Case allocation method based on dynamic priority and related device

Abstract

The embodiment of the invention provides a case allocation method based on dynamic priority and a related device, and relates to the technical field of computers. The method comprises the following steps: adjusting the number of cases allowed to enter the system in unit time according to the current allowed rate of the system; the index value of each to-be-audited case entering the system under each evaluation dimension is obtained, the dynamic weight of each evaluation dimension is obtained, and the dynamic weight of each evaluation dimension is dynamically updated through reinforcement learning with maximized revenue and processing efficiency after risk adjustment as target functions; and calculating a priority score of each to-be-audited case based on the dynamic weight of each evaluation dimension and each index value of each to-be-audited case, and allocating each to-be-audited case according to the priority score of each to-be-audited case. Thus, the fundamental defects that in the prior art, queue management is static, and real-time changes cannot be responded are overcome, the problems of resource waste and risk improvement are avoided, and accurate distribution of cases is achieved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a case allocation method and related apparatus based on dynamic priority. Background Technology

[0002] Currently, the manual review modules of financial transaction systems generally experience traffic fluctuations: peak request periods occur at fixed times each day (e.g., morning), leading to backlogs in the review queue, resulting in excessively long waiting times and a degraded user experience; while at other times (e.g., afternoon), review resources are idle, resulting in low overall resource utilization. Here, "case" refers to a transaction request awaiting manual review, such as a loan application.

[0003] In existing technologies, when allocating cases for transaction review, cases are usually allocated and processed according to First-In-First-Out (FIFO) or fixed review priority (e.g., fixed review priority configured according to the severity of customer complaints or customer level).

[0004] However, the above case allocation method results in a static queue management system that cannot respond to real-time changes in traffic fluctuations and case characteristics. At the same time, it is prone to resource misallocation. High-value, low-risk cases may be lost due to long queues, while high-risk cases may be processed quickly, leading to resource waste and increased risk. Summary of the Invention

[0005] This application provides a case allocation method and related apparatus based on dynamic priority to avoid resource waste and risk escalation, and to achieve accurate case allocation.

[0006] In a first aspect, embodiments of this application provide a case allocation method based on dynamic priority, the method comprising: Adjust the number of cases allowed to enter the system per unit time based on the system's current allowed rate; Obtain the indicator values ​​of each pending case entering the system under each evaluation dimension, and obtain the dynamic weights of each evaluation dimension. The dynamic weights of each evaluation dimension are dynamically updated through reinforcement learning with the objective function of maximizing risk-adjusted returns and processing efficiency. Based on the dynamic weights of each evaluation dimension and the indicator values ​​of each case pending review, the priority score of each case pending review is calculated, and the cases pending review are allocated according to their priority scores.

[0007] In an optional embodiment, before adjusting the number of cases allowed to enter the system per unit time based on the system's current allowed rate, the method further includes: Based on the current status data and current traffic prediction results of the system, determine the candidate allowable rate of the system; When the candidate allowed rate is less than the maximum allowed rate, the candidate allowed rate is used as the current allowed rate; When the candidate allowed rate is greater than the maximum allowed rate, the maximum allowed rate will be used as the current allowed rate.

[0008] In one optional embodiment, the current status data includes: basic processing capacity, which represents the overall rate of each online auditor at present; Based on the system's current state data and current traffic prediction results, the candidate permissible rates for the system are determined, including: Adjust the safety margin coefficient based on the current traffic forecast results; Based on the safety margin coefficient and basic processing capacity, candidate allowable rates are determined.

[0009] In one optional embodiment, obtaining the dynamic weights of each evaluation dimension includes: The trained strategy model is invoked to generate dynamic weights for each evaluation dimension based on the case status of each case pending review.

[0010] In an optional embodiment, the method further includes: Based on the training sample set, the policy model to be trained is iteratively trained with the objective function of maximizing risk-adjusted return and processing efficiency to obtain a trained policy model. During each iteration of training, the following operations are performed: The strategy model to be trained is invoked, and the dynamic weights of each evaluation dimension are determined based on the case status of each sample case. Each sample case is processed according to the dynamic weight of the samples in each assessment dimension; When the reward generation conditions are met, the reward for the dynamic weights of the samples in each evaluation dimension is determined based on the objective function and constraint penalties, and the parameters of the policy model to be trained are adjusted based on the reward.

[0011] In one optional embodiment, the cases to be reviewed are allocated according to their priority scores, including: For each pending case, perform the following operations: determine the target priority level that matches the priority score of a pending case, and add the pending case to the priority queue corresponding to the target priority level.

[0012] In one alternative embodiment, the evaluation dimensions include any combination of risk dimension, monetary dimension, user value dimension, and waiting time dimension.

[0013] Secondly, embodiments of this application also provide a case allocation device based on dynamic priority, the device comprising: The traffic shaping module is used to adjust the number of cases allowed to enter the system per unit time based on the system's current allowed rate. The acquisition module is used to acquire the indicator values ​​of each pending case entering the system under each evaluation dimension, as well as the dynamic weights of each evaluation dimension. The dynamic weights of each evaluation dimension are dynamically updated through reinforcement learning with the objective function of maximizing risk-adjusted returns and processing efficiency. The allocation module is used to calculate the priority score of each pending case based on the dynamic weights of each evaluation dimension and the indicator values ​​of each pending case, and to allocate each pending case according to its priority score.

[0014] In an optional embodiment, before adjusting the number of cases allowed to enter the system per unit time based on the system's current allowed rate, the traffic shaping module is further configured to: Based on the current status data and current traffic prediction results of the system, determine the candidate allowable rate of the system; When the candidate allowed rate is less than the maximum allowed rate, the candidate allowed rate is used as the current allowed rate; When the candidate allowed rate is greater than the maximum allowed rate, the maximum allowed rate will be used as the current allowed rate.

[0015] In one optional embodiment, the current status data includes: basic processing capacity, which represents the overall rate of each online auditor at present; When determining the candidate allowable rate of the system based on the current state data and current traffic prediction results, the traffic shaping module is also used for: Adjust the safety margin coefficient based on the current traffic forecast results; Based on the safety margin coefficient and basic processing capacity, candidate allowable rates are determined.

[0016] In an optional embodiment, when obtaining the dynamic weights of each evaluation dimension, the acquisition module is further configured to: The trained strategy model is invoked to generate dynamic weights for each evaluation dimension based on the case status of each case pending review.

[0017] In an optional embodiment, the apparatus further includes a training module, which is used for: Based on the training sample set, the policy model to be trained is iteratively trained with the objective function of maximizing risk-adjusted return and processing efficiency to obtain a trained policy model. During each iteration of training, the following operations are performed: The strategy model to be trained is invoked, and the dynamic weights of each evaluation dimension are determined based on the case status of each sample case. Each sample case is processed according to the dynamic weight of the samples in each assessment dimension; When the reward generation conditions are met, the reward for the dynamic weights of the samples in each evaluation dimension is determined based on the objective function and constraint penalties, and the parameters of the policy model to be trained are adjusted based on the reward.

[0018] In an optional embodiment, when allocating cases to be reviewed according to their priority scores, the allocation module is further configured to: For each pending case, perform the following operations: determine the target priority level that matches the priority score of a pending case, and add the pending case to the priority queue corresponding to the target priority level.

[0019] In one alternative embodiment, the evaluation dimensions include any combination of risk dimension, monetary dimension, user value dimension, and waiting time dimension.

[0020] Thirdly, embodiments of this application also provide an electronic device, including: Processor; and Stored program memory, The program includes instructions that, when executed by the processor, cause the processor to perform the case allocation method based on dynamic priority as described in the first aspect.

[0021] Fourthly, embodiments of this application also provide a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute the case allocation method based on dynamic priority as described in the first aspect.

[0022] Fifthly, this application provides a computer program product that, when invoked by a computer, causes the computer to execute the steps of the dynamic priority-based case allocation method as described in the first aspect.

[0023] The beneficial effects of this application are as follows: In the case allocation method based on dynamic priority provided in this application embodiment, the number of cases allowed to enter the system per unit time is adjusted according to the current allowed rate of the system. First, the indicator values ​​of each case to be reviewed entering the system under each evaluation dimension are obtained, as well as the dynamic weights of each evaluation dimension. The dynamic weights of each evaluation dimension are dynamically updated through reinforcement learning with the objective function of maximizing risk-adjusted returns and processing efficiency. Then, based on the dynamic weights of each evaluation dimension and the indicator values ​​of each case to be reviewed, the priority score of each case to be reviewed is calculated, and the cases to be reviewed are allocated according to their priority scores. In this way, cases are smoothly imported into the system according to the current allowed rate of the system, and dynamic priority scores are calculated to allocate each case to be reviewed. By introducing dynamic weights and combining them with reinforcement learning for optimization, the priority of cases is no longer static or based on simple rules, but can be adaptively adjusted according to global business objectives (risk-adjusted returns and processing efficiency). This solves the fundamental defects of static queue management in the prior art, which cannot respond to real-time changes, avoids resource waste and risk escalation, and achieves accurate case allocation.

[0024] Furthermore, other features and advantages of this application will be set forth in the following description and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described herein are used to provide a further understanding of this application, constitute a part of this application, and do not constitute an improper limitation of this application. In the accompanying drawings: Figure 1 This is a schematic diagram of an optional system architecture applicable to the embodiments of this application; Figure 2 A schematic diagram illustrating the implementation process of a case allocation method based on dynamic priority, provided for an embodiment of this application; Figure 3 A logical diagram illustrating a case allocation method based on dynamic priority provided in an embodiment of this application; Figure 4 A schematic diagram of a case allocation device based on dynamic priority provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0026] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While some embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this application. It should be understood that the drawings and embodiments of this application are for illustrative purposes only and are not intended to limit the scope of protection of this application.

[0027] It should be understood that the steps described in the method embodiments of this application may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this application is not limited in this respect.

[0028] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc., mentioned in this application are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0029] It should be noted that the terms "a" and "a plurality of" used in this application are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0030] The names of the messages or information exchanged between multiple devices in the embodiments of this application are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0031] The following explanations of some terms used in the embodiments of this application are provided to facilitate understanding by those skilled in the art.

[0032] (1) Traffic Tide Phenomenon: refers to the regular and large fluctuation of the number of case requests in the transaction review system within a specific time period (such as one day), like the rise and fall of tides.

[0033] (2) Risk coefficient: A value between 0 and 1, output by an independent risk control model, used to quantify the probability of potential fraud or default risk in a transaction. In this scheme, a lower risk coefficient will increase the priority of the case.

[0034] Based on the above explanations of terms and related terminology, the design concept of the embodiments of this application will be briefly introduced below: Currently, the manual review modules of financial transaction systems generally experience traffic fluctuations: peak request periods occur at fixed times each day (e.g., morning), leading to backlogs in the review queue, resulting in excessively long waiting times and a degraded user experience; while at other times (e.g., afternoon), review resources are idle, resulting in low overall resource utilization. Here, "case" refers to a transaction request awaiting manual review, such as a loan application.

[0035] In existing technologies, when allocating cases for transaction review, cases of the selected users are usually first processed by batch rejection or approval based on fixed user screening rules (e.g., batch rejection of users with a history of overdue payments, and batch approval of users whose debt is below a preset first threshold and whose outstanding loan amount is below a preset second threshold). Then, cases of the remaining users are allocated and processed according to FIFO or fixed review priority (e.g., fixed review priority configured according to the severity of customer complaints or customer level).

[0036] However, using the aforementioned case allocation method, user pre-screening relies on experience to set single-dimensional or statically combined user screening rules, neglecting the dynamic balance between risk and timeliness. Meanwhile, queue management is static, primarily based on FIFO or fixed review priorities, failing to respond to real-time traffic fluctuations and case characteristics. It is also prone to resource misallocation; high-value, low-risk cases may be lost due to long queues, while high-risk cases may be processed quickly, leading to resource waste and increased risk.

[0037] In view of this, this application provides a case allocation method and related apparatus based on dynamic priority. First, the number of cases allowed to enter the system per unit time is adjusted according to the current allowed rate of the system. Then, the indicator values ​​of each case awaiting review in the system under each evaluation dimension are obtained, as well as the dynamic weights of each evaluation dimension. The dynamic weights of each evaluation dimension are dynamically updated through reinforcement learning with the objective function of maximizing risk-adjusted returns and processing efficiency. Finally, based on the dynamic weights of each evaluation dimension and the indicator values ​​of each case awaiting review, a priority score is calculated for each case awaiting review, and the cases are allocated according to their priority scores. In this way, cases are smoothly imported into the system based on the current allowed rate, and dynamic priority scores are calculated for allocation. By introducing dynamic weights and combining them with reinforcement learning for optimization, the priority of cases is no longer static or based on simple rules, but can adaptively adjust according to global business objectives (risk-adjusted returns and processing efficiency). This solves the fundamental defects of static queue management in existing technologies, which cannot respond to real-time changes, avoids resource waste and risk escalation, and achieves accurate case allocation.

[0038] In particular, the preferred embodiments of this application will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only 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.

[0039] See Figure 1 As shown, it is a schematic diagram of an optional system architecture applicable to the embodiments of this application. The system architecture may include: a control center 101, an intelligent traffic shaper 102, a dynamic priority calculation engine 103, and an elastic audit resource pool 104.

[0040] The control center 101 includes: a prediction module 1011, a real-time monitoring module 1012, and a parameter optimization module 1013. The prediction module 1011 predicts future case volume based on historical data; the real-time monitoring module 1012 monitors the current status of each pending case; and the parameter optimization module 1013 finds the globally optimal balance between conflicting business objectives (such as increased loan disbursement, low risk, and fast processing) and dynamically updates the dynamic weights of each evaluation dimension.

[0041] The intelligent flow shaper 102 is the first line of defense in the system. Instead of simply rejecting requests, it smoothly guides cases into the review system through prediction and monitoring, dynamically controlling the rate and flow of cases entering the review system to achieve smooth processing and prevent overload.

[0042] The dynamic priority calculation engine 103 calculates a priority score in real time for each case passing through the intelligent traffic shaper 102, rather than simply sorting by time.

[0043] The Elastic Review Resource Pool 104 supports generating flexible scheduling suggestions for reviewers based on predicted traffic; it supports routing cases to different queues (high priority, normal priority, and buffer queue) based on their priority scores and assigning them to review groups with different skill levels; for cases in the buffer queue that have been waiting for a certain period of time and have extremely low risk, automatic review rules can be triggered for processing, further freeing up manpower.

[0044] The following describes the case allocation method based on dynamic priority provided by the exemplary embodiments of this application in conjunction with the above-described system architecture and with reference to the accompanying drawings. It should be noted that the above-described system architecture is only shown to facilitate understanding of the spirit and principles of this application, and the embodiments of this application are not limited in any way in this respect.

[0045] See Figure 2The diagram shown illustrates the implementation flow of a case allocation method based on dynamic priority provided in this application. Taking a server as an example, the specific implementation flow of this method is as follows: S20: Adjust the number of cases allowed to enter the system per unit time based on the current allowed rate of the system.

[0046] The allowed rate is defined as the number of cases allowed to enter the system per unit time, and the system can be an auditing system.

[0047] In this embodiment of the application, the current allowed rate of the system is determined, and then the number of cases allowed to enter the system per unit time is adjusted according to the current allowed rate of the system.

[0048] Optionally, in this application embodiment, a possible implementation is provided for determining the current allowed rate of the system, specifically by performing the following operations: S200: Based on the current state data and current traffic prediction results of the system, determine the candidate allowed rate of the system.

[0049] In this embodiment of the application, the current traffic prediction result is: the predicted number of current cases based on historical traffic data.

[0050] The Facebook Prophet time series forecasting algorithm is used to obtain the current traffic forecast results. Specifically, the Facebook Prophet time series forecasting algorithm predicts the future number of cases (e.g., the number of cases per hour in the next 24 hours) based on historical traffic data (e.g., traffic data from the past 30 days) and can identify the morning peak and the afternoon trough.

[0051] Additionally, it is worth noting that in this embodiment of the application, flexible scheduling suggestions can be generated for auditors based on the predicted future case volume.

[0052] In this embodiment of the application, the current status data includes: basic processing capacity, which represents the overall rate of each online auditor. The current status data also includes: current queue length, number of online auditors, and other real-time system load.

[0053] In this embodiment, the basic processing capacity is calculated in real time based on the efficiency of online auditors. Optionally, the current processing rate of each online auditor is obtained based on their historical processing data, and then the rates of all online auditors are summed to obtain the basic processing capacity.

[0054] Optionally, in this application embodiment, a possible implementation is provided for determining the candidate allowed rate of the system based on the current state data and current traffic prediction results, specifically by performing the following operations: S2000: Adjust the safety margin coefficient based on the current traffic forecast results.

[0055] The larger the current traffic forecast result (the predicted number of current cases), the smaller the safety margin coefficient.

[0056] In this embodiment, the safety margin coefficient is adjusted based on the current traffic forecast result. Optionally, if the current traffic forecast result shows an increase in the number of cases compared to the previous moment, the safety margin coefficient is decreased; if the current traffic forecast result shows a decrease in the number of cases compared to the previous moment, the safety margin coefficient is increased. Optionally, the safety margin coefficient can also be determined using the calculation formula: Safety Margin Coefficient = Base Coefficient + Current Traffic Forecast Result × Adjustment Factor, where the base coefficient is an empirical value (e.g., 0.1).

[0057] S2001: Determine candidate allowable rates based on safety margin coefficient and basic processing capacity.

[0058] In this embodiment, the safety margin coefficient is increased by 1 to obtain the intermediate coefficient, and then the product of the intermediate coefficient and the basic processing capacity is calculated to obtain the candidate allowable rate.

[0059] Optionally, in this embodiment, the mathematical expression for the candidate allowed rate is: Candidate allowed rate = Basic processing capacity × (1 + safety margin coefficient).

[0060] In this way, by dynamically linking the safety margin coefficient with the current traffic prediction result, the system's allowable rate is no longer arbitrary or fixed. When the current traffic prediction result is large, the system automatically reduces the safety margin coefficient, behaving more conservatively to ensure stability; when the current traffic prediction result is small, the safety margin coefficient is increased, behaving more aggressively to improve efficiency. This can more effectively alleviate traffic surges, thereby improving user experience and resource utilization.

[0061] S201: When the candidate allowed rate is less than the maximum allowed rate, the candidate allowed rate is used as the current allowed rate.

[0062] The maximum permissible rate represents the maximum capacity of the system, which is a preset and relatively stable system limit parameter.

[0063] In this embodiment of the application, it is determined whether the candidate allowed rate is less than the maximum allowed rate. If so, the candidate allowed rate is taken as the current allowed rate.

[0064] S202: When the candidate allowed rate is greater than the maximum allowed rate, the maximum allowed rate shall be used as the current allowed rate.

[0065] In this embodiment of the application, it is determined whether the candidate allowed rate is greater than the maximum allowed rate. If so, the maximum allowed rate is taken as the current allowed rate.

[0066] Optionally, in this embodiment of the application, when the candidate allowed rate is equal to the maximum allowed rate, the candidate allowed rate or the maximum allowed rate is used as the current allowed rate.

[0067] In this way, by introducing the comparison and selection between candidate allowed rates and maximum allowed rates, the system's flexibility under normal fluctuations is ensured (using candidate allowed rates), while a safety valve is set for extreme traffic scenarios (using maximum allowed rates), fundamentally preventing the risk of system overload and crash, and ensuring service stability.

[0068] Additionally, it is worth noting that in this embodiment, the process of determining the current allowed rate of the system can be triggered at fixed times (e.g., every hour, every minute, etc.) or when conditions are met (e.g., when a change in the system's state data is detected). This embodiment does not impose any restrictions on this.

[0069] S21: Obtain the indicator values ​​of each pending case entering the system under each evaluation dimension, and obtain the dynamic weights of each evaluation dimension.

[0070] The dynamic weights of each evaluation dimension are dynamically updated through reinforcement learning, with the objective function of maximizing risk-adjusted returns and processing efficiency; they are not fixed.

[0071] In this embodiment, the evaluation dimensions include any combination of risk, monetary, user value, and waiting time dimensions. That is, each evaluation dimension may include risk and monetary dimensions, or risk, monetary, and user value dimensions, or risk, monetary, user value, and waiting time dimensions, etc. For ease of understanding, the following example uses evaluation dimensions including risk, monetary, user value, and waiting time. The indicator value for the risk dimension is the risk coefficient, for the monetary dimension it is the monetary coefficient, for the user value dimension it is the user value coefficient, and for the waiting time dimension it is the waiting time coefficient. The dynamic weights for the risk dimension and monetary dimension are w1, w2, w3, and w4 respectively.

[0072] Optionally, in this embodiment, the risk coefficient of the pending case is used to quantify the probability of potential fraud or default risk in a transaction. It is output by the risk control model and ranges from 0 to 1. The lower the risk coefficient, the lower the risk and the higher the priority of the pending case. The amount coefficient of the pending case is the ratio of the natural logarithm of the transaction amount of the pending case to the logarithm of the maximum amount threshold, normalized to 0-1. The larger the amount coefficient, the higher the priority of the pending case. The user value coefficient of the user corresponding to the pending case is output by the user value model and ranges from 0 to 1. The larger the user value coefficient, the higher the priority of the pending case. The user value model can calculate the user value coefficient based on the user's historical transactions, loyalty, etc., and this embodiment does not impose any restrictions on this. The waiting time coefficient of the pending case is the ratio between the waiting time of the pending case and the maximum waiting time threshold. The maximum value of the waiting time coefficient is 1. The longer the waiting time coefficient, the higher the priority of the pending case, preventing it from being "starved" (i.e., left unused).

[0073] Optionally, in this embodiment of the application, a trained strategy model is invoked to generate dynamic weights for each evaluation dimension based on the case status of each case to be reviewed.

[0074] In this way, the strategy model avoids the inadequacy of fixed parameters and ensures that high-quality dynamic weights that conform to the global optimal goal can be made in different times and different scenarios, so that the system can make high-quality allocation decisions.

[0075] In this embodiment of the application, based on the training sample set, the policy model to be trained is iteratively trained with the objective function of maximizing risk-adjusted returns and processing efficiency to obtain a trained policy model.

[0076] The mathematical expression for the objective function is: ∑(loan amount × (1 - risk coefficient)) / (average processing time)^0.5.

[0077] Among them, the loan amount refers to the loan amount of a sample case, the risk coefficient refers to the risk coefficient of a sample case, and the average processing time refers to the ratio of the total processing time of each sample case to the total number of sample cases.

[0078] Optionally, in this embodiment of the application, the following operations are performed during one iteration of training: SA1: Call the strategy model to be trained and determine the dynamic weights of each evaluation dimension based on the case status of each sample case.

[0079] The case status includes, but is not limited to: the distribution of cases based on different amounts, risk levels, user value levels, waiting times, the number of currently online reviewers, and processing speed.

[0080] For example, suppose the case status of each sample case is as follows: a large number of high-value, medium-risk cases suddenly flood the queue, while the waiting time for several high-value users' cases begins to lengthen. In this situation, the agent's actions are: significantly increasing the dynamic weight w2 corresponding to the amount dimension, moderately increasing the dynamic weight w3 corresponding to the user value dimension and the dynamic weight w4 corresponding to the waiting time dimension, and decreasing the dynamic weight w1 corresponding to the risk dimension, thus obtaining the sample dynamic weights for each evaluation dimension. This means that the system temporarily favors processing cases involving large amounts and high-value users, and is slightly more lenient towards risk.

[0081] SA2: Process each sample case according to the dynamic weights of each evaluation dimension.

[0082] In this embodiment of the application, the priority score of each sample case is calculated according to the dynamic weight of the samples in each evaluation dimension, and each case to be reviewed is allocated and processed.

[0083] SA3: When the reward generation conditions are met, the reward for the dynamic weights of samples in each evaluation dimension is determined based on the objective function and constraint penalties, and the parameters of the policy model to be trained are adjusted based on the reward.

[0084] The reward generation conditions can be met at the end of the parameter update cycle or when all sample cases are processed; this embodiment does not impose any restrictions on this. Constraint penalties refer to penalties imposed when constraint conditions are not met. Constraint conditions include one or a combination of the following: the overall bad debt rate is less than a preset bad debt rate threshold; the processing time for a preset percentage (XX%) of cases is less than a preset time limit.

[0085] For example, the system processes cases according to the new dynamic weights of each evaluation dimension, resulting in a significant increase in loan amounts and a reduction in user waiting time. However, due to a slightly more lenient approach to risk, a small number of medium-risk cases may be approved. In this case, the objective function generates a large positive reward due to the significant increase in loan amounts. However, if the bad debt rate reaches the warning line, a negative penalty will be imposed, resulting in the final reward.

[0086] In this way, the reward mechanism design uses high-level business objectives such as risk-adjusted returns and processing efficiency as a compass for model optimization, finding the globally optimal balance between conflicting business objectives (such as more loans, lower risk, and faster processing), which can continuously improve business performance. The introduction of constraints and penalties ensures that the learning process will not violate basic business rules in pursuit of rewards, thus ensuring the security and compliance of the system behavior.

[0087] In this embodiment, by performing multiple rounds of iterative training on the policy model to be trained, the agent gradually learns what actions to take under what conditions (i.e., how to adjust the dynamic weights of each risk dimension) to most likely obtain the maximum cumulative reward in the long term.

[0088] In addition, it is worth noting that the trained policy model can be continuously optimized in the embodiments of this application.

[0089] S22: Based on the dynamic weights of each evaluation dimension and the indicator values ​​of each case to be reviewed, calculate the priority score of each case to be reviewed, and allocate each case to be reviewed according to its priority score.

[0090] In this embodiment of the application, the following operations are performed for each case to be reviewed: according to the dynamic weight of each evaluation dimension, the indicator values ​​of a case to be reviewed are weighted and summed to obtain the priority score of the case to be reviewed.

[0091] Optionally, in this embodiment of the application, the mathematical expression for the priority score of a case pending review is: w1×(1-risk coefficient)+w2×amount coefficient+w3×user value coefficient+w4×user waiting time coefficient.

[0092] In this way, based on multi-dimensional coefficients and using dynamic weights to calculate priority scores in real time, this comprehensive and dynamic evaluation can more comprehensively and accurately depict the overall value and urgency of a case pending review compared to a single dimension (such as time only) or fixed rules and fixed weights.

[0093] In this embodiment of the application, after obtaining the priority score of a case to be reviewed, a target priority level that matches the priority score of the case to be reviewed is determined, and the case to be reviewed is added to the priority queue corresponding to the target priority level.

[0094] For example, priority levels include: high priority, medium priority, and low priority. The priority queue corresponding to high priority is a high-priority queue, the priority queue corresponding to medium priority is a normal priority queue, and the priority queue corresponding to low priority is a buffer queue. If the priority score of a case to be reviewed meets the high-priority condition, the target priority level of the case to be reviewed is high priority, and the case to be reviewed is added to the high-priority queue. If the priority score of a case to be reviewed meets the medium priority condition, the target priority level of the case to be reviewed is medium priority, and the case to be reviewed is added to the normal priority queue. If the priority score of a case to be reviewed meets the low priority condition, the target priority level of the case to be reviewed is low priority, and the case to be reviewed is added to the buffer queue. Furthermore, it is worth noting that in this embodiment, the type of priority level is not limited. For example, priority levels may also include: super high priority, high priority, medium priority, and low priority.

[0095] Optionally, in this embodiment of the application, each priority queue corresponds to a different review group.

[0096] For example, the high-priority queue corresponds to the expert review group, the ordinary-priority queue corresponds to the regular review group, and the buffer queue corresponds to the off-peak review group. The off-peak review group consists of reviewers during their off-peak hours. Furthermore, for cases waiting in the buffer queue for more than a certain period of time and with extremely low risk, automatic review rules can be triggered for processing, further freeing up manpower.

[0097] By categorizing cases into different queues based on their priority scores, the system lays the foundation for differentiated processing. This allows the system to route high-priority cases to more skilled and experienced auditors (expert review teams), ensuring both the quality and efficiency of important cases while optimizing the allocation of human resources.

[0098] Based on the above embodiments, see Figure 3 The diagram shown is a logical illustration of a case allocation method based on dynamic priority provided in this application embodiment. The specific implementation process is as follows: First, as transaction cases flow in, the intelligent traffic shaper adjusts the number of cases allowed to enter the system per unit time based on the system's current allowed rate. Then, the dynamic priority calculation engine obtains the indicator values ​​of each pending case under various evaluation dimensions (calculating risk coefficient, calculating amount coefficient, calculating user value coefficient, calculating waiting time coefficient) and the dynamic weights of each evaluation dimension. Based on the dynamic weights of each evaluation dimension and the indicator values ​​of each pending case, the priority score of each pending case is calculated. Then, according to the priority scores of each pending case, case routing is performed, and each pending case is added to the corresponding priority queue (high priority queue, normal priority queue, and cache queue). Finally, the review results are output.

[0099] Furthermore, based on the same technical concept, embodiments of this application provide a case allocation device based on dynamic priority, which is used to implement the above-described method flow of embodiments of this application. For example, see [link to relevant documentation]. Figure 4 As shown, the case allocation device 400 based on dynamic priority may include: a flow shaping module 401, an acquisition module 402, an allocation module 403, and a training module 404, wherein: The flow shaping module 401 is used to adjust the number of cases allowed to enter the system per unit time according to the current allowed rate of the system. The acquisition module 402 is used to acquire the indicator values ​​of each pending case entering the system under each evaluation dimension, and to acquire the dynamic weights of each evaluation dimension. The dynamic weights of each evaluation dimension are dynamically updated through reinforcement learning with the objective function of maximizing risk-adjusted returns and processing efficiency. The allocation module 403 is used to calculate the priority score of each case to be reviewed based on the dynamic weights of each evaluation dimension and the indicator values ​​of each case to be reviewed, and to allocate each case to be reviewed according to the priority score of each case to be reviewed.

[0100] In an alternative embodiment, before adjusting the number of cases allowed to enter the system per unit time based on the system's current allowed rate, the traffic shaping module 401 is further configured to: Based on the current status data and current traffic prediction results of the system, determine the candidate allowable rate of the system; When the candidate allowed rate is less than the maximum allowed rate, the candidate allowed rate is used as the current allowed rate; When the candidate allowed rate is greater than the maximum allowed rate, the maximum allowed rate will be used as the current allowed rate.

[0101] In one optional embodiment, the current status data includes: basic processing capacity, which represents the overall rate of each online auditor at present; When determining the candidate allowable rate of the system based on the current state data and the current traffic prediction results, the traffic shaping module 401 is also used for: Adjust the safety margin coefficient based on the current traffic forecast results; Based on the safety margin coefficient and basic processing capacity, candidate allowable rates are determined.

[0102] In an optional embodiment, when obtaining the dynamic weights of each evaluation dimension, the acquisition module 402 is further configured to: The trained strategy model is invoked to generate dynamic weights for each evaluation dimension based on the case status of each case pending review.

[0103] In an optional embodiment, the apparatus further includes a training module 404, which is used for: Based on the training sample set, the policy model to be trained is iteratively trained with the objective function of maximizing risk-adjusted return and processing efficiency to obtain a trained policy model. During each iteration of training, the following operations are performed: The strategy model to be trained is invoked, and the dynamic weights of each evaluation dimension are determined based on the case status of each sample case. Each sample case is processed according to the dynamic weight of the samples in each assessment dimension; When the reward generation conditions are met, the reward for the dynamic weights of the samples in each evaluation dimension is determined based on the objective function and constraint penalties, and the parameters of the policy model to be trained are adjusted based on the reward.

[0104] In an optional embodiment, when allocating cases to be reviewed according to their priority scores, the allocation module 403 is further configured to: For each pending case, perform the following operations: determine the target priority level that matches the priority score of a pending case, and add the pending case to the priority queue corresponding to the target priority level.

[0105] In one alternative embodiment, the evaluation dimensions include any combination of risk dimension, monetary dimension, user value dimension, and waiting time dimension.

[0106] Based on the description of the method and apparatus embodiments above, an exemplary embodiment of the present invention also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, which, when executed by the at least one processor, causes the electronic device to perform the method according to an embodiment of the present invention.

[0107] This application also provides a non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to an embodiment of this application.

[0108] This application also provides a computer program product, including a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to an embodiment of this application.

[0109] See Figure 5 The diagram shown below illustrates the structure of an electronic device 500 that can serve as a server or client in this application, and is an example of a hardware device that can be applied to various aspects of this application. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the application described and / or claimed herein.

[0110] like Figure 5 As shown, the electronic device 500 includes a computing unit 501, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. The RAM 503 may also store various programs and data required for the operation of the device 500. The computing unit 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0111] Multiple components in electronic device 500 are connected to I / O interface 505, including: input unit 506, output unit 507, storage unit 508, and communication unit 509. Input unit 506 can be any type of device capable of inputting information to electronic device 500. Input unit 506 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of electronic device. Output unit 507 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 508 may include, but is not limited to, disks and optical discs. Communication unit 509 allows electronic device 500 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers and / or chipsets, such as Bluetooth devices, WiFi devices, worldwide interoperability for microwave access (WiMax) devices, cellular communication devices, and / or the like.

[0112] The computing unit 501 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above. For example, in some embodiments, the above-described dynamic priority-based case allocation method can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as storage unit 508. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 500 via ROM 502 and / or communication unit 509. In some embodiments, the computing unit 501 can be configured to perform the above-described dynamic priority-based case allocation method by any other suitable means (e.g., by means of firmware).

[0113] The program code used to implement the methods of this application may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0114] In the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, erasable programmable read-only memory (EPROM) or flash memory, optical fibers, compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0115] As used in this application, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device, PLD) used to provide machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal used to provide machine instructions and / or data to a programmable processor.

[0116] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a cathode ray tube (CRT) or liquid crystal display (LCD) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0117] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0118] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.

[0119] Furthermore, it should be understood that the above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of the invention. Therefore, any equivalent variations made in accordance with the claims of this invention are still within the scope of this application.

Claims

1. A case allocation method based on dynamic priority, characterized in that, include: Adjust the number of cases allowed to enter the system per unit time based on the system's current allowed rate; The system obtains the indicator values ​​of each pending case entering the system under each evaluation dimension, and obtains the dynamic weights of each evaluation dimension. The dynamic weights of each evaluation dimension are dynamically updated through reinforcement learning with the objective function of maximizing risk-adjusted returns and processing efficiency. Based on the dynamic weights of each evaluation dimension and the indicator values ​​of each case pending review, a priority score is calculated for each case pending review, and the cases pending review are allocated according to their priority scores.

2. The method as described in claim 1, characterized in that, Before adjusting the number of cases allowed to enter the system per unit time based on the system's current allowed rate, the method further includes: Based on the current state data and current traffic prediction results of the system, the candidate allowed rate of the system is determined; When the candidate allowed rate is less than the maximum allowed rate, the candidate allowed rate is taken as the current allowed rate; When the candidate allowed rate is greater than the maximum allowed rate, the maximum allowed rate is taken as the current allowed rate.

3. The method as described in claim 2, characterized in that, The current status data includes: basic processing capacity, which represents the overall speed of each online auditor at present; The step of determining the candidate allowed rate of the system based on the current state data and current traffic prediction results includes: Adjust the safety margin coefficient based on the current traffic prediction results; The candidate allowed rate is determined based on the safety margin coefficient and the basic processing capacity.

4. The method as described in claim 1, characterized in that, The process of obtaining the dynamic weights of each evaluation dimension includes: The trained strategy model is invoked to generate dynamic weights for each evaluation dimension based on the case status of each case to be reviewed.

5. The method as described in claim 4, characterized in that, The method further includes: Based on the training sample set, the policy model to be trained is iteratively trained with the objective function of maximizing risk-adjusted return and processing efficiency to obtain the trained policy model. During each iteration of training, the following operations are performed: The strategy model to be trained is invoked, and the dynamic weights of each evaluation dimension are determined based on the case status of each sample case. Each sample case is processed according to the sample dynamic weights of each evaluation dimension. When the reward generation conditions are met, the reward for the dynamic weights of the samples in each evaluation dimension is determined based on the objective function and the constraint penalty, and the parameters of the policy model to be trained are adjusted based on the reward.

6. The method as described in claim 1, characterized in that, The allocation of cases to be reviewed according to their priority scores includes: For each case pending review, the following operations are performed: determine the target priority level that matches the priority score of a case pending review, and add the case pending review to the priority queue corresponding to the target priority level.

7. The method according to any one of claims 1-6, characterized in that, The evaluation dimensions include any combination of the following: risk dimension, monetary dimension, user value dimension, and waiting time dimension.

8. A case allocation device based on dynamic priority, characterized in that, include: The traffic shaping module is used to adjust the number of cases allowed to enter the system per unit time according to the current allowed rate of the system. The acquisition module is used to acquire the indicator values ​​of each pending case entering the system under each evaluation dimension, and to acquire the dynamic weights of each evaluation dimension. The dynamic weights of each evaluation dimension are dynamically updated through reinforcement learning with the objective function of maximizing risk-adjusted returns and processing efficiency. The allocation module is used to calculate the priority score of each case to be reviewed based on the dynamic weights of each evaluation dimension and the indicator values ​​of each case to be reviewed, and to allocate each case to be reviewed according to the priority score of each case to be reviewed.

9. An electronic device, comprising: processor; as well as Stored program memory, The program includes instructions that, when executed by the processor, cause the processor to perform the method as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-7.

Images