, so that the global optimal benefit can be taken as an optimization target; meanwhile, the sum of the resource amount of the cycle to be optimized corresponding to each cycle should not exceed the total amount of the cycle resources, so that the second model can be trained by taking the sum as a constraint.

Similarly, the constrained optimization target in the second model may also be represented by a markov decision process, and the optimization target may be specifically represented by the following formula:

wherein, J _T Representing the maximum sum of the rewards available for each period, Π representing the decision of the second model, i.e. the parameters of the second model, s _t State information indicating the t-th period, i.e., user information and environment information, s indicates input sample user information and sample environment information, g _t Indicating the amount of resources that can be consumed in the t-th cycle, g indicating the total amount of cycle resources input, a _t Representing the action taken by the second model, i.e. the amount of periodic resources to be optimized, allocated to the t-th period, a _t Is the second model according to s _t Obtaining; gamma ray _t The weighting of the t-th period is represented, and the actual effects reflected by the awards with the same size under different environments are different, so that a preset weighting can be added to each period to balance the actual effects reflected by the awards under each period. The meaning of the above formula is to find the parameter of the second model that can maximize the sum of the rewards of each period under the condition of the input sample user information, the sample environment information, the total amount of the period resources and the amount of the period resources to be optimized distributed to each period obtained according to the state information.

In addition, sometimes, users may be in a user group, and each user in the user group may wish to bring maximum benefit to the user group when using the service. By definition of users in this specification, the user group in this specification does not refer to a group consisting of a plurality of individuals using services, but may refer to a large-scale business or organization, for example, the user group may refer to fast food brands such as kendirk, mcdonald's labor, and the users in the user group may be a plurality of shops around the fast food brand.

Thus, the amount of user resources for each user within the user group may be limited by the total amount of resources owned by the user group. At this time, the resource amount of the user may be determined according to the total resource amount of the user group. Specifically, a sample user group where the sample user is located can be determined; acquiring information of the sample user group and total resource amount of the sample user group; inputting the sample user information, the sample user group information and the total resource amount into a pre-trained third model to determine the user resource amount of the sample user through the third model.

Similarly, the third model used in the above process is also pre-trained, and the method for pre-training the third model specifically may be to obtain sample user information and determine a sample user group where the sample user is located; acquiring information of the sample user group and total resource amount of the sample user group; inputting the sample user information, the information of the sample user group and the total resource amount into a third model to be trained so as to determine the user resource amount to be optimized of the sample user through the third model; determining the rewards obtained by the sample users under the condition that the user resource amount of the sample users is the user resource amount to be optimized; and training the third model by taking the maximum sum of the rewards obtained by each sample user in the sample user group as an optimization target and the constraint that the sum of the resource amount of the user to be optimized of each sample user in the sample user group is not more than the total resource amount.

In the process of training the third model, the third model determines the amount of user resources to be optimized for each sample user in the sample user group. And assuming that the user resource amount of the sample user is the user resource amount to be optimized determined by the third model, and determining the reward available to the sample user under the assumption. At this time, for the sample users in the sample user group, the globally optimal profit is the maximum profit of the sample user group, that is, the sum of the rewards obtained by each sample user in the sample user group is the maximum, so that the global optimal profit can be taken as an optimization target; meanwhile, the sum of the resource amounts of the users to be optimized corresponding to each sample user should not exceed the total resource amount of the sample user group, so that the third model can be trained by taking the sum as a constraint.

Similarly, the constrained optimization target in the third model may also be represented by a markov decision process, and the optimization target may be specifically represented by the following formula:

wherein, J _K Representing the maximum value of the reward that a sample group of users can obtain, Π representing the decision of the third model, i.e. the parameters of the third model, s _k State information indicating the kth sample user, i.e., user information and environment information, s indicates the input sample user information and sample environment information, g _k Denotes the amount of resources that can be consumed in the k-th cycle, g denotes the total amount of resources of the input sample user group, a _k Representing the actions taken by the third model, i.e. the amount of user resources to be optimized allocated to the kth sample user, a _k Is the third model according to s _k Obtained of gamma _k Representing the weight of the kth sample user. The meaning of the formula is that the parameter of the third model which can maximize the reward of the sample user group is searched under the condition of the input sample user information, the sample environment information, the total resource amount and the user resource amount to be optimized which is obtained according to the state information and is distributed to each sample user.

In addition, besides the resource amount as a constraint mentioned in the specification, other constraints can exist, and the pertinence of the model is further strengthened. For example, a constraint may also be that the return on investment is greater than a specified threshold, which may be arbitrarily set. Besides the general use of resource amount, return on investment and the like, the method can be applied to the constraint of most services, and can also have different and unique constraints aiming at different services. For example, when the service is delivered instantly, the constraint may be that the delivery duration is less than a specified duration; when the service is information recommendation, the constraint can be that the cost of thousands of people is less than the specified cost, and the like.

The present specification further provides a service providing method corresponding to the model training method, and fig. 2 is a schematic flow chart of the service providing method provided by the present specification, which specifically includes the following steps:

s200: acquiring user information, environment information and user resource amount of the user;

all the steps in the service providing method provided in this specification can be implemented by any electronic device with a computing function, such as a terminal, a server, and the like.

The specific implementation method of this step is similar to step S100, and this description is not repeated here.

S202: acquiring each first model corresponding to each service according to the user information;

as mentioned above, each user provides a model for determining the amount of allocated resources in advance before using the services, and there is a corresponding model for each service. Therefore, it is necessary to determine each first model corresponding to each service provided by the user according to the information of the user.

S204: inputting the user information, the environment information and the user resource quantity into a first model corresponding to each service required by the user, so as to determine the resource quantity allocated to the service by the user through the first model, wherein the first model is obtained according to any method shown in fig. 1;

the first model employed in the method may be a first model trained according to the model training method shown in fig. 1. The user information, the environment information, and the user resource amount obtained in step S200 are input into each trained first model, so that the resource amount allocated to each service by the user can be obtained through each first model.

S206: and aiming at each service, determining a user for distributing the resource amount to the service, selecting a target user according to the determined resource amount distributed to the service by each user, providing the service to the target user, and deducting the resource amount distributed to the service by the target user.

When providing a service to a user, the user to which the resource amount is allocated to the service may be determined first, and a target user may be selected according to the determined resource amount allocated to the service by each user. Specifically, the first several users with the highest resource amount allocated to the service may be determined as target users, and the number of the target users may be determined according to specific service content. After the target user is determined, the service can be provided to the target user, and the resource amount allocated to the service by the target user is correspondingly deducted.

Based on the same idea, the model training method provided by the present specification further provides a corresponding model training device, as shown in fig. 3.

Fig. 3 is a schematic diagram of a model training apparatus provided in this specification, which specifically includes:

the sample acquisition module 300 is used for acquiring sample user information, sample environment information and user resource amount of the sample user;

a resource amount determining module 302, configured to, for each service required by the sample user, input the sample user information, the sample environment information, and the user resource amount into a first model to be trained corresponding to the service, so as to determine, through the first model, an amount of resource to be optimized allocated to the service by the sample user;

the reward determining module 304 is used for determining the reward obtained by the sample user under each business when the resource amount distributed by the sample user to each business is the resource amount to be optimized corresponding to each business;

the first training module 306 trains each first model by taking the maximum sum of the rewards obtained by the sample user under each business as an optimization target and taking the constraint that the sum of the resource amount to be optimized distributed by the sample user to each business is not more than the user resource amount.

In an alternative embodiment:

the resource amount determining module 302 is specifically configured to, for each service required by the sample user for using the service each time, input sample user information, sample environment information, and user resource amount of the sample user when using the service this time into a first model to be trained corresponding to the service, so as to determine, through the first model, an amount of resource to be optimized allocated to the service when using the service this time by the sample user.

In an alternative embodiment:

the acquired sample acquiring module 300 is specifically configured to determine a plurality of preset periods and a total amount of period resources of the plurality of periods; determining the period of the current moment as the current period; and inputting the sample user information, the sample environment information and the total amount of the periodic resources into a pre-trained second model, and determining the amount of the periodic resources allocated to the current period by the sample user through the second model to be used as the user resource amount of the sample user.

In an alternative embodiment:

the acquired sample acquiring module 300 is specifically configured to determine a sample user group where the sample user is located; acquiring information of the sample user group and total resource amount of the sample user group; inputting the sample user information, the sample user group information and the total resource amount into a pre-trained third model to determine the user resource amount of the sample user through the third model.

In an alternative embodiment:

the apparatus further includes a second training module 308, specifically configured to obtain sample user information and sample environment information; determining a plurality of preset periods and the total amount of period resources of the periods; determining the period of the current time as the current period; inputting the sample user information, the sample environment information and the total cycle resource amount into a second model to be trained, so as to determine the cycle resource amount to be optimized, which is distributed to the current cycle by the sample user, through the second model; determining the reward of the sample user in the current period under the condition that the resource amount distributed by the sample user to the current period is the resource amount of the period to be optimized; and training the second model by taking the maximum sum of the rewards in each period as an optimization target and taking the constraint that the sum of the resource amount of the period to be optimized corresponding to each period is not more than the total resource amount of the period.

In an alternative embodiment:

the device further comprises a third training module 310, which is specifically configured to obtain sample user information and determine a sample user group where the sample user is located; acquiring information of the sample user group and total resource amount of the sample user group; inputting the sample user information, the information of the sample user group and the total resource amount into a third model to be trained so as to determine the user resource amount to be optimized of the sample user through the third model; determining the rewards obtained by the sample users under the condition that the user resource amount of the sample users is the user resource amount to be optimized; and training the third model by taking the maximum sum of the rewards obtained by each sample user in the sample user group as an optimization target and the constraint that the sum of the resource amount of the user to be optimized of each sample user in the sample user group is not more than the total resource amount.

Based on the same idea, the service providing method provided by the present specification further provides a corresponding service providing apparatus, as shown in fig. 4.

Fig. 4 is a schematic diagram of a service providing apparatus provided in this specification, which specifically includes:

an information obtaining module 400, for obtaining user information, environment information, and user resource amount of the user;

a model obtaining module 402, configured to obtain, according to the user information, each first model corresponding to each service;

an allocating module 404, configured to input, for each service required by the user, the user information, the environment information, and the user resource amount into a first model corresponding to the service, so as to determine, through the first model, a resource amount allocated to the service by the user, where the first model is obtained according to the method shown in fig. 1;

the executing module 406 determines, for each service, a user that allocates a resource amount to the service, selects a target user according to the determined resource amount allocated to the service by each user, provides the service to the target user, and deducts the resource amount allocated to the service by the target user.

The present specification also provides a computer-readable storage medium having stored thereon a computer program, the computer program being operable to execute the model training method provided in fig. 1 above.

This specification also provides a schematic block diagram of the electronic device shown in fig. 5. As shown in fig. 5, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, but may also include hardware required for other services. The processor reads a corresponding computer program from the non-volatile memory into the memory and then runs the computer program to implement the model training method described in fig. 1 above. Of course, besides the software implementation, the present specification does not exclude other implementations, such as logic devices or a combination of software and hardware, and the like, that is, the execution subject of the following processing flow is not limited to each logic unit, and may be hardware or logic devices.

It should be noted that all actions of acquiring signals, information or data in the present application are performed under the premise of complying with the corresponding data protection regulation policy of the country of the location and obtaining the authorization given by the owner of the corresponding device.

In the 90 s of the 20 th century, improvements in a technology could clearly distinguish between improvements in hardware (e.g., improvements in circuit structures such as diodes, transistors, switches, etc.) and improvements in software (improvements in process flow). However, as technology advances, many of today's process flow improvements have been seen as direct improvements in hardware circuit architecture. Designers almost always obtain the corresponding hardware circuit structure by programming an improved method flow into the hardware circuit. Thus, it cannot be said that an improvement in the process flow cannot be realized by hardware physical modules. For example, a Programmable Logic Device (PLD), such as a Field Programmable Gate Array (FPGA), is an integrated circuit whose Logic functions are determined by programming the Device by a user. A digital system is "integrated" on a PLD by the designer's own programming without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Furthermore, nowadays, instead of manually making an Integrated Circuit chip, such Programming is often implemented by "logic compiler" software, which is similar to a software compiler used in program development and writing, but the original code before compiling is also written by a specific Programming Language, which is called Hardware Description Language (HDL), and HDL is not only one but many, such as abel (advanced Boolean Expression Language), ahdl (alternate Hardware Description Language), traffic, pl (core universal Programming Language), HDCal (jhdware Description Language), lang, Lola, HDL, laspam, hardward Description Language (vhr Description Language), vhal (Hardware Description Language), and vhigh-Language, which are currently used in most common. It will also be apparent to those skilled in the art that hardware circuitry that implements the logical method flows can be readily obtained by merely slightly programming the method flows into an integrated circuit using the hardware description languages described above.

The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic for the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller as pure computer readable program code, the same functionality can be implemented by logically programming method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Such a controller may thus be regarded as a hardware component and the means for performing the various functions included therein may also be regarded as structures within the hardware component. Or even means for performing the functions may be regarded as being both a software module for performing the method and a structure within a hardware component.

The systems, devices, modules or units illustrated in the above embodiments may be implemented by a computer chip or an entity, or by a product with certain functions. One typical implementation device is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functions of the various elements may be implemented in the same one or more software and/or hardware implementations of the present description.

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.

Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.

It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

As will be appreciated by one skilled in the art, embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

This description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.

All the embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.

The above description is only an example of the present specification, and is not intended to limit the present specification. Various modifications and alterations to this description will become apparent to those skilled in the art. Any modification, equivalent replacement, improvement or the like made within the spirit and principle of the present specification should be included in the scope of the claims of the present application.

Claims

1. A method of model training, comprising:

2. The method according to claim 1, wherein for each service required by the sample user, inputting the sample user information, the sample environment information, and the user resource amount into a first model to be trained corresponding to the service, so as to determine, through the first model, an amount of resource to be optimized allocated to the service by the sample user, specifically comprising:

3. The method of claim 1, wherein obtaining the user resource amount of the sample user specifically comprises:

determining the period of the current time as the current period;

4. The method of claim 1, wherein obtaining the user resource amount of the sample user specifically comprises:

determining a sample user group where the sample user is located;

5. The method of claim 3, wherein pre-training the second model comprises:

determining the period of the current time as the current period;

6. The method of claim 4, wherein pre-training the third model specifically comprises:

7. A service providing method is characterized by comprising the following steps:

inputting the user information, the environment information and the user resource quantity into a first model corresponding to each service required by the user so as to determine the resource quantity allocated to the service by the user through the first model, wherein the first model is obtained according to any one of the methods in claims 1-6;

8. A model training apparatus, comprising:

the system comprises a sample acquisition module, a data processing module and a data processing module, wherein the sample acquisition module is used for acquiring sample user information, sample environment information and user resource quantity of a sample user;

the resource quantity determining module is used for inputting the sample user information, the sample environment information and the user resource quantity into a first model to be trained corresponding to the service aiming at each service required by the sample user so as to determine the resource quantity to be optimized distributed to the service by the sample user through the first model;

9. A service providing apparatus, comprising:

an allocation module, which inputs the user information, the environment information and the user resource amount into a first model corresponding to each service required by the user, so as to determine the resource amount allocated to the service by the user through the first model, wherein the first model is obtained according to any method of claims 1-6;

10. A computer-readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the method of any of the preceding claims 1 to 7.

11. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any of claims 1 to 7 when executing the program.