Transaction event risk management and control method and device

Abstract

The embodiment of the invention discloses a transaction event risk management and control method and device, and the method introduces a large language model to replace the decisive role of expert experience in a risk management and control evaluation system, and the large language model becomes a reward model for optimizing a transaction event risk management and control model in a reinforcement learning framework under the guidance of a prompt text. And a full-automatic transaction event risk management and control model reinforcement learning optimization link based on large language model feedback can be realized. The transaction event risk management and control device disclosed by the embodiment of the specification also has the above beneficial effects.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a transaction event risk management method and device. Background Art

[0002] Transaction event risk control refers to the output of control solutions for transaction events with potential risks, such as failures and reminders, through risk control models. Among them, the risk control model needs to be continuously improved and optimized through the results of risk control evaluation. The current risk control evaluation mainly relies on the control evaluation system driven solely by expert experience, which requires manual and continuous extraction of high-value risk-related features and reasonable and strict control logic. This process will occupy a large amount of professional human resources, and the control evaluation system that is highly dependent on expert experience increases the complexity of optimizing and adjusting the risk control model. Summary of the invention

[0003] One or more embodiments of the present specification provide a method and device for risk management of transaction events. The method utilizes a large language model to replace expert experience to generate feedback signals for risk management strategies, thereby updating the risk management model, overcoming the problem that the optimization of existing risk management models is highly dependent on expert experience.

[0004] In the first aspect, a training method for a transaction event risk management model is provided, including: Get a sample of transaction events; Inputting the transaction event sample into a transaction event risk management model, and using the transaction event risk management model to determine a risk management strategy for the transaction event sample; Constructing a prompt text according to the transaction event sample and the risk management strategy; Inputting the prompt text into a pre-trained large language model to obtain an evaluation result of the large language model on the applicability of the risk management strategy to the transaction event sample; The evaluation result is fed back to the transaction event risk management model as an incentive signal, and a reinforcement strategy is adopted to update the parameters of the transaction event risk management model until a target transaction event risk management model is obtained.

[0005] As an optional implementation manner of the method of the first aspect, constructing a prompt text according to the transaction event sample and the risk management strategy specifically includes: According to the transaction event sample and the risk management and control strategy, a preset prompt text template is used to generate the prompt text.

[0006] Specifically, the method further includes: Fine-tune the prompt text according to the response result after the risk management and control strategy is executed.

[0007] As an optional implementation manner of the method described in the first aspect, the risk management strategy includes a management decision on whether to manage the transaction event sample, and a management plan to be adopted when it is determined to manage the transaction event sample.

[0008] As an optional implementation manner of the method of the first aspect, the method further includes: Obtain historical control data, and construct binary sample data consisting of historical transaction events and corresponding risk control strategies based on the historical control data; Scoring the binary sample data based on expert experience, and using the scoring result as a label for the binary sample data; the scoring result is used to characterize the applicability of the risk control strategy in the binary sample data to the corresponding historical transaction event; The large language model is fine-tuned using the bigram sample data.

[0009] In the second aspect, a transaction event risk management method is provided, including: Get the target transaction event; Inputting the target transaction event into a transaction event risk management model to obtain a risk management strategy for the target transaction event; the transaction event risk management model is pre-trained using the above-mentioned transaction event risk management model training method; Execute the risk management strategy.

[0010] In a third aspect, a training device for a transaction risk management model is provided, comprising: A first data acquisition module configured to acquire transaction event samples; A strategy generation module is configured to input the transaction event sample into a transaction event risk management model, obtain the transaction event risk management model to determine a risk management strategy for the transaction event sample; A prompt text generation module, configured to construct a prompt text according to the transaction event sample and the risk management strategy; an incentive module, configured to input the prompt text into a pre-trained large language model to obtain an evaluation result of the large language model on the applicability of the risk management strategy to the transaction event sample; The training module is configured to feed back the evaluation result as an incentive signal to the transaction event risk management model, and adopt a reinforcement strategy to update the parameters of the transaction event risk management model until a target transaction event risk management model is obtained.

[0011] As an optional implementation manner of the device described in the third aspect, the prompt text generation module is specifically used to generate the prompt text using a preset prompt text template according to the transaction event sample and the risk management strategy.

[0012] Specifically, the prompt text generation module is further used to fine-tune the prompt text according to the response result after the risk management and control strategy is executed.

[0013] As an optional implementation manner of the device described in the third aspect, the risk management strategy includes a management decision on whether to manage the transaction event sample, and a management scheme to be adopted when it is determined to manage the transaction event sample.

[0014] As an optional implementation manner of the device according to the third aspect, the training module is further used for: Obtain historical control data, and construct binary sample data consisting of historical transaction events and corresponding risk control strategies based on the historical control data; Scoring the binary sample data based on expert experience, and using the scoring result as a label for the binary sample data; the scoring result is used to characterize the applicability of the risk control strategy in the binary sample data to the corresponding historical transaction event; The large language model is fine-tuned using the bigram sample data.

[0015] In a fourth aspect, a transaction event risk management device is provided, comprising: A second data acquisition module is configured to acquire a target transaction event; a risk control module configured to input the target transaction event into a transaction event risk control model to obtain a risk control strategy for the target transaction event; the transaction event risk control model is pre-trained using the training method of the transaction event risk control model; An execution module is configured to execute the risk management strategy.

[0016] In a fifth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the processor executes the above-mentioned training method of the transaction event risk management model, or executes the above-mentioned transaction event risk management method.

[0017] In a sixth aspect, an electronic device is provided, including: One or more processors; and a memory associated with the one or more processors, the memory being used to store program instructions, which, when read and executed by the one or more processors, enable the electronic device to execute the above-mentioned training method for the transaction event risk management model, or to execute the above-mentioned transaction event risk management method.

[0018] The beneficial effect of the transaction event risk management method described in one or more embodiments of this specification is that the method introduces a large language model to replace the decisive role of expert experience in the risk management evaluation system. The large language model becomes a reward model for optimizing the transaction event risk management model in the reinforcement learning framework under the guidance of the prompt text, and can realize a fully automated transaction event risk management model reinforcement learning optimization link based on the feedback of the large language model. The transaction event risk management device described in the embodiments of this specification also has the above beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 A flowchart of a method for training a transaction event risk management model provided in one or more embodiments of this specification.

[0021] Figure 2 A schematic diagram of the content structure of a prompt text template provided for one or more embodiments of this specification.

[0022] Figure 3 A schematic diagram of a fine-tuning process of a large language model provided in one or more embodiments of this specification.

[0023] Figure 4 A schematic diagram of the workflow of a large language model feedback link provided for one or more embodiments of this specification.

[0024] Figure 5 A flowchart of a transaction event risk management method provided in one or more embodiments of this specification.

[0025] Figure 6 A schematic diagram of the structure of a training device for a transaction event risk management model provided in one or more embodiments of this specification.

[0026] Figure 7A schematic diagram of the structure of a transaction event risk management device provided in one or more embodiments of this specification.

[0027] Figure 8 A schematic diagram of the structure of an electronic device provided in one or more embodiments of this specification. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0029] It should be noted that: in other embodiments, the steps of the corresponding method are not necessarily performed in the order shown and described in this specification. In some other embodiments, the steps included in the method may be more or less than those described in this specification. In addition, a single step described in this specification may be decomposed into multiple steps for description in other embodiments; and multiple steps described in this specification may be combined into a single step for description in other embodiments.

[0030] Those skilled in the art will appreciate that the terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms of "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0031] In the field of transaction event risk management, risk management output refers to the output of management plans for transaction events with potential risks, such as failures, reminders, etc., through risk management models. Management evaluation refers to the quantitative evaluation of the quality of the management plans output by the risk management model. In the entire process of transaction event risk management, the risk management model is crucial. In order to ensure the effectiveness of the risk management plan for transaction events in the risk management model, it is necessary to continuously improve and optimize the risk management model through the risk management evaluation results. The current risk management plan output and risk management evaluation results mainly rely on a management output system driven solely by expert experience. It is necessary to continuously refine high-value risk-related features and reasonable and strict management logic manually. This process will occupy a large amount of professional human resources. The current risk management evaluation results are mainly derived from the actual online management products, subsequent risks, customer complaint reporting, and business demand logic. The following problems may exist: 1. The current situation of high reliance on expert experience increases the complexity of optimization and adjustment, and may give rise to suboptimal decision-making problems: Scoring is based on logical deduction of business needs, which is highly dependent on expert experience. Once the needs change, the evaluation logic needs to be re-formulated, which not only significantly increases the difficulty and time-consuming of optimization iteration, but also causes the system to lag in responding to the ever-changing real-time needs; at the same time, the formulation of the evaluation plan needs to fully consider the overall situation, otherwise it is easy to fall into the local optimal trap.

[0032] 2. Hidden cases that are difficult to quantify affect the accuracy of evaluation: Under the existing evaluation mechanism, there are many transaction events that result in the output of "failed" products under control. These transaction events become hidden cases that are difficult to be effectively quantified and evaluated, thus posing a challenge to accurately quantifying the overall effect of "failed" control products.

[0033] In view of this, one or more embodiments of the present specification propose a transaction event risk management method and device, which can use a large language model instead of expert experience to generate feedback signals for risk management strategies, thereby updating the risk management model to at least partially overcome the above-mentioned technical problems.

[0034] The transaction event risk management method and device described in one or more embodiments of this specification will be further described in detail below in conjunction with the drawings and specific embodiments of the specification, but this detailed description does not constitute a limitation on the embodiments of this specification.

[0035] Please refer to Figure 1 , Figure 1 The following is a flow chart of a training method for a transaction event risk management model proposed in one or more embodiments of this specification. Figure 1 As shown, the method may include steps S100 to S108.

[0036] S100: Obtain transaction event samples.

[0037] In some implementations, feature extraction may be performed on the collected transaction events to obtain transaction event features, and then the transaction event features are used as the transaction event samples described above.

[0038] The above-mentioned transaction event samples should include risky transaction events and ordinary transaction events.

[0039] S102: Input the transaction event samples into the transaction event risk management model, and use the transaction event risk management model to determine the risk management strategy for the transaction event samples.

[0040] The above-mentioned transaction event risk control model is a decision model of risk control strategy, which can select the corresponding risk control strategy from the pre-constructed risk control strategy space as the decision output according to the input transaction event sample. The above-mentioned risk control strategy space refers to a set of pre-constructed risk control strategies for different types of risk transaction events. The risk control strategy may include a control decision on whether to control the transaction event sample, and the specific control scheme adopted when it is determined to control the transaction event sample.

[0041] S104: Construct a prompt text based on the transaction event sample and risk management strategy.

[0042] In a large language model (LLM), a prompt is the initial input text provided when interacting with a large language model. It plays the role of guiding the model to generate relevant output or perform specific tasks. A prompt can be an instruction, a question, the beginning of a conversation, or a narrative context setting, the purpose of which is to stimulate the model to continue to generate coherent, reasonable, or targeted text responses based on the given context information.

[0043] In this step S104, a prompt text may be constructed based on the characteristics of the transaction event, the control decision on whether to control the transaction event sample, and the characteristics of the specific control plan to guide the large language model to understand the target to be evaluated.

[0044] In some implementations, a pre-built prompt text template may be used to generate the prompt text. Figure 2 , Figure 2 A schematic diagram of the content structure of a prompt text template is provided as an example. Figure 2 In the prompt text template shown, the following contents need to be filled in: a detailed description of the transaction event characteristics, such as transaction type, transaction amount, transaction frequency, accounts involved, etc.; the relevance of the control plan characteristics to this transaction event, such as control intensity, output frequency, user disturbance level, etc.; whether the control plan has been output for intervention (yes or no), if so, it is necessary to further describe how the control plan responds to this transaction event; examples of excellent control plans. The above content is used to guide the large language model to understand the goals that need to be evaluated, and to guide the large language model to provide fact-based and logically reasonable scoring feedback. Specifically, it is to guide the large language model to generate evaluation results based on the prompt content, including the possibility of risk and customer complaints, scores and their corresponding explanations.

[0045] The required prompt text can be quickly generated using prompt text templates, but fixed templates may cause the large language model to over-adapt to certain specific prompts. In order to improve the generalization ability and diversity of the results generated by the large language model and to maximize the potential of the large language model, in some embodiments, the generated prompt text can also be fine-tuned according to the response results after the execution of the risk management strategy. The method of fine-tuning the prompt text can be adaptively selected according to needs, and this embodiment does not limit this. For example, AutoPrompt, Prompt Tuning, Prefix Tuning and other methods can be used to learn, adjust, and optimize the prompt text without changing the main parameters of the large language model and ensuring the controllability of the large language model.

[0046] S106: Input the prompt text into a pre-trained large language model to obtain an evaluation result of the large language model on the applicability of the risk management strategy to the transaction event sample.

[0047] Before using the above-mentioned large language model, you can construct prompt text in advance based on expert experience and excellent historical control data to fine-tune the large language model so that it can reasonably score the (risky transaction event, risk control strategy) binary group.

[0048] Please refer to Figure 3 , Figure 3 The fine-tuning process of the large language model in the fine-tuning stage is schematically shown, and the fine-tuning process includes: Obtain historical control data, and construct binary sample data consisting of historical transaction events and corresponding risk control strategies based on the historical control data; Scoring the binary sample data based on expert experience, and using the scoring result as a label for the binary sample data; the scoring result is used to characterize the applicability of the risk control strategy in the binary sample data to the corresponding historical transaction event; The constructed bigram sample data is used to fine-tune the parameters of the large language model until a large language model that meets the preset conditions is obtained.

[0049] The fine-tuned large language model can be used as the large language model feedback link. Please refer to Figure 4 The input of the feedback link of the large language model is the binary data of (transaction event, risk management strategy), and the output is the evaluation result for the binary data, which serves as the incentive signal (reward).

[0050] S108: Feedback the evaluation results as incentive signals to the transaction event risk management model, and adopt a reinforcement strategy to update the parameters of the transaction event risk management model until the target transaction event risk management model is obtained.

[0051] The above-mentioned transaction event risk management model is a reinforcement learning model. In each decision, the transaction event risk management model selects the corresponding risk management strategy from the pre-built risk management strategy space as the decision output, and obtains the evaluation result of the large language model for the risk management strategy as the incentive signal. Based on the incentive signal, the parameters of the transaction event risk management model are updated, so that the next decision action can obtain higher incentives. Repeat the above iterative process until a target transaction event risk management model that meets the requirements is obtained. The target transaction event risk management model can select the most appropriate risk management strategy for the input transaction event in each decision-making process.

[0052] The above is Figure 1 The specific implementation process of the training method of the transaction event risk control model shown in the figure. This method introduces a large language model to replace the decisive role of expert experience in the evaluation scheme, and converts expert experience into an important reference for optimizing prompts to assist in improving the control evaluation system. Only by improving prompts can new needs be responded to in a timely manner, and the optimization and iteration of existing schemes can be optimized without retraining the evaluation model, which greatly reduces the manpower investment and time cost. At the same time, the above training method uses historical excellent control sample data as an example to construct prompts, giving full play to the powerful understanding and reasoning ability of the large language model to mine and refine the intrinsic characteristics and information associated with those transaction events and risk control strategies that are difficult to directly summarize in human language, so as to achieve more agile and more accurate incentive signal output. Under the guidance of prompts, the large language model becomes an incentive model in the framework of reinforcement learning optimization of transaction event risk control model, which can realize a fully automated transaction event risk control model reinforcement learning optimization link based on large language model feedback.

[0053] Based on the transaction event risk management model trained by the above-mentioned transaction event risk management model training method, a transaction event risk management method is also proposed in one or more embodiments of this specification.

[0054] like Figure 5 As shown, the method may include steps S500 to S504.

[0055] S500: Obtain a target transaction event.

[0056] S502: Input the target transaction event into the transaction event risk management model to obtain a risk management strategy for the target transaction event.

[0057] S504: Execute risk management strategy.

[0058] The above-mentioned transaction event risk management method can select the best risk management strategy for the target transaction event based on the pre-trained transaction event risk management model.

[0059] Corresponding to the training method of the transaction event risk management model described above, one or more embodiments of this specification also propose a training device for the transaction event risk management model. Figure 6 , Figure 6 The structure diagram of a training device for a transaction event risk management model proposed in one or more embodiments of this specification is shown. The device can be used to implement the training method for the transaction event risk management model described above. It should be noted that the training method for the transaction event risk management model described in one or more embodiments of this application can rely on Figure 6 The training device for the trading event risk management model shown is implemented, but not limited to this device.

[0060] like Figure 6 As shown, the training device of the transaction event risk management model includes: The first data acquisition module 601 is configured to acquire transaction event samples.

[0061] The strategy generation module 602 is configured to input the transaction event sample into the transaction event risk management model, obtain the transaction event risk management model to determine the risk management strategy for the transaction event sample.

[0062] The prompt text generation module 603 is configured to construct a prompt text according to the transaction event sample and the risk management and control strategy.

[0063] The incentive module 604 is configured to input the prompt text into a pre-trained large language model to obtain an evaluation result of the large language model on the applicability of the risk management strategy to the transaction event sample.

[0064] The training module 605 is configured to feed back the evaluation results as an incentive signal to the transaction event risk management model, and adopt a reinforcement strategy to update the parameters of the transaction event risk management model until the target transaction event risk management model is obtained.

[0065] With respect to the above-mentioned first data acquisition module 601, the transaction event samples acquired by the first data acquisition module 601 should include risky transaction events and ordinary transaction events.

[0066] In some implementations, the first data acquisition module 601 may also perform feature extraction on the collected transaction events to obtain transaction event features, and then use the transaction event features as the above-mentioned transaction event samples.

[0067] With respect to the above-mentioned strategy generation module 602, after the strategy generation module 602 inputs the transaction event sample into the transaction event risk management model, the transaction event risk management model can select the corresponding risk management strategy from the pre-constructed risk management strategy space as the decision output according to the input transaction event sample. The above-mentioned risk management strategy space refers to a set of pre-constructed risk management strategies for different types of risk transaction events. The risk management strategy obtained by the strategy generation module 602 from the transaction event risk management model may include a management decision on whether to control the transaction event sample, and a specific management scheme adopted when determining to control the transaction event sample.

[0068] Regarding the above-mentioned prompt text generation module 603, the prompt text generation module 603 can construct a prompt text according to the characteristics of the transaction event, the control decision on whether to control the transaction event sample, and the characteristics of the specific control plan, so as to guide the large language model to understand the target to be evaluated.

[0069] In some implementations, the prompt text generation module 603 may generate the prompt text using a pre-built prompt text template according to the transaction event sample and the risk management strategy, for example: Figure 2 The prompt text template shown.

[0070] In some implementations, the prompt text generation module 603 can also fine-tune the generated prompt text according to the response result after the risk management strategy is executed. The method of fine-tuning the prompt text can be adaptively selected according to the needs, and this embodiment does not limit this. For example, the prompt text generation module 603 can use AutoPrompt, Prompt Tuning, Prefix Tuning and other methods to learn, adjust and optimize the prompt text without changing the main parameters of the large language model and ensuring the controllability of the large language model.

[0071] Regarding the above-mentioned incentive module 604, the incentive module 604 is mainly composed of a pre-trained large language model. The large language model can reasonably score the (risk transaction event, risk management strategy) binary group.

[0072] With respect to the above-mentioned training module 605, the training module 605 is mainly used to train the transaction event risk management model using a reinforcement learning strategy. In each decision, the transaction event risk management model selects the corresponding risk management strategy from the pre-constructed risk management strategy space as the decision output, and obtains the evaluation result of the large language model for the risk management strategy as an incentive signal, and updates the parameters of the transaction event risk management model based on the incentive signal, so that the next decision action can obtain a higher incentive. Repeat the above iterative process until a target transaction event risk management model that meets the requirements is obtained, and the target transaction event risk management model can select the most appropriate risk management strategy for the input transaction event in each decision-making process.

[0073] In some implementations, the training module 605 can also be used to complete the pre-training of the large language model. Specifically, the training module 605 can be used Figure 3 The process shown in the figure is used to fine-tune the large language model. The specific steps include: Obtain historical control data, and construct binary sample data consisting of historical transaction events and corresponding risk control strategies based on the historical control data; Scoring the binary sample data based on expert experience, and using the scoring result as a label for the binary sample data; the scoring result is used to characterize the applicability of the risk control strategy in the binary sample data to the corresponding historical transaction event; The constructed bigram sample data is used to fine-tune the parameters of the large language model until a large language model that meets the preset conditions is obtained.

[0074] The fine-tuned large language model can be used as the large language model feedback link. Please refer to Figure 4 The input of the feedback link of the large language model is the binary data of (transaction event, risk management strategy), and the output is the evaluation result for the binary data, which serves as the incentive signal (reward).

[0075] For the above-mentioned training device of a transaction event risk management model, taking a module as an example of a software functional unit, the first data acquisition module 601 may include code running on a computing instance. Among them, the computing instance may include at least one of a physical host (computing device), a virtual machine, and a container. Further, the above-mentioned computing instance may be one or more. For example, the first data acquisition module 601 may include code running on multiple hosts / virtual machines / containers. The multiple hosts / virtual machines / containers used to run the code may be distributed in the same region (region) or in different regions. Furthermore, the multiple hosts / virtual machines / containers used to run the code may be distributed in the same availability zone (AZ) or in different AZs, each AZ including one data center or multiple data centers with close geographical locations. Among them, usually a region may include multiple AZs.

[0076] Similarly, multiple hosts / virtual machines / containers used to run the code can be distributed in the same virtual private cloud (VPC) or in multiple VPCs. Usually, a VPC is set up in a region. For cross-region communication between two VPCs in the same region and between VPCs in different regions, a communication gateway needs to be set up in each VPC to achieve interconnection between VPCs through the communication gateway.

[0077] As an example of a hardware functional unit, the first data acquisition module 601 may include at least one computing device, such as a server, etc. Alternatively, the first data acquisition module 601 may also be a device implemented using an application-specific integrated circuit (ASIC) or a programmable logic device (PLD). The PLD may be a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL) or any combination thereof.

[0078] The multiple computing devices included in the first data acquisition module 601 can be distributed in the same region or in different regions. The multiple computing devices included in the first data acquisition module 601 can be distributed in the same AZ or in different AZs. Similarly, the multiple computing devices included in the first data acquisition module 601 can be distributed in the same VPC or in multiple VPCs. The multiple computing devices can be any combination of computing devices such as servers, ASICs, PLDs, CPLDs, FPGAs, and GALs.

[0079] In other embodiments, the first data acquisition module 601 can be used to execute any step in the training method of the above-mentioned transaction event risk management model, the strategy generation module 602 can be used to execute any step in the training method of the above-mentioned transaction event risk management model, the prompt text generation module 603 can be used to execute any step in the training method of the above-mentioned transaction event risk management model, the incentive module 604 can be used to execute any step in the training method of the above-mentioned transaction event risk management model, and the training module 605 can be used to execute any step in the training method of the above-mentioned transaction event risk management model. The steps that the first data acquisition module 601, the strategy generation module 602, the prompt text generation module 603, the incentive module 604 and the training module 605 are responsible for implementing can be specified as needed, and the first data acquisition module 601, the strategy generation module 602, the prompt text generation module 603, the incentive module 604 and the training module 605 respectively implement different steps in the training method of the above-mentioned transaction event risk management model to realize all the functions of the training device of the above-mentioned transaction event risk management model.

[0080] In this implementation, the training device for the transaction event risk management model can also be applied to computing devices such as computers and servers, or to a computing device cluster including at least one computing device, so as to realize the specific functions of the training device for the transaction event risk management model.

[0081] Corresponding to the above-mentioned transaction event risk management method, one or more embodiments of this specification also propose a transaction event risk management device. Figure 7 , Figure 7 1 is a schematic diagram of the structure of a transaction event risk management device proposed in one or more embodiments of this specification. The device can be used to implement the above-mentioned transaction event risk management method. It should be noted that the transaction event risk management method described in one or more embodiments of this application can rely on Figure 7 The transaction event risk management and control device shown is implemented, but not limited to this device.

[0082] like Figure 7As shown, the transaction event risk management and control device includes: The second data acquisition module 701 is configured to acquire a target transaction event.

[0083] The risk management module 702 is configured to input the target transaction event into the transaction event risk management model to obtain a risk management strategy for the target transaction event.

[0084] The execution module 703 is configured to execute the risk management and control strategy output by the risk management and control module 702 .

[0085] For the above-mentioned transaction event risk management and control device, taking a module as an example of a software functional unit, the second data acquisition module 701 may include code running on a computing instance. Among them, the computing instance may include at least one of a physical host (computing device), a virtual machine, and a container. Further, the above-mentioned computing instance may be one or more. For example, the second data acquisition module 701 may include code running on multiple hosts / virtual machines / containers. The multiple hosts / virtual machines / containers used to run the code may be distributed in the same region (region) or in different regions. Furthermore, the multiple hosts / virtual machines / containers used to run the code may be distributed in the same availability zone (AZ) or in different AZs, each AZ including a data center or multiple data centers with close geographical locations. Among them, usually a region may include multiple AZs.

[0086] Similarly, multiple hosts / virtual machines / containers used to run the code can be distributed in the same virtual private cloud (VPC) or in multiple VPCs. Usually, a VPC is set up in a region. For cross-region communication between two VPCs in the same region and between VPCs in different regions, a communication gateway needs to be set up in each VPC to achieve interconnection between VPCs through the communication gateway.

[0087] As an example of a hardware functional unit, the second data acquisition module 701 may include at least one computing device, such as a server, etc. Alternatively, the second data acquisition module 701 may also be a device implemented using an application-specific integrated circuit (ASIC) or a programmable logic device (PLD). The PLD may be a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL) or any combination thereof.

[0088] The multiple computing devices included in the second data acquisition module 701 can be distributed in the same region or in different regions. The multiple computing devices included in the second data acquisition module 701 can be distributed in the same AZ or in different AZs. Similarly, the multiple computing devices included in the second data acquisition module 701 can be distributed in the same VPC or in multiple VPCs. The multiple computing devices can be any combination of computing devices such as servers, ASICs, PLDs, CPLDs, FPGAs, and GALs.

[0089] In other embodiments, the second data acquisition module 701 can be used to execute any step in the above-mentioned transaction event risk management method, the risk management module 702 can be used to execute any step in the above-mentioned transaction event risk management method, and the execution module 703 can be used to execute any step in the above-mentioned transaction event risk management method. The steps that the second data acquisition module 701, the risk management module 702, and the execution module 703 are responsible for implementing can be specified as needed, and the second data acquisition module 701, the risk management module 702, and the execution module 703 respectively implement different steps in the above-mentioned transaction event risk management method to realize all the functions of the above-mentioned transaction event risk management device.

[0090] In this implementation, the transaction event risk management device can also be applied to computing devices such as computers and servers, or to a computing device cluster including at least one computing device, to realize the specific functions of the transaction event risk management device.

[0091] In some embodiments, an electronic device is also provided. Figure 8, the electronic device includes: a bus 801, a processor 802, a memory 803 and a communication interface 804. The processor 802, the memory 803 and the communication interface 804 communicate with each other through the bus 801. The electronic device can be a server or a terminal device. It should be understood that the present application does not limit the number of processors and memories in the electronic device.

[0092] The bus 801 may be a peripheral component interconnect (PCI) bus, or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8 The bus 801 is represented by only one line, but it does not mean that there is only one bus or one type of bus. The bus 801 may include a path for transmitting information between various components of the electronic device (for example, the processor 802, the memory 803, and the communication interface 804).

[0093] The processor 802 may include any one or more processors such as a CPU, a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0094] The memory 803 may include a volatile memory, such as a random access memory (RAM). The memory 803 may also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0095] The memory 803 stores executable program codes, and the processor 802 executes the executable program codes to implement the aforementioned training method of the transaction event risk management model, or to implement the aforementioned transaction event risk management method.

[0096] The communication interface 804 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the electronic device and other devices or a communication network.

[0097] In some embodiments, a computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, the processor executes the above-mentioned training method of the transaction event risk management model, or executes the above-mentioned transaction event risk management method.

[0098] The computer-readable storage medium may be any available medium that can be stored in an electronic device or a data storage device such as a data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk). The computer-readable storage medium includes instructions that instruct the electronic device to execute the above-mentioned training method for the transaction event risk management model, or to execute the above-mentioned transaction event risk management method.

[0099] It is to be understood that the structure illustrated in the embodiments of this specification does not constitute a specific limitation on the system of the embodiments of this specification. In other embodiments of the specification, the above system may include more or fewer components than shown in the figure, or combine some components, or split some components, or arrange the components differently. The components shown in the figure may be implemented in hardware, software, or a combination of software and hardware.

[0100] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0101] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0102] It should be noted that the above examples are only specific embodiments of the present invention, and the present invention is obviously not limited to the above examples, and there are many similar variations. All variations directly derived or associated from the contents disclosed by the technicians in this field should fall within the protection scope of the present invention.

Claims

1. A training method for a transaction event risk management model, comprising: Get a sample of transaction events; Inputting the transaction event sample into a transaction event risk management model, and using the transaction event risk management model to determine a risk management strategy for the transaction event sample; Constructing a prompt text according to the transaction event sample and the risk management strategy; Inputting the prompt text into a pre-trained large language model to obtain an evaluation result of the large language model on the applicability of the risk management strategy to the transaction event sample; The evaluation result is fed back to the transaction event risk management model as an incentive signal, and a reinforcement strategy is adopted to update the parameters of the transaction event risk management model until a target transaction event risk management model is obtained.

2. The method according to claim 1, constructing a prompt text according to the transaction event sample and the risk management strategy, specifically comprising: According to the transaction event sample and the risk management and control strategy, a preset prompt text template is used to generate the prompt text.

3. The method according to claim 2, further comprising: Fine-tune the prompt text according to the response result after the risk management and control strategy is executed.

4. According to the method described in claim 1, the risk management strategy includes a management decision on whether to manage the transaction event sample, and a management scheme to be adopted when it is determined to manage the transaction event sample.

5. The method according to claim 1, further comprising: Obtain historical control data, and construct binary sample data consisting of historical transaction events and corresponding risk control strategies based on the historical control data; Scoring the binary sample data based on expert experience, and using the scoring result as a label of the binary sample data; The scoring result is used to characterize the applicability of the risk management strategy in the binary sample data to the corresponding historical transaction event; The large language model is fine-tuned using the bigram sample data.

6. A transaction event risk management method, comprising: Get the target transaction event; Inputting the target transaction event into a transaction event risk management and control model to obtain a risk management and control strategy for the target transaction event; The transaction event risk management model is obtained by pre-training using the method described in any one of claims 1 to 5; Execute the risk management strategy.

7. A training device for a transaction event risk management model, comprising: A first data acquisition module configured to acquire transaction event samples; A strategy generation module is configured to input the transaction event sample into a transaction event risk management model, obtain the transaction event risk management model to determine a risk management strategy for the transaction event sample; A prompt text generation module, configured to construct a prompt text according to the transaction event sample and the risk management strategy; an incentive module, configured to input the prompt text into a pre-trained large language model to obtain an evaluation result of the large language model on the applicability of the risk management strategy to the transaction event sample; The training module is configured to feed back the evaluation result as an incentive signal to the transaction event risk management model, and adopt a reinforcement strategy to update the parameters of the transaction event risk management model until a target transaction event risk management model is obtained.

8. According to the device of claim 7, the prompt text generation module is specifically used to generate the prompt text using a preset prompt text template according to the transaction event sample and the risk management strategy.

9. According to the device of claim 8, the prompt text generation module is also used to fine-tune the prompt text according to the response result after the risk management and control strategy is executed.

10. The device according to claim 7, wherein the risk management strategy includes a management decision on whether to manage the transaction event sample, and a management scheme to be adopted when it is determined to manage the transaction event sample.

11. The apparatus according to claim 7, wherein the training module is further configured to: Obtain historical control data, and construct binary sample data consisting of historical transaction events and corresponding risk control strategies based on the historical control data; Scoring the binary sample data based on expert experience, and using the scoring result as a label of the binary sample data; The scoring result is used to characterize the applicability of the risk management strategy in the binary sample data to the corresponding historical transaction event; The large language model is fine-tuned using the bigram sample data.

12. A transaction event risk management and control device, comprising: A second data acquisition module is configured to acquire a target transaction event; A risk control module configured to input the target transaction event into a transaction event risk control model to obtain a risk control strategy for the target transaction event; The transaction event risk management model is obtained by pre-training using the method described in any one of claims 1 to 5; An execution module is configured to execute the risk management strategy.

13. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor executes the method according to any one of claims 1 to 5, or the method according to claim 6.

14. An electronic device comprising: one or more processors; And a memory associated with the one or more processors, the memory being used to store program instructions, wherein when the program instructions are read and executed by the one or more processors, the electronic device executes the method as claimed in any one of claims 1 to 5, or executes the method as claimed in claim 6.

Images