WO2020181907A1 - 决策优化方法及装置 - Google Patents

决策优化方法及装置 Download PDF

Info

Publication number
WO2020181907A1
WO2020181907A1 PCT/CN2020/070492 CN2020070492W WO2020181907A1 WO 2020181907 A1 WO2020181907 A1 WO 2020181907A1 CN 2020070492 W CN2020070492 W CN 2020070492W WO 2020181907 A1 WO2020181907 A1 WO 2020181907A1
Authority
WO
WIPO (PCT)
Prior art keywords
decision
variable
decision feature
feature variable
combination
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2020/070492
Other languages
English (en)
French (fr)
Inventor
刘星
许辽萨
易灿
王维强
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Alibaba Group Holding Ltd
Original Assignee
Alibaba Group Holding Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Alibaba Group Holding Ltd filed Critical Alibaba Group Holding Ltd
Publication of WO2020181907A1 publication Critical patent/WO2020181907A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0635—Risk analysis of enterprise or organisation activities

Definitions

  • the present disclosure generally relates to the field of computer technology, and more specifically, to methods and devices for decision optimization.
  • service providers hope to provide users with customized services to maximize the user experience, protect the rights of users, or achieve other business goals. For example, in a marketing scenario, maximize the registration conversion rate or retention rate of customers. In transaction risk management and control, the risk of users being defrauded of each transaction fund is minimized.
  • the essence of this kind of demand is to make personalized decisions for different users, and this kind of personalized decision can achieve the optimal decision effect, so as to maximize the decision purpose.
  • the present disclosure provides a method and device for decision optimization.
  • personalized decision-making optimization can be realized for different users.
  • a method for decision optimization including: determining the contribution of each decision feature variable in a prediction model to the prediction result under user feature data, the prediction model being created based on the decision goal ,
  • the prediction model includes decision feature variables and non-decision feature variables; based on the determined contribution of each decision feature variable to the prediction result, construct a combination of decision feature variables to be optimized; for each of the constructed combination of decision feature variables The variable value of the decision feature variable is optimized to make the corresponding prediction result of the prediction model the best; and based on the variable value of each decision feature variable in the combination of decision feature variables obtained after the optimization processing Decision optimization processing.
  • determining the contribution of each decision feature variable in the prediction model to the prediction result of the user feature data includes: using an explanatory model to determine that each decision feature variable in the prediction model is in the user feature data Contribution of the predicted results under.
  • the interpretation model includes one of the following interpretation models: a Shap value model, a LIME model, and a DeepLift model.
  • constructing the decision feature variable combination to be optimized may include: based on the prediction result of each decision feature variable determined Contribution, sorting the various decision feature variables; and selecting a predetermined number of decision feature variables with greater contribution from the various decision feature variables after the ranking, as a combination of decision feature variables to be optimized.
  • performing optimization processing on the value of each decision feature variable in the constructed combination of decision feature variables may include: using one of the following optimization algorithms To optimize the value of each decision feature variable in the constructed decision feature variable combination: particle swarm algorithm, genetic algorithm, annealing algorithm.
  • performing optimization processing on the variable value of each decision feature variable in the constructed decision feature variable combination may include: within a predetermined value range of the decision variable, The variable value of each decision feature variable in the decision feature variable combination is optimized.
  • an apparatus for decision optimization including: a contribution degree determination unit configured to determine the contribution degree of the prediction result of each decision feature variable in the prediction model under user feature data, the The prediction model is created based on the decision objective, and the prediction model includes decision feature variables and non-decision feature variables; the decision feature variable combination construction unit is configured to construct the predicted result based on the determined contribution of each decision feature variable Optimized decision feature variable combination; the optimization processing unit is configured to perform optimization processing on the variable value of each decision feature variable in the constructed decision feature variable combination, so that the corresponding prediction result of the prediction model is the best ; And a decision optimization unit, configured to perform decision optimization processing according to the value of each decision feature variable in the combination of decision feature variables obtained after optimization processing.
  • the contribution degree determination unit is configured to use an interpretation model to determine the prediction result contribution degree of each decision feature variable in the prediction model under user feature data.
  • the interpretation model includes one of the following interpretation models: a Shap value model, a LIME model, and a DeepLift model.
  • the decision feature variable combination construction unit includes: a sorting module configured to contribute to each decision feature variable based on the determined contribution to the prediction result of each decision feature variable Sorting; and a feature selection module configured to select a predetermined number of decision feature variables with a larger contribution from the decision feature variables after the ranking, as a combination of decision feature variables to be optimized.
  • the optimization processing unit is configured to use one of the following optimization algorithms to select the variable of each decision feature variable in the constructed combination of decision feature variables Value optimization processing: particle swarm algorithm, genetic algorithm, annealing algorithm.
  • the optimization processing unit is configured to: within a predetermined value range of the decision variable, perform a variable value evaluation on each decision feature variable in the constructed decision feature variable combination. Optimization processing.
  • a computing device including: at least one processor, and a memory coupled with the at least one processor, the memory storing instructions, when the instructions are used by the at least one processor During execution, the at least one processor is caused to execute the method for decision optimization as described above.
  • a non-transitory machine-readable storage medium which stores executable instructions that, when executed, cause the machine to execute the method for decision optimization as described above.
  • Fig. 1 shows a flowchart of a decision optimization method according to an embodiment of the present disclosure
  • FIG. 2 shows a flowchart of an example of a process of determining the contribution degree of the prediction result of a decision feature variable according to an embodiment of the present disclosure
  • FIG. 3 shows a flowchart of an example of a process of constructing a decision feature variable combination according to an embodiment of the present disclosure
  • FIG. 4 shows a flowchart of an example of the on-line switching process of the personalized decision mechanism according to an embodiment of the present disclosure
  • Figure 5 shows a block diagram of a decision-making optimization device according to an embodiment of the present disclosure
  • Figure 6 shows a block diagram of an example of a decision feature variable combination construction unit according to an embodiment of the present disclosure
  • Fig. 7 shows a block diagram of a computing device for decision optimization according to an embodiment of the present disclosure.
  • the term “including” and its variants means open terms, meaning “including but not limited to.”
  • the term “based on” means “based at least in part on.”
  • the terms “one embodiment” and “an embodiment” mean “at least one embodiment.”
  • the term “another embodiment” means “at least one other embodiment.”
  • the terms “first”, “second”, etc. may refer to different or the same objects. Other definitions can be included below, either explicit or implicit. Unless clearly indicated in the context, the definition of a term is consistent throughout the specification.
  • Fig. 1 shows a flowchart of a decision optimization method according to an embodiment of the present disclosure.
  • the contribution degree of each decision feature variable in the prediction model to the prediction result under the user feature data is determined.
  • the prediction model is created based on the decision objective, and the prediction model includes decision feature variables and non-decision feature variables.
  • the user characteristic data is user characteristic data corresponding to decision characteristic variables and non-decision characteristic variables.
  • the decision target is the number of users registered to convert
  • the decision target is the sum of the user’s losses due to fraudulent transactions.
  • a suitable feature variable group is created to establish a prediction model, so as to predict the decision goal based on user feature data, for example, predict whether a user will convert or predict the amount of fraud in a user transaction.
  • non-decision characteristic variable refers to characteristic variables that the decision maker cannot interfere, such as the user's age, personality, and past behavior history.
  • the term “decision characteristic variable” refers to the characteristic variable that the decision-making party can interfere in the change.
  • the characteristic variables such as discounts and rights issued to the user, or in the transaction
  • characteristic variables such as risk reminders output to the user.
  • the prediction model may be created in advance based on the decision target using training data, or may be created in real time using training data after the decision target is obtained. Similarly, the decision target may be input in advance or in real time.
  • the training data used for prediction model training should cover user groups with as many features as possible, and try to select as many decision variables as possible to ensure the richness of the training data dimensions .
  • an explanatory model may be used to determine the contribution of each decision feature variable in the prediction model to the prediction result of the user feature data.
  • the interpretation model includes one of the following interpretation models: Shap value model, LIME model and DeepLift model. How to use the interpretation model to determine the contribution of each decision feature variable in the prediction model to the prediction result of the user feature data will be described below with reference to FIG. 2.
  • the decision feature variable combination to be optimized is constructed. How to construct the combination of decision feature variables to be optimized based on the determined contribution of the predicted result of each decision feature variable will be described below with reference to the example shown in FIG. 3.
  • the variable value of each decision feature variable in the constructed decision feature variable combination is optimized, so that the corresponding prediction result of the prediction model is the best good.
  • performing optimization processing on the value of each decision feature variable in the constructed combination of decision feature variables may include: using one of the following optimization algorithms to perform optimization on the constructed
  • the variable value of each decision feature variable in the decision feature variable combination is optimized for processing: particle swarm algorithm, genetic algorithm, annealing algorithm.
  • the optimization algorithm used can be determined based on the type of the decision feature variable to be optimized. For example, when the decision feature variable to be optimized is a categorical variable, the optimization algorithm preferably uses a particle swarm algorithm, such as a discrete particle swarm algorithm.
  • performing optimization processing on the value of each decision feature variable in the constructed decision feature variable combination may include: within a predetermined value range of the decision variable, The variable value of each decision feature variable in the feature variable combination is optimized.
  • the value range of the decision variable can be appropriately adjusted according to the actual situation.
  • the decision optimization process is performed according to the variable value of each decision feature variable in the decision feature variable combination obtained after optimization processing .
  • the variable value of each decision characteristic variable is used to optimize and adjust the decision mechanism used in the decision engine.
  • FIG. 2 shows a flowchart of an example of a process of determining the contribution degree of the prediction result of a decision feature variable according to an embodiment of the present disclosure.
  • a decision feature variable is selected as the initial current decision feature variable.
  • a decision feature variable can be randomly selected from each decision feature variable as the initial current decision feature variable.
  • the operations from block 220 to block 270 are performed until the corresponding prediction result contribution degree is determined for all decision feature variables.
  • the prediction result corresponding to any combination of the remaining decision feature variables other than the current decision feature variable in the decision feature variable combination is determined.
  • the prediction model is used to calculate the input to the prediction model corresponding to any combination of features V2, V3, V4 The model prediction results in the case of user characteristic data.
  • the prediction result corresponding to any combination of feature variables that includes the current decision feature variable in the combination of decision feature variables is determined.
  • the feature variable combination described here is a feature variable combination obtained by adding a feature variable V1 to each feature combination in the block 220. Then, in block 240, the model prediction result determined in block 220 (the model prediction result without the current decision feature variable) and the corresponding model prediction result determined in block 230 (the current decision feature variable is present) are calculated. Model prediction results) difference.
  • the difference between the prediction results of each model is determined, in block 250, the difference of all the model prediction results is averaged to obtain the prediction result contribution degree of the current decision feature variable. Then, at block 260, it is determined whether there are unprocessed decision feature variables. If it exists, in block 270, a decision feature variable from the unprocessed decision feature variables is selected as the next current decision feature variable, and then returns to block 220, and the operations from blocks 220 to 270 are performed again.
  • FIG. 3 shows a flowchart of an example of a process of constructing a combination of decision feature variables according to an embodiment of the present disclosure.
  • the decision feature variables are sorted based on the determined contribution of each decision feature variable.
  • a predetermined number of decision feature variables with a larger contribution degree are selected from the sorted decision feature variables as the combination of decision feature variables to be optimized.
  • the predetermined number may be determined based on actual application scenarios or other suitable conditions.
  • other suitable methods may also be used to construct the combination of decision feature variables to be optimized.
  • the user characteristic data of the user is used to determine the prediction result contribution of each decision feature variable in the prediction model, and the decision feature to be optimized is constructed based on the determined prediction result Variable combinations, and an optimization algorithm is used to determine the value of each decision feature variable in the decision feature variable combination.
  • the decision optimization method because different user characteristic data can obtain different prediction result contributions of each decision characteristic variable, different combinations of decision characteristic variables can be constructed, and thus the optimized decision characteristic for different users The combination of variables will also be different to achieve personalized decision-making optimization.
  • the decision optimization method of the present disclosure by appropriately adjusting the value range of the decision feature variable according to the actual situation, and within the value range of the decision feature variable, the variable of each decision feature variable in the constructed decision feature variable combination The value is selected for optimization processing, which can improve the optimization effect, thereby improving the decision optimization effect.
  • FIG. 4 shows a flowchart of an example of the on-line switching process of the personalized decision mechanism according to an embodiment of the present disclosure.
  • the inflowing customer data to be decided is randomly assigned to the original decision-making mechanism or optimization.
  • the subsequent personalized decision-making mechanism processes, and then makes decisions, and records the corresponding decision-making effects.
  • the decision-making effect of the original decision-making mechanism or the optimized personalized decision-making mechanism is evaluated. If the decision-making effect of the original decision-making mechanism is better, the original decision-making mechanism is maintained. If the optimized personalized decision-making mechanism is better Good, then use the optimized personalized decision-making mechanism to replace the original decision-making mechanism, thereby ensuring the stability of the decision-making business.
  • FIG. 5 shows a block diagram of a decision optimization device 500 according to an embodiment of the present disclosure.
  • the decision optimization device 500 includes a contribution degree determination unit 510, a decision feature variable combination construction unit 520, an optimization processing unit 530, and a decision optimization unit 540.
  • the contribution degree determining unit 510 is configured to determine the prediction result contribution degree of each decision feature variable in the prediction model under the user feature data.
  • the prediction model is created based on the decision goal, and the prediction model includes decision feature variables and non-decision Characteristic variables.
  • the operation of the contribution degree determination unit 510 may refer to the operation of the block 110 described above with reference to FIG. 1 and the operation described with reference to FIG. 2.
  • the decision feature variable combination construction unit 520 is configured to construct a decision feature variable combination to be optimized based on the determined contribution degree of the prediction result of each decision feature variable.
  • the operation of the decision feature variable combination construction unit 520 may refer to the operation of the block 120 described above with reference to FIG. 1 and the operation described with reference to FIG. 3.
  • the optimization processing unit 530 is configured to perform optimization processing on the variable value of each decision feature variable in the constructed decision feature variable combination, so that the corresponding prediction result of the prediction model is the best.
  • the optimization processing unit 530 is configured to use one of the following optimization algorithms to perform optimization processing on the value of each decision feature variable in the constructed combination of decision feature variables : Particle swarm algorithm, genetic algorithm, annealing algorithm.
  • the optimization processing unit 530 is configured to perform optimization processing on the variable value of each decision feature variable in the constructed decision feature variable combination within a predetermined value range of the decision variable.
  • the operation of the optimization processing unit 530 may refer to the operation of the block 130 described above with reference to FIG. 1.
  • the decision optimization unit 540 is configured to perform decision optimization processing according to the value of each decision feature variable in the decision feature variable combination obtained after optimization processing.
  • the operation of the decision optimization unit 540 may refer to the operation of the block 140 described above with reference to FIG. 1.
  • the contribution degree determination unit 510 may be configured to use an interpretation model to determine the prediction result contribution degree of each decision feature variable in the prediction model under the user feature data.
  • the interpretation model may include one of the following interpretation models: Shap value model, LIME model, and DeepLift model.
  • FIG. 6 shows a block diagram of an example of the decision feature variable combination construction unit 520 according to an embodiment of the present disclosure.
  • the decision feature variable combination construction unit 520 includes a ranking module 521 and a feature selection module 523.
  • the ranking module 521 is configured to rank the decision feature variables based on the determined contribution of each decision feature variable.
  • the feature selection module 523 is configured to select a predetermined number of decision feature variables with a large contribution from the sorted decision feature variables, as a combination of decision feature variables to be optimized.
  • the above decision-making optimization device can be implemented by hardware, or by software or a combination of hardware and software.
  • FIG. 7 shows a hardware structure diagram of a computing device 700 for decision optimization according to an embodiment of the present disclosure.
  • the computing device 700 may include at least one processor 710, memory 720, memory 730, and communication interface 740, and at least one processor 710, memory 720, memory 730, and communication interface 740 are connected together via a bus 760.
  • At least one processor 710 executes at least one computer readable instruction (ie, the above-mentioned element implemented in the form of software) stored or encoded in the memory.
  • computer-executable instructions are stored in the memory, which when executed, cause at least one processor 710 to: determine the prediction result contribution degree of each decision feature variable in the prediction model under the user feature data, the prediction model It is created based on the decision objective, and the prediction model includes decision feature variables and non-decision feature variables; based on the determined contribution of each decision feature variable to the prediction result, the combination of decision feature variables to be optimized is constructed; The variable value of each decision feature variable in the feature variable combination is optimized to make the corresponding prediction result of the prediction model the best; and according to the decision feature variable in the decision feature variable combination obtained after the optimization processing The value of the variable is used for decision-making optimization processing.
  • the computing device 700 may include, but is not limited to: personal computers, server computers, workstations, desktop computers, laptop computers, notebook computers, mobile computing devices, smart phones, tablet computers, cellular phones, personal digital assistants (PDA), handheld devices, messaging devices, wearable computing devices, consumer electronic devices, etc.
  • PDA personal digital assistants
  • a program product such as a non-transitory machine-readable medium.
  • the non-transitory machine-readable medium may have instructions (ie, the above-mentioned elements implemented in the form of software), which when executed by a machine, cause the machine to execute the various embodiments described above in conjunction with FIGS. 1-6 in the various embodiments of the present disclosure. Operation and function.
  • a system or device equipped with a readable storage medium may be provided, and the software program code for realizing the function of any one of the above embodiments is stored on the readable storage medium, and the computer or device of the system or device The processor reads out and executes the instructions stored in the readable storage medium.
  • a program product such as a non-transitory machine-readable medium.
  • the non-transitory machine-readable medium may have instructions (ie, the above-mentioned elements implemented in the form of software), which when executed by a machine, cause the machine to execute the various embodiments described above in conjunction with FIGS. 1-6 in the various embodiments of the present disclosure. Operation and function.
  • a system or device equipped with a readable storage medium may be provided, and the software program code for realizing the function of any one of the above embodiments is stored on the readable storage medium, and the computer or device of the system or device The processor reads out and executes the instructions stored in the readable storage medium.
  • the program code itself read from the readable medium can realize the function of any one of the above embodiments, so the machine readable code and the readable storage medium storing the machine readable code constitute the present invention a part of.
  • Examples of readable storage media include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD-RW), magnetic tape, Volatile memory card and ROM.
  • the program code can be downloaded from a server computer or cloud via a communication network.
  • the device structure described in the foregoing embodiments may be a physical structure or a logical structure. That is, some units may be implemented by the same physical entity, or some units may be implemented by multiple physical entities, or may be implemented by multiple physical entities. Some components in independent devices are implemented together.
  • the hardware unit or module can be implemented mechanically or electrically.
  • a hardware unit, module or processor may include permanent dedicated circuits or logic (such as a dedicated processor, FPGA or ASIC) to complete the corresponding operation.
  • the hardware unit or processor may also include programmable logic or circuits (such as general-purpose processors or other programmable processors), which may be temporarily set by software to complete corresponding operations.
  • the specific implementation mode mechanical method, or dedicated permanent circuit, or temporarily set circuit

Landscapes

  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Engineering & Computer Science (AREA)
  • Strategic Management (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Economics (AREA)
  • Operations Research (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Marketing (AREA)
  • Educational Administration (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种决策优化方法及装置。该方法包括：确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度（110），所述预测模型是基于决策目标创建的，所述预测模型包括决策特征变量和非决策特征变量；基于所确定出的各个决策特征变量的预测结果贡献度，构建待优化的决策特征变量组合（120）；对决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理（130），以使得所述预测模型的对应预测结果最佳；以及根据优化处理后各个决策特征变量的变量取值来进行决策优化处理（140）。利用该方法，可以实现针对用户的个性化决策优化。

Description

决策优化方法及装置 技术领域
本公开通常涉及计算机技术领域，更具体地，涉及用于决策优化的方法及装置。
背景技术
在很多商业场景中，服务提供方希望为用户提供定制化服务，以最大可能地提高用户的体验，保障用户的权益，或者达到其他商业目标。例如，在营销场景中，最大化客户的注册转化率或留存率。在交易风险管控中，最小化用户每笔交易资金被诈骗的风险。这类需求本质是针对不同的用户进行个性化决策，并且这种个性化决策能够实现最优决策效果，从而最大化地实现决策目的。
发明内容
鉴于上述，本公开提供了一种用于决策优化的方法及装置。利用该方法及装置，能够针对不同的用户实现个性化决策优化。
根据本公开的一个方面，提供了一种用于决策优化的方法，包括：确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度，所述预测模型是基于决策目标创建的，所述预测模型包括决策特征变量和非决策特征变量；基于所确定出的各个决策特征变量的预测结果贡献度，构建待优化的决策特征变量组合；对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理，以使得所述预测模型的对应预测结果最佳；以及根据经过寻优处理后得到的决策特征变量组合中的各个决策特征变量的变量取值来进行决策优化处理。
可选地，在上述方面的一个示例中，确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度包括：使用解释模型来确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度。
可选地，在上述方面的一个示例中，所述解释模型包括下述解释模型中的一种：Shap value模型、LIME模型和DeepLift模型。
可选地，在上述方面的一个示例中，基于所确定出的各个决策特征变量的预测结果贡献度，构建待优化的决策特征变量组合可以包括：基于所确定出的各个决策特征变 量的预测结果贡献度，对所述各个决策特征变量进行排序；以及从所述排序后的各个决策特征变量中选择贡献度较大的预定数目个决策特征变量，作为待优化的决策特征变量组合。
可选地，在上述方面的一个示例中，对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理可以包括：使用下述寻优算法中的一种寻优算法来对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理：粒子群算法、遗传算法、退火算法。
可选地，在上述方面的一个示例中，对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理可以包括：在预定决策变量取值范围内，对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理。
根据本公开的另一方面，提供一种用于决策优化的装置，包括：贡献度确定单元，被配置为确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度，所述预测模型是基于决策目标创建的，所述预测模型包括决策特征变量和非决策特征变量；决策特征变量组合构建单元，被配置为基于所确定出的各个决策特征变量的预测结果贡献度，构建待优化的决策特征变量组合；寻优处理单元，被配置为对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理，以使得所述预测模型的对应预测结果最佳；以及决策优化单元，被配置为根据经过寻优处理后得到的决策特征变量组合中的各个决策特征变量的变量取值来进行决策优化处理。
可选地，在上述方面的一个示例中，所述贡献度确定单元被配置为：使用解释模型来确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度。
可选地，在上述方面的一个示例中，所述解释模型包括下述解释模型中的一种：Shap value模型、LIME模型和DeepLift模型。
可选地，在上述方面的一个示例中，所述决策特征变量组合构建单元包括：排序模块，被配置为基于所确定出的各个决策特征变量的预测结果贡献度，对所述各个决策特征变量进行排序；以及特征选择模块，被配置为从所述排序后的各个决策特征变量中选择贡献度较大的预定数目个决策特征变量，作为待优化的决策特征变量组合。
可选地，在上述方面的一个示例中，所述寻优处理单元被配置为：使用下述寻优算法中的一种来对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理：粒子群算法、遗传算法、退火算法。
可选地，在上述方面的一个示例中，所述寻优处理单元被配置为：在预定决策变量取值范围内，对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理。
根据本公开的另一方面，提供一种计算设备，包括：至少一个处理器，以及与所述至少一个处理器耦合的存储器，所述存储器存储指令，当所述指令被所述至少一个处理器执行时，使得所述至少一个处理器执行如上所述的用于决策优化的方法。
根据本公开的另一方面，提供一种非暂时性机器可读存储介质，其存储有可执行指令，所述指令当被执行时使得所述机器执行如上所述的用于决策优化的方法。
附图说明
通过参照下面的附图，可以实现对于本公开内容的本质和优点的进一步理解。在附图中，类似组件或特征可以具有相同的附图标记。
图1示出了根据本公开的实施例的决策优化方法的流程图；
图2示出了根据本公开的实施例的决策特征变量的预测结果贡献度确定过程的一个示例的流程图；
图3示出了根据本公开的实施例的决策特征变量组合构建过程的一个示例的流程图；
图4示出了根据本公开的实施例的个性化决策机制上线切换过程的一个示例的流程图；
图5示出了根据本公开的实施例的决策优化装置的方框图；
图6示出了根据本公开的实施例的决策特征变量组合构建单元的一个示例的方框图；
图7示出了根据本公开的实施例的用于决策优化的计算设备的方框图。
具体实施方式
现在将参考示例实施方式讨论本文描述的主题。应该理解，讨论这些实施方式只是为了使得本领域技术人员能够更好地理解从而实现本文描述的主题，并非是对权利要求书中所阐述的保护范围、适用性或者示例的限制。可以在不脱离本公开内容的保护范 围的情况下，对所讨论的元素的功能和排列进行改变。各个示例可以根据需要，省略、替代或者添加各种过程或组件。例如，所描述的方法可以按照与所描述的顺序不同的顺序来执行，以及各个步骤可以被添加、省略或者组合。另外，相对一些示例所描述的特征在其它例子中也可以进行组合。
如本文中使用的，术语“包括”及其变型表示开放的术语，含义是“包括但不限于”。术语“基于”表示“至少部分地基于”。术语“一个实施例”和“一实施例”表示“至少一个实施例”。术语“另一个实施例”表示“至少一个其他实施例”。术语“第一”、“第二”等可以指代不同的或相同的对象。下面可以包括其他的定义，无论是明确的还是隐含的。除非上下文中明确地指明，否则一个术语的定义在整个说明书中是一致的。
下面将结合附图来详细描述根据本公开的实施例的决策优化方法及装置。
图1示出了根据本公开的实施例的决策优化方法的流程图。
如图1所示，在块110，确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度。这里，所述预测模型是基于决策目标创建的，并且所述预测模型包括决策特征变量和非决策特征变量。所述用户特征数据是与决策特征变量和非决策特征变量对应的用户特征数据。
例如，假设在营销场景中，决策目标是用户注册转化人数，在交易风险控制场景中，决策目标是用户因诈骗交易损失的金额总和。针对此决策目标，创建合适的特征变量组来建立预测模型，从而基于用户特征数据来预测决策目标，比如，预测用户是否转化，或预测用户交易中因欺诈损失的金额。在本公开中，术语“非决策特征变量”是指决策方无法干涉的特征变量，比如，用户的年龄、性格、过去的行为历史等。术语“决策特征变量”是指决策方能够干涉改变的特征变量，比如，在营销场景中，在决策目标是用户注册转化人数的情况下的发放给用户的优惠和权益等特征变量，或者在交易风险控制场景中，输出给用户的风险提醒等特征变量。
此外，在本公开中，所述预测模型可以事先基于决策目标来利用训练数据创建的，也可以是在获取到决策目标后利用训练数据实时创建的。同样，所述决策目标可以是事先输入的或者实时输入的。另外，为了提高预测模型的准确率，被使用来进行预测模型训练的训练数据应该覆盖具有尽可能多特征的用户群体，并且尝试尽可能多的决策变量取值，从而确保训练数据维度的丰富性。
在本公开的一个示例中，可以使用解释模型来确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度。所述解释模型包括下述解释模型中的一种：Shap value模型、LIME模型和DeepLift模型。关于如何使用解释模型来确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度，将在下面参照图2来进行说明。
在如上确定出各个决策特征变量的预测结果贡献度后，在块120，基于所确定出的各个决策特征变量的预测结果贡献度，构建待优化的决策特征变量组合。关于如何基于所确定出的各个决策特征变量的预测结果贡献度来构建待优化的决策特征变量组合，将在下面结合图3中示出的示例进行说明。
在构建出待优化的决策特征变量组合后，在块130，对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理，以使得所述预测模型的对应预测结果最佳。
例如，在一个示例中，对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理可以包括：使用下述寻优算法中的一种寻优算法来对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理：粒子群算法、遗传算法、退火算法。这里，可以基于待进行寻优处理的决策特征变量的类型来确定所使用的寻优算法。比如，在待寻优处理的决策特征变量是分类型变量的情况下，寻优算法优选使用粒子群算法，比如，离散粒子群算法。
此外，在本公开的另一示例中，对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理可以包括：在预定决策变量取值范围内，对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理。另外，为了使得寻优效果更佳，可以根据实际情况来合适调整决策变量取值范围的大小。
在确定出决策特征变量组合中的各个决策特征变量的变量取值后，在块140，根据经过寻优处理后得到的决策特征变量组合中的各个决策特征变量的变量取值来进行决策优化处理。例如，利用该各个决策特征变量的变量取值来对决策引擎中所使用的决策机制进行优化调整。
图2示出了根据本公开的实施例的决策特征变量的预测结果贡献度确定过程的一个示例的流程图。
如图2所示，针对预测模型中的各个决策特征变量，首先，在块210，选择一个决 策特征变量来作为初始当前决策特征变量。例如，可以从各个决策特征变量中随机选择一个决策特征变量来作为初始当前决策特征变量。
然后，针对该当前决策特征变量，执行块220到块270的操作，直到针对所有决策特征变量都确定出对应的预测结果贡献度。具体地，在块220，确定决策特征变量组合中除了该当前决策特征变量之外的剩余决策特征变量的任何组合所对应的预测结果。例如，假设预测模型存在4个决策特征变量V1，V2，V3，V4，以及V1是当前决策特征变量，则使用预测模型来计算在预测模型的输入为V2，V3，V4的任意特征组合所对应的用户特征数据的情况下的模型预测结果。接着，在块230，确定决策特征变量组合中包含该当前决策特征变量的任何特征变量组合所对应的预测结果。这里所述的特征变量组合是针对块220中的每种特征组合都增加特征变量V1所得到的特征变量组合。然后，在块240，计算块220中确定出的模型预测结果(无该当前决策特征变量时的模型预测结果)与在块230中确定出的对应模型预测结果(有该当前决策特征变量时的模型预测结果)之差。
在确定出各个模型预测结果之差后，在块250，对所有模型预测结果差进行求平均，以得到该当前决策特征变量的预测结果贡献度。然后，在块260，确定是否存在未处理的决策特征变量。如果存在，则在块270中，从未处理的决策特征变量中选择一个决策特征变量作为下一当前决策特征变量，然后回到块220，重新执行块220到块270的操作。
图3示出了根据本公开的实施例的决策特征变量组合构建过程的一个示例的流程图。
如图3所示，首先，在块310，基于所确定出的各个决策特征变量的预测结果贡献度，对所述各个决策特征变量进行排序。然后，在块320，从所述排序后的各个决策特征变量中选择贡献度较大的预定数目个决策特征变量，作为待优化的决策特征变量组合。这里，所述预定数目可以基于实际应用场景或者其他合适条件来确定。此外，在本公开的其他示例中，也可以采用其他合适的方式来构建待优化的决策特征变量组合。
利用本公开的决策优化方法，针对不同用户，通过使用该用户的用户特征数据来确定预测模型中的各个决策特征变量的预测结果贡献度，基于所确定出的预测结果来构建待优化的决策特征变量组合，并且使用寻优算法来确定该决策特征变量组合中的各个决策特征变量的变量取值。在该决策优化方法中，由于针对不同的用户特征数据可以得到各个决策特征变量的不同预测结果贡献度，从而可以构建出不同的决策特征变量组合， 由此针对不同的用户，所优化的决策特征变量组合也会不同，进而实现个性化决策优化。
利用本公开的决策优化方法，通过根据实际情况来合适调整决策特征变量取值范围的大小，并且在决策特征变量取值范围内，对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理，由此可以提升寻优效果，进而提高决策优化效果。
图4示出了根据本公开的实施例的个性化决策机制上线切换过程的一个示例的流程图。
如图4所示，在如上确定出经过优化后的个性化决策机制与引擎后，在进行在线决策时，对流入的待决策的客户数据，随机地将该决策任务分配给原决策机制或优化后的个性化决策机制进行处理，进而做出决策，并记录对应的决策效果。在这样运行一段时候后，对原决策机制或优化后的个性化决策机制的决策效果进行评估，如果原决策机制的决策效果更好，则保持原决策机制，如果优化后的个性化决策机制较好，则利用优化后的个性化决策机制来替换原决策机制，由此保证决策业务的平稳性。
图5示出了根据本公开的实施例的决策优化装置500的方框图。如图5所示，决策优化装置500包括贡献度确定单元510、决策特征变量组合构建单元520、寻优处理单元530和决策优化单元540。
贡献度确定单元510被配置为确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度，所述预测模型是基于决策目标创建的，所述预测模型包括决策特征变量和非决策特征变量。贡献度确定单元510的操作可以参考上面参照图1描述的块110的操作以及参照图2描述的操作。
决策特征变量组合构建单元520被配置为基于所确定出的各个决策特征变量的预测结果贡献度，构建待优化的决策特征变量组合。决策特征变量组合构建单元520的操作可以参考上面参照图1描述的块120的操作以及参照图3描述的操作。
寻优处理单元530被配置为对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理，以使得所述预测模型的对应预测结果最佳。在本公开的一个示例中，寻优处理单元530被配置为：使用下述寻优算法中的一种来对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理：粒子群算法、遗传算法、退火算法。在本公开的另一示例中，寻优处理单元530被配置为：在预定决策变量取值范围内，对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理。寻优处理单元530的操作可以参考上面参照图1描述的块130的操作。
决策优化单元540被配置为根据经过寻优处理后得到的决策特征变量组合中的各个决策特征变量的变量取值来进行决策优化处理。决策优化单元540的操作可以参考上面参照图1描述的块140的操作。
在本公开的一个示例中，贡献度确定单元510可以被配置为：使用解释模型来确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度。所述解释模型可以包括下述解释模型中的一种：Shap value模型、LIME模型和DeepLift模型。
图6示出了根据本公开的实施例的决策特征变量组合构建单元520的一个示例的方框图。如图6所示，决策特征变量组合构建单元520包括排序模块521和特征选择模块523。
排序模块521被配置为基于所确定出的各个决策特征变量的预测结果贡献度，对所述各个决策特征变量进行排序。
特征选择模块523被配置为从所述排序后的各个决策特征变量中选择贡献度较大的预定数目个决策特征变量，作为待优化的决策特征变量组合。
如上参照图1到图6，对根据本公开的决策优化方法及装置的实施例进行了描述。上面的决策优化装置可以采用硬件实现，也可以采用软件或者硬件和软件的组合来实现。
图7示出了根据本公开的实施例的用于决策优化的计算设备700的硬件结构图。如图7所示，计算设备700可以包括至少一个处理器710、存储器720、内存730和通信接口740，并且至少一个处理器710、存储器720、内存730和通信接口740经由总线760连接在一起。至少一个处理器710执行在存储器中存储或编码的至少一个计算机可读指令(即，上述以软件形式实现的元素)。
在一个实施例中，在存储器中存储计算机可执行指令，其当执行时使得至少一个处理器710：确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度，所述预测模型是基于决策目标创建的，所述预测模型包括决策特征变量和非决策特征变量；基于所确定出的各个决策特征变量的预测结果贡献度，构建待优化的决策特征变量组合；对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理，以使得所述预测模型的对应预测结果最佳；以及根据经过寻优处理后得到的决策特征变量组合中的各个决策特征变量的变量取值来进行决策优化处理。
应该理解，在存储器中存储的计算机可执行指令当执行时使得至少一个处理器710进行本公开的各个实施例中以上结合图1-6描述的各种操作和功能。
在本公开中，计算设备700可以包括但不限于：个人计算机、服务器计算机、工作站、桌面型计算机、膝上型计算机、笔记本计算机、移动计算设备、智能电话、平板计算机、蜂窝电话、个人数字助理(PDA)、手持装置、消息收发设备、可佩戴计算设备、消费电子设备等等。
根据一个实施例，提供了一种比如非暂时性机器可读介质的程序产品。非暂时性机器可读介质可以具有指令(即，上述以软件形式实现的元素)，该指令当被机器执行时，使得机器执行本公开的各个实施例中以上结合图1-6描述的各种操作和功能。具体地，可以提供配有可读存储介质的系统或者装置，在该可读存储介质上存储着实现上述实施例中任一实施例的功能的软件程序代码，且使该系统或者装置的计算机或处理器读出并执行存储在该可读存储介质中的指令。
根据一个实施例，提供了一种比如非暂时性机器可读介质的程序产品。非暂时性机器可读介质可以具有指令(即，上述以软件形式实现的元素)，该指令当被机器执行时，使得机器执行本公开的各个实施例中以上结合图1-6描述的各种操作和功能。具体地，可以提供配有可读存储介质的系统或者装置，在该可读存储介质上存储着实现上述实施例中任一实施例的功能的软件程序代码，且使该系统或者装置的计算机或处理器读出并执行存储在该可读存储介质中的指令。
在这种情况下，从可读介质读取的程序代码本身可实现上述实施例中任何一项实施例的功能，因此机器可读代码和存储机器可读代码的可读存储介质构成了本发明的一部分。
可读存储介质的实施例包括软盘、硬盘、磁光盘、光盘(如CD-ROM、CD-R、CD-RW、DVD-ROM、DVD-RAM、DVD-RW、DVD-RW)、磁带、非易失性存储卡和ROM。可选择地，可以由通信网络从服务器计算机上或云上下载程序代码。
本领域技术人员应当理解，上面公开的各个实施例可以在不偏离发明实质的情况下做出各种变形和修改。因此，本发明的保护范围应当由所附的权利要求书来限定。
需要说明的是，上述各流程和各系统结构图中不是所有的步骤和单元都是必须的，可以根据实际的需要忽略某些步骤或单元。各步骤的执行顺序不是固定的，可以根据需要进行确定。上述各实施例中描述的装置结构可以是物理结构，也可以是逻辑结构，即，有些单元可能由同一物理实体实现，或者，有些单元可能分由多个物理实体实现，或者，可以由多个独立设备中的某些部件共同实现。
以上各实施例中，硬件单元或模块可以通过机械方式或电气方式实现。例如，一个硬件单元、模块或处理器可以包括永久性专用的电路或逻辑(如专门的处理器，FPGA或ASIC)来完成相应操作。硬件单元或处理器还可以包括可编程逻辑或电路(如通用处理器或其它可编程处理器)，可以由软件进行临时的设置以完成相应操作。具体的实现方式(机械方式、或专用的永久性电路、或者临时设置的电路)可以基于成本和时间上的考虑来确定。
上面结合附图阐述的具体实施方式描述了示例性实施例，但并不表示可以实现的或者落入权利要求书的保护范围的所有实施例。在整个本说明书中使用的术语“示例性”意味着“用作示例、实例或例示”，并不意味着比其它实施例“优选”或“具有优势”。出于提供对所描述技术的理解的目的，具体实施方式包括具体细节。然而，可以在没有这些具体细节的情况下实施这些技术。在一些实例中，为了避免对所描述的实施例的概念造成难以理解，公知的结构和装置以框图形式示出。
本公开内容的上述描述被提供来使得本领域任何普通技术人员能够实现或者使用本公开内容。对于本领域普通技术人员来说，对本公开内容进行的各种修改是显而易见的，并且，也可以在不脱离本公开内容的保护范围的情况下，将本文所定义的一般性原理应用于其它变型。因此，本公开内容并不限于本文所描述的示例和设计，而是与符合本文公开的原理和新颖性特征的最广范围相一致。

Claims (14)

  1. 一种用于决策优化的方法，包括：
    确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度，所述预测模型是基于决策目标创建的，所述预测模型包括决策特征变量和非决策特征变量；
    基于所确定出的各个决策特征变量的预测结果贡献度，构建待优化的决策特征变量组合；
    对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理，以使得所述预测模型的对应预测结果最佳；以及
    根据经过寻优处理后得到的决策特征变量组合中的各个决策特征变量的变量取值来进行决策优化处理。
  2. 如权利要求1所述的方法，其中，确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度包括：
    使用解释模型来确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度。
  3. 如权利要求2所述的方法，其中，所述解释模型包括下述解释模型中的一种：
    Shap value模型、LIME模型和DeepLift模型。
  4. 如权利要求1所述的方法，其中，基于所确定出的各个决策特征变量的预测结果贡献度，构建待优化的决策特征变量组合包括：
    基于所确定出的各个决策特征变量的预测结果贡献度，对所述各个决策特征变量进行排序；以及
    从所述排序后的各个决策特征变量中选择贡献度较大的预定数目个决策特征变量，作为待优化的决策特征变量组合。
  5. 如权利要求1所述的方法，其中，对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理包括：
    使用下述寻优算法中的一种寻优算法来对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理：
    粒子群算法、遗传算法、退火算法。
  6. 如权利要求1所述的方法，其中，对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理包括：
    在预定决策变量取值范围内，对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理。
  7. 一种用于决策优化的装置，包括：
    贡献度确定单元，被配置为确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度，所述预测模型是基于决策目标创建的，所述预测模型包括决策特征变量和非决策特征变量；
    决策特征变量组合构建单元，被配置为基于所确定出的各个决策特征变量的预测结果贡献度，构建待优化的决策特征变量组合；
    寻优处理单元，被配置为对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理，以使得所述预测模型的对应预测结果最佳；以及
    决策优化单元，被配置为根据经过寻优处理后得到的决策特征变量组合中的各个决策特征变量的变量取值来进行决策优化处理。
  8. 如权利要求7所述的装置，其中，所述贡献度确定单元被配置为：
    使用解释模型来确定预测模型中的各个决策特征变量在用户特征数据下的预测结果贡献度。
  9. 如权利要求8所述的装置，其中，所述解释模型包括下述解释模型中的一种：
    Shap value模型、LIME模型和DeepLift模型。
  10. 如权利要求7所述的装置，其中，所述决策特征变量组合构建单元包括：
    排序模块，被配置为基于所确定出的各个决策特征变量的预测结果贡献度，对所述各个决策特征变量进行排序；以及
    特征选择模块，被配置为从所述排序后的各个决策特征变量中选择贡献度较大的预定数目个决策特征变量，作为待优化的决策特征变量组合。
  11. 如权利要求7所述的装置，其中，所述寻优处理单元被配置为：
    使用下述寻优算法中的一种来对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理：
    粒子群算法、遗传算法、退火算法。
  12. 如权利要求7所述的装置，其中，所述寻优处理单元被配置为：
    在预定决策变量取值范围内，对所构建的决策特征变量组合中的各个决策特征变量的变量取值进行寻优处理。
  13. 一种计算设备，包括：
    至少一个处理器，以及
    与所述至少一个处理器耦合的存储器，所述存储器存储指令，当所述指令被所述至少一个处理器执行时，使得所述至少一个处理器执行如权利要求1到6中任一所述的方 法。
  14. 一种非暂时性机器可读存储介质，其存储有可执行指令，所述指令当被执行时使得所述机器执行如权利要求1到6中任一所述的方法。
PCT/CN2020/070492 2019-03-08 2020-01-06 决策优化方法及装置 Ceased WO2020181907A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910176389.8 2019-03-08
CN201910176389.8A CN110046799B (zh) 2019-03-08 2019-03-08 决策优化方法及装置

Publications (1)

Publication Number Publication Date
WO2020181907A1 true WO2020181907A1 (zh) 2020-09-17

Family

ID=67274727

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/070492 Ceased WO2020181907A1 (zh) 2019-03-08 2020-01-06 决策优化方法及装置

Country Status (2)

Country Link
CN (1) CN110046799B (zh)
WO (1) WO2020181907A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112766586A (zh) * 2021-01-26 2021-05-07 绍兴文理学院 基于分步渐进策略的多目标优化排考方法及系统
CN117710066A (zh) * 2024-02-05 2024-03-15 厦门傲凡科技股份有限公司 金融客户推荐方法及系统

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110046799B (zh) * 2019-03-08 2021-09-10 创新先进技术有限公司 决策优化方法及装置
CN111461892B (zh) * 2020-03-31 2021-07-06 支付宝(杭州)信息技术有限公司 用于风险识别模型的衍生变量选择方法和装置
CN111262887B (zh) * 2020-04-26 2020-08-28 腾讯科技（深圳）有限公司 基于对象特征的网络风险检测方法、装置、设备及介质
CN113570260B (zh) * 2021-07-30 2024-09-13 贝壳找房(北京)科技有限公司 任务分配方法和计算机可读存储介质、电子设备
CN113590691B (zh) * 2021-08-04 2024-06-28 浙江网商银行股份有限公司 目标对象处理方法以及装置
CN113902450A (zh) * 2021-12-08 2022-01-07 四川新网银行股份有限公司 一种可解释的欺诈交易识别方法及装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108665277A (zh) * 2017-03-27 2018-10-16 阿里巴巴集团控股有限公司 一种信息处理方法及装置
CN109036568A (zh) * 2018-09-03 2018-12-18 浪潮软件集团有限公司 一种基于朴素贝叶斯算法的预测模型的建立方法
CN109243172A (zh) * 2018-07-25 2019-01-18 华南理工大学 基于遗传算法优化lstm神经网络的交通流预测方法
CN109343019A (zh) * 2018-11-05 2019-02-15 中国矿业大学(北京) 一种地质雷达数据解释方法及装置
CN110046799A (zh) * 2019-03-08 2019-07-23 阿里巴巴集团控股有限公司 决策优化方法及装置

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106022473B (zh) * 2016-05-23 2019-03-05 大连理工大学 一种融合粒子群和遗传算法的基因调控网络构建方法
CN107703749A (zh) * 2017-10-10 2018-02-16 常州大学 一种基于自适应遗传算法的pid参数优化方法
CN109271695B (zh) * 2018-09-07 2022-12-23 中南大学 基于神经网络的多目标天线设计方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108665277A (zh) * 2017-03-27 2018-10-16 阿里巴巴集团控股有限公司 一种信息处理方法及装置
CN109243172A (zh) * 2018-07-25 2019-01-18 华南理工大学 基于遗传算法优化lstm神经网络的交通流预测方法
CN109036568A (zh) * 2018-09-03 2018-12-18 浪潮软件集团有限公司 一种基于朴素贝叶斯算法的预测模型的建立方法
CN109343019A (zh) * 2018-11-05 2019-02-15 中国矿业大学(北京) 一种地质雷达数据解释方法及装置
CN110046799A (zh) * 2019-03-08 2019-07-23 阿里巴巴集团控股有限公司 决策优化方法及装置

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112766586A (zh) * 2021-01-26 2021-05-07 绍兴文理学院 基于分步渐进策略的多目标优化排考方法及系统
CN117710066A (zh) * 2024-02-05 2024-03-15 厦门傲凡科技股份有限公司 金融客户推荐方法及系统
CN117710066B (zh) * 2024-02-05 2024-05-10 厦门傲凡科技股份有限公司 金融客户推荐方法及系统

Also Published As

Publication number Publication date
CN110046799B (zh) 2021-09-10
CN110046799A (zh) 2019-07-23

Similar Documents

Publication Publication Date Title
WO2020181907A1 (zh) 决策优化方法及装置
JP6912591B2 (ja) リスク管理制御方法及びデバイス
US8781915B2 (en) Recommending items to users utilizing a bi-linear collaborative filtering model
US10637990B1 (en) Call center load balancing and routing management
US20090043593A1 (en) Event Prediction
US20190035015A1 (en) Method and apparatus for obtaining a stable credit score
CN104965844A (zh) 信息处理方法及装置
US12412093B2 (en) Machine learning based approach for identification of extremely rare events in high-dimensional space
CN110930226A (zh) 一种金融产品推荐方法、装置、电子设备及存储介质
US20220405621A1 (en) Controlling reachability in a collaboratively filtered recommender
WO2020063116A1 (zh) 一种风险保障产品的推送方法、装置及电子设备
CN111915315B (zh) 认证方式推荐方法、装置、电子设备和可读存储介质
CN112967044A (zh) 一种支付业务的处理方法及装置
CN110020728B (zh) 业务模型强化学习方法以及装置
CN113554448A (zh) 一种用户流失预测方法、装置及电子设备
CN114201687A (zh) 信息推荐方法、装置、电子设备及存储介质
CN110046233A (zh) 问题派发方法及装置
US12567232B2 (en) Certification-based robust training by refining decision boundary
CN116645196A (zh) 信用预测处理方法及装置
CN111798282A (zh) 一种信息处理方法、终端及存储介质
CN117010954A (zh) 商家权益处理方法及装置
KR20130065778A (ko) 신용정보 관리 방법 및 시스템
Gao et al. A BI‐LEVEL DECISION MODEL FOR CUSTOMER CHURN ANALYSIS
US12259943B2 (en) Systems and methods for optimizing and harmonizing sequences of communications over different mediums
CN121365995A (zh) 基于贝叶斯网络的客户状态预测方法、装置、设备和介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20770905

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20770905

Country of ref document: EP

Kind code of ref document: A1