An order receiving end quota allocation method, device, equipment and storage medium

Abstract

The application relates to the technical field of Internet, in particular to a method and device for allocating quotas for order-accepting ends, equipment and a storage medium, which are used to solve the problem that a traditional quota allocation scheme is not conducive to enhancing the stickiness between a platform and order-accepting ends. The method comprises the following steps: quantifying the behavior of an order-accepting end under a target quota based on historical behavior data of the order-accepting end, obtaining a first reward value of the order-accepting end under the target quota, obtaining a UCB value of the order-accepting end under the target quota by using a UCB algorithm based on the first reward value of the order-accepting end under the target quota, a second reward value of order-accepting ends in a cluster to which the order-accepting end belongs under the target quota, and a third reward value of order-accepting ends in the platform under the target quota, and determining a cancellation quota of the order-accepting end in a current allocation period based on the UCB values under multiple cancellation quotas; and the method can realize accurate, stable and interpretable quota allocation and enhance the stickiness between the platform and the order-accepting ends.

Description

Technical Field

[0001] This application relates to the field of Internet technology, and in particular to a method, apparatus, device and storage medium for allocating credit limits on an order receiving terminal. Background Technology

[0002] In ride-hailing operations, drivers inevitably encounter scenarios where orders need to be cancelled. To retain drivers and enhance the stickiness between the platform and drivers, ride-hailing platforms typically provide each driver with a certain amount of cancellation credit without penalty. In traditional credit allocation scenarios, static driver stratification and the traditional Upper Confidence Bound (UCB) algorithm are often used to allocate cancellation credit to drivers. However, in actual operation, it has been found that both methods have certain drawbacks: For example, static driver stratification divides drivers based on historical static metrics and sets a fixed cancellation limit for each stratum. This stratification method cannot reflect real-time changes in driver behavior. When driver behavior changes abruptly, such as when they suddenly start maliciously cancelling or frequently making calls, the platform's response is delayed, which is not conducive to enhancing the stickiness between the platform and drivers. Another example is that when the traditional UCB algorithm reduces the cancellation limit to lower the cancellation rate, it will cause driver dissatisfaction. Moreover, for new drivers without historical behavior data, the algorithm has a high cancellation rate during the exploration period, which reduces passenger experience and driver retention. Summary of the Invention

[0003] This application provides a method, apparatus, device, and storage medium for allocating credit limits to order-receiving terminals, which can reduce the responsible cancellation rate of order-receiving terminals while controlling the incoming complaint rate to prevent an increase, thereby achieving accurate, robust, and explainable credit limit allocation and enhancing the stickiness between the platform and order-receiving terminals.

[0004] The specific technical solutions provided in this application are as follows: In a first aspect, embodiments of this application provide a method for allocating credit limits on an order-receiving end, including: Based on the historical behavior data of the order-receiving end, the behavior of the order-receiving end under the target amount is quantified to obtain the first reward value of the order-receiving end under the target amount. The order-receiving end is the order-receiving end that needs to be allocated a cancellation amount, and the target amount is any one of multiple cancellation amounts within the platform. Based on the first reward value of the order-receiving end under the target amount, the second reward value of the order-receiving end within the cluster to which the order-receiving end belongs under the target amount, and the third reward value of the order-receiving end within the platform under the target amount, the UCB value of the order-receiving end under the target amount is obtained using the confidence upper bound UCB algorithm. The cancellation limit corresponding to the largest UCB value among the multiple cancellation limits obtained by the order receiving end is determined as the cancellation limit of the order receiving end in the current allocation period.

[0005] In the above method, when determining the cancellation limit for the current allocation period of the order-taking end, not only the first reward value of the order-taking end's behavior is considered, but also the second reward value of the cluster, which efficiently shares the exploration experience of similar order-taking ends. In addition, the third reward value of the order-taking end's behavior within the platform is also considered, so as to achieve accurate, stable and explainable limit allocation. This makes it possible for the order-taking end to reduce the possibility of responsible cancellations and complaint calls when using the allocated cancellation limit, thereby enhancing the stickiness between the platform and the order-taking end.

[0006] In one possible implementation, quantifying the behavior of the order-receiving terminal under the target amount based on its historical behavior data to obtain a first reward value for the order-receiving terminal under the target amount includes: Based on the historical behavior data, determine the number of completed orders and the number of complaints made by the order-receiving end under the target amount; The total reward value is obtained based on the number of completed items, the number of complaints, and the reward coefficient. The ratio of the total reward value to the cumulative number of orders received by the order-receiving end under the target amount is used as the first reward value for the order-receiving end under the target amount.

[0007] The above method comprehensively considers the order completion status and complaint status of the order-receiving end on the platform to determine the first reward value, which represents the behavioral characteristics of the order-receiving end. This allows the cancellation limit determined based on the first reward value to reduce the responsible cancellation rate of the order-receiving end while controlling the incoming complaint rate to prevent it from rising or falling.

[0008] In one possible implementation, the reward coefficient is an adjustable coefficient, and the initial value of the reward coefficient is a preset value; if the individual incoming complaint rate in the previous allocation period of the order receiving end is greater than the target incoming complaint rate, then the reward coefficient is determined based on the preset value, the preset adjustment step size, the individual incoming complaint rate, and the target incoming complaint rate.

[0009] In the above method, by setting the reward coefficient as an adjustable coefficient, when the individual incoming complaint rate of the order-receiving end rises abnormally, such as when the order-receiving end frequently makes malicious incoming complaints, the platform can respond in a timely manner and adjust the cancellation limit of the order-receiving end in a timely manner, so as to adaptively balance the responsible cancellation rate and the incoming complaint rate of the order-receiving end, ensuring that the responsible cancellation rate decreases while the incoming complaint rate does not increase.

[0010] In one possible implementation, the step of obtaining the UCB value of the order-receiving end under the target amount based on the first reward value of the order-receiving end under the target amount, the second reward value of the order-receiving end within the cluster to which the order-receiving end belongs under the target amount, and the third reward value of the order-receiving end within the platform under the target amount, using the confidence upper bound UCB algorithm, includes: Based on the first reward value, the cumulative number of orders received by the order-receiving terminal under the target amount, and the total number of orders on the platform, a first UCB value is obtained; The second UCB value is obtained based on the second reward value, the cumulative number of orders of all order-receiving terminals in the cluster under the target amount, and the total number of orders of the platform; The third UCB value is obtained based on the third reward value, the cumulative number of orders from all order-receiving terminals within the platform under the target cancellation limit, and the total number of orders on the platform. Based on the first UCB value, the second UCB value, the third UCB value, and their respective empirical weights, the UCB value of the order receiving end under the target amount is obtained.

[0011] The above method utilizes a three-tiered progressive UCB architecture of "global-cluster-individual" to achieve efficient sharing of experience among similar order-taking clients and fully personalized quota allocation. In particular, new order-taking clients do not need to explore from scratch and can directly inherit clustering experience and global prior knowledge to achieve accurate, robust, and interpretable quota allocation.

[0012] In one possible implementation, the individual experience weight of the first UCB value is the ratio of the cumulative number of orders at the order-receiving end to the first value, and is less than 1; The clustering empirical weight of the second UCB value is determined based on the ratio and the first coefficient; The global empirical weight of the third UCB value is determined based on the ratio and the second coefficient, wherein the first coefficient is not less than the second coefficient.

[0013] In the above method, the experience weight of each order receiving end is determined according to the cumulative number of orders. This allows new order receiving ends to rely more on global and cluster experience in the early stages, and gradually transition to individual experience as orders accumulate, thereby realizing dynamic adjustment of the weight. The individual experience weight increases with the number of orders.

[0014] In one possible implementation, if the cluster to which the order-receiving end belongs changes, the process of obtaining the second UCB value based on the second reward value, the cumulative number of orders from all order-receiving ends within the cluster under the target amount, and the total number of orders on the platform includes: Using a preset attenuation coefficient, the cumulative number of orders, the number of completed orders, and the number of complaints of the order receiving end belonging to the original cluster under the target amount are respectively converted to obtain the first number of orders, the first number of completed orders, and the first number of complaints. Based on the first number of completed orders, the first number of complaints, the second number of completed orders and the second number of complaints within the new cluster to which the order-receiving end belongs under the target amount, and the reward coefficient, the total reward value of the new cluster is obtained; The ratio of the total reward value of the new cluster to the number of second orders of the new cluster under the target amount is used as the second reward value of the new cluster. The number of second orders is the sum of the cumulative number of orders of the new cluster's internal order terminal under the target amount and the number of first orders. The second UCB value is obtained based on the second reward value of the new cluster, the second number of orders, and the total number of orders on the platform.

[0015] The above method introduces a decay migration mechanism, that is, when the behavior of the order taker changes and the cluster affiliation changes, the historical experience of the order taker in the original cluster is converted according to the decay coefficient and then migrated to the shared statistics of the new cluster to avoid wasting experience.

[0016] In one possible implementation, the order-receiving end can be any order-receiving end within the platform. Before quantifying the behavior of the order-receiving end under the target amount based on its historical behavior data to obtain the first reward value of the order-receiving end under the target amount, the method further includes: The conditions for canceling the credit limit update are determined to be met. These conditions include reaching a preset update time or the usage duration of the current canceled credit limit on the receiving end reaching a usage duration threshold. Initiate the cancellation quota processing procedure to allocate cancellation quotas to the order-receiving end; The historical behavior data is obtained, and based on the historical behavior data, it is determined that the cumulative number of orders on the order receiving end is less than a first value and greater than a second value.

[0017] The above method uses a three-tiered progressive UCB architecture to determine the cancellation limit for order-receiving terminals within the current allocation period for those terminals whose cumulative order count reaches a certain number (i.e., less than the first value but greater than the second value), thereby achieving fully personalized limit allocation.

[0018] In one possible implementation, after acquiring the historical behavior data, the method further includes: Based on the historical behavior data, it is determined that the cumulative number of orders received by the order-receiving end is not less than the first value; Based on the historical behavior data, the behavior of the order-receiving end under the historical quota is quantified to obtain the fourth reward value of the order-receiving end under the historical quota, where the historical quota is any one of the multiple allocated cancellation quotas of the order-receiving end. The target UCB value is obtained based on the fourth reward value of the order receiving end under the historical limit, the cumulative number of orders of the order receiving end under the historical limit, and the total number of orders of the platform; The cancellation quota corresponding to the largest UCB value among the multiple allocated cancellation quotas is determined as the cancellation quota within the current allocation period of the order receiving end.

[0019] The above method, for order-receiving terminals with a cumulative number of orders not less than the first value, i.e. experienced order-receiving terminals, allocates a cancellation quota that balances cancellation rate and incoming complaint rate based on their own historical behavior data left on the platform, thereby improving the stickiness between the order-receiving terminal and the platform.

[0020] In one possible implementation, after acquiring the historical behavior data, the method further includes: Based on the historical behavior data, it is determined that the cumulative number of orders received by the order receiving end is not greater than the second value; When the cumulative number of orders on the order receiving end is not greater than the third value, the minimum cancellation limit among the multiple cancellation limits is determined as the cancellation limit of the current allocation period of the order receiving end; When the cumulative number of orders on the order receiving end is greater than the third value, if the order receiving end is the selected target order receiving end, then a cancellation limit is randomly selected from the multiple cancellation limits other than the minimum limit, and determined as the cancellation limit for the current allocation period of the order receiving end; if the order receiving end is not the target order receiving end, then the cancellation limit corresponding to the largest UCB value among the cancellation limits used by the order receiving ends in the cluster to which the order receiving end belongs is determined as the cancellation limit for the current allocation period of the order receiving end.

[0021] In the above methods, for order-receiving terminals whose cumulative order count is no greater than the second value, i.e., order-receiving terminals lacking experience, a minimum guarantee amount is adopted depending on the situation to ensure a smooth transition and accumulate initial data; or, the above-mentioned various cancellation limit determination methods are adopted to realize personalized limit attempts on the basis of the minimum guarantee.

[0022] Secondly, embodiments of this application provide a quota allocation device for an order receiving end, comprising: The reward value calculation unit is used to quantify the behavior of the order-receiving terminal under the target amount based on the historical behavior data of the order-receiving terminal, and obtain the first reward value of the order-receiving terminal under the target amount. The order-receiving terminal is the order-receiving terminal that needs to be allocated a cancellation quota, and the target quota is any one of multiple cancellation quotas in the platform. The confirmation unit is used to obtain the UCB value of the order receiving end under the target amount based on the first reward value of the order receiving end under the target amount, the second reward value of the order receiving end within the cluster to which the order receiving end belongs under the target amount, and the third reward value of the order receiving end within the platform under the target amount, using the confidence upper bound UCB algorithm. The quota allocation unit is used to determine the cancellation quota corresponding to the largest UCB value among the multiple cancellation quotas obtained by the order receiving end as the cancellation quota within the current allocation period of the order receiving end.

[0023] Thirdly, embodiments of this application provide an electronic device, including: Memory is used to store computer programs or instructions; A processor for executing a computer program or instructions in the memory, such that the method described in any of the first aspects is performed.

[0024] Fourthly, embodiments of this application provide a computer-readable storage medium that, when instructions in the storage medium are executed by a processor, enables the processor to perform the method described in any one of the first aspects above.

[0025] Fifthly, embodiments of this application provide a computer program product comprising: computer program code, which, when executed on a computer, causes the computer to perform the method described in any one of the first aspects.

[0026] Furthermore, the technical effects of any of the implementation methods in the second to fifth aspects can be found in the technical effects of different implementation methods in the first aspect, and will not be repeated here.

[0027] 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

[0028] Figure 1 This is a schematic diagram illustrating an application scenario of an optional order-receiving end quota allocation method in some embodiments of this application; Figure 2 This is a schematic diagram of the architecture of a quota allocation terminal in some embodiments of this application; Figure 3This is a flowchart illustrating a method for allocating credit limits on an order-receiving end in an embodiment of this application. Figure 4 This is a flowchart illustrating a method for triggering the cancellation of quota allocation update in an embodiment of this application; Figure 5 This is a flowchart illustrating a method for determining a first reward value in an embodiment of this application; Figure 6 This is a flowchart illustrating a method for determining a UCB value in an embodiment of this application. Figure 7 This is a flowchart illustrating another method for allocating credit limits on the order receiving end in an embodiment of this application. Figure 8 This is a flowchart illustrating another method for allocating credit limits on the order receiving end in an embodiment of this application; Figure 9 This is a schematic diagram of the logical architecture of a quota allocation device for an order receiving end in an embodiment of this application; Figure 10 This is a schematic diagram of the physical architecture of the electronic device in the embodiments of this application. Detailed Implementation

[0029] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0030] It should be noted that the terms "first," "second," "third," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in sequences other than those illustrated or described herein.

[0031] The data collection, dissemination, and use in this application all comply with relevant national laws and regulations.

[0032] Ride-hailing, short for online-booked taxi service, refers to the business activity of providing off-the-cruise taxi services by building a service platform based on internet technology, integrating supply and demand information, and using qualified vehicles and drivers. Users can select vehicle type and driver through mobile application software to quickly hail a ride and enjoy travel services.

[0033] In ride-hailing operations, drivers inevitably encounter scenarios where orders need to be cancelled. To enhance the stickiness between the platform and drivers, ride-hailing platforms typically provide each driver with a certain amount of cancellation credit without penalty, known as the cancellation credit (hereinafter referred to as the cancellation credit). In traditional credit allocation scenarios, static driver stratification and the traditional Upper Confidence Bound (UCB) algorithm are often used to allocate the cancellation credit to drivers. However, in actual operation, it has been found that both methods have certain drawbacks and are not conducive to enhancing the stickiness between the platform and drivers.

[0034] In this embodiment of the application, to solve the above problems, a method for allocating credit limits for order-receiving terminals is provided. Based on the historical behavior data of the order-receiving terminal, the behavior of the order-receiving terminal under the target credit limit is quantified to obtain a first reward value for the order-receiving terminal under the target credit limit. Here, the order-receiving terminal is the order-receiving terminal that needs to be allocated a cancellation credit limit, and the target credit limit is any one of multiple cancellation credit limits within the platform. Based on the first reward value of the order-receiving terminal under the target credit limit, the second reward value of the order-receiving terminal under the target credit limit within the cluster to which the order-receiving terminal belongs, and the third reward value of the order-receiving terminal under the target credit limit within the platform, the UCB algorithm is used to obtain the UCB value of the aforementioned order-receiving terminal under the target credit limit. Then, the cancellation credit limit corresponding to the largest UCB value among the obtained UCB values ​​of the order-receiving terminal under multiple cancellation credit limits is determined as the cancellation credit limit for the order-receiving terminal in the current allocation period.

[0035] By using the above method, the historical behavior data of the order-taking end is used to quantify the behavior of the order-taking end under the target amount, resulting in a first reward value that represents the comprehensive behavioral characteristics of the order-taking end under the target amount. Then, using the UCB algorithm, the cancellation limit for the order-taking end in the current allocation period is determined by combining the first reward value of the order-taking end's own behavior, the second reward value of the behavior of order-taking ends within the cluster (similar order-taking ends), and the third reward value of the behavior of order-taking ends within the platform. It can be seen that the determination of the cancellation limit not only considers the comprehensive behavioral characteristics of the order-taking end itself, but also the second reward value of the cluster, which is to say, the efficient sharing of exploration experience among similar order-taking ends, and the third reward value of the behavior of order-taking ends within the platform. This achieves accurate, robust, and interpretable limit allocation, which reduces the possibility of responsible cancellations and complaint calls when the order-taking end uses the allocated cancellation limit, thereby enhancing the stickiness between the platform and the order-taking end.

[0036] The preferred embodiments of this application will be further described in detail below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for illustration and explanation of this application and are not intended to limit this application. Furthermore, the embodiments of this application and the features in the embodiments can be combined with each other without conflict.

[0037] See Figure 1 , Figure 1 This is a scenario diagram illustrating an optional order-receiving terminal quota allocation method provided in an embodiment of this application. The scenario includes multiple order-receiving terminals 10 and one quota allocation terminal 20, wherein each order-receiving terminal 10 and the quota allocation terminal 20 can be wirelessly connected.

[0038] like Figure 1 As shown, the credit limit allocation terminal 20 can pre-deploy cancellation credit limit update conditions. Based on the pre-set cancellation credit limit update conditions, the credit limit allocation terminal 20 determines whether the conditions are met. If the conditions are not met, it records the behavioral data of order completion, cancellation, and incoming complaints of the order receiving terminals within the platform in real time. If the conditions are met, it initiates the corresponding processing flow to allocate cancellation credit to the order receiving terminals within the platform and issues new cancellation credit to the order receiving terminals before operation to ensure normal operation.

[0039] In some embodiments, the quota allocation terminal 20 may be an electronic device, such as a server, an in-vehicle edge computing node, etc. For example, the server may be a standalone physical server, a server cluster composed of multiple physical servers, or a distributed system that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0040] Of course, the methods provided in the embodiments of this application are not limited to those described above. Figure 1 The application scenarios shown can also be used in other possible application scenarios, and this application does not limit them.

[0041] See also some possible implementations. Figure 2 As shown, the aforementioned quota allocation terminal 20 includes an input module, a dual-objective reward calculation module, a three-layer progressive UCB decision module, and an output module; wherein, The input module is used to determine whether the conditions for canceling the credit limit update are met, and when it is determined that the conditions for canceling the credit limit update are not met, to record the behavior data of each order receiving end in real time, and when it is determined that the conditions for canceling the credit limit update are met, to start the cancellation credit limit processing flow and obtain the historical behavior data of the order receiving end; The dual-objective reward calculation module is equipped with a preset dual-objective reward function. This module is used to quantify the behavior of the historical behavior data under each cancellation limit based on the aforementioned dual-objective reward function, and obtain the reward value under each cancellation limit. Specifically, if the historical behavior data belongs to the order-receiving end, the reward value obtained is the reward value of the order-receiving end under the corresponding cancellation limit, which will be recorded as the first reward value; if the historical behavior data belongs to the order-receiving end within the cluster to which the order-receiving end belongs, the reward value obtained is the reward value of the cluster under the corresponding cancellation limit, which will be recorded as the second reward value; if the historical behavior data belongs to the order-receiving end within the platform, the reward value obtained is the reward value of the platform under the corresponding cancellation limit, which will be recorded as the third reward value. The three-layer progressive UCB decision module deploys a "global-cluster-individual" three-layer progressive UCB architecture, which can be built based on the classic UCB algorithm. Each layer independently calculates its corresponding UCB value, and the final UCB value is obtained through dynamic weight fusion for use in cancellation quota allocation decisions. Specifically, this module is used to obtain the UCB value of the order taker under the corresponding cancellation quota based on the reward value; and to determine the cancellation quota corresponding to the largest UCB value among multiple cancellation quotas for the order taker as the cancellation quota for the order taker in the current allocation period. The output module is used to output the cancellation quota within the current allocation period of the order receiving end.

[0042] In this embodiment of the application, the aforementioned dual-objective reward function is used to specifically quantify order processing and incoming complaint behavior at the order receiving end. This function can be expressed by the following formula:

[0043] in, Total reward value; This represents the order completion reward value for the order-receiving end, specifically determined based on the first unit reward value and the number of completed orders on the order-receiving end; This represents the incoming line reward value at the order receiving end, specifically based on the second unit reward value and the reward coefficient (). The number of complaints from the order-receiving end is determined.

[0044] For example, taking a ride-hailing operation scenario, suppose the first unit reward value is 1, and the second unit reward value is 1. Then, when a driver completes an order, then... =1 1=1; if a driver submits a complaint upon entering the line, the complaint count increases by 1. )= = .

[0045] In this embodiment, the two business objectives of "reducing the at-fault cancellation rate" and "controlling the complaint inbound rate" are transformed into a unified optimization objective through the aforementioned dual-objective reward function (wherein: for a single order, a completed order with no complaint inbound is valued at 1, and a cancelled order with a complaint inbound is valued at 1). (The intermediate states are: completed orders but complaints received or cancelled orders but no complaints received). This aims to reduce the responsible cancellation rate on the order receiving side while controlling the complaint rate from rising.

[0046] In this embodiment, the "global-cluster-individual" three-layer progressive UCB architecture achieves experience sharing among similar order-taking terminals through dynamic clustering, solving the problems of cold start difficulties and information silos for new order-taking terminals, while also realizing fully personalized cancellation quota allocation for order-taking terminals.

[0047] In this embodiment of the application, the above-mentioned cancellation limit update conditions include reaching a preset update time, or the usage duration of the current cancellation limit on the receiving end reaches a usage duration threshold.

[0048] For example, in the context of ride-hailing operations, taking the preset update time of 2:00 AM daily in the cancellation credit limit update conditions as an example, these cancellation credit limit update conditions can be presented using a T+1 batch processing update mechanism, which can be as follows: 1. Record driver behavior data such as order completion, cancellation, and incoming complaints in real time throughout the entire operating period on day T; 2. On T+1 day (2:00 AM), start the batch processing of cancellation quotas, aggregate the behavior data of all drivers on the platform on T day, and calculate the daily reward value of each driver's action under each cancellation quota according to the preset dual-objective reward function, such as the first reward value of the aforementioned driver under the target quota; 3. Update the statistical parameters of the three-layer progressive UCB architecture on T+1 day (e.g., 4 AM), including the number of attempts, cancellation amount and corresponding UCB value of the individual layer, cluster layer and global layer; 4. The new daily quota for each driver will be issued on T+1 day (e.g., 4 AM), and the new quota will take effect.

[0049] For example, in the context of ride-hailing operations, taking a usage duration threshold of 3 days as an example for canceling credit limit updates, the specific implementation method in this application embodiment can be as follows: 1. Monitor the duration of each driver's current cancellation limit on the platform; 2. If it is determined that the cancellation limit used by any driver has not been used for 3 days, record the driver's behavior data such as order completion, cancellation, and complaint initiation throughout the entire operating period in real time; 3. If it is determined that the cancellation quota used by the driver has been used for 3 days, the cancellation quota allocation process for the driver will be initiated. The daily reward value of the driver's actions under each cancellation quota will be calculated according to the preset dual-objective reward function. For example, the first reward value of the driver under the target quota mentioned above. The statistical parameters of the three-layer progressive UCB architecture will be updated, including the number of attempts, cancellation quota and corresponding UCB value of the individual layer, cluster layer and global layer, so as to obtain the driver's new quota for the day. The new quota will take effect.

[0050] In this embodiment of the application, in order to ensure stability and smoothness, it is preferable to update the clustering results of the clustering layer once a month. That is, based on the driver's basic attributes and the structured data of four dimensions, namely the number of orders accepted, cancellation rate, responsible cancellation rate and line entry rate, the drivers in the platform are dynamically clustered by KMeans every month to obtain the clustering results and update them.

[0051] It should be noted that the specific values ​​in the examples above are merely illustrative and are not intended to be limiting. By introducing the aforementioned conditions for canceling credit limit updates, the interference of a single, accidental event on the decision to cancel the credit limit is eliminated.

[0052] The following describes the quota allocation method for the order-receiving end in this application embodiment from a specific implementation perspective. (See reference...) Figure 3 As shown in the embodiments of this application, the specific process of this method is as follows: Step 200: Based on the historical behavior data of the order-receiving end, quantify the behavior of the order-receiving end under the target amount to obtain the first reward value of the order-receiving end under the target amount. Here, the order-receiving end is the order-receiving end that needs to be allocated a cancellation amount, and the target amount is any one of multiple cancellation amounts within the platform.

[0053] In this embodiment of the application, the aforementioned order-receiving terminal is any order-receiving terminal within the platform. Therefore, before executing step 200, please refer to... Figure 4 As shown, the following process also needs to be performed: Step 300: Determine if the cancellation limit update conditions are met, wherein the cancellation limit update conditions include reaching the preset update time, or the usage duration of the current cancellation limit on the receiving end reaches the usage duration threshold; Step 310: Initiate the cancellation credit limit processing procedure to cancel the credit limit allocation for the above-mentioned order receiving end; Step 320: Obtain the historical behavior data of the order receiving end, and based on the historical behavior data, determine that the cumulative number of orders of the above order receiving end is less than the first value and greater than the second value.

[0054] In this embodiment, the aforementioned historical behavior data includes order acceptance, cancellation, and incoming complaint data of the order-receiving terminal. During step 320, the aforementioned input module retrieves the historical behavior data of the order-receiving terminal from the platform's order system and / or user behavior database. Based on the historical behavior data, it determines that the cumulative number of orders for the order-receiving terminal is less than a first value and greater than a second value, thereby triggering step 200.

[0055] In steps 300-320, in this embodiment, for order-receiving terminals within the platform whose accumulated order count reaches a certain number (i.e., less than the first value but greater than the second value), the aforementioned three-tier progressive UCB architecture is used to determine their cancellation quota within the current allocation period. For newly registered order-receiving terminals or those with a small accumulated order count (not exceeding the aforementioned second value), a preset cold-start guarantee strategy is adopted. Specific processing details are provided later. Figure 7 The process is illustrated below. For order takers with a large number of accumulated orders, such as those with a first value, they are considered experienced order takers. In this embodiment, the cancellation limit for the current allocation period is mainly allocated based on the historical behavior data of such order takers. The first value can be regarded as a threshold for measuring whether the order taker is an experienced order taker, and can be set according to business needs, such as taking a value of 100. The specific processing flow is detailed later. Figure 8 The process is shown below.

[0056] The reason for adopting the aforementioned three-tiered progressive UCB architecture to determine the cancellation limit within the current allocation period for order-receiving terminals that have accumulated a certain number of orders is that, although the platform has collected some behavioral data on these terminals compared to those with a cumulative order count not exceeding the second highest value, the data volume is still relatively small and cannot accurately reflect the characteristics of these terminals. Therefore, allocating cancellation limits based on the accumulated experience of similar terminals and other terminals within the platform is more conducive to enhancing the stickiness between the platform and these terminals. The cancellation limit allocation process for these terminals will be explained below.

[0057] In this embodiment of the application, when performing step 200, refer to... Figure 5 As shown, the specific process is as follows: Step 2001: Based on the historical behavior data of the order-receiving terminal, determine the number of completed orders and the number of complaints made by the order-receiving terminal under the target amount.

[0058] In this embodiment of the application, since the above-mentioned historical behavior data includes the order acceptance, cancellation, and incoming complaint data of the order receiving end, when executing step 2001, the dual-target reward calculation module extracts the number of completed orders and the number of complaints of the order receiving end under each cancellation limit from the historical behavior data of the order receiving end.

[0059] Step 2002: Based on the number of completed orders, the number of complaints, and the reward coefficient, obtain the total reward value.

[0060] In this embodiment of the application, when executing step 2002, the number of completed items and the number of complaints are substituted into the aforementioned dual-objective reward function, and the total reward value is obtained by combining the first unit reward value, the second unit reward value, and the reward coefficient.

[0061] Step 2003: The ratio of the total reward value to the cumulative number of orders received by the order-receiving end under the target amount is used as the first reward value for that order-receiving end under the target amount.

[0062] For example, taking the order receiving end as order receiving end d and the target amount as the cancellation amount a as an example, the first reward value can be recorded as: The total reward value under the cancellation limit 'a' is The cumulative number of orders is ,but .

[0063] In this embodiment of the application, in order to achieve a dynamic balance between the two objectives at different business stages, the above-mentioned reward coefficient can be set as an adjustable coefficient, and the initial value of the reward coefficient is a preset value; if the personal incoming complaint rate in the previous allocation period of the order receiving end is greater than the target incoming complaint rate, the reward coefficient is determined based on the above-mentioned preset value, preset adjustment step size, personal incoming complaint rate and target incoming complaint rate.

[0064] Thus, by setting the above reward coefficient ( λ is an adjustable coefficient. When the individual complaint rate on the order-taking end rises abnormally, such as when the order-taking end maliciously cancels orders or frequently makes malicious complaints, the platform can respond in a timely manner and adjust the λ value according to the real-time situation to ensure that while pursuing a decrease in the responsible cancellation rate, the complaint rate is controlled to prevent an increase, thus achieving a balance between the two.

[0065] In one possible implementation, the specific adjustment method of the above-mentioned reward coefficient can be expressed by the following formula:

[0066] in, Indicates the current The value can be initially set to 1; Indicates the adjusted Values; complaint_rate_t represents the individual inbound complaint rate at the order receiving end; complaint_target represents the target inbound complaint rate, which can be determined according to the specific business requirements; This indicates the preset adjustment step size.

[0067] In the embodiments of this application, The value can range from 0.1 to 0.3, and the specific value can be determined according to business needs to ensure that the adjustment of the reward coefficient is sensitive and stable. For example, assuming that the individual complaint rate keeps doubling, by setting... Adjusted to The multiple is used to determine the total number of days x that have actually passed, thus deriving the above-mentioned range of values ​​that are sensitive to adjustments in the reward coefficient and have stable daily changes. Taking the above-mentioned multiple of 2 as an example, then 2 = (1+ ) x This leads to the derivation x = ln2 / ln(1+ With the goal of significantly adjusting the problem within one week and maintaining stable daily changes, the existing algorithm was used to calculate that: when α=0.1, x=7.3; when α=0.2, x=3.8; when α=0.3, x=2.6; therefore, α takes values ​​from 0.1 to 0.3.

[0068] Step 210: Based on the first reward value of the order-receiving end under the target amount, the second reward value of the order-receiving end within the cluster to which the order-receiving end belongs under the target amount, and the third reward value of the order-receiving end within the platform under the target amount, the UCB algorithm is used to obtain the UCB value of the order-receiving end under the target amount.

[0069] In this embodiment of the application, step 210 is implemented using the aforementioned three-layer progressive UCB architecture of "global-cluster-individual". See also... Figure 2 As shown, the aforementioned three-layer progressive UCB decision module includes a strategy layer, a fusion layer, and a decision layer. The strategy layer includes an individual layer, a clustering layer, and a global layer. The fusion layer is used to fuse the UCB values ​​determined by the individual layer, the clustering layer, and the global layer. The decision layer is used to determine the cancellation limit for the current allocation period of the order receiving end based on the UCB values ​​under multiple cancellation limits.

[0070] In specific implementation, when performing step 210, refer to... Figure 6 As shown, the specific process is as follows: Step 2101: Based on the first reward value of the order receiving end under the target amount, the cumulative number of orders of the order receiving end under the target amount, and the total number of orders of the platform, obtain the first UCB value.

[0071] In this embodiment, step 2101 is implemented using the individual layer of the aforementioned three-layer progressive UCB architecture. Each order-receiving end in the individual layer maintains independent UCB statistics, recording its actual performance under different cancellation limits, thus enabling personalized fine-tuning.

[0072] Taking the order-receiving end as order-receiving end d and the target amount as cancellation amount a as an example, in specific implementation, when executing step 2101, the first reward value of order-receiving end d under cancellation amount a, the cumulative number of orders of order-receiving end d under cancellation amount a, and the total number of orders on the platform are substituted into the following formula, and combined with the historical behavior of order-receiving end d under cancellation amount a, the first UCB value is obtained:

[0073] in, This indicates that the order-receiving end d canceled the credit limit. The first UCB value below; This indicates that the order-receiving end d canceled the credit limit. The first reward value is obtained by averaging the total reward values ​​calculated using the aforementioned bi-objective reward function. The exploration coefficient is represented in the embodiments of this application. The possible values ​​are the optimal values ​​provided by the relevant derivations of the classic UCB algorithm, such as 2; T represents the total number of orders across all order-receiving terminals within the platform under all credit limits, i.e., the total number of orders on the platform; This indicates that the order-receiving end d canceled the credit limit. The cumulative number of orders placed; To prevent the removal of zero smoothing terms, the value in this embodiment can be 1, so that the UCB values ​​under all cancellation limits are on the same order of magnitude, thus ensuring a balance between exploration and utilization, and making subsequent decisions more robust.

[0074] Step 2102: Based on the second reward value of the order-receiving end within the above cluster under the target amount, the cumulative number of orders of all order-receiving ends within the cluster under the target amount, and the total number of orders on the platform, obtain the second UCB value.

[0075] In this embodiment, step 2102 is implemented using the clustering layer in the aforementioned three-layer progressive UCB architecture. The clustering layer comprises multiple clusters, specifically obtained through dynamic KMeans clustering based on the basic attributes of the order-taking end, combined with its four-dimensional structured data including order count, cancellation rate, responsible cancellation rate, and call-in rate. Order-taking ends within the same cluster share UCB experience, enabling knowledge transfer between similar order-taking ends.

[0076] Taking the order-receiving end as order-receiving end d and the target amount as cancellation amount a as an example, in specific implementation, when executing step 2102, the second reward value of order-receiving end d under cancellation amount a, the cumulative number of orders of all order-receiving ends in the cluster under cancellation amount a, and the total number of orders on the platform are all substituted into the following formula. Combined with the historical behavior of all order-receiving ends in the cluster under cancellation amount a, the second UCB value is obtained:

[0077] in, This indicates that the order-receiving terminal d belongs to the same cluster where the order-receiving terminal canceled its quota. The second UCB value below; This indicates that the order-receiving terminal d belongs to the same cluster where the order-receiving terminal canceled its quota. The second reward value below; This indicates that all order-receiving terminals within the cluster to which order-receiving terminal d belongs have cancelled their quotas. The cumulative number of orders placed; , , As mentioned above, there will be no further explanation.

[0078] It should be noted that, when calculating the second reward value, similar to the method for determining the first reward value, the cluster can be regarded as a receiving end. After counting the number of individual orders and the number of complaints, the total reward value of the cluster calculated by the aforementioned bi-objective reward function is substituted into the average value to obtain the second reward value of the cluster under the cancellation limit a.

[0079] Considering that the cluster to which the order taker belongs may not remain unchanged, this application provides a decay migration mechanism. When the behavior of the order taker changes and the cluster affiliation changes, the historical experience of the order taker in the original cluster is converted according to the decay coefficient and migrated to the shared statistics of the new cluster to avoid wasting experience.

[0080] In some preferred embodiments, the value range of the above-mentioned attenuation coefficient can be (0.5-0.9) to ensure that when the behavior of the receiving end changes, the historical experience gradually fades (e.g., 1 to 7 migrations), but does not drop to zero abruptly, and does not excessively interfere with the new clustering.

[0081] Based on the aforementioned attenuation migration mechanism, before executing step 2102, it is determined that the cluster to which the receiving end belongs has not changed.

[0082] In other embodiments, if it is determined that the cluster to which the receiving end belongs has changed, then the following operations are specifically performed when executing step 2102: Operation 1: Using a preset attenuation coefficient, calculate the cumulative number of orders, the number of completed orders, and the number of complaints for the order receiving end under the target amount, respectively, to obtain the first number of orders, the first number of completed orders, and the first number of complaints; Operation 2: Based on the number of completed orders, the number of complaints, the number of completed orders and the number of complaints within the new cluster to which the order-taking end belongs, and the aforementioned reward coefficient, the total reward value of the new cluster is obtained. Operation 3: The ratio of the total reward value of the new cluster to the number of second orders in the new cluster under the target amount is used as the second reward value of the new cluster. The number of second orders is the sum of the cumulative number of orders and the number of first orders in the new cluster under the target amount. Operation 4: Based on the second reward value of the new cluster, the second number of orders, and the total number of orders on the platform, the second UCB value is obtained.

[0083] In operation two, the number of completed orders, the number of complaints, the number of completed orders and the number of complaints under the target amount for the new cluster to which the order-taking end belongs are substituted into the aforementioned dual-objective reward function. Combined with the reward coefficient, the total reward value of the new cluster is obtained, which is the total reward value of the new cluster under the target amount. This new cluster is the cluster to which the order-taking end currently belongs. It should be noted that if the order data of the order-taking end also exists in the new cluster, the reward value of this part of the order data can be calculated as normal order data. If there are other order-taking ends in the new cluster whose clusters have changed, the same processing method as for the order-taking end is still used, and the historical order data belonging to the original cluster is converted and migrated to the new cluster. This will not be elaborated here.

[0084] In step four, the second reward value of the new cluster, the second number of orders, and the total number of orders on the platform are substituted into the formula for calculating the second UCB value to obtain the aforementioned UCB value.

[0085] Step 2103: Based on the third reward value of the order-taking end within the platform under the target amount, the cumulative number of orders of all order-taking ends within the platform under the target cancellation amount, and the total number of orders of the platform, obtain the third UCB value.

[0086] In this embodiment of the application, when performing step 2103, the global layer in the above-mentioned three-layer progressive UCB architecture is used. This global layer is used to aggregate the historical experience of all order-receiving terminals within the platform, calculate the global UCB value—the third UCB value—under each cancellation limit, and provide basic prior knowledge for new order-receiving terminals.

[0087] Taking the order-receiving end as order-receiving end d and the target amount as cancellation amount a as an example, in specific implementation, when executing step 2103, the third reward value of the order-receiving end under cancellation amount a, the cumulative number of orders of all order-receiving ends under cancellation amount a, and the total number of orders of the platform are all substituted into the following formula. Combined with the historical behavior of all order-receiving ends under cancellation amount a, the third UCB value is obtained:

[0088] in, This indicates that the order-taking end within the platform has cancelled the quota. The third UCB value below; This indicates that the order-taking end within the platform has cancelled the quota. The third reward value below; This indicates that all order-receiving terminals within the platform have cancelled their quotas. The cumulative number of orders placed; , , As mentioned above, there will be no further explanation.

[0089] It should be noted that, when calculating the third reward value, similar to the method used to determine the first reward value, the platform can be regarded as an order-receiving end. The number of completed orders and the number of complaints under the cancellation limit a are counted, and then substituted into the total reward value of the platform calculated by the aforementioned dual-objective reward function. The average value is then taken to obtain the third reward value of the platform under the cancellation limit a.

[0090] Step 2104: Based on the first UCB value, the second UCB value, the third UCB value, and their respective empirical weights, obtain the UCB value of the order receiving end under the target amount.

[0091] In the embodiments of this application, please refer to Figure 2 As shown, when executing step 2104, the first UCB value, the second UCB value, the third UCB value, and their respective experience weights are weighted and summed through the fusion layer to obtain the UCB value of the order receiving end under the target amount.

[0092] In this embodiment of the application, taking the order receiving end as order receiving end d and the target amount as the cancellation amount a as an example, the above weighted sum can be expressed by the following formula:

[0093] in, This indicates that the order-receiving end d has cancelled the credit limit. UCB value below; , , These are individual experience weights, cluster experience weights, and global experience weights, respectively.

[0094] In some embodiments of this application, the individual experience weight of the aforementioned first UCB value is... The first value can be the ratio of the cumulative number of orders received by the order-receiving end to the aforementioned first value, and it must be less than 1. This first value is a critical value for measuring whether the order-receiving end is an experienced order-receiving end, and it can be set based on specific business needs; the second UCB value is the clustering experience weight. The global empirical weight of the third UCB value is determined based on this ratio and the first coefficient. The coefficient is determined based on the ratio and the second coefficient, wherein the first coefficient is not less than the second coefficient.

[0095] In specific implementation, when the aforementioned first value is 100, the above... , , They can be expressed by the following formulas: ;

[0096] in, This represents the cumulative number of orders received by the order-receiving end d.

[0097] because So, when the cumulative number of orders on order-receiving end d reaches 100, the weight... The weight is set to 1; when the cumulative number of orders is less than 100, the weight is adjusted accordingly. The calculation is based on a ratio of 1 / 100. This method allows for greater reliance on global and clustering experience when allocating cancellation quotas to order-receiving terminals with a smaller cumulative number of orders. As orders accumulate, it gradually transitions to a system primarily driven by individual experience, enabling dynamic adjustment of weights. The weight of individual experience increases with the number of orders.

[0098] In this embodiment of the application, by setting This allows for dynamic adjustments based on the experience of the order-taking end, providing a clear business explanation logic for the decision-making process. For example, "The more mature the order-taking end, the more confident they are. When they lack experience, they should refer to similar order-taking ends and the overall average level." From a business perspective, it is easy to understand "why this order-taking end is given 2 times instead of 3 times," thereby significantly improving the credibility and acceptability of credit allocation.

[0099] Step 220: Among the UCB values ​​obtained for the order receiving end under multiple cancellation limits, the cancellation limit corresponding to the largest UCB value is determined as the cancellation limit for the order receiving end in the current allocation period.

[0100] In this embodiment of the application, when performing step 220, refer to Figure 2 As shown, the decision-making layer selects the maximum UCB value from the UCB values ​​of the order receiving end under multiple cancellation limits, and determines the cancellation limit corresponding to the maximum UCB value as the cancellation limit of the order receiving end in the current allocation period, and outputs it by the aforementioned output module.

[0101] The following description uses a specific embodiment as an example.

[0102] For example, in a ride-hailing operation scenario, the total cumulative number of orders for driver A can be obtained based on driver A's historical order data. The total number of orders on the platform is 50; the total number of orders T is 10,000; the first unit reward value is 1; the second unit reward value is 1; and the reward coefficient λ is the initial value (e.g., 1).

[0103] Assume that driver A has historically been allocated cancellation limits of 2, 3, and 4; driver A's cumulative order count, completed orders, and complaint history under these multiple cancellation limits on the platform are shown in the table below: Table 1. Summary of Driver A's Cumulative Order Count, Completed Orders, and On-line Complaints

[0104] Since driver A's cumulative order count of 50 is greater than 30, a three-layer progressive UCB architecture is used to calculate the UCB values ​​of the individual layer, cluster layer, and global layer under different cancellation limits.

[0105] First, calculate the reward value for each level.

[0106] 1. Individual level

[0107] In specific implementation, based on the aforementioned method for determining the dual-objective reward function R, that is, for a single order: No complaint after completion: R=1; Complaint after completion with complaint: R=0; No complaint after cancellation: R=0; Complaint after cancellation with complaint: R=-1; Combining the data in Table 1, the total reward value for driver A under multiple cancellation limits is calculated and denoted as... .

[0108] Since the first unit reward value is 1, the second unit reward value is 1, and the reward coefficient is 1 in this example, the following simplified formula can be used to calculate the total reward value under different cancellation amounts: Regarding cancellation limit 4: From Table 1, we know that the number of completed orders is 20-3=17, and there were no complaints received, meaning the number of complaints is 0. Therefore, substituting the number of completed orders (17), the number of complaints (0), the first unit reward value (1), the second unit reward value (1), and the reward coefficient (1) into the aforementioned bi-objective reward function, we obtain the total reward value for driver A under cancellation limit 4, i.e. =17 1-1 (1) 0 = 17; By following this pattern, you can obtain the total reward value for canceling quota 3, canceling quota 2, and canceling quota 1. Then, the total reward value under the same cancellation amount will be... Divide by the cumulative number of orders under the cancellation limit a Then you will receive the first reward value under that cancellation limit. The specific results are shown in the table below: Table 2. Summary of Driver A's Cumulative Orders, Completed Orders, On-line Complaints, and Reward Value

[0109] 2. Clustering layer

[0110] In practice, the number of completed transactions and the number of complaints for each cancellation limit are determined for the cluster to which driver A belongs. The number of completed transactions and the number of complaints for any cancellation limit, along with the first unit reward value 1, the second unit reward value 1, and the reward coefficient 1, are substituted into the aforementioned bi-objective reward function to calculate the total reward value for the cluster under that cancellation limit, denoted as... Then divide by the cumulative number of orders for that cluster under the cancellation limit, and record as: Then, the second reward value for the cluster to which driver A belongs under the cancellation limit is obtained. The results are shown in the table below: Table 3. Summary of Cumulative Order Count and Second Reward Value for Driver A's Cluster

[0111] 3. Global layer

[0112] In practice, the number of completed transactions and the number of complaints are determined for each cancellation limit within the platform. The number of completed transactions and the number of complaints for any cancellation limit, along with the first unit reward value 1, the second unit reward value 1, and the reward coefficient 1, are substituted into the aforementioned bi-objective reward function to calculate the platform's total reward value for that cancellation limit, denoted as... Divide this by the cumulative number of orders on the platform under this cancellation limit, and record it as... Then you will receive the platform's third reward value for each cancellation amount. The results are shown in the table below: Table 4. Summary of Cumulative Order Count and Third-Party Reward Value on the Platform

[0113] Then, the UCB value of each layer is calculated.

[0114] Assuming exploration coefficient The value is 2. The value is 1. Where lnT = ln10000 = 9.21, then, regarding the cancellation of quota 4: 1. Individual level: =0.85, =20; ; 2. Clustering layer: =0.70, =400; ; 3. Global layer: =0.70, =4000; ; Weighted fusion: Based on the aforementioned weight calculation formula, the individual experience weight is obtained. = min(1, / 100) = 0.5; Clustering empirical weights = (1- ) × 0.7 = 0.35; Global empirical weight = (1- )×0.3 = 0.15.

[0115] Therefore, the UCB value for canceling credit limit 4 is: = 0.5×2.17 + 0.35×1.00 + 0.15×0.80 = 1.555.

[0116] For cancellation amounts 3, 2, and 1: Following the calculation method for cancellation amount 4, cancellation amounts 3, 2, and 1 are calculated sequentially, and the results are summarized in the table below: Table 5. Summary of Driver A's final UCB values ​​under different cancellation amounts

[0117] Finally, based on the data in Table 5, the cancellation limit corresponding to the largest UCB value (i.e., 3.689) is selected as the cancellation limit for driver A in the current allocation period, i.e., the cancellation limit is 1.

[0118] In this embodiment, since any of the aforementioned reward values ​​for the cancellation quota of driver A within the current allocation period takes into account driver A's order cancellations and incoming complaint behavior, the two conflicting business objectives of "reducing the at-fault cancellation rate" and "controlling the incoming complaint rate" can be transformed into a unified optimization objective, ensuring that while pursuing a decrease in the at-fault cancellation rate of drivers, the incoming complaint rate is controlled to not increase. When determining the cancellation quota of driver A within the current allocation period, the second reward value of clustering is considered to efficiently share the exploration experience of similar drivers, and the third reward value of the behavior of drivers within the platform is considered to achieve accurate, robust, and explainable quota allocation, thereby enhancing the stickiness between the platform and driver A.

[0119] The following describes the aforementioned cold start guarantee strategy and two processing procedures: determining the cancellation limit within the current allocation cycle based solely on the historical behavior data of the order-receiving end.

[0120] First, we will introduce the cold start safety strategy, which includes a pure safety strategy and a light exploration strategy.

[0121] For the pure minimum guarantee strategy, for order-receiving terminals whose cumulative order count is no greater than the second value mentioned above, and for order-receiving terminals whose cumulative order count is no greater than the third value, the minimum guarantee amount among multiple cancellation limits will be determined as the cancellation limit for that order-receiving terminal in the current allocation period.

[0122] For example, suppose the first value is 100, the second value is 30, and the third value is 10. Then, the pure minimum guarantee strategy targets order takers with 0-10 orders, uniformly allocating a minimum guarantee amount, such as 3 times, to ensure a smooth transition and accumulate initial data.

[0123] For the light exploration strategy, among the order-receiving terminals with a cumulative order count not exceeding the second value mentioned above, those with a cumulative order count exceeding the third value are statistically analyzed. A certain percentage (e.g., 70%) of the total number of order-receiving terminals are then assigned the cancellation quota corresponding to the largest UCB value among the cancellation quotas used by order-receiving terminals within the cluster to which the order-receiving terminal belongs, based on the UCB value of the cluster. The remaining percentage (e.g., 30%) of the total number of order-receiving terminals are then randomly assigned any cancellation quota other than the minimum cancellation quota mentioned above, and this is assigned to them for the current allocation period.

[0124] For example, suppose the first value is 100, the second value is 30, and the third value is 10. Then, the light exploration strategy targets order takers with 11-30 orders. For 70% of the total number of order takers in this category, the clustering layer UCB value is used, that is, based on the experience of similar order takers, the cancellation limit within the current allocation cycle is determined. For the remaining 30% of the order takers in this category, a limit is randomly selected from the exploration pool, such as {1,2,4}, to explore, so as to start personalized attempts on the basis of the minimum limit.

[0125] It should be noted that when allocating cancellation limits based on the UCB value of the clustering layer, on the one hand, the collected behavioral data of this type of order-taking client focuses on their behavior under the guaranteed minimum limit. In other words, a large portion of the data from this type of client consists of behavior under the guaranteed minimum limit, with fewer data points under other cancellation limits. Therefore, although cancellation limits are allocated according to the UCB value of the clustering layer, the final allocated cancellation limits tend to favor the guaranteed minimum limit. On the other hand, this stage is a period of adjustment between the order-taking client and the platform, and excessive attention is unnecessary. From the platform's operational perspective, the focus is on collecting behavioral data from the order-taking clients. Therefore, in a preferred embodiment, when allocating cancellation limits based on the UCB value of the clustering layer, the guaranteed minimum limit can be directly allocated to this type of order-taking client.

[0126] In practice, after obtaining the historical behavior data of the order-receiving end in step 320, refer to... Figure 7 As shown, the following process can also be executed: Step 700: If, based on the historical behavior data, it is determined that the cumulative number of orders on the order receiving end is not greater than the second value mentioned above, then the above cold start guarantee strategy is activated.

[0127] Step 710: Based on the above cold start guarantee strategy, determine whether the cumulative number of orders on the order receiving end is not greater than the third value. If yes, proceed to step 720; otherwise, proceed to step 730.

[0128] Step 720: Determine the minimum cancellation limit among the multiple cancellation limits within the platform as the cancellation limit for the current allocation period of the order receiving end.

[0129] Step 730: When the order receiving end is the selected target order receiving end, randomly select a cancellation limit from multiple cancellation limits other than the minimum limit and determine it as the cancellation limit for the current allocation period of the order receiving end; when the order receiving end is not the above-mentioned target order receiving end, determine the cancellation limit corresponding to the largest UCB value among the cancellation limits used by order receiving ends in the cluster to which the order receiving end belongs as the cancellation limit for the current allocation period of the order receiving end.

[0130] If the order-receiving end in step 730 is not the target order-receiving end mentioned above, based on the historical behavior data of all order-receiving ends in the cluster to which the order-receiving end belongs, under multiple cancellation limits, the UCB value under each cancellation limit is obtained through the above-mentioned bi-objective reward function. Then, the cancellation limit corresponding to the largest UCB value is selected as the cancellation limit for the current allocation period of the order-receiving end.

[0131] In other embodiments, after performing step 320 above to obtain the historical behavior data of the order-receiving end, refer to Figure 8 As shown, the following process can also be executed: Step 800: Based on the historical behavior data, determine that the cumulative number of orders on the order receiving end is not less than the first value mentioned above.

[0132] For example, assuming the first value is 100, the second value is 30, and the third value is 10, then, when it is determined that the cumulative number of orders on the order-receiving end is not less than 100, combined with the above three-layer progressive UCB architecture... The weight of personal experience is set to 1, which means that when determining the cancellation limit of the order receiving end in the current allocation period, personal experience is the primary factor, triggering the execution step 810.

[0133] Step 810: Based on the historical behavior data, quantify the behavior of the order-receiving end under the historical quota to obtain the fourth reward value of the order-receiving end under the historical quota, wherein the historical quota is any one of the multiple allocated cancellation quotas of the order-receiving end.

[0134] In this embodiment of the application, when performing step 810, steps 2001 to 2003 described above can be performed in order to obtain the fourth reward value, which will not be repeated here.

[0135] Step 820: Based on the fourth reward value of the order receiving end under the historical quota, the cumulative number of orders of the order receiving end under the historical quota, and the total number of orders of the platform, obtain the target UCB value.

[0136] In this embodiment of the application, when performing step 820, the calculation method of the UCB value of the aforementioned individual layer can be referred to to obtain the target UCB value, which will not be repeated here.

[0137] Step 830: Among the target UCB values ​​obtained under multiple allocated cancellation quotas, the cancellation quota corresponding to the largest UCB value is determined as the cancellation quota within the current allocation period of the order receiving end.

[0138] In this way, by executing steps 800 to 830, when the cumulative number of orders on the order-receiving end is not less than the first value, that is, when the cumulative number of orders on a certain order-receiving end is large enough, a cancellation quota that balances the cancellation rate and the incoming complaint rate is allocated to the order-receiving end based on its own historical behavior data left on the platform, thereby increasing the stickiness between the order-receiving end and the platform and contributing to the stable operation of the platform.

[0139] The following example summarizes and explains the different allocation methods mentioned above. For instance, let's assume the first value is 100, the second value is 30, and the third value is 10.

[0140] In scenario one, when it is determined that the cumulative number of orders on the order receiving end is not less than 100, based on the historical behavior data of the order receiving end, the target UCB value of the order receiving end under different allocated cancellation quotas is determined through the aforementioned dual-objective reward function. Then, the maximum UCB value is selected from them, and the cancellation quota corresponding to the maximum UCB value is determined as the cancellation quota of the order receiving end in the current allocation period.

[0141] In scenario two, when the cumulative number of orders on the order-receiving end is less than 100 but greater than 30, the above three-layer progressive UCB architecture is adopted to obtain the UCB value of the order-receiving end under multiple cancellation limits within the platform. Then, the largest UCB value is selected from them, and the cancellation limit corresponding to the largest UCB value is determined as the cancellation limit of the order-receiving end in the current allocation period.

[0142] Scenario 3: If the cumulative number of orders on the order-receiving end is determined to be no more than 30, the cold start guarantee strategy is triggered. The cancellation limit for the current allocation period on the order-receiving end is determined according to the cold start guarantee strategy.

[0143] In sub-case one, when it is determined that the cumulative number of orders on the order receiving end is no more than 10, the minimum cancellation limit among multiple cancellation limits, such as 3 times, is determined as the cancellation limit within the current allocation cycle of that order receiving end. In sub-scenario two, when the cumulative number of orders received by a receiving end is greater than 10, if the receiving end is a selected target receiving end, such as one within the top 30%, then a cancellation limit is randomly selected from multiple cancellation limits other than the minimum guarantee limit, and determined as the cancellation limit for the current allocation period of that receiving end; if the receiving end is not a target receiving end, such as one within the top 70%, then according to the UCB value of the cluster, the cancellation limit corresponding to the largest UCB value among the cancellation limits used by receiving ends in the cluster to which the receiving end belongs is determined as the cancellation limit for the current allocation period of that receiving end.

[0144] Based on the same inventive concept, see [reference] Figure 9 As shown in the figure, this application embodiment provides a quota allocation device for an order receiving end, including: The reward value calculation unit 910 is used to quantify the behavior of the order receiving terminal under the target amount based on the historical behavior data of the order receiving terminal, and obtain the first reward value of the order receiving terminal under the target amount. The order receiving terminal is the order receiving terminal that needs to be allocated a cancellation amount, and the target amount is any one of multiple cancellation amounts in the platform. The determining unit 920 is used to obtain the UCB value of the order receiving end under the target amount based on the first reward value of the order receiving end under the target amount, the second reward value of the order receiving end within the cluster to which the order receiving end belongs under the target amount, and the third reward value of the order receiving end within the platform under the target amount, using the confidence upper bound UCB algorithm. The quota allocation unit 930 is used to determine the cancellation quota corresponding to the largest UCB value among the multiple cancellation quotas obtained by the order receiving end as the cancellation quota within the current allocation period of the order receiving end.

[0145] In one possible implementation, the reward value calculation unit 910 is specifically used for: Based on the historical behavior data, determine the number of completed orders and the number of complaints made by the order-receiving end under the target amount; The total reward value is obtained based on the number of completed items, the number of complaints, and the reward coefficient. The ratio of the total reward value to the cumulative number of orders received by the order-receiving end under the target amount is used as the first reward value for the order-receiving end under the target amount.

[0146] In one possible implementation, the reward coefficient is an adjustable coefficient, and the initial value of the reward coefficient is a preset value; if the individual incoming complaint rate in the previous allocation period of the order receiving end is greater than the target incoming complaint rate, then the reward coefficient is determined based on the preset value, the preset adjustment step size, the individual incoming complaint rate, and the target incoming complaint rate.

[0147] In one possible implementation, the determining unit 920 is specifically used for: Based on the first reward value, the cumulative number of orders received by the order-receiving terminal under the target amount, and the total number of orders on the platform, a first UCB value is obtained; The second UCB value is obtained based on the second reward value, the cumulative number of orders of all order-receiving terminals in the cluster under the target amount, and the total number of orders of the platform; The third UCB value is obtained based on the third reward value, the cumulative number of orders from all order-receiving terminals within the platform under the target cancellation limit, and the total number of orders on the platform. Based on the first UCB value, the second UCB value, the third UCB value, and their respective empirical weights, the UCB value of the order receiving end under the target amount is obtained.

[0148] In one possible implementation, the individual experience weight of the first UCB value is the ratio of the cumulative number of orders at the order-receiving end to the first value, and is less than 1; The clustering empirical weight of the second UCB value is determined based on the ratio and the first coefficient; The global empirical weight of the third UCB value is determined based on the ratio and the second coefficient, wherein the first coefficient is not less than the second coefficient.

[0149] In one possible implementation, if the cluster to which the receiving end belongs changes, the determining unit 920 is specifically used for: Using a preset attenuation coefficient, the cumulative number of orders, the number of completed orders, and the number of complaints of the order receiving end belonging to the original cluster under the target amount are respectively converted to obtain the first number of orders, the first number of completed orders, and the first number of complaints. Based on the first number of completed orders, the first number of complaints, the second number of completed orders and the second number of complaints within the new cluster to which the order-receiving end belongs under the target amount, and the reward coefficient, the total reward value of the new cluster is obtained; The ratio of the total reward value of the new cluster to the number of second orders of the new cluster under the target amount is used as the second reward value of the new cluster. The number of second orders is the sum of the cumulative number of orders of the new cluster's internal order terminal under the target amount and the number of first orders. The second UCB value is obtained based on the second reward value of the new cluster, the second number of orders, and the total number of orders on the platform.

[0150] In one possible implementation, the order-receiving end is any order-receiving end within the platform. Before quantifying the behavior of the order-receiving end under the target amount based on its historical behavior data to obtain the first reward value of the order-receiving end under the target amount, the quota allocation unit 930 is further configured to: The conditions for canceling the credit limit update are determined to be met. These conditions include reaching a preset update time or the usage duration of the current canceled credit limit on the receiving end reaching a usage duration threshold. Initiate the cancellation quota processing procedure to allocate cancellation quotas to the order-receiving end; The historical behavior data is obtained, and based on the historical behavior data, it is determined that the cumulative number of orders on the order receiving end is less than a first value and greater than a second value.

[0151] In one possible implementation, after acquiring the historical behavior data, the quota allocation unit 930 is further configured to: Based on the historical behavior data, it is determined that the cumulative number of orders received by the order-receiving end is not less than the first value; Based on the historical behavior data of the order receiving terminal, the behavior of the order receiving terminal under the historical quota is quantified to obtain the fourth reward value of the order receiving terminal under the historical quota, wherein the historical quota is any one of the multiple allocated cancellation quotas of the order receiving terminal; The target UCB value is obtained based on the fourth reward value of the order receiving end under the historical limit, the cumulative number of orders of the order receiving end under the historical limit, and the total number of orders of the platform; The cancellation quota corresponding to the largest UCB value among the multiple allocated cancellation quotas is determined as the cancellation quota within the current allocation period of the order receiving end.

[0152] In one possible implementation, after acquiring the historical behavior data, the quota allocation unit 930 is further configured to: Based on the historical behavior data, it is determined that the cumulative number of orders received by the order receiving end is not greater than the second value; When the cumulative number of orders on the order receiving end is not greater than the third value, the minimum cancellation limit among the multiple cancellation limits is determined as the cancellation limit of the current allocation period of the order receiving end; When the cumulative number of orders on the order receiving end is greater than the third value, if the order receiving end is the selected target order receiving end, then a cancellation limit is randomly selected from the multiple cancellation limits other than the minimum limit, and determined as the cancellation limit for the current allocation period of the order receiving end; if the order receiving end is not the target order receiving end, then the cancellation limit corresponding to the largest UCB value among the cancellation limits used by the order receiving ends in the cluster to which the order receiving end belongs is determined as the cancellation limit for the current allocation period of the order receiving end.

[0153] Based on the same inventive concept, this application provides an electronic device, which is described below with reference to... Figure 10 This application describes an electronic device, such as a cloud server, according to this embodiment. Figure 10 The electronic device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0154] like Figure 10 As shown, the electronic device may include, but is not limited to: Memory 1001 is used to store computer programs or instructions; Processor 1002 is configured to execute computer programs or instructions in memory 1001 such that any of the methods described in the above embodiments is performed.

[0155] In some embodiments, memory 1001 may include a readable medium in the form of volatile memory, such as random access memory (RAM) and / or cache memory, and may further include read-only memory (ROM).

[0156] In some embodiments, processor 1002 may include one or more central processing units (CPUs) or digital processing units, etc., for executing computer programs or instructions in memory 1001, such that any of the methods described in the above embodiments is executed.

[0157] It should be noted that the specific connection medium between the memory 1001 and the processor 1002 is not limited in the embodiments of this application. The embodiments of this application... Figure 10 In this diagram, the memory 1001 and processor 1002 are connected via bus 1003. The connections between other components are merely illustrative and not intended to be limiting. Bus 1003 can be categorized as an address bus, data bus, control bus, etc. For ease of illustration, Figure 10 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0158] Based on the same inventive concept, embodiments of this application provide a computer-readable storage medium that, when executed by a processor, enables the processor to perform any of the methods described above. Since the principle by which the computer-readable storage medium solves the problem is similar to the quota allocation method at the order receiving end, the implementation of the computer-readable storage medium can be found in the implementation of the method, and repeated details will not be elaborated further.

[0159] Based on the same inventive concept, this application also provides a computer program product, which includes computer program code. When the computer program code is run on a computer, it causes the computer to execute any of the methods described above. Since the principle by which the above-described computer program product solves the problem is similar to the quota allocation method of the order receiving end, the implementation of the above-described computer program product can refer to the implementation of the method, and the repeated parts will not be described again.

[0160] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can 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, etc.) containing computer-usable program code.

[0161] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can 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 one or more blocks of the flowchart illustrations and / or one or more blocks of the block diagrams.

[0162] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flows in a flowchart and / or one or more blocks in a block diagram.

[0163] 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 one or more processes in the flowchart and / or one or more blocks in the block diagram.

[0164] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for allocating credit limits at the order receiving end, characterized in that, include: Based on the historical behavior data of the order-receiving end, the behavior of the order-receiving end under the target amount is quantified to obtain the first reward value of the order-receiving end under the target amount. The order-receiving end is the order-receiving end that needs to be allocated a cancellation amount, and the target amount is any one of multiple cancellation amounts within the platform. Based on the first reward value of the order-receiving end under the target amount, the second reward value of the order-receiving end within the cluster to which the order-receiving end belongs under the target amount, and the third reward value of the order-receiving end within the platform under the target amount, the UCB value of the order-receiving end under the target amount is obtained using the confidence upper bound UCB algorithm. The cancellation limit corresponding to the largest UCB value among the multiple cancellation limits obtained by the order receiving end is determined as the cancellation limit of the order receiving end in the current allocation period.

2. The method as described in claim 1, characterized in that, The process of quantifying the behavior of the order-receiving terminal under the target amount based on its historical behavior data to obtain a first reward value for the order-receiving terminal under the target amount includes: Based on the historical behavior data, determine the number of completed orders and the number of complaints made by the order-receiving end under the target amount; The total reward value is obtained based on the number of completed items, the number of complaints, and the reward coefficient. The ratio of the total reward value to the cumulative number of orders received by the order-receiving end under the target amount is used as the first reward value for the order-receiving end under the target amount.

3. The method as described in claim 2, characterized in that, The reward coefficient is an adjustable coefficient, and the initial value of the reward coefficient is a preset value. If the personal inbound complaint rate in the previous allocation period of the order receiving end is greater than the target inbound complaint rate, the reward coefficient is determined based on the preset value, the preset adjustment step size, the personal inbound complaint rate, and the target inbound complaint rate.

4. The method as described in claim 2, characterized in that, The step of obtaining the UCB value of the order-receiving end under the target amount based on the first reward value of the order-receiving end under the target amount, the second reward value of the order-receiving end within the cluster to which the order-receiving end belongs under the target amount, and the third reward value of the order-receiving end within the platform under the target amount, using the confidence upper bound UCB algorithm, includes: Based on the first reward value, the cumulative number of orders received by the order-receiving terminal under the target amount, and the total number of orders on the platform, a first UCB value is obtained; The second UCB value is obtained based on the second reward value, the cumulative number of orders of all order-receiving terminals in the cluster under the target amount, and the total number of orders of the platform; The third UCB value is obtained based on the third reward value, the cumulative number of orders from all order-receiving terminals within the platform under the target cancellation limit, and the total number of orders on the platform. Based on the first UCB value, the second UCB value, the third UCB value, and their respective empirical weights, the UCB value of the order receiving end under the target amount is obtained.

5. The method as described in claim 4, characterized in that, The individual experience weight of the first UCB value is the ratio of the cumulative number of orders at the order-receiving end to the first value, and is less than 1; The clustering empirical weight of the second UCB value is determined based on the ratio and the first coefficient; The global empirical weight of the third UCB value is determined based on the ratio and the second coefficient, wherein the first coefficient is not less than the second coefficient.

6. The method as described in claim 4, characterized in that, If the cluster to which the order-receiving end belongs changes, the second UCB value is obtained based on the second reward value, the cumulative number of orders from all order-receiving ends within the cluster under the target amount, and the total number of orders on the platform, including: Using a preset attenuation coefficient, the cumulative number of orders, the number of completed orders, and the number of complaints of the order receiving end belonging to the original cluster under the target amount are respectively converted to obtain the first number of orders, the first number of completed orders, and the first number of complaints. Based on the first number of completed orders, the first number of complaints, the second number of completed orders and the second number of complaints within the new cluster to which the order-receiving end belongs under the target amount, and the reward coefficient, the total reward value of the new cluster is obtained; The ratio of the total reward value of the new cluster to the number of second orders of the new cluster under the target amount is used as the second reward value of the new cluster. The number of second orders is the sum of the cumulative number of orders of the new cluster's internal order terminal under the target amount and the number of first orders. The second UCB value is obtained based on the second reward value of the new cluster, the second number of orders, and the total number of orders on the platform.

7. The method according to any one of claims 1-6, characterized in that, The order-receiving end can be any order-receiving end within the platform. Before quantifying the behavior of the order-receiving end under the target amount based on its historical behavior data to obtain the first reward value of the order-receiving end under the target amount, the method further includes: The conditions for canceling the credit limit update are determined to be met. These conditions include reaching a preset update time or the usage duration of the current canceled credit limit on the receiving end reaching a usage duration threshold. Initiate the cancellation quota processing procedure to allocate cancellation quotas to the order-receiving end; The historical behavior data is obtained, and based on the historical behavior data, it is determined that the cumulative number of orders on the order receiving end is less than a first value and greater than a second value.

8. The method as described in claim 7, characterized in that, After acquiring the historical behavior data, the process further includes: Based on the historical behavior data, it is determined that the cumulative number of orders received by the order-receiving end is not less than the first value; Based on the historical behavior data, the behavior of the order-receiving end under the historical quota is quantified to obtain the fourth reward value of the order-receiving end under the historical quota, where the historical quota is any one of the multiple allocated cancellation quotas of the order-receiving end. The target UCB value is obtained based on the fourth reward value of the order receiving end under the historical limit, the cumulative number of orders of the order receiving end under the historical limit, and the total number of orders of the platform; The cancellation quota corresponding to the largest UCB value among the multiple allocated cancellation quotas is determined as the cancellation quota within the current allocation period of the order receiving end.

9. The method as described in claim 7, characterized in that, After acquiring the historical behavior data, the process further includes: Based on the historical behavior data, it is determined that the cumulative number of orders received by the order receiving end is not greater than the second value; When the cumulative number of orders on the order receiving end is not greater than the third value, the minimum cancellation limit among the multiple cancellation limits is determined as the cancellation limit of the current allocation period of the order receiving end; When the cumulative number of orders on the order receiving end is greater than the third value, if the order receiving end is the selected target order receiving end, then a cancellation limit is randomly selected from the multiple cancellation limits other than the minimum limit, and determined as the cancellation limit for the current allocation period of the order receiving end; if the order receiving end is not the target order receiving end, then the cancellation limit corresponding to the largest UCB value among the cancellation limits used by the order receiving ends in the cluster to which the order receiving end belongs is determined as the cancellation limit for the current allocation period of the order receiving end.

10. A quota allocation device for a receiving end, characterized in that, include: The reward value calculation unit is used to quantify the behavior of the order-receiving terminal under the target amount based on the historical behavior data of the order-receiving terminal, and obtain the first reward value of the order-receiving terminal under the target amount. The order-receiving terminal is the order-receiving terminal that needs to be allocated a cancellation quota, and the target quota is any one of multiple cancellation quotas in the platform. The determining unit is used to obtain the UCB value of the order receiving end under the target amount based on the first reward value of the order receiving end under the target amount, the second reward value of the order receiving end within the cluster to which the order receiving end belongs under the target amount, and the third reward value of the order receiving end within the platform under the target amount, using the confidence upper bound UCB algorithm. The quota allocation unit is used to determine the cancellation quota corresponding to the largest UCB value among the multiple cancellation quotas obtained by the order receiving end as the cancellation quota within the current allocation period of the order receiving end.

11. An electronic device, characterized in that, include: Memory is used to store computer programs or instructions; A processor for executing a computer program or instructions in the memory such that the method described in any one of claims 1-9 is performed.

12. A computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor, the processor is able to perform the method as described in any one of claims 1-9.

Images